While unmapping an ODP writable page, the dirty bit of the page is set. In
order to do so, the head of the compound page is found.
Currently, the compound head is found even on non-writable pages, where it is
never used, leading to unnecessary cpu barrier that impacts performance.
This patch moves the search for the compound head to be done only when needed.
Signed-off-by: Guy Shapiro <guysh@mellanox.com>
Acked-by: Shachar Raindel <raindel@mellanox.com>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Currently, while mapping or unmapping pages for ODP, the umem mutex is locked
and unlocked once for each page. Such lock/unlock operation take few tens to
hundreds of nsecs. This makes a significant impact when mapping or unmapping few
MBs of memory.
To avoid this, the mutex should be locked only once per operation, and not per
page.
Signed-off-by: Guy Shapiro <guysh@mellanox.com>
Acked-by: Shachar Raindel <raindel@mellanox.com>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Get the actual (non-mapped) ip/tcp address of the connecting peer from
the port mapper
Also setup the passive side endpoint to correctly display the actual
and mapped addresses for the new connection.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Get the actual (non-mapped) ip/tcp address of the connecting peer from
the port mapper and report the address info to the user space application
at the time of connection establishment
Signed-off-by: Tatyana Nikolova <tatyana.e.nikolova@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Add functionality to enable the port mapper on the passive side to provide to its
clients the actual (non-mapped) ip/tcp address information of the connecting peer
1) Adding remote_info_cb() to process the address info of the connecting peer
The address info is provided by the user space port mapper service when
the connection is initiated by the peer
2) Adding a hash list to store the remote address info
3) Adding functionality to add/remove the remote address info
After the info has been provided to the port mapper client,
it is removed from the hash list
Signed-off-by: Tatyana Nikolova <tatyana.e.nikolova@intel.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Currently the iw_cxgb4 implementation requires the qp and cq qid densities
to match as well as the qp and cq id ranges. So fail a device open if
the device configuration doesn't meet the requirements.
The reason for these restictions has to do with the fact that IQ qid X
has a UGTS register in the same bar2 page as EQ qid X. Thus both qids
need to be allocated to the same user process for security reasons.
The logic that does this (the qpid allocator in iw_cxgb4/resource.c)
handles this but requires the above restrictions.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
For T5, we must not use the kdb/kgts registers, in order avoid db drops
under extreme loads.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
- get_dma_mr() was using ~0UL which is should be ~0ULL. This causes the
DMA MR to get setup incorrectly in hardware.
- wr_log_show() needed a 64b divide function div64_u64() instead of
doing
division directly.
- fixed warnings about recasting a pointer to a u64
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When accepting a new IPv4 connect to an IPv6 socket, the CMA tries to
canonize the address family to IPv4, but does not properly process
the listening sockaddr to get the listening port, and does not properly
set the address family of the canonized sockaddr.
Fixes: e51060f08a ("IB: IP address based RDMA connection manager")
Cc: <stable@vger.kernel.org>
Reported-By: Yotam Kenneth <yotamke@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Tested-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Pull fourth vfs update from Al Viro:
"d_inode() annotations from David Howells (sat in for-next since before
the beginning of merge window) + four assorted fixes"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
RCU pathwalk breakage when running into a symlink overmounting something
fix I_DIO_WAKEUP definition
direct-io: only inc/dec inode->i_dio_count for file systems
fs/9p: fix readdir()
VFS: assorted d_backing_inode() annotations
VFS: fs/inode.c helpers: d_inode() annotations
VFS: fs/cachefiles: d_backing_inode() annotations
VFS: fs library helpers: d_inode() annotations
VFS: assorted weird filesystems: d_inode() annotations
VFS: normal filesystems (and lustre): d_inode() annotations
VFS: security/: d_inode() annotations
VFS: security/: d_backing_inode() annotations
VFS: net/: d_inode() annotations
VFS: net/unix: d_backing_inode() annotations
VFS: kernel/: d_inode() annotations
VFS: audit: d_backing_inode() annotations
VFS: Fix up some ->d_inode accesses in the chelsio driver
VFS: Cachefiles should perform fs modifications on the top layer only
VFS: AF_UNIX sockets should call mknod on the top layer only
Pull SCSI target updates from Nicholas Bellinger:
"Lots of activity in target land the last months.
