linux/kernel/padata.c

1159 lines
28 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
/*
* padata.c - generic interface to process data streams in parallel
*
* See Documentation/core-api/padata.rst for more information.
*
* Copyright (C) 2008, 2009 secunet Security Networks AG
* Copyright (C) 2008, 2009 Steffen Klassert <steffen.klassert@secunet.com>
*
padata: add basic support for multithreaded jobs Sometimes the kernel doesn't take full advantage of system memory bandwidth, leading to a single CPU spending excessive time in initialization paths where the data scales with memory size. Multithreading naturally addresses this problem. Extend padata, a framework that handles many parallel yet singlethreaded jobs, to also handle multithreaded jobs by adding support for splitting up the work evenly, specifying a minimum amount of work that's appropriate for one helper thread to do, load balancing between helpers, and coordinating them. This is inspired by work from Pavel Tatashin and Steve Sistare. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-5-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:43 +00:00
* Copyright (c) 2020 Oracle and/or its affiliates.
* Author: Daniel Jordan <daniel.m.jordan@oracle.com>
*/
padata: add basic support for multithreaded jobs Sometimes the kernel doesn't take full advantage of system memory bandwidth, leading to a single CPU spending excessive time in initialization paths where the data scales with memory size. Multithreading naturally addresses this problem. Extend padata, a framework that handles many parallel yet singlethreaded jobs, to also handle multithreaded jobs by adding support for splitting up the work evenly, specifying a minimum amount of work that's appropriate for one helper thread to do, load balancing between helpers, and coordinating them. This is inspired by work from Pavel Tatashin and Steve Sistare. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-5-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:43 +00:00
#include <linux/completion.h>
#include <linux/export.h>
#include <linux/cpumask.h>
#include <linux/err.h>
#include <linux/cpu.h>
#include <linux/padata.h>
#include <linux/mutex.h>
#include <linux/sched.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 08:04:11 +00:00
#include <linux/slab.h>
#include <linux/sysfs.h>
#include <linux/rcupdate.h>
padata: add basic support for multithreaded jobs Sometimes the kernel doesn't take full advantage of system memory bandwidth, leading to a single CPU spending excessive time in initialization paths where the data scales with memory size. Multithreading naturally addresses this problem. Extend padata, a framework that handles many parallel yet singlethreaded jobs, to also handle multithreaded jobs by adding support for splitting up the work evenly, specifying a minimum amount of work that's appropriate for one helper thread to do, load balancing between helpers, and coordinating them. This is inspired by work from Pavel Tatashin and Steve Sistare. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-5-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:43 +00:00
#define PADATA_WORK_ONSTACK 1 /* Work's memory is on stack */
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
struct padata_work {
struct work_struct pw_work;
struct list_head pw_list; /* padata_free_works linkage */
void *pw_data;
};
static DEFINE_SPINLOCK(padata_works_lock);
static struct padata_work *padata_works;
static LIST_HEAD(padata_free_works);
padata: add basic support for multithreaded jobs Sometimes the kernel doesn't take full advantage of system memory bandwidth, leading to a single CPU spending excessive time in initialization paths where the data scales with memory size. Multithreading naturally addresses this problem. Extend padata, a framework that handles many parallel yet singlethreaded jobs, to also handle multithreaded jobs by adding support for splitting up the work evenly, specifying a minimum amount of work that's appropriate for one helper thread to do, load balancing between helpers, and coordinating them. This is inspired by work from Pavel Tatashin and Steve Sistare. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-5-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:43 +00:00
struct padata_mt_job_state {
spinlock_t lock;
struct completion completion;
struct padata_mt_job *job;
int nworks;
int nworks_fini;
unsigned long chunk_size;
};
static void padata_free_pd(struct parallel_data *pd);
padata: add basic support for multithreaded jobs Sometimes the kernel doesn't take full advantage of system memory bandwidth, leading to a single CPU spending excessive time in initialization paths where the data scales with memory size. Multithreading naturally addresses this problem. Extend padata, a framework that handles many parallel yet singlethreaded jobs, to also handle multithreaded jobs by adding support for splitting up the work evenly, specifying a minimum amount of work that's appropriate for one helper thread to do, load balancing between helpers, and coordinating them. This is inspired by work from Pavel Tatashin and Steve Sistare. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-5-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:43 +00:00
static void __init padata_mt_helper(struct work_struct *work);
static int padata_index_to_cpu(struct parallel_data *pd, int cpu_index)
{
int cpu, target_cpu;
target_cpu = cpumask_first(pd->cpumask.pcpu);
for (cpu = 0; cpu < cpu_index; cpu++)
target_cpu = cpumask_next(target_cpu, pd->cpumask.pcpu);
return target_cpu;
}
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
static int padata_cpu_hash(struct parallel_data *pd, unsigned int seq_nr)
{
/*
* Hash the sequence numbers to the cpus by taking
* seq_nr mod. number of cpus in use.
*/
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
int cpu_index = seq_nr % cpumask_weight(pd->cpumask.pcpu);
return padata_index_to_cpu(pd, cpu_index);
}
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
static struct padata_work *padata_work_alloc(void)
{
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
struct padata_work *pw;
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
lockdep_assert_held(&padata_works_lock);
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
if (list_empty(&padata_free_works))
return NULL; /* No more work items allowed to be queued. */
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
pw = list_first_entry(&padata_free_works, struct padata_work, pw_list);
list_del(&pw->pw_list);
return pw;
}
/*
* This function is marked __ref because this function may be optimized in such
* a way that it directly refers to work_fn's address, which causes modpost to
* complain when work_fn is marked __init. This scenario was observed with clang
* LTO, where padata_work_init() was optimized to refer directly to
* padata_mt_helper() because the calls to padata_work_init() with other work_fn
* values were eliminated or inlined.
*/
static void __ref padata_work_init(struct padata_work *pw, work_func_t work_fn,
void *data, int flags)
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
{
padata: add basic support for multithreaded jobs Sometimes the kernel doesn't take full advantage of system memory bandwidth, leading to a single CPU spending excessive time in initialization paths where the data scales with memory size. Multithreading naturally addresses this problem. Extend padata, a framework that handles many parallel yet singlethreaded jobs, to also handle multithreaded jobs by adding support for splitting up the work evenly, specifying a minimum amount of work that's appropriate for one helper thread to do, load balancing between helpers, and coordinating them. This is inspired by work from Pavel Tatashin and Steve Sistare. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-5-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:43 +00:00
if (flags & PADATA_WORK_ONSTACK)
INIT_WORK_ONSTACK(&pw->pw_work, work_fn);
else
INIT_WORK(&pw->pw_work, work_fn);
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
pw->pw_data = data;
}
padata: add basic support for multithreaded jobs Sometimes the kernel doesn't take full advantage of system memory bandwidth, leading to a single CPU spending excessive time in initialization paths where the data scales with memory size. Multithreading naturally addresses this problem. Extend padata, a framework that handles many parallel yet singlethreaded jobs, to also handle multithreaded jobs by adding support for splitting up the work evenly, specifying a minimum amount of work that's appropriate for one helper thread to do, load balancing between helpers, and coordinating them. This is inspired by work from Pavel Tatashin and Steve Sistare. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-5-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:43 +00:00
static int __init padata_work_alloc_mt(int nworks, void *data,
struct list_head *head)
{
int i;
spin_lock(&padata_works_lock);
/* Start at 1 because the current task participates in the job. */
for (i = 1; i < nworks; ++i) {
struct padata_work *pw = padata_work_alloc();
if (!pw)
break;
padata_work_init(pw, padata_mt_helper, data, 0);
list_add(&pw->pw_list, head);
}
spin_unlock(&padata_works_lock);
return i;
}
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
static void padata_work_free(struct padata_work *pw)
{
lockdep_assert_held(&padata_works_lock);
list_add(&pw->pw_list, &padata_free_works);
}
padata: add basic support for multithreaded jobs Sometimes the kernel doesn't take full advantage of system memory bandwidth, leading to a single CPU spending excessive time in initialization paths where the data scales with memory size. Multithreading naturally addresses this problem. Extend padata, a framework that handles many parallel yet singlethreaded jobs, to also handle multithreaded jobs by adding support for splitting up the work evenly, specifying a minimum amount of work that's appropriate for one helper thread to do, load balancing between helpers, and coordinating them. This is inspired by work from Pavel Tatashin and Steve Sistare. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-5-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:43 +00:00
static void __init padata_works_free(struct list_head *works)
{
struct padata_work *cur, *next;
if (list_empty(works))
return;
spin_lock(&padata_works_lock);
list_for_each_entry_safe(cur, next, works, pw_list) {
list_del(&cur->pw_list);
padata_work_free(cur);
}
spin_unlock(&padata_works_lock);
}
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
static void padata_parallel_worker(struct work_struct *parallel_work)
{
struct padata_work *pw = container_of(parallel_work, struct padata_work,
pw_work);
struct padata_priv *padata = pw->pw_data;
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
local_bh_disable();
padata->parallel(padata);
spin_lock(&padata_works_lock);
padata_work_free(pw);
spin_unlock(&padata_works_lock);
local_bh_enable();
}
/**
* padata_do_parallel - padata parallelization function
*
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
* @ps: padatashell
* @padata: object to be parallelized
* @cb_cpu: pointer to the CPU that the serialization callback function should
* run on. If it's not in the serial cpumask of @pinst
* (i.e. cpumask.cbcpu), this function selects a fallback CPU and if
* none found, returns -EINVAL.
*
* The parallelization callback function will run with BHs off.
* Note: Every object which is parallelized by padata_do_parallel
* must be seen by padata_do_serial.
*
* Return: 0 on success or else negative error code.