The highlights include:
- Convert fabric drivers tree-wide to target_register_template() (hch
+ bart)
- iser-target hardening fixes + v1.0 improvements (sagi)
- Convert iscsi_thread_set usage to kthread.h + kill
iscsi_target_tq.c (sagi + nab)
- Add support for T10-PI WRITE_STRIP + READ_INSERT operation (mkp +
sagi + nab)
- DIF fixes for CONFIG_DEBUG_SG=y + UNMAP file emulation (akinobu +
sagi + mkp)
- Extended TCMU ABI v2 for future BIDI + DIF support (andy + ilias)
- Fix COMPARE_AND_WRITE handling for NO_ALLLOC drivers (hch + nab)
Thanks to everyone who contributed this round with new features,
bug-reports, fixes, cleanups and improvements.
Looking forward, it's currently shaping up to be a busy v4.2 as well"
* 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending: (69 commits)
target: Put TCMU under a new config option
target: Version 2 of TCMU ABI
target: fix tcm_mod_builder.py
target/file: Fix UNMAP with DIF protection support
target/file: Fix SG table for prot_buf initialization
target/file: Fix BUG() when CONFIG_DEBUG_SG=y and DIF protection enabled
target: Make core_tmr_abort_task() skip TMFs
target/sbc: Update sbc_dif_generate pr_debug output
target/sbc: Make internal DIF emulation honor ->prot_checks
target/sbc: Return INVALID_CDB_FIELD if DIF + sess_prot_type disabled
target: Ensure sess_prot_type is saved across session restart
target/rd: Don't pass incomplete scatterlist entries to sbc_dif_verify_*
target: Remove the unused flag SCF_ACK_KREF
target: Fix two sparse warnings
target: Fix COMPARE_AND_WRITE with SG_TO_MEM_NOALLOC handling
target: simplify the target template registration API
target: simplify target_xcopy_init_pt_lun
target: remove the unused SCF_CMD_XCOPY_PASSTHROUGH flag
target/rd: reduce code duplication in rd_execute_rw()
tcm_loop: fixup tpgt string to integer conversion
...
Pull networking fixes from David Miller:
1) Fix verifier memory corruption and other bugs in BPF layer, from
Alexei Starovoitov.
2) Add a conservative fix for doing BPF properly in the BPF classifier
of the packet scheduler on ingress. Also from Alexei.
3) The SKB scrubber should not clear out the packet MARK and security
label, from Herbert Xu.
4) Fix oops on rmmod in stmmac driver, from Bryan O'Donoghue.
5) Pause handling is not correct in the stmmac driver because it
doesn't take into consideration the RX and TX fifo sizes. From
Vince Bridgers.
6) Failure path missing unlock in FOU driver, from Wang Cong.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (44 commits)
net: dsa: use DEVICE_ATTR_RW to declare temp1_max
netns: remove BUG_ONs from net_generic()
IB/ipoib: Fix ndo_get_iflink
sfc: Fix memcpy() with const destination compiler warning.
altera tse: Fix network-delays and -retransmissions after high throughput.
net: remove unused 'dev' argument from netif_needs_gso()
act_mirred: Fix bogus header when redirecting from VLAN
inet_diag: fix access to tcp cc information
tcp: tcp_get_info() should fetch socket fields once
net: dsa: mv88e6xxx: Add missing initialization in mv88e6xxx_set_port_state()
skbuff: Do not scrub skb mark within the same name space
Revert "net: Reset secmark when scrubbing packet"
bpf: fix two bugs in verification logic when accessing 'ctx' pointer
bpf: fix bpf helpers to use skb->mac_header relative offsets
stmmac: Configure Flow Control to work correctly based on rxfifo size
stmmac: Enable unicast pause frame detect in GMAC Register 6
stmmac: Read tx-fifo-depth and rx-fifo-depth from the devicetree
stmmac: Add defines and documentation for enabling flow control
stmmac: Add properties for transmit and receive fifo sizes
stmmac: fix oops on rmmod after assigning ip addr
...
Currently, iflink of the parent interface was always accessed, even
when interface didn't have a parent and hence we crashed there.
Handle the interface types properly: for a child interface, return
the ifindex of the parent, for parent interface, return its ifindex.
For child devices, make sure to set the parent pointer prior to
invoking register_netdevice(), this allows the new ndo to be called
by the stack immediately after the child device is registered.
Fixes: 5aa7add8f1 ('infiniband/ipoib: implement ndo_get_iflink')
Reported-by: Honggang Li <honli@redhat.com>
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Honggang Li <honli@redhat.com>
Reviewed-By: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>+
Acked-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
set_filter_wr is requesting __GFP_NOFAIL allocation although it can return
ENOMEM without any problems obviously (t4_l2t_set_switching does that
already). So the non-failing requirement is too strong without any
obvious reason. Drop __GFP_NOFAIL and reorganize the code to have the
failure paths easier.
The same applies to _c4iw_write_mem_dma_aligned which uses __GFP_NOFAIL
and then checks the return value and returns -ENOMEM on failure. This
doesn't make any sense what so ever. Either the allocation cannot fail or
it can.
del_filter_wr seems to be safe as well because the filter entry is not
marked as pending and the return value is propagated up the stack up to
c4iw_destroy_listen.
Signed-off-by: Michal Hocko <mhocko@suse.cz>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Hariprasad S <hariprasad@chelsio.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
In some rare cases, IO operations may be not aligned to page
boundaries. This prevents iser from performing fast memory
registration. In order to overcome that iser uses a bounce
buffer to carry the transaction. We basically allocate a buffer
in the size of the transaction and perform a copy.
The buffer allocation using kmalloc is too restrictive since it
requires higher order (atomic) allocations for large transactions
(which may result in memory exhaustion fairly fast for some workloads).
We rewrite the bounce buffer code path to allocate scattered pages
and perform a copy between the transaction sg and the bounce sg.
Reported-by: Alex Lyakas <alex@zadarastorage.com>
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
In singleton scatterlists, DMA memory registration code
is taken both for Fastreg and FMR code paths. Move it to
a function.
This patch does not change any functionality.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Adir Lev <adirl@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Instead of passing ib_sge as output variable, we pass the mem_reg
pointer to have the routines fill the rkey as well. This reduces
code duplication and extra assignments. This is a preparation step
to unify some registration logics together. Also, pass iser_fast_reg_mr
the fastreg descriptor directly.
This patch does not change any functionality.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Adir Lev <adirl@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
No need to keep lkey, va, len variables, we can keep
them as struct ib_sge. This will help when we change the
memory registration logic.
This patch does not change any functionality.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Adir Lev <adirl@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Memory regions are resources that are saved
in the device caches. Increase the probability for
a cache hit by adding the MRU descriptor to pool
head.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Make iser_[create|destroy]_fastreg_desc shorter, more
readable and easily extendable.
This patch does not change any functionality.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Adir Lev <adirl@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Instead of open-coding connection fastreg pool get/put,
we introduce iser_reg_desc[get|put] helpers.
We aren't setting these static as this will be a per-device
routine later on. Also, cleanup iser_unreg_rdma_mem_fastreg
a bit.
This patch does not change any functionality.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Adir Lev <adirl@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
No need for these two separate. Keep it in a single routine
like in the fastreg case. This will also make iser_reg_page_vec
closer to iser_fast_reg_mr arguments. This is a preparation
step for registration flow refactor.
This patch does not change any functionality.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Adir Lev <adirl@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This struct members other than struct iser_mem_reg are unused,
so remove it altogether.
This patch does not change any functionality.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Adir Lev <adirl@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Buffer length was assigned twice, and no reason to set va to
io_addr and then add the offset, just set va to io_addr + offset.
This patch does not change any functionality.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Adir Lev <adirl@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
As memory registration/de-registration methods, lets
move them to their natural location. While we're at it,
make iser_reg_page_vec routine static.
This patch does not change any functionality.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
No need to pass that, we can take it from the task.
In a later stage, this function will be invoked
according to a device capability.