*/
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
int padata_do_parallel(struct padata_shell *ps,
struct padata_priv *padata, int *cb_cpu)
{
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
struct padata_instance *pinst = ps->pinst;
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
int i, cpu, cpu_index, err;
struct parallel_data *pd;
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
struct padata_work *pw;
rcu_read_lock_bh();
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
pd = rcu_dereference_bh(ps->pd);
err = -EINVAL;
if (!(pinst->flags & PADATA_INIT) || pinst->flags & PADATA_INVALID)
goto out;
if (!cpumask_test_cpu(*cb_cpu, pd->cpumask.cbcpu)) {
if (cpumask_empty(pd->cpumask.cbcpu))
goto out;
/* Select an alternate fallback CPU and notify the caller. */
cpu_index = *cb_cpu % cpumask_weight(pd->cpumask.cbcpu);
cpu = cpumask_first(pd->cpumask.cbcpu);
for (i = 0; i < cpu_index; i++)
cpu = cpumask_next(cpu, pd->cpumask.cbcpu);
*cb_cpu = cpu;
}
crypto: pcrypt - Fix hungtask for PADATA_RESET We found a hungtask bug in test_aead_vec_cfg as follows: INFO: task cryptomgr_test:391009 blocked for more than 120 seconds. "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. Call trace: __switch_to+0x98/0xe0 __schedule+0x6c4/0xf40 schedule+0xd8/0x1b4 schedule_timeout+0x474/0x560 wait_for_common+0x368/0x4e0 wait_for_completion+0x20/0x30 wait_for_completion+0x20/0x30 test_aead_vec_cfg+0xab4/0xd50 test_aead+0x144/0x1f0 alg_test_aead+0xd8/0x1e0 alg_test+0x634/0x890 cryptomgr_test+0x40/0x70 kthread+0x1e0/0x220 ret_from_fork+0x10/0x18 Kernel panic - not syncing: hung_task: blocked tasks For padata_do_parallel, when the return err is 0 or -EBUSY, it will call wait_for_completion(&wait->completion) in test_aead_vec_cfg. In normal case, aead_request_complete() will be called in pcrypt_aead_serial and the return err is 0 for padata_do_parallel. But, when pinst->flags is PADATA_RESET, the return err is -EBUSY for padata_do_parallel, and it won't call aead_request_complete(). Therefore, test_aead_vec_cfg will hung at wait_for_completion(&wait->completion), which will cause hungtask. The problem comes as following: (padata_do_parallel) | rcu_read_lock_bh(); | err = -EINVAL; | (padata_replace) | pinst->flags |= PADATA_RESET; err = -EBUSY | if (pinst->flags & PADATA_RESET) | rcu_read_unlock_bh() | return err In order to resolve the problem, we replace the return err -EBUSY with -EAGAIN, which means parallel_data is changing, and the caller should call it again. v3: remove retry and just change the return err. v2: introduce padata_try_do_parallel() in pcrypt_aead_encrypt and pcrypt_aead_decrypt to solve the hungtask. Signed-off-by: Lu Jialin <lujialin4@huawei.com> Signed-off-by: Guo Zihua <guozihua@huawei.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2023-09-04 13:33:41 +00:00
err = -EBUSY;
if ((pinst->flags & PADATA_RESET))
goto out;
refcount_inc(&pd->refcnt);
padata->pd = pd;
padata->cb_cpu = *cb_cpu;
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
spin_lock(&padata_works_lock);
padata->seq_nr = ++pd->seq_nr;
pw = padata_work_alloc();
spin_unlock(&padata_works_lock);
if (!pw) {
/* Maximum works limit exceeded, run in the current task. */
padata->parallel(padata);
}
rcu_read_unlock_bh();
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
if (pw) {
padata: add basic support for multithreaded jobs Sometimes the kernel doesn't take full advantage of system memory bandwidth, leading to a single CPU spending excessive time in initialization paths where the data scales with memory size. Multithreading naturally addresses this problem. Extend padata, a framework that handles many parallel yet singlethreaded jobs, to also handle multithreaded jobs by adding support for splitting up the work evenly, specifying a minimum amount of work that's appropriate for one helper thread to do, load balancing between helpers, and coordinating them. This is inspired by work from Pavel Tatashin and Steve Sistare. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-5-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:43 +00:00
padata_work_init(pw, padata_parallel_worker, padata, 0);
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
queue_work(pinst->parallel_wq, &pw->pw_work);
}
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
return 0;
out:
rcu_read_unlock_bh();
return err;
}
EXPORT_SYMBOL(padata_do_parallel);
/*
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
* padata_find_next - Find the next object that needs serialization.
*
* Return:
* * A pointer to the control struct of the next object that needs
* serialization, if present in one of the percpu reorder queues.
* * NULL, if the next object that needs serialization will
* be parallel processed by another cpu and is not yet present in
* the cpu's reorder queue.
*/
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
static struct padata_priv *padata_find_next(struct parallel_data *pd,
bool remove_object)
{
struct padata_priv *padata;
struct padata_list *reorder;
int cpu = pd->cpu;
reorder = per_cpu_ptr(pd->reorder_list, cpu);
padata: avoid race in reordering Under extremely heavy uses of padata, crashes occur, and with list debugging turned on, this happens instead: [87487.298728] WARNING: CPU: 1 PID: 882 at lib/list_debug.c:33 __list_add+0xae/0x130 [87487.301868] list_add corruption. prev->next should be next (ffffb17abfc043d0), but was ffff8dba70872c80. (prev=ffff8dba70872b00). [87487.339011] [<ffffffff9a53d075>] dump_stack+0x68/0xa3 [87487.342198] [<ffffffff99e119a1>] ? console_unlock+0x281/0x6d0 [87487.345364] [<ffffffff99d6b91f>] __warn+0xff/0x140 [87487.348513] [<ffffffff99d6b9aa>] warn_slowpath_fmt+0x4a/0x50 [87487.351659] [<ffffffff9a58b5de>] __list_add+0xae/0x130 [87487.354772] [<ffffffff9add5094>] ? _raw_spin_lock+0x64/0x70 [87487.357915] [<ffffffff99eefd66>] padata_reorder+0x1e6/0x420 [87487.361084] [<ffffffff99ef0055>] padata_do_serial+0xa5/0x120 padata_reorder calls list_add_tail with the list to which its adding locked, which seems correct: spin_lock(&squeue->serial.lock); list_add_tail(&padata->list, &squeue->serial.list); spin_unlock(&squeue->serial.lock); This therefore leaves only place where such inconsistency could occur: if padata->list is added at the same time on two different threads. This pdata pointer comes from the function call to padata_get_next(pd), which has in it the following block: next_queue = per_cpu_ptr(pd->pqueue, cpu); padata = NULL; reorder = &next_queue->reorder; if (!list_empty(&reorder->list)) { padata = list_entry(reorder->list.next, struct padata_priv, list); spin_lock(&reorder->lock); list_del_init(&padata->list); atomic_dec(&pd->reorder_objects); spin_unlock(&reorder->lock); pd->processed++; goto out; } out: return padata; I strongly suspect that the problem here is that two threads can race on reorder list. Even though the deletion is locked, call to list_entry is not locked, which means it's feasible that two threads pick up the same padata object and subsequently call list_add_tail on them at the same time. The fix is thus be hoist that lock outside of that block. Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com> Acked-by: Steffen Klassert <steffen.klassert@secunet.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-23 11:24:43 +00:00
spin_lock(&reorder->lock);
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
if (list_empty(&reorder->list)) {
spin_unlock(&reorder->lock);
return NULL;
}
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
padata = list_entry(reorder->list.next, struct padata_priv, list);
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
/*
* Checks the rare case where two or more parallel jobs have hashed to
* the same CPU and one of the later ones finishes first.
*/
if (padata->seq_nr != pd->processed) {
padata: avoid race in reordering Under extremely heavy uses of padata, crashes occur, and with list debugging turned on, this happens instead: [87487.298728] WARNING: CPU: 1 PID: 882 at lib/list_debug.c:33 __list_add+0xae/0x130 [87487.301868] list_add corruption. prev->next should be next (ffffb17abfc043d0), but was ffff8dba70872c80. (prev=ffff8dba70872b00). [87487.339011] [<ffffffff9a53d075>] dump_stack+0x68/0xa3 [87487.342198] [<ffffffff99e119a1>] ? console_unlock+0x281/0x6d0 [87487.345364] [<ffffffff99d6b91f>] __warn+0xff/0x140 [87487.348513] [<ffffffff99d6b9aa>] warn_slowpath_fmt+0x4a/0x50 [87487.351659] [<ffffffff9a58b5de>] __list_add+0xae/0x130 [87487.354772] [<ffffffff9add5094>] ? _raw_spin_lock+0x64/0x70 [87487.357915] [<ffffffff99eefd66>] padata_reorder+0x1e6/0x420 [87487.361084] [<ffffffff99ef0055>] padata_do_serial+0xa5/0x120 padata_reorder calls list_add_tail with the list to which its adding locked, which seems correct: spin_lock(&squeue->serial.lock); list_add_tail(&padata->list, &squeue->serial.list); spin_unlock(&squeue->serial.lock); This therefore leaves only place where such inconsistency could occur: if padata->list is added at the same time on two different threads. This pdata pointer comes from the function call to padata_get_next(pd), which has in it the following block: next_queue = per_cpu_ptr(pd->pqueue, cpu); padata = NULL; reorder = &next_queue->reorder; if (!list_empty(&reorder->list)) { padata = list_entry(reorder->list.next, struct padata_priv, list); spin_lock(&reorder->lock); list_del_init(&padata->list); atomic_dec(&pd->reorder_objects); spin_unlock(&reorder->lock); pd->processed++; goto out; } out: return padata; I strongly suspect that the problem here is that two threads can race on reorder list. Even though the deletion is locked, call to list_entry is not locked, which means it's feasible that two threads pick up the same padata object and subsequently call list_add_tail on them at the same time. The fix is thus be hoist that lock outside of that block. Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com> Acked-by: Steffen Klassert <steffen.klassert@secunet.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-03-23 11:24:43 +00:00
spin_unlock(&reorder->lock);
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
return NULL;
}
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
if (remove_object) {
list_del_init(&padata->list);
++pd->processed;
pd->cpu = cpumask_next_wrap(cpu, pd->cpumask.pcpu, -1, false);
}
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
spin_unlock(&reorder->lock);
return padata;
}
static void padata_reorder(struct parallel_data *pd)
{
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
struct padata_instance *pinst = pd->ps->pinst;
int cb_cpu;
struct padata_priv *padata;
struct padata_serial_queue *squeue;
struct padata_list *reorder;
/*
* We need to ensure that only one cpu can work on dequeueing of
* the reorder queue the time. Calculating in which percpu reorder
* queue the next object will arrive takes some time. A spinlock
* would be highly contended. Also it is not clear in which order
* the objects arrive to the reorder queues. So a cpu could wait to
* get the lock just to notice that there is nothing to do at the
* moment. Therefore we use a trylock and let the holder of the lock
* care for all the objects enqueued during the holdtime of the lock.
*/
if (!spin_trylock_bh(&pd->lock))
return;
while (1) {
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
padata = padata_find_next(pd, true);
/*
* If the next object that needs serialization is parallel
* processed by another cpu and is still on it's way to the
* cpu's reorder queue, nothing to do for now.