This patch does not change any functionality.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Adir Lev <adirl@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
No need to keep two iser_data_buf structures just in case we use
mem copy. We can avoid that just by adding a pointer to the original
sg. So keep only two iser_data_buf per command (data and protection)
and pass the relevant data_buf to bounce buffer routine.
This patch does not change any functionality.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Adir Lev <adirl@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This code was added before we had protection data length
calculation (in iser_send_command), so we needed to calc
the sg data length from the sg itself. This is not needed
anymore.
This patch does not change any functionality.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Adir Lev <adirl@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This length miss-calculation may cause a silent data corruption
in the DIX case and cause the device to reference unmapped area.
Fixes: d77e65350f ('libiscsi, iser: Adjust data_length to include protection information')
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Fast registration and local invalidate work requests can
also fail. We should call error completion handler for them.
Reported-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
In case the user unloaded ib_iser while ep_connect is in
progress, we need to destroy the endpoint although ep_disconnect
wasn't invoked (we detect this by the iser conn state != DOWN).
However, if we got an REJECTED/UNREACHABLE CM event we move the
connection state to DOWN which will prevent us from destroying
the endpoint in the module unload stage. Fix this by setting the
connection state to TERMINATING in iser_conn_error so we can still
destroy the endpoint at unload stage.
Reported-by: Ariel Nahum <arieln@mellanox.com>
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The driver already defined the pr_format, it just hadn't
been converted to use pr_info, pr_warn, and pr_err instead
of the equivalent printks. Convert so that messages from
the driver are now properly tagged with their driver name
and can be more easily debugged.
In addition, a number of these printk's were not newline
terminated, so fix that at the same time.
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This change slightly reduces the time needed to log in.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Reviewed-by: David Dillow <dave@thedillows.org>
Cc: Sebastian Parschauer <sebastian.riemer@profitbricks.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Since ib_dma_map_single can fail use ib_dma_mapping_error to check
for errors.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Acked-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Hello,
When an application using XRCs abruptly terminates, the mmaped pages
of the CQ buffers are leaked.
This comes from the fact that when resources are released in
ib_uverbs_cleanup_ucontext(), we fail to release the CQs because their
refcount is not 0.
When creating an XRC SRQ, we increment the associated CQ refcount.
This refcount is only decremented when the SRQ is released.
Therefore we need to release the SRQs prior to the CQs to make sure
that all references to the CQs are gone before trying to release these.
Signed-off-by: Sebastien Dugue <sebastien.dugue@bull.net>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The current code decreases from the mss size (which is the gso_size
from the kernel skb) the size of the packet headers.
It shouldn't do that because the mss that comes from the stack
(e.g IPoIB) includes only the tcp payload without the headers.
The result is indication to the HW that each packet that the HW sends
is smaller than what it could be, and too many packets will be sent
for big messages.
An easy way to demonstrate one more aspect of the problem is by
configuring the ipoib mtu to be less than 2*hlen (2*56) and then
run app sending big TCP messages. This will tell the HW to send packets
with giant (negative value which under unsigned arithmetics becomes
a huge positive one) length and the QP moves to SQE state.
Fixes: b832be1e40 ('IB/mlx4: Add IPoIB LSO support')
Reported-by: Matthew Finlay <matt@mellanox.com>
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
After Doug Ledford's changes there is no need in that bit, it's
semantic becomes subset of the IPOIB_FLAG_OPER_UP bit.
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Whenever there is no path->ah to the destination, keep only defined
number of skb's. Otherwise there are cases that the driver can keep
infinite list of skb's.
For example, when one device want to send unicast arp to the destination,
and from some reason the SM doesn't respond, the driver currently keeps
all the skb's. If that unicast arp traffic stopped, all these skb's
are kept by the path object till the interface is down.
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
As the result of a completion error the QP can moved to SQE state by
the hardware. Since it's not the Error state, there are no flushes
and hence the driver doesn't know about that.
The fix creates a task that after completion with error which is not a
flush tracks the QP state and if it is in SQE state moves it back to RTS.
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Update the cached broadcast record in the priv object after every new
join of this broadcast domain group.