*/
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
if (!padata)
break;
cb_cpu = padata->cb_cpu;
squeue = per_cpu_ptr(pd->squeue, cb_cpu);
spin_lock(&squeue->serial.lock);
list_add_tail(&padata->list, &squeue->serial.list);
spin_unlock(&squeue->serial.lock);
queue_work_on(cb_cpu, pinst->serial_wq, &squeue->work);
}
spin_unlock_bh(&pd->lock);
/*
* The next object that needs serialization might have arrived to
* the reorder queues in the meantime.
padata: use smp_mb in padata_reorder to avoid orphaned padata jobs Testing padata with the tcrypt module on a 5.2 kernel... # modprobe tcrypt alg="pcrypt(rfc4106(gcm(aes)))" type=3 # modprobe tcrypt mode=211 sec=1 ...produces this splat: INFO: task modprobe:10075 blocked for more than 120 seconds. Not tainted 5.2.0-base+ #16 modprobe D 0 10075 10064 0x80004080 Call Trace: ? __schedule+0x4dd/0x610 ? ring_buffer_unlock_commit+0x23/0x100 schedule+0x6c/0x90 schedule_timeout+0x3b/0x320 ? trace_buffer_unlock_commit_regs+0x4f/0x1f0 wait_for_common+0x160/0x1a0 ? wake_up_q+0x80/0x80 { crypto_wait_req } # entries in braces added by hand { do_one_aead_op } { test_aead_jiffies } test_aead_speed.constprop.17+0x681/0xf30 [tcrypt] do_test+0x4053/0x6a2b [tcrypt] ? 0xffffffffa00f4000 tcrypt_mod_init+0x50/0x1000 [tcrypt] ... The second modprobe command never finishes because in padata_reorder, CPU0's load of reorder_objects is executed before the unlocking store in spin_unlock_bh(pd->lock), causing CPU0 to miss CPU1's increment: CPU0 CPU1 padata_reorder padata_do_serial LOAD reorder_objects // 0 INC reorder_objects // 1 padata_reorder TRYLOCK pd->lock // failed UNLOCK pd->lock CPU0 deletes the timer before returning from padata_reorder and since no other job is submitted to padata, modprobe waits indefinitely. Add a pair of full barriers to guarantee proper ordering: CPU0 CPU1 padata_reorder padata_do_serial UNLOCK pd->lock smp_mb() LOAD reorder_objects INC reorder_objects smp_mb__after_atomic() padata_reorder TRYLOCK pd->lock smp_mb__after_atomic is needed so the read part of the trylock operation comes after the INC, as Andrea points out. Thanks also to Andrea for help with writing a litmus test. Fixes: 16295bec6398 ("padata: Generic parallelization/serialization interface") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: <stable@vger.kernel.org> Cc: Andrea Parri <andrea.parri@amarulasolutions.com> Cc: Boqun Feng <boqun.feng@gmail.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Paul E. McKenney <paulmck@linux.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: linux-arch@vger.kernel.org Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-07-16 16:32:53 +00:00
*
* Ensure reorder queue is read after pd->lock is dropped so we see
* new objects from another task in padata_do_serial. Pairs with
* smp_mb in padata_do_serial.
*/
padata: use smp_mb in padata_reorder to avoid orphaned padata jobs Testing padata with the tcrypt module on a 5.2 kernel... # modprobe tcrypt alg="pcrypt(rfc4106(gcm(aes)))" type=3 # modprobe tcrypt mode=211 sec=1 ...produces this splat: INFO: task modprobe:10075 blocked for more than 120 seconds. Not tainted 5.2.0-base+ #16 modprobe D 0 10075 10064 0x80004080 Call Trace: ? __schedule+0x4dd/0x610 ? ring_buffer_unlock_commit+0x23/0x100 schedule+0x6c/0x90 schedule_timeout+0x3b/0x320 ? trace_buffer_unlock_commit_regs+0x4f/0x1f0 wait_for_common+0x160/0x1a0 ? wake_up_q+0x80/0x80 { crypto_wait_req } # entries in braces added by hand { do_one_aead_op } { test_aead_jiffies } test_aead_speed.constprop.17+0x681/0xf30 [tcrypt] do_test+0x4053/0x6a2b [tcrypt] ? 0xffffffffa00f4000 tcrypt_mod_init+0x50/0x1000 [tcrypt] ... The second modprobe command never finishes because in padata_reorder, CPU0's load of reorder_objects is executed before the unlocking store in spin_unlock_bh(pd->lock), causing CPU0 to miss CPU1's increment: CPU0 CPU1 padata_reorder padata_do_serial LOAD reorder_objects // 0 INC reorder_objects // 1 padata_reorder TRYLOCK pd->lock // failed UNLOCK pd->lock CPU0 deletes the timer before returning from padata_reorder and since no other job is submitted to padata, modprobe waits indefinitely. Add a pair of full barriers to guarantee proper ordering: CPU0 CPU1 padata_reorder padata_do_serial UNLOCK pd->lock smp_mb() LOAD reorder_objects INC reorder_objects smp_mb__after_atomic() padata_reorder TRYLOCK pd->lock smp_mb__after_atomic is needed so the read part of the trylock operation comes after the INC, as Andrea points out. Thanks also to Andrea for help with writing a litmus test. Fixes: 16295bec6398 ("padata: Generic parallelization/serialization interface") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: <stable@vger.kernel.org> Cc: Andrea Parri <andrea.parri@amarulasolutions.com> Cc: Boqun Feng <boqun.feng@gmail.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Paul E. McKenney <paulmck@linux.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: linux-arch@vger.kernel.org Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-07-16 16:32:53 +00:00
smp_mb();
reorder = per_cpu_ptr(pd->reorder_list, pd->cpu);
if (!list_empty(&reorder->list) && padata_find_next(pd, false))
queue_work(pinst->serial_wq, &pd->reorder_work);
}
static void invoke_padata_reorder(struct work_struct *work)
{
struct parallel_data *pd;
local_bh_disable();
pd = container_of(work, struct parallel_data, reorder_work);
padata_reorder(pd);
local_bh_enable();
}
static void padata_serial_worker(struct work_struct *serial_work)
{
struct padata_serial_queue *squeue;
struct parallel_data *pd;
LIST_HEAD(local_list);
int cnt;
local_bh_disable();
squeue = container_of(serial_work, struct padata_serial_queue, work);
pd = squeue->pd;
spin_lock(&squeue->serial.lock);
list_replace_init(&squeue->serial.list, &local_list);
spin_unlock(&squeue->serial.lock);
cnt = 0;
while (!list_empty(&local_list)) {
struct padata_priv *padata;
padata = list_entry(local_list.next,
struct padata_priv, list);
list_del_init(&padata->list);
padata->serial(padata);
cnt++;
}
local_bh_enable();
if (refcount_sub_and_test(cnt, &pd->refcnt))
padata_free_pd(pd);
}
/**
* padata_do_serial - padata serialization function
*
* @padata: object to be serialized.
*
* padata_do_serial must be called for every parallelized object.
* The serialization callback function will run with BHs off.
*/
void padata_do_serial(struct padata_priv *padata)
{
struct parallel_data *pd = padata->pd;
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
int hashed_cpu = padata_cpu_hash(pd, padata->seq_nr);
struct padata_list *reorder = per_cpu_ptr(pd->reorder_list, hashed_cpu);
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
struct padata_priv *cur;
struct list_head *pos;
spin_lock(&reorder->lock);
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
/* Sort in ascending order of sequence number. */
list_for_each_prev(pos, &reorder->list) {
cur = list_entry(pos, struct padata_priv, list);
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
if (cur->seq_nr < padata->seq_nr)
break;
}
list_add(&padata->list, pos);
spin_unlock(&reorder->lock);
padata: use smp_mb in padata_reorder to avoid orphaned padata jobs Testing padata with the tcrypt module on a 5.2 kernel... # modprobe tcrypt alg="pcrypt(rfc4106(gcm(aes)))" type=3 # modprobe tcrypt mode=211 sec=1 ...produces this splat: INFO: task modprobe:10075 blocked for more than 120 seconds. Not tainted 5.2.0-base+ #16 modprobe D 0 10075 10064 0x80004080 Call Trace: ? __schedule+0x4dd/0x610 ? ring_buffer_unlock_commit+0x23/0x100 schedule+0x6c/0x90 schedule_timeout+0x3b/0x320 ? trace_buffer_unlock_commit_regs+0x4f/0x1f0 wait_for_common+0x160/0x1a0 ? wake_up_q+0x80/0x80 { crypto_wait_req } # entries in braces added by hand { do_one_aead_op } { test_aead_jiffies } test_aead_speed.constprop.17+0x681/0xf30 [tcrypt] do_test+0x4053/0x6a2b [tcrypt] ? 0xffffffffa00f4000 tcrypt_mod_init+0x50/0x1000 [tcrypt] ... The second modprobe command never finishes because in padata_reorder, CPU0's load of reorder_objects is executed before the unlocking store in spin_unlock_bh(pd->lock), causing CPU0 to miss CPU1's increment: CPU0 CPU1 padata_reorder padata_do_serial LOAD reorder_objects // 0 INC reorder_objects // 1 padata_reorder TRYLOCK pd->lock // failed UNLOCK pd->lock CPU0 deletes the timer before returning from padata_reorder and since no other job is submitted to padata, modprobe waits indefinitely. Add a pair of full barriers to guarantee proper ordering: CPU0 CPU1 padata_reorder padata_do_serial UNLOCK pd->lock smp_mb() LOAD reorder_objects INC reorder_objects smp_mb__after_atomic() padata_reorder TRYLOCK pd->lock smp_mb__after_atomic is needed so the read part of the trylock operation comes after the INC, as Andrea points out. Thanks also to Andrea for help with writing a litmus test. Fixes: 16295bec6398 ("padata: Generic parallelization/serialization interface") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: <stable@vger.kernel.org> Cc: Andrea Parri <andrea.parri@amarulasolutions.com> Cc: Boqun Feng <boqun.feng@gmail.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Paul E. McKenney <paulmck@linux.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: linux-arch@vger.kernel.org Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-07-16 16:32:53 +00:00
/*
* Ensure the addition to the reorder list is ordered correctly
padata: use smp_mb in padata_reorder to avoid orphaned padata jobs Testing padata with the tcrypt module on a 5.2 kernel... # modprobe tcrypt alg="pcrypt(rfc4106(gcm(aes)))" type=3 # modprobe tcrypt mode=211 sec=1 ...produces this splat: INFO: task modprobe:10075 blocked for more than 120 seconds. Not tainted 5.2.0-base+ #16 modprobe D 0 10075 10064 0x80004080 Call Trace: ? __schedule+0x4dd/0x610 ? ring_buffer_unlock_commit+0x23/0x100 schedule+0x6c/0x90 schedule_timeout+0x3b/0x320 ? trace_buffer_unlock_commit_regs+0x4f/0x1f0 wait_for_common+0x160/0x1a0 ? wake_up_q+0x80/0x80 { crypto_wait_req } # entries in braces added by hand { do_one_aead_op } { test_aead_jiffies } test_aead_speed.constprop.17+0x681/0xf30 [tcrypt] do_test+0x4053/0x6a2b [tcrypt] ? 0xffffffffa00f4000 tcrypt_mod_init+0x50/0x1000 [tcrypt] ... The second modprobe command never finishes because in padata_reorder, CPU0's load of reorder_objects is executed before the unlocking store in spin_unlock_bh(pd->lock), causing CPU0 to miss CPU1's increment: CPU0 CPU1 padata_reorder padata_do_serial LOAD reorder_objects // 0 INC reorder_objects // 1 padata_reorder TRYLOCK pd->lock // failed UNLOCK pd->lock CPU0 deletes the timer before returning from padata_reorder and since no other job is submitted to padata, modprobe waits indefinitely. Add a pair of full barriers to guarantee proper ordering: CPU0 CPU1 padata_reorder padata_do_serial UNLOCK pd->lock smp_mb() LOAD reorder_objects INC reorder_objects smp_mb__after_atomic() padata_reorder TRYLOCK pd->lock smp_mb__after_atomic is needed so the read part of the trylock operation comes after the INC, as Andrea points out. Thanks also to Andrea for help with writing a litmus test. Fixes: 16295bec6398 ("padata: Generic parallelization/serialization interface") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: <stable@vger.kernel.org> Cc: Andrea Parri <andrea.parri@amarulasolutions.com> Cc: Boqun Feng <boqun.feng@gmail.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Paul E. McKenney <paulmck@linux.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: linux-arch@vger.kernel.org Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-07-16 16:32:53 +00:00
* with the trylock of pd->lock in padata_reorder. Pairs with smp_mb
* in padata_reorder.
*/
smp_mb();
padata: use smp_mb in padata_reorder to avoid orphaned padata jobs Testing padata with the tcrypt module on a 5.2 kernel... # modprobe tcrypt alg="pcrypt(rfc4106(gcm(aes)))" type=3 # modprobe tcrypt mode=211 sec=1 ...produces this splat: INFO: task modprobe:10075 blocked for more than 120 seconds. Not tainted 5.2.0-base+ #16 modprobe D 0 10075 10064 0x80004080 Call Trace: ? __schedule+0x4dd/0x610 ? ring_buffer_unlock_commit+0x23/0x100 schedule+0x6c/0x90 schedule_timeout+0x3b/0x320 ? trace_buffer_unlock_commit_regs+0x4f/0x1f0 wait_for_common+0x160/0x1a0 ? wake_up_q+0x80/0x80 { crypto_wait_req } # entries in braces added by hand { do_one_aead_op } { test_aead_jiffies } test_aead_speed.constprop.17+0x681/0xf30 [tcrypt] do_test+0x4053/0x6a2b [tcrypt] ? 0xffffffffa00f4000 tcrypt_mod_init+0x50/0x1000 [tcrypt] ... The second modprobe command never finishes because in padata_reorder, CPU0's load of reorder_objects is executed before the unlocking store in spin_unlock_bh(pd->lock), causing CPU0 to miss CPU1's increment: CPU0 CPU1 padata_reorder padata_do_serial LOAD reorder_objects // 0 INC reorder_objects // 1 padata_reorder TRYLOCK pd->lock // failed UNLOCK pd->lock CPU0 deletes the timer before returning from padata_reorder and since no other job is submitted to padata, modprobe waits indefinitely. Add a pair of full barriers to guarantee proper ordering: CPU0 CPU1 padata_reorder padata_do_serial UNLOCK pd->lock smp_mb() LOAD reorder_objects INC reorder_objects smp_mb__after_atomic() padata_reorder TRYLOCK pd->lock smp_mb__after_atomic is needed so the read part of the trylock operation comes after the INC, as Andrea points out. Thanks also to Andrea for help with writing a litmus test. Fixes: 16295bec6398 ("padata: Generic parallelization/serialization interface") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: <stable@vger.kernel.org> Cc: Andrea Parri <andrea.parri@amarulasolutions.com> Cc: Boqun Feng <boqun.feng@gmail.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Paul E. McKenney <paulmck@linux.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: linux-arch@vger.kernel.org Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-07-16 16:32:53 +00:00
padata_reorder(pd);
}
EXPORT_SYMBOL(padata_do_serial);
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
static int padata_setup_cpumasks(struct padata_instance *pinst)
{
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
struct workqueue_attrs *attrs;
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
int err;
attrs = alloc_workqueue_attrs();
if (!attrs)
return -ENOMEM;
/* Restrict parallel_wq workers to pd->cpumask.pcpu. */
cpumask_copy(attrs->cpumask, pinst->cpumask.pcpu);
err = apply_workqueue_attrs(pinst->parallel_wq, attrs);
free_workqueue_attrs(attrs);
return err;
}
padata: add basic support for multithreaded jobs Sometimes the kernel doesn't take full advantage of system memory bandwidth, leading to a single CPU spending excessive time in initialization paths where the data scales with memory size. Multithreading naturally addresses this problem. Extend padata, a framework that handles many parallel yet singlethreaded jobs, to also handle multithreaded jobs by adding support for splitting up the work evenly, specifying a minimum amount of work that's appropriate for one helper thread to do, load balancing between helpers, and coordinating them. This is inspired by work from Pavel Tatashin and Steve Sistare. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-5-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:43 +00:00
static void __init padata_mt_helper(struct work_struct *w)
{
struct padata_work *pw = container_of(w, struct padata_work, pw_work);
struct padata_mt_job_state *ps = pw->pw_data;
struct padata_mt_job *job = ps->job;
bool done;
spin_lock(&ps->lock);
while (job->size > 0) {
unsigned long start, size, end;
start = job->start;
/* So end is chunk size aligned if enough work remains. */
size = roundup(start + 1, ps->chunk_size) - start;
size = min(size, job->size);
end = start + size;
job->start = end;
job->size -= size;
spin_unlock(&ps->lock);
job->thread_fn(start, end, job->fn_arg);
spin_lock(&ps->lock);
}
++ps->nworks_fini;
done = (ps->nworks_fini == ps->nworks);
spin_unlock(&ps->lock);
if (done)
complete(&ps->completion);
}
/**
* padata_do_multithreaded - run a multithreaded job
* @job: Description of the job.
*
* See the definition of struct padata_mt_job for more details.
*/
void __init padata_do_multithreaded(struct padata_mt_job *job)
{
/* In case threads finish at different times. */
static const unsigned long load_balance_factor = 4;
struct padata_work my_work, *pw;
struct padata_mt_job_state ps;
LIST_HEAD(works);
int nworks;
if (job->size == 0)
return;
/* Ensure at least one thread when size < min_chunk. */
nworks = max(job->size / max(job->min_chunk, job->align), 1ul);
padata: add basic support for multithreaded jobs Sometimes the kernel doesn't take full advantage of system memory bandwidth, leading to a single CPU spending excessive time in initialization paths where the data scales with memory size. Multithreading naturally addresses this problem. Extend padata, a framework that handles many parallel yet singlethreaded jobs, to also handle multithreaded jobs by adding support for splitting up the work evenly, specifying a minimum amount of work that's appropriate for one helper thread to do, load balancing between helpers, and coordinating them. This is inspired by work from Pavel Tatashin and Steve Sistare. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-5-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:43 +00:00
nworks = min(nworks, job->max_threads);
if (nworks == 1) {
/* Single thread, no coordination needed, cut to the chase. */
job->thread_fn(job->start, job->start + job->size, job->fn_arg);
return;
}
spin_lock_init(&ps.lock);
init_completion(&ps.completion);
ps.job = job;
ps.nworks = padata_work_alloc_mt(nworks, &ps, &works);
ps.nworks_fini = 0;
/*
* Chunk size is the amount of work a helper does per call to the
* thread function. Load balance large jobs between threads by
* increasing the number of chunks, guarantee at least the minimum
* chunk size from the caller, and honor the caller's alignment.