These values are needed for the port configuration (MTU size) and to
all the new multicast (non-broadcast) join requests initial parameters.
For example, SM starts with 2K MTU for all the fabric, and after that it
restarts (or handover to new SM) with new port configuration of 4K MTU.
Without using the new values, the driver will keep its old configuration
of 2K and will not apply the new configuration of 4K.
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The current code in the RX flow uses two sg entries for each incoming
packet, the first one was for the IB headers and the second for the rest
of the data, that causes two dma map/unmap and two allocations, and few
more actions that were done at the data path.
Use only one linear skb on each incoming packet, for the data (IB
headers and payload), that reduces the packet processing in the
data-path (only one skb, no frags, the first frag was not used anyway,
less memory allocations) and the dma handling (only one dma map/unmap
over each incoming packet instead of two map/unmap per each incoming packet).
After commit 73d3fe6d1c ("gro: fix aggregation for skb using frag_list") from
Eric Dumazet, we will get full aggregation for large packets.
When running bandwidth tests before and after the (over the card's numa node),
using "netperf -H 1.1.1.3 -T -t TCP_STREAM", the results before are ~12Gbs before
and after ~16Gbs on my setup (Mellanox's ConnectX3).
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
We needed the mcast_mutex when we had to prevent the join completion
callback from having the value it stored in mcast->mc overwritten
by a delayed return from ib_sa_join_multicast. By storing the return
of ib_sa_join_multicast in an intermediate variable, we prevent a
delayed return from ib_sa_join_multicast overwriting the valid
contents of mcast->mc, and we no longer need a mutex to force the
join callback to run after the return of ib_sa_join_multicast. This
allows us to do away with the mutex entirely and protect our critical
sections with a just a spinlock instead. This is highly desirable
as there were some places where we couldn't use a mutex because the
code was not allowed to sleep, and so we were currently using a mix
of mutex and spinlock to protect what we needed to protect. Now we
only have a spin lock and the locking complexity is greatly reduced.
Signed-off-by: Doug Ledford <dledford@redhat.com>
Allow the ipoib layer to attempt to join all outstanding multicast
groups at once. The ib_sa layer will serialize multiple attempts to
join the same group, but will process attempts to join different groups
in parallel. Take advantage of that.
In order to make this happen, change the mcast_join_thread to loop
through all needed joins, sending a join request for each one that we
still need to join. There are a few special cases we handle though:
1) Don't attempt to join anything but the broadcast group until the join
of the broadcast group has succeeded.
2) No longer restart the join task at the end of completion handling.
If we completed successfully, we are done. The join task now needs kicked
either by mcast_send or mcast_restart_task or mcast_start_thread, but
should not need started anytime else except when scheduling a backoff
attempt to rejoin.
3) No longer use separate join/completion routines for regular and
sendonly joins, pass them all through the same routine and just do the
right thing based on the SENDONLY join flag.
4) Only try to join a SENDONLY join twice, then drop the packets and
quit trying. We leave the mcast group in the list so that if we get a
new packet, all that we have to do is queue up the packet and restart
the join task and it will automatically try to join twice and then
either send or flush the queue again.
Signed-off-by: Doug Ledford <dledford@redhat.com>
Commit a9c8ba5884 ("IPoIB: Fix usage of uninitialized multicast
objects") added a new flag MCAST_JOIN_STARTED, but was not very strict
in how it was used. We didn't always initialize the completion struct
before we set the flag, and we didn't always call complete on the
completion struct from all paths that complete it. And when we did
complete it, sometimes we continued to touch the mcast entry after
the completion, opening us up to possible use after free issues.
This made it less than totally effective, and certainly made its use
confusing. And in the flush function we would use the presence of this
flag to signal that we should wait on the completion struct, but we never
cleared this flag, ever.
In order to make things clearer and aid in resolving the rtnl deadlock
bug I've been chasing, I cleaned this up a bit.