*/
ps.chunk_size = job->size / (ps.nworks * load_balance_factor);
ps.chunk_size = max(ps.chunk_size, job->min_chunk);
ps.chunk_size = roundup(ps.chunk_size, job->align);
list_for_each_entry(pw, &works, pw_list)
queue_work(system_unbound_wq, &pw->pw_work);
/* Use the current thread, which saves starting a workqueue worker. */
padata_work_init(&my_work, padata_mt_helper, &ps, PADATA_WORK_ONSTACK);
padata_mt_helper(&my_work.pw_work);
/* Wait for all the helpers to finish. */
wait_for_completion(&ps.completion);
destroy_work_on_stack(&my_work.pw_work);
padata_works_free(&works);
}
static void __padata_list_init(struct padata_list *pd_list)
{
INIT_LIST_HEAD(&pd_list->list);
spin_lock_init(&pd_list->lock);
}
/* Initialize all percpu queues used by serial workers */
static void padata_init_squeues(struct parallel_data *pd)
{
int cpu;
struct padata_serial_queue *squeue;
for_each_cpu(cpu, pd->cpumask.cbcpu) {
squeue = per_cpu_ptr(pd->squeue, cpu);
squeue->pd = pd;
__padata_list_init(&squeue->serial);
INIT_WORK(&squeue->work, padata_serial_worker);
}
}
/* Initialize per-CPU reorder lists */
static void padata_init_reorder_list(struct parallel_data *pd)
{
int cpu;
struct padata_list *list;
for_each_cpu(cpu, pd->cpumask.pcpu) {
list = per_cpu_ptr(pd->reorder_list, cpu);
__padata_list_init(list);
}
}
/* Allocate and initialize the internal cpumask dependend resources. */
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
static struct parallel_data *padata_alloc_pd(struct padata_shell *ps)
{
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
struct padata_instance *pinst = ps->pinst;
struct parallel_data *pd;
pd = kzalloc(sizeof(struct parallel_data), GFP_KERNEL);
if (!pd)
goto err;
pd->reorder_list = alloc_percpu(struct padata_list);
if (!pd->reorder_list)
goto err_free_pd;
pd->squeue = alloc_percpu(struct padata_serial_queue);
if (!pd->squeue)
goto err_free_reorder_list;
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
pd->ps = ps;
if (!alloc_cpumask_var(&pd->cpumask.pcpu, GFP_KERNEL))
goto err_free_squeue;
if (!alloc_cpumask_var(&pd->cpumask.cbcpu, GFP_KERNEL))
goto err_free_pcpu;
cpumask_and(pd->cpumask.pcpu, pinst->cpumask.pcpu, cpu_online_mask);
cpumask_and(pd->cpumask.cbcpu, pinst->cpumask.cbcpu, cpu_online_mask);
padata_init_reorder_list(pd);
padata_init_squeues(pd);
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
pd->seq_nr = -1;
refcount_set(&pd->refcnt, 1);
spin_lock_init(&pd->lock);
padata: initialize pd->cpu with effective cpumask Exercising CPU hotplug on a 5.2 kernel with recent padata fixes from cryptodev-2.6.git in an 8-CPU kvm guest... # modprobe tcrypt alg="pcrypt(rfc4106(gcm(aes)))" type=3 # echo 0 > /sys/devices/system/cpu/cpu1/online # echo c > /sys/kernel/pcrypt/pencrypt/parallel_cpumask # modprobe tcrypt mode=215 ...caused the following crash: BUG: kernel NULL pointer dereference, address: 0000000000000000 #PF: supervisor read access in kernel mode #PF: error_code(0x0000) - not-present page PGD 0 P4D 0 Oops: 0000 [#1] SMP PTI CPU: 2 PID: 134 Comm: kworker/2:2 Not tainted 5.2.0-padata-base+ #7 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.12.0-<snip> Workqueue: pencrypt padata_parallel_worker RIP: 0010:padata_reorder+0xcb/0x180 ... Call Trace: padata_do_serial+0x57/0x60 pcrypt_aead_enc+0x3a/0x50 [pcrypt] padata_parallel_worker+0x9b/0xe0 process_one_work+0x1b5/0x3f0 worker_thread+0x4a/0x3c0 ... In padata_alloc_pd, pd->cpu is set using the user-supplied cpumask instead of the effective cpumask, and in this case cpumask_first picked an offline CPU. The offline CPU's reorder->list.next is NULL in padata_reorder because the list wasn't initialized in padata_init_pqueues, which only operates on CPUs in the effective mask. Fix by using the effective mask in padata_alloc_pd. Fixes: 6fc4dbcf0276 ("padata: Replace delayed timer with immediate workqueue in padata_reorder") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-08-08 16:05:35 +00:00
pd->cpu = cpumask_first(pd->cpumask.pcpu);
INIT_WORK(&pd->reorder_work, invoke_padata_reorder);
return pd;
err_free_pcpu:
free_cpumask_var(pd->cpumask.pcpu);
err_free_squeue:
free_percpu(pd->squeue);
err_free_reorder_list:
free_percpu(pd->reorder_list);
err_free_pd:
kfree(pd);
err:
return NULL;
}
static void padata_free_pd(struct parallel_data *pd)
{
free_cpumask_var(pd->cpumask.pcpu);
free_cpumask_var(pd->cpumask.cbcpu);
free_percpu(pd->reorder_list);
free_percpu(pd->squeue);
kfree(pd);
}
static void __padata_start(struct padata_instance *pinst)
{
pinst->flags |= PADATA_INIT;
}
static void __padata_stop(struct padata_instance *pinst)
{
if (!(pinst->flags & PADATA_INIT))
return;
pinst->flags &= ~PADATA_INIT;
synchronize_rcu();
}
/* Replace the internal control structure with a new one. */
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
static int padata_replace_one(struct padata_shell *ps)
{
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
struct parallel_data *pd_new;
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
pd_new = padata_alloc_pd(ps);
if (!pd_new)
return -ENOMEM;
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
ps->opd = rcu_dereference_protected(ps->pd, 1);
rcu_assign_pointer(ps->pd, pd_new);
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
return 0;
}
padata: validate cpumask without removed CPU during offline Configuring an instance's parallel mask without any online CPUs... echo 2 > /sys/kernel/pcrypt/pencrypt/parallel_cpumask echo 0 > /sys/devices/system/cpu/cpu1/online ...makes tcrypt mode=215 crash like this: divide error: 0000 [#1] SMP PTI CPU: 4 PID: 283 Comm: modprobe Not tainted 5.4.0-rc8-padata-doc-v2+ #2 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20191013_105130-anatol 04/01/2014 RIP: 0010:padata_do_parallel+0x114/0x300 Call Trace: pcrypt_aead_encrypt+0xc0/0xd0 [pcrypt] crypto_aead_encrypt+0x1f/0x30 do_mult_aead_op+0x4e/0xdf [tcrypt] test_mb_aead_speed.constprop.0.cold+0x226/0x564 [tcrypt] do_test+0x28c2/0x4d49 [tcrypt] tcrypt_mod_init+0x55/0x1000 [tcrypt] ... cpumask_weight() in padata_cpu_hash() returns 0 because the mask has no CPUs. The problem is __padata_remove_cpu() checks for valid masks too early and so doesn't mark the instance PADATA_INVALID as expected, which would have made padata_do_parallel() return error before doing the division. Fix by introducing a second padata CPU hotplug state before CPUHP_BRINGUP_CPU so that __padata_remove_cpu() sees the online mask without @cpu. No need for the second argument to padata_replace() since @cpu is now already missing from the online mask. Fixes: 33e54450683c ("padata: Handle empty padata cpumasks") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Eric Biggers <ebiggers@kernel.org> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-12-03 19:31:10 +00:00
static int padata_replace(struct padata_instance *pinst)
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
{
struct padata_shell *ps;
int err = 0;
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
pinst->flags |= PADATA_RESET;
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
list_for_each_entry(ps, &pinst->pslist, list) {
err = padata_replace_one(ps);
if (err)
break;
}
synchronize_rcu();
list_for_each_entry_continue_reverse(ps, &pinst->pslist, list)
if (refcount_dec_and_test(&ps->opd->refcnt))
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
padata_free_pd(ps->opd);
pinst->flags &= ~PADATA_RESET;
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
return err;
}
/* If cpumask contains no active cpu, we mark the instance as invalid. */
static bool padata_validate_cpumask(struct padata_instance *pinst,
const struct cpumask *cpumask)
{
if (!cpumask_intersects(cpumask, cpu_online_mask)) {
pinst->flags |= PADATA_INVALID;
return false;
}
pinst->flags &= ~PADATA_INVALID;
return true;
}
static int __padata_set_cpumasks(struct padata_instance *pinst,
cpumask_var_t pcpumask,
cpumask_var_t cbcpumask)
{
int valid;
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
int err;
valid = padata_validate_cpumask(pinst, pcpumask);
if (!valid) {
__padata_stop(pinst);
goto out_replace;
}
valid = padata_validate_cpumask(pinst, cbcpumask);
if (!valid)
__padata_stop(pinst);
out_replace:
cpumask_copy(pinst->cpumask.pcpu, pcpumask);
cpumask_copy(pinst->cpumask.cbcpu, cbcpumask);
padata: validate cpumask without removed CPU during offline Configuring an instance's parallel mask without any online CPUs... echo 2 > /sys/kernel/pcrypt/pencrypt/parallel_cpumask echo 0 > /sys/devices/system/cpu/cpu1/online ...makes tcrypt mode=215 crash like this: divide error: 0000 [#1] SMP PTI CPU: 4 PID: 283 Comm: modprobe Not tainted 5.4.0-rc8-padata-doc-v2+ #2 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20191013_105130-anatol 04/01/2014 RIP: 0010:padata_do_parallel+0x114/0x300 Call Trace: pcrypt_aead_encrypt+0xc0/0xd0 [pcrypt] crypto_aead_encrypt+0x1f/0x30 do_mult_aead_op+0x4e/0xdf [tcrypt] test_mb_aead_speed.constprop.0.cold+0x226/0x564 [tcrypt] do_test+0x28c2/0x4d49 [tcrypt] tcrypt_mod_init+0x55/0x1000 [tcrypt] ... cpumask_weight() in padata_cpu_hash() returns 0 because the mask has no CPUs. The problem is __padata_remove_cpu() checks for valid masks too early and so doesn't mark the instance PADATA_INVALID as expected, which would have made padata_do_parallel() return error before doing the division. Fix by introducing a second padata CPU hotplug state before CPUHP_BRINGUP_CPU so that __padata_remove_cpu() sees the online mask without @cpu. No need for the second argument to padata_replace() since @cpu is now already missing from the online mask. Fixes: 33e54450683c ("padata: Handle empty padata cpumasks") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Eric Biggers <ebiggers@kernel.org> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-12-03 19:31:10 +00:00
err = padata_setup_cpumasks(pinst) ?: padata_replace(pinst);
if (valid)
__padata_start(pinst);
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
return err;
}
/**
* padata_set_cpumask - Sets specified by @cpumask_type cpumask to the value
* equivalent to @cpumask.
* @pinst: padata instance
* @cpumask_type: PADATA_CPU_SERIAL or PADATA_CPU_PARALLEL corresponding
* to parallel and serial cpumasks respectively.