1) Remove the MCAST_JOIN_STARTED flag entirely
2) Change MCAST_FLAG_BUSY so it now only means a join is in-flight
3) Test mcast->mc directly to see if we have completed
ib_sa_join_multicast (using IS_ERR_OR_NULL)
4) Make sure that before setting MCAST_FLAG_BUSY we always initialize
the mcast->done completion struct
5) Make sure that before calling complete(&mcast->done), we always clear
the MCAST_FLAG_BUSY bit
6) Take the mcast_mutex before we call ib_sa_multicast_join and also
take the mutex in our join callback. This forces
ib_sa_multicast_join to return and set mcast->mc before we process
the callback. This way, our callback can safely clear mcast->mc
if there is an error on the join and we will do the right thing as
a result in mcast_dev_flush.
7) Because we need the mutex to synchronize mcast->mc, we can no
longer call mcast_sendonly_join directly from mcast_send and
instead must add sendonly join processing to the mcast_join_task
8) Make MCAST_RUN mean that we have a working mcast subsystem, not that
we have a running task. We know when we need to reschedule our
join task thread and don't need a flag to tell us.
9) Add a helper for rescheduling the join task thread
A number of different races are resolved with these changes. These
races existed with the old MCAST_FLAG_BUSY usage, the
MCAST_JOIN_STARTED flag was an attempt to address them, and while it
helped, a determined effort could still trip things up.
One race looks something like this:
Thread 1 Thread 2
ib_sa_join_multicast (as part of running restart mcast task)
alloc member
call callback
ifconfig ib0 down
wait_for_completion
callback call completes
wait_for_completion in
mcast_dev_flush completes
mcast->mc is PTR_ERR_OR_NULL
so we skip ib_sa_leave_multicast
return from callback
return from ib_sa_join_multicast
set mcast->mc = return from ib_sa_multicast
We now have a permanently unbalanced join/leave issue that trips up the
refcounting in core/multicast.c
Another like this:
Thread 1 Thread 2 Thread 3
ib_sa_multicast_join
ifconfig ib0 down
priv->broadcast = NULL
join_complete
wait_for_completion
mcast->mc is not yet set, so don't clear
return from ib_sa_join_multicast and set mcast->mc
complete
return -EAGAIN (making mcast->mc invalid)
call ib_sa_multicast_leave
on invalid mcast->mc, hang
forever
By holding the mutex around ib_sa_multicast_join and taking the mutex
early in the callback, we force mcast->mc to be valid at the time we
run the callback. This allows us to clear mcast->mc if there is an
error and the join is going to fail. We do this before we complete
the mcast. In this way, mcast_dev_flush always sees consistent state
in regards to mcast->mc membership at the time that the
wait_for_completion() returns.
Signed-off-by: Doug Ledford <dledford@redhat.com>
Various places in the IPoIB code had a deadlock related to flushing
the ipoib workqueue. Now that we have per device workqueues and a
specific flush workqueue, there is no longer a deadlock issue with
flushing the device specific workqueues and we can do so unilaterally.
Signed-off-by: Doug Ledford <dledford@redhat.com>
During my recent work on the rtnl lock deadlock in the IPoIB driver, I
saw that even once I fixed the apparent races for a single device, as
soon as that device had any children, new races popped up. It turns
out that this is because no matter how well we protect against races
on a single device, the fact that all devices use the same workqueue,
and flush_workqueue() flushes *everything* from that workqueue means
that we would also have to prevent all races between different devices
(for instance, ipoib_mcast_restart_task on interface ib0 can race with
ipoib_mcast_flush_dev on interface ib0.8002, resulting in a deadlock on
the rtnl_lock).
There are several possible solutions to this problem:
Make carrier_on_task and mcast_restart_task try to take the rtnl for
some set period of time and if they fail, then bail. This runs the
real risk of dropping work on the floor, which can end up being its
own separate kind of deadlock.
Set some global flag in the driver that says some device is in the
middle of going down, letting all tasks know to bail. Again, this can
drop work on the floor.
Or the method this patch attempts to use, which is when we bring an
interface up, create a workqueue specifically for that interface, so
that when we take it back down, we are flushing only those tasks
associated with our interface. In addition, keep the global
workqueue, but now limit it to only flush tasks. In this way, the
flush tasks can always flush the device specific work queues without
having deadlock issues.
Signed-off-by: Doug Ledford <dledford@redhat.com>