* @cpumask: the cpumask to use
*
* Return: 0 on success or negative error code
*/
int padata_set_cpumask(struct padata_instance *pinst, int cpumask_type,
cpumask_var_t cpumask)
{
struct cpumask *serial_mask, *parallel_mask;
int err = -EINVAL;
cpus_read_lock();
mutex_lock(&pinst->lock);
switch (cpumask_type) {
case PADATA_CPU_PARALLEL:
serial_mask = pinst->cpumask.cbcpu;
parallel_mask = cpumask;
break;
case PADATA_CPU_SERIAL:
parallel_mask = pinst->cpumask.pcpu;
serial_mask = cpumask;
break;
default:
goto out;
}
err = __padata_set_cpumasks(pinst, parallel_mask, serial_mask);
out:
mutex_unlock(&pinst->lock);
cpus_read_unlock();
return err;
}
EXPORT_SYMBOL(padata_set_cpumask);
#ifdef CONFIG_HOTPLUG_CPU
static int __padata_add_cpu(struct padata_instance *pinst, int cpu)
{
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
int err = 0;
if (cpumask_test_cpu(cpu, cpu_online_mask)) {
padata: validate cpumask without removed CPU during offline Configuring an instance's parallel mask without any online CPUs... echo 2 > /sys/kernel/pcrypt/pencrypt/parallel_cpumask echo 0 > /sys/devices/system/cpu/cpu1/online ...makes tcrypt mode=215 crash like this: divide error: 0000 [#1] SMP PTI CPU: 4 PID: 283 Comm: modprobe Not tainted 5.4.0-rc8-padata-doc-v2+ #2 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20191013_105130-anatol 04/01/2014 RIP: 0010:padata_do_parallel+0x114/0x300 Call Trace: pcrypt_aead_encrypt+0xc0/0xd0 [pcrypt] crypto_aead_encrypt+0x1f/0x30 do_mult_aead_op+0x4e/0xdf [tcrypt] test_mb_aead_speed.constprop.0.cold+0x226/0x564 [tcrypt] do_test+0x28c2/0x4d49 [tcrypt] tcrypt_mod_init+0x55/0x1000 [tcrypt] ... cpumask_weight() in padata_cpu_hash() returns 0 because the mask has no CPUs. The problem is __padata_remove_cpu() checks for valid masks too early and so doesn't mark the instance PADATA_INVALID as expected, which would have made padata_do_parallel() return error before doing the division. Fix by introducing a second padata CPU hotplug state before CPUHP_BRINGUP_CPU so that __padata_remove_cpu() sees the online mask without @cpu. No need for the second argument to padata_replace() since @cpu is now already missing from the online mask. Fixes: 33e54450683c ("padata: Handle empty padata cpumasks") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Eric Biggers <ebiggers@kernel.org> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-12-03 19:31:10 +00:00
err = padata_replace(pinst);
if (padata_validate_cpumask(pinst, pinst->cpumask.pcpu) &&
padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
__padata_start(pinst);
}
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
return err;
}
static int __padata_remove_cpu(struct padata_instance *pinst, int cpu)
{
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
int err = 0;
padata: validate cpumask without removed CPU during offline Configuring an instance's parallel mask without any online CPUs... echo 2 > /sys/kernel/pcrypt/pencrypt/parallel_cpumask echo 0 > /sys/devices/system/cpu/cpu1/online ...makes tcrypt mode=215 crash like this: divide error: 0000 [#1] SMP PTI CPU: 4 PID: 283 Comm: modprobe Not tainted 5.4.0-rc8-padata-doc-v2+ #2 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20191013_105130-anatol 04/01/2014 RIP: 0010:padata_do_parallel+0x114/0x300 Call Trace: pcrypt_aead_encrypt+0xc0/0xd0 [pcrypt] crypto_aead_encrypt+0x1f/0x30 do_mult_aead_op+0x4e/0xdf [tcrypt] test_mb_aead_speed.constprop.0.cold+0x226/0x564 [tcrypt] do_test+0x28c2/0x4d49 [tcrypt] tcrypt_mod_init+0x55/0x1000 [tcrypt] ... cpumask_weight() in padata_cpu_hash() returns 0 because the mask has no CPUs. The problem is __padata_remove_cpu() checks for valid masks too early and so doesn't mark the instance PADATA_INVALID as expected, which would have made padata_do_parallel() return error before doing the division. Fix by introducing a second padata CPU hotplug state before CPUHP_BRINGUP_CPU so that __padata_remove_cpu() sees the online mask without @cpu. No need for the second argument to padata_replace() since @cpu is now already missing from the online mask. Fixes: 33e54450683c ("padata: Handle empty padata cpumasks") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Eric Biggers <ebiggers@kernel.org> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-12-03 19:31:10 +00:00
if (!cpumask_test_cpu(cpu, cpu_online_mask)) {
if (!padata_validate_cpumask(pinst, pinst->cpumask.pcpu) ||
!padata_validate_cpumask(pinst, pinst->cpumask.cbcpu))
__padata_stop(pinst);
padata: validate cpumask without removed CPU during offline Configuring an instance's parallel mask without any online CPUs... echo 2 > /sys/kernel/pcrypt/pencrypt/parallel_cpumask echo 0 > /sys/devices/system/cpu/cpu1/online ...makes tcrypt mode=215 crash like this: divide error: 0000 [#1] SMP PTI CPU: 4 PID: 283 Comm: modprobe Not tainted 5.4.0-rc8-padata-doc-v2+ #2 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20191013_105130-anatol 04/01/2014 RIP: 0010:padata_do_parallel+0x114/0x300 Call Trace: pcrypt_aead_encrypt+0xc0/0xd0 [pcrypt] crypto_aead_encrypt+0x1f/0x30 do_mult_aead_op+0x4e/0xdf [tcrypt] test_mb_aead_speed.constprop.0.cold+0x226/0x564 [tcrypt] do_test+0x28c2/0x4d49 [tcrypt] tcrypt_mod_init+0x55/0x1000 [tcrypt] ... cpumask_weight() in padata_cpu_hash() returns 0 because the mask has no CPUs. The problem is __padata_remove_cpu() checks for valid masks too early and so doesn't mark the instance PADATA_INVALID as expected, which would have made padata_do_parallel() return error before doing the division. Fix by introducing a second padata CPU hotplug state before CPUHP_BRINGUP_CPU so that __padata_remove_cpu() sees the online mask without @cpu. No need for the second argument to padata_replace() since @cpu is now already missing from the online mask. Fixes: 33e54450683c ("padata: Handle empty padata cpumasks") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Eric Biggers <ebiggers@kernel.org> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-12-03 19:31:10 +00:00
err = padata_replace(pinst);
}
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
return err;
}
static inline int pinst_has_cpu(struct padata_instance *pinst, int cpu)
{
return cpumask_test_cpu(cpu, pinst->cpumask.pcpu) ||
cpumask_test_cpu(cpu, pinst->cpumask.cbcpu);
}
static int padata_cpu_online(unsigned int cpu, struct hlist_node *node)
{
struct padata_instance *pinst;
int ret;
pinst = hlist_entry_safe(node, struct padata_instance, cpu_online_node);
if (!pinst_has_cpu(pinst, cpu))
return 0;
mutex_lock(&pinst->lock);
ret = __padata_add_cpu(pinst, cpu);
mutex_unlock(&pinst->lock);
return ret;
}
padata: validate cpumask without removed CPU during offline Configuring an instance's parallel mask without any online CPUs... echo 2 > /sys/kernel/pcrypt/pencrypt/parallel_cpumask echo 0 > /sys/devices/system/cpu/cpu1/online ...makes tcrypt mode=215 crash like this: divide error: 0000 [#1] SMP PTI CPU: 4 PID: 283 Comm: modprobe Not tainted 5.4.0-rc8-padata-doc-v2+ #2 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20191013_105130-anatol 04/01/2014 RIP: 0010:padata_do_parallel+0x114/0x300 Call Trace: pcrypt_aead_encrypt+0xc0/0xd0 [pcrypt] crypto_aead_encrypt+0x1f/0x30 do_mult_aead_op+0x4e/0xdf [tcrypt] test_mb_aead_speed.constprop.0.cold+0x226/0x564 [tcrypt] do_test+0x28c2/0x4d49 [tcrypt] tcrypt_mod_init+0x55/0x1000 [tcrypt] ... cpumask_weight() in padata_cpu_hash() returns 0 because the mask has no CPUs. The problem is __padata_remove_cpu() checks for valid masks too early and so doesn't mark the instance PADATA_INVALID as expected, which would have made padata_do_parallel() return error before doing the division. Fix by introducing a second padata CPU hotplug state before CPUHP_BRINGUP_CPU so that __padata_remove_cpu() sees the online mask without @cpu. No need for the second argument to padata_replace() since @cpu is now already missing from the online mask. Fixes: 33e54450683c ("padata: Handle empty padata cpumasks") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Eric Biggers <ebiggers@kernel.org> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-12-03 19:31:10 +00:00
static int padata_cpu_dead(unsigned int cpu, struct hlist_node *node)
{
struct padata_instance *pinst;
int ret;
pinst = hlist_entry_safe(node, struct padata_instance, cpu_dead_node);
if (!pinst_has_cpu(pinst, cpu))
return 0;
mutex_lock(&pinst->lock);
ret = __padata_remove_cpu(pinst, cpu);
mutex_unlock(&pinst->lock);
return ret;
}
static enum cpuhp_state hp_online;
#endif
static void __padata_free(struct padata_instance *pinst)
{
#ifdef CONFIG_HOTPLUG_CPU
cpuhp_state_remove_instance_nocalls(CPUHP_PADATA_DEAD,
&pinst->cpu_dead_node);
cpuhp_state_remove_instance_nocalls(hp_online, &pinst->cpu_online_node);
#endif
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
WARN_ON(!list_empty(&pinst->pslist));
free_cpumask_var(pinst->cpumask.pcpu);
free_cpumask_var(pinst->cpumask.cbcpu);
destroy_workqueue(pinst->serial_wq);
destroy_workqueue(pinst->parallel_wq);
kfree(pinst);
}
#define kobj2pinst(_kobj) \
container_of(_kobj, struct padata_instance, kobj)
#define attr2pentry(_attr) \
container_of(_attr, struct padata_sysfs_entry, attr)
static void padata_sysfs_release(struct kobject *kobj)
{
struct padata_instance *pinst = kobj2pinst(kobj);
__padata_free(pinst);
}
struct padata_sysfs_entry {
struct attribute attr;
ssize_t (*show)(struct padata_instance *, struct attribute *, char *);
ssize_t (*store)(struct padata_instance *, struct attribute *,
const char *, size_t);
};
static ssize_t show_cpumask(struct padata_instance *pinst,
struct attribute *attr, char *buf)
{
struct cpumask *cpumask;
ssize_t len;
mutex_lock(&pinst->lock);
if (!strcmp(attr->name, "serial_cpumask"))
cpumask = pinst->cpumask.cbcpu;
else
cpumask = pinst->cpumask.pcpu;
len = snprintf(buf, PAGE_SIZE, "%*pb\n",
nr_cpu_ids, cpumask_bits(cpumask));
mutex_unlock(&pinst->lock);
return len < PAGE_SIZE ? len : -EINVAL;
}
static ssize_t store_cpumask(struct padata_instance *pinst,
struct attribute *attr,
const char *buf, size_t count)
{
cpumask_var_t new_cpumask;
ssize_t ret;
int mask_type;
if (!alloc_cpumask_var(&new_cpumask, GFP_KERNEL))
return -ENOMEM;
ret = bitmap_parse(buf, count, cpumask_bits(new_cpumask),
nr_cpumask_bits);
if (ret < 0)
goto out;
mask_type = !strcmp(attr->name, "serial_cpumask") ?
PADATA_CPU_SERIAL : PADATA_CPU_PARALLEL;
ret = padata_set_cpumask(pinst, mask_type, new_cpumask);
if (!ret)
ret = count;
out:
free_cpumask_var(new_cpumask);
return ret;
}
#define PADATA_ATTR_RW(_name, _show_name, _store_name) \
static struct padata_sysfs_entry _name##_attr = \
__ATTR(_name, 0644, _show_name, _store_name)
#define PADATA_ATTR_RO(_name, _show_name) \
static struct padata_sysfs_entry _name##_attr = \
__ATTR(_name, 0400, _show_name, NULL)
PADATA_ATTR_RW(serial_cpumask, show_cpumask, store_cpumask);
PADATA_ATTR_RW(parallel_cpumask, show_cpumask, store_cpumask);
/*
* Padata sysfs provides the following objects:
* serial_cpumask [RW] - cpumask for serial workers
* parallel_cpumask [RW] - cpumask for parallel workers
*/
static struct attribute *padata_default_attrs[] = {
&serial_cpumask_attr.attr,
&parallel_cpumask_attr.attr,
NULL,
};
ATTRIBUTE_GROUPS(padata_default);
static ssize_t padata_sysfs_show(struct kobject *kobj,
struct attribute *attr, char *buf)
{
struct padata_instance *pinst;
struct padata_sysfs_entry *pentry;
ssize_t ret = -EIO;
pinst = kobj2pinst(kobj);
pentry = attr2pentry(attr);
if (pentry->show)
ret = pentry->show(pinst, attr, buf);
return ret;
}
static ssize_t padata_sysfs_store(struct kobject *kobj, struct attribute *attr,
const char *buf, size_t count)
{
struct padata_instance *pinst;
struct padata_sysfs_entry *pentry;
ssize_t ret = -EIO;
pinst = kobj2pinst(kobj);
pentry = attr2pentry(attr);
if (pentry->show)
ret = pentry->store(pinst, attr, buf, count);
return ret;
}
static const struct sysfs_ops padata_sysfs_ops = {
.show = padata_sysfs_show,
.store = padata_sysfs_store,
};
static const struct kobj_type padata_attr_type = {
.sysfs_ops = &padata_sysfs_ops,
.default_groups = padata_default_groups,
.release = padata_sysfs_release,
};
/**
* padata_alloc - allocate and initialize a padata instance
* @name: used to identify the instance
*
* Return: new instance on success, NULL on error
*/
struct padata_instance *padata_alloc(const char *name)
{
struct padata_instance *pinst;
pinst = kzalloc(sizeof(struct padata_instance), GFP_KERNEL);
if (!pinst)
goto err;
padata: unbind parallel jobs from specific CPUs Padata binds the parallel part of a job to a single CPU and round-robins over all CPUs in the system for each successive job. Though the serial parts rely on per-CPU queues for correct ordering, they're not necessary for parallel work, and it improves performance to run the job locally on NUMA machines and let the scheduler pick the CPU within a node on a busy system. So, make the parallel workqueue unbound. Update the parallel workqueue's cpumask when the instance's parallel cpumask changes. Now that parallel jobs no longer run on max_active=1 workqueues, two or more parallel works that hash to the same CPU may run simultaneously, finish out of order, and so be serialized out of order. Prevent this by keeping the works sorted on the reorder list by sequence number and checking that in the reordering logic. padata_get_next becomes padata_find_next so it can be reused for the end of padata_reorder, where it's used to avoid uselessly queueing work when the next job by sequence number isn't finished yet but a later job that hashed to the same CPU has. The ENODATA case in padata_find_next no longer makes sense because parallel jobs aren't bound to specific CPUs. The EINPROGRESS case takes care of the scenario where a parallel job is potentially running on the same CPU as padata_find_next, and with only one error code left, just use NULL instead. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Tejun Heo <tj@kernel.org> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-09-06 01:40:28 +00:00
pinst->parallel_wq = alloc_workqueue("%s_parallel", WQ_UNBOUND, 0,
name);
if (!pinst->parallel_wq)
goto err_free_inst;
cpus_read_lock();
pinst->serial_wq = alloc_workqueue("%s_serial", WQ_MEM_RECLAIM |
WQ_CPU_INTENSIVE, 1, name);
if (!pinst->serial_wq)
goto err_put_cpus;
if (!alloc_cpumask_var(&pinst->cpumask.pcpu, GFP_KERNEL))
goto err_free_serial_wq;
if (!alloc_cpumask_var(&pinst->cpumask.cbcpu, GFP_KERNEL)) {
free_cpumask_var(pinst->cpumask.pcpu);
goto err_free_serial_wq;
}
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
INIT_LIST_HEAD(&pinst->pslist);
cpumask_copy(pinst->cpumask.pcpu, cpu_possible_mask);
cpumask_copy(pinst->cpumask.cbcpu, cpu_possible_mask);
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
if (padata_setup_cpumasks(pinst))
goto err_free_masks;
__padata_start(pinst);
kobject_init(&pinst->kobj, &padata_attr_type);
mutex_init(&pinst->lock);
#ifdef CONFIG_HOTPLUG_CPU
cpuhp_state_add_instance_nocalls_cpuslocked(hp_online,
&pinst->cpu_online_node);
padata: validate cpumask without removed CPU during offline Configuring an instance's parallel mask without any online CPUs... echo 2 > /sys/kernel/pcrypt/pencrypt/parallel_cpumask echo 0 > /sys/devices/system/cpu/cpu1/online ...makes tcrypt mode=215 crash like this: divide error: 0000 [#1] SMP PTI CPU: 4 PID: 283 Comm: modprobe Not tainted 5.4.0-rc8-padata-doc-v2+ #2 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20191013_105130-anatol 04/01/2014 RIP: 0010:padata_do_parallel+0x114/0x300 Call Trace: pcrypt_aead_encrypt+0xc0/0xd0 [pcrypt] crypto_aead_encrypt+0x1f/0x30 do_mult_aead_op+0x4e/0xdf [tcrypt] test_mb_aead_speed.constprop.0.cold+0x226/0x564 [tcrypt] do_test+0x28c2/0x4d49 [tcrypt] tcrypt_mod_init+0x55/0x1000 [tcrypt] ... cpumask_weight() in padata_cpu_hash() returns 0 because the mask has no CPUs. The problem is __padata_remove_cpu() checks for valid masks too early and so doesn't mark the instance PADATA_INVALID as expected, which would have made padata_do_parallel() return error before doing the division. Fix by introducing a second padata CPU hotplug state before CPUHP_BRINGUP_CPU so that __padata_remove_cpu() sees the online mask without @cpu. No need for the second argument to padata_replace() since @cpu is now already missing from the online mask. Fixes: 33e54450683c ("padata: Handle empty padata cpumasks") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Eric Biggers <ebiggers@kernel.org> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-12-03 19:31:10 +00:00
cpuhp_state_add_instance_nocalls_cpuslocked(CPUHP_PADATA_DEAD,
&pinst->cpu_dead_node);
#endif
cpus_read_unlock();
return pinst;
err_free_masks:
free_cpumask_var(pinst->cpumask.pcpu);
free_cpumask_var(pinst->cpumask.cbcpu);
err_free_serial_wq:
destroy_workqueue(pinst->serial_wq);
err_put_cpus:
cpus_read_unlock();
destroy_workqueue(pinst->parallel_wq);
err_free_inst:
kfree(pinst);
err:
return NULL;
}
EXPORT_SYMBOL(padata_alloc);
/**
* padata_free - free a padata instance
*
* @pinst: padata instance to free
*/
void padata_free(struct padata_instance *pinst)
{
kobject_put(&pinst->kobj);
}
EXPORT_SYMBOL(padata_free);
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
/**
* padata_alloc_shell - Allocate and initialize padata shell.
*
* @pinst: Parent padata_instance object.
*
* Return: new shell on success, NULL on error
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
*/
struct padata_shell *padata_alloc_shell(struct padata_instance *pinst)
{
struct parallel_data *pd;
struct padata_shell *ps;
ps = kzalloc(sizeof(*ps), GFP_KERNEL);
if (!ps)
goto out;
ps->pinst = pinst;
cpus_read_lock();
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
pd = padata_alloc_pd(ps);
cpus_read_unlock();
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
if (!pd)
goto out_free_ps;
mutex_lock(&pinst->lock);
RCU_INIT_POINTER(ps->pd, pd);
list_add(&ps->list, &pinst->pslist);
mutex_unlock(&pinst->lock);
return ps;
out_free_ps:
kfree(ps);
out:
return NULL;
}
EXPORT_SYMBOL(padata_alloc_shell);
/**
* padata_free_shell - free a padata shell
*
* @ps: padata shell to free
*/
void padata_free_shell(struct padata_shell *ps)
{
padata: Fix refcnt handling in padata_free_shell() In a high-load arm64 environment, the pcrypt_aead01 test in LTP can lead to system UAF (Use-After-Free) issues. Due to the lengthy analysis of the pcrypt_aead01 function call, I'll describe the problem scenario using a simplified model: Suppose there's a user of padata named `user_function` that adheres to the padata requirement of calling `padata_free_shell` after `serial()` has been invoked, as demonstrated in the following code: ```c struct request { struct padata_priv padata; struct completion *done; }; void parallel(struct padata_priv *padata) { do_something(); } void serial(struct padata_priv *padata) { struct request *request = container_of(padata, struct request, padata); complete(request->done); } void user_function() { DECLARE_COMPLETION(done) padata->parallel = parallel; padata->serial = serial; padata_do_parallel(); wait_for_completion(&done); padata_free_shell(); } ``` In the corresponding padata.c file, there's the following code: ```c static void padata_serial_worker(struct work_struct *serial_work) { ... cnt = 0; while (!list_empty(&local_list)) { ... padata->serial(padata); cnt++; } local_bh_enable(); if (refcount_sub_and_test(cnt, &pd->refcnt)) padata_free_pd(pd); } ``` Because of the high system load and the accumulation of unexecuted softirq at this moment, `local_bh_enable()` in padata takes longer to execute than usual. Subsequently, when accessing `pd->refcnt`, `pd` has already been released by `padata_free_shell()`, resulting in a UAF issue with `pd->refcnt`. The fix is straightforward: add `refcount_dec_and_test` before calling `padata_free_pd` in `padata_free_shell`. Fixes: 07928d9bfc81 ("padata: Remove broken queue flushing") Signed-off-by: WangJinchao <wangjinchao@xfusion.com> Acked-by: Daniel Jordan <daniel.m.jordan@oracle.com> Acked-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2023-10-16 01:15:21 +00:00
struct parallel_data *pd;
if (!ps)
return;
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
mutex_lock(&ps->pinst->lock);
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
list_del(&ps->list);
padata: Fix refcnt handling in padata_free_shell() In a high-load arm64 environment, the pcrypt_aead01 test in LTP can lead to system UAF (Use-After-Free) issues. Due to the lengthy analysis of the pcrypt_aead01 function call, I'll describe the problem scenario using a simplified model: Suppose there's a user of padata named `user_function` that adheres to the padata requirement of calling `padata_free_shell` after `serial()` has been invoked, as demonstrated in the following code: ```c struct request { struct padata_priv padata; struct completion *done; }; void parallel(struct padata_priv *padata) { do_something(); } void serial(struct padata_priv *padata) { struct request *request = container_of(padata, struct request, padata); complete(request->done); } void user_function() { DECLARE_COMPLETION(done) padata->parallel = parallel; padata->serial = serial; padata_do_parallel(); wait_for_completion(&done); padata_free_shell(); } ``` In the corresponding padata.c file, there's the following code: ```c static void padata_serial_worker(struct work_struct *serial_work) { ... cnt = 0; while (!list_empty(&local_list)) { ... padata->serial(padata); cnt++; } local_bh_enable(); if (refcount_sub_and_test(cnt, &pd->refcnt)) padata_free_pd(pd); } ``` Because of the high system load and the accumulation of unexecuted softirq at this moment, `local_bh_enable()` in padata takes longer to execute than usual. Subsequently, when accessing `pd->refcnt`, `pd` has already been released by `padata_free_shell()`, resulting in a UAF issue with `pd->refcnt`. The fix is straightforward: add `refcount_dec_and_test` before calling `padata_free_pd` in `padata_free_shell`. Fixes: 07928d9bfc81 ("padata: Remove broken queue flushing") Signed-off-by: WangJinchao <wangjinchao@xfusion.com> Acked-by: Daniel Jordan <daniel.m.jordan@oracle.com> Acked-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2023-10-16 01:15:21 +00:00
pd = rcu_dereference_protected(ps->pd, 1);
if (refcount_dec_and_test(&pd->refcnt))
padata_free_pd(pd);
mutex_unlock(&ps->pinst->lock);
crypto: pcrypt - Avoid deadlock by using per-instance padata queues If the pcrypt template is used multiple times in an algorithm, then a deadlock occurs because all pcrypt instances share the same padata_instance, which completes requests in the order submitted. That is, the inner pcrypt request waits for the outer pcrypt request while the outer request is already waiting for the inner. This patch fixes this by allocating a set of queues for each pcrypt instance instead of using two global queues. In order to maintain the existing user-space interface, the pinst structure remains global so any sysfs modifications will apply to every pcrypt instance. Note that when an update occurs we have to allocate memory for every pcrypt instance. Should one of the allocations fail we will abort the update without rolling back changes already made. The new per-instance data structure is called padata_shell and is essentially a wrapper around parallel_data. Reproducer: #include <linux/if_alg.h> #include <sys/socket.h> #include <unistd.h> int main() { struct sockaddr_alg addr = { .salg_type = "aead", .salg_name = "pcrypt(pcrypt(rfc4106-gcm-aesni))" }; int algfd, reqfd; char buf[32] = { 0 }; algfd = socket(AF_ALG, SOCK_SEQPACKET, 0); bind(algfd, (void *)&addr, sizeof(addr)); setsockopt(algfd, SOL_ALG, ALG_SET_KEY, buf, 20); reqfd = accept(algfd, 0, 0); write(reqfd, buf, 32); read(reqfd, buf, 16); } Reported-by: syzbot+56c7151cad94eec37c521f0e47d2eee53f9361c4@syzkaller.appspotmail.com Fixes: 5068c7a883d1 ("crypto: pcrypt - Add pcrypt crypto parallelization wrapper") Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Tested-by: Eric Biggers <ebiggers@kernel.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-11-26 07:58:45 +00:00
kfree(ps);
}
EXPORT_SYMBOL(padata_free_shell);
void __init padata_init(void)
{
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
unsigned int i, possible_cpus;
#ifdef CONFIG_HOTPLUG_CPU
int ret;
ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "padata:online",
padata: validate cpumask without removed CPU during offline Configuring an instance's parallel mask without any online CPUs... echo 2 > /sys/kernel/pcrypt/pencrypt/parallel_cpumask echo 0 > /sys/devices/system/cpu/cpu1/online ...makes tcrypt mode=215 crash like this: divide error: 0000 [#1] SMP PTI CPU: 4 PID: 283 Comm: modprobe Not tainted 5.4.0-rc8-padata-doc-v2+ #2 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20191013_105130-anatol 04/01/2014 RIP: 0010:padata_do_parallel+0x114/0x300 Call Trace: pcrypt_aead_encrypt+0xc0/0xd0 [pcrypt] crypto_aead_encrypt+0x1f/0x30 do_mult_aead_op+0x4e/0xdf [tcrypt] test_mb_aead_speed.constprop.0.cold+0x226/0x564 [tcrypt] do_test+0x28c2/0x4d49 [tcrypt] tcrypt_mod_init+0x55/0x1000 [tcrypt] ... cpumask_weight() in padata_cpu_hash() returns 0 because the mask has no CPUs. The problem is __padata_remove_cpu() checks for valid masks too early and so doesn't mark the instance PADATA_INVALID as expected, which would have made padata_do_parallel() return error before doing the division. Fix by introducing a second padata CPU hotplug state before CPUHP_BRINGUP_CPU so that __padata_remove_cpu() sees the online mask without @cpu. No need for the second argument to padata_replace() since @cpu is now already missing from the online mask. Fixes: 33e54450683c ("padata: Handle empty padata cpumasks") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Eric Biggers <ebiggers@kernel.org> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-12-03 19:31:10 +00:00
padata_cpu_online, NULL);
if (ret < 0)
goto err;
hp_online = ret;
padata: validate cpumask without removed CPU during offline Configuring an instance's parallel mask without any online CPUs... echo 2 > /sys/kernel/pcrypt/pencrypt/parallel_cpumask echo 0 > /sys/devices/system/cpu/cpu1/online ...makes tcrypt mode=215 crash like this: divide error: 0000 [#1] SMP PTI CPU: 4 PID: 283 Comm: modprobe Not tainted 5.4.0-rc8-padata-doc-v2+ #2 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20191013_105130-anatol 04/01/2014 RIP: 0010:padata_do_parallel+0x114/0x300 Call Trace: pcrypt_aead_encrypt+0xc0/0xd0 [pcrypt] crypto_aead_encrypt+0x1f/0x30 do_mult_aead_op+0x4e/0xdf [tcrypt] test_mb_aead_speed.constprop.0.cold+0x226/0x564 [tcrypt] do_test+0x28c2/0x4d49 [tcrypt] tcrypt_mod_init+0x55/0x1000 [tcrypt] ... cpumask_weight() in padata_cpu_hash() returns 0 because the mask has no CPUs. The problem is __padata_remove_cpu() checks for valid masks too early and so doesn't mark the instance PADATA_INVALID as expected, which would have made padata_do_parallel() return error before doing the division. Fix by introducing a second padata CPU hotplug state before CPUHP_BRINGUP_CPU so that __padata_remove_cpu() sees the online mask without @cpu. No need for the second argument to padata_replace() since @cpu is now already missing from the online mask. Fixes: 33e54450683c ("padata: Handle empty padata cpumasks") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Eric Biggers <ebiggers@kernel.org> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2019-12-03 19:31:10 +00:00
ret = cpuhp_setup_state_multi(CPUHP_PADATA_DEAD, "padata:dead",
NULL, padata_cpu_dead);
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
if (ret < 0)
goto remove_online_state;
#endif
possible_cpus = num_possible_cpus();
padata_works = kmalloc_array(possible_cpus, sizeof(struct padata_work),
GFP_KERNEL);
if (!padata_works)
goto remove_dead_state;
for (i = 0; i < possible_cpus; ++i)
list_add(&padata_works[i].pw_list, &padata_free_works);
return;
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
remove_dead_state:
#ifdef CONFIG_HOTPLUG_CPU
cpuhp_remove_multi_state(CPUHP_PADATA_DEAD);
remove_online_state:
cpuhp_remove_multi_state(hp_online);
err:
#endif
padata: allocate work structures for parallel jobs from a pool padata allocates per-CPU, per-instance work structs for parallel jobs. A do_parallel call assigns a job to a sequence number and hashes the number to a CPU, where the job will eventually run using the corresponding work. This approach fit with how padata used to bind a job to each CPU round-robin, makes less sense after commit bfde23ce200e6 ("padata: unbind parallel jobs from specific CPUs") because a work isn't bound to a particular CPU anymore, and isn't needed at all for multithreaded jobs because they don't have sequence numbers. Replace the per-CPU works with a preallocated pool, which allows sharing them between existing padata users and the upcoming multithreaded user. The pool will also facilitate setting NUMA-aware concurrency limits with later users. The pool is sized according to the number of possible CPUs. With this limit, MAX_OBJ_NUM no longer makes sense, so remove it. If the global pool is exhausted, a parallel job is run in the current task instead to throttle a system trying to do too much in parallel. Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Josh Triplett <josh@joshtriplett.org> Cc: Alexander Duyck <alexander.h.duyck@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: David Hildenbrand <david@redhat.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Pavel Machek <pavel@ucw.cz> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Robert Elliott <elliott@hpe.com> Cc: Shile Zhang <shile.zhang@linux.alibaba.com> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: Steven Sistare <steven.sistare@oracle.com> Cc: Tejun Heo <tj@kernel.org> Cc: Zi Yan <ziy@nvidia.com> Link: http://lkml.kernel.org/r/20200527173608.2885243-4-daniel.m.jordan@oracle.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 22:59:39 +00:00
pr_warn("padata: initialization failed\n");
}