The tegra_smmu_group_get was added to group devices in different
SWGROUPs and it'd return a NULL group pointer upon a mismatch at
tegra_smmu_find_group(), so for most of clients/devices, it very
likely would mismatch and need a fallback generic_device_group().
But now tegra_smmu_group_get handles devices in same SWGROUP too,
which means that it would allocate a group for every new SWGROUP
or would directly return an existing one upon matching a SWGROUP,
i.e. any device will go through this function.
So possibility of having a NULL group pointer in device_group()
is upon failure of either devm_kzalloc() or iommu_group_alloc().
In either case, calling generic_device_group() no longer makes a
sense. Especially for devm_kzalloc() failing case, it'd cause a
problem if it fails at devm_kzalloc() yet succeeds at a fallback
generic_device_group(), because it does not create a group->list
for other devices to match.
This patch simply unwraps the function to clean it up.
Signed-off-by: Nicolin Chen <nicoleotsuka@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reviewed-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20201125101013.14953-2-nicoleotsuka@gmail.com
Signed-off-by: Will Deacon <will@kernel.org>
Robin Murphy pointed out that if the arm-smmu driver probes before
the qcom_scm driver, we may call qcom_scm_qsmmu500_wait_safe_toggle()
before the __scm is initialized.
Now, getting this to happen is a bit contrived, as in my efforts it
required enabling asynchronous probing for both drivers, moving the
firmware dts node to the end of the dtsi file, as well as forcing a
long delay in the qcom_scm_probe function.
With those tweaks we ran into the following crash:
[ 2.631040] arm-smmu 15000000.iommu: Stage-1: 48-bit VA -> 48-bit IPA
[ 2.633372] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000
...
[ 2.633402] [0000000000000000] user address but active_mm is swapper
[ 2.633409] Internal error: Oops: 96000005 [#1] PREEMPT SMP
[ 2.633415] Modules linked in:
[ 2.633427] CPU: 5 PID: 117 Comm: kworker/u16:2 Tainted: G W 5.10.0-rc1-mainline-00025-g272a618fc36-dirty #3971
[ 2.633430] Hardware name: Thundercomm Dragonboard 845c (DT)
[ 2.633448] Workqueue: events_unbound async_run_entry_fn
[ 2.633456] pstate: 80c00005 (Nzcv daif +PAN +UAO -TCO BTYPE=--)
[ 2.633465] pc : qcom_scm_qsmmu500_wait_safe_toggle+0x78/0xb0
[ 2.633473] lr : qcom_smmu500_reset+0x58/0x78
[ 2.633476] sp : ffffffc0105a3b60
...
[ 2.633567] Call trace:
[ 2.633572] qcom_scm_qsmmu500_wait_safe_toggle+0x78/0xb0
[ 2.633576] qcom_smmu500_reset+0x58/0x78
[ 2.633581] arm_smmu_device_reset+0x194/0x270
[ 2.633585] arm_smmu_device_probe+0xc94/0xeb8
[ 2.633592] platform_drv_probe+0x58/0xa8
[ 2.633597] really_probe+0xec/0x398
[ 2.633601] driver_probe_device+0x5c/0xb8
[ 2.633606] __driver_attach_async_helper+0x64/0x88
[ 2.633610] async_run_entry_fn+0x4c/0x118
[ 2.633617] process_one_work+0x20c/0x4b0
[ 2.633621] worker_thread+0x48/0x460
[ 2.633628] kthread+0x14c/0x158
[ 2.633634] ret_from_fork+0x10/0x18
[ 2.633642] Code: a9034fa0 d0007f73 29107fa0 91342273 (f9400020)
To avoid this, this patch adds a check on qcom_scm_is_available() in
the qcom_smmu_impl_init() function, returning -EPROBE_DEFER if its
not ready.
This allows the driver to try to probe again later after qcom_scm has
finished probing.
Reported-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
Reviewed-by: Robin Murphy <robin.murphy@arm.com>
Cc: Robin Murphy <robin.murphy@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Andy Gross <agross@kernel.org>
Cc: Maulik Shah <mkshah@codeaurora.org>
Cc: Bjorn Andersson <bjorn.andersson@linaro.org>
Cc: Saravana Kannan <saravanak@google.com>
Cc: Marc Zyngier <maz@kernel.org>
Cc: Lina Iyer <ilina@codeaurora.org>
Cc: iommu@lists.linux-foundation.org
Cc: linux-arm-msm <linux-arm-msm@vger.kernel.org>
Link: https://lore.kernel.org/r/20201112220520.48159-1-john.stultz@linaro.org
Signed-off-by: Will Deacon <will@kernel.org>
The sva_bind() function allows devices to access process address spaces
using a PASID (aka SSID).
(1) bind() allocates or gets an existing MMU notifier tied to the
(domain, mm) pair. Each mm gets one PASID.
(2) Any change to the address space calls invalidate_range() which sends
ATC invalidations (in a subsequent patch).
(3) When the process address space dies, the release() notifier disables
the CD to allow reclaiming the page tables. Since release() has to
be light we do not instruct device drivers to stop DMA here, we just
ignore incoming page faults from this point onwards.
To avoid any event 0x0a print (C_BAD_CD) we disable translation
without clearing CD.V. PCIe Translation Requests and Page Requests
are silently denied. Don't clear the R bit because the S bit can't
be cleared when STALL_MODEL==0b10 (forced), and clearing R without
clearing S is useless. Faulting transactions will stall and will be
aborted by the IOPF handler.
(4) After stopping DMA, the device driver releases the bond by calling
unbind(). We release the MMU notifier, free the PASID and the bond.
Three structures keep track of bonds:
* arm_smmu_bond: one per {device, mm} pair, the handle returned to the
device driver for a bind() request.
* arm_smmu_mmu_notifier: one per {domain, mm} pair, deals with ATS/TLB
invalidations and clearing the context descriptor on mm exit.
* arm_smmu_ctx_desc: one per mm, holds the pinned ASID and pgd.
Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Link: https://lore.kernel.org/r/20201106155048.997886-4-jean-philippe@linaro.org
Signed-off-by: Will Deacon <will@kernel.org>
AMD IOMMU requires 4k-aligned pages for the event log, the PPR log,
and the completion wait write-back regions. However, when allocating
the pages, they could be part of large mapping (e.g. 2M) page.
This causes #PF due to the SNP RMP hardware enforces the check based
on the page level for these data structures.
So, fix by calling set_memory_4k() on the allocated pages.
Fixes: c69d89aff3 ("iommu/amd: Use 4K page for completion wait write-back semaphore")
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Cc: Brijesh Singh <brijesh.singh@amd.com>
Link: https://lore.kernel.org/r/20201105145832.3065-1-suravee.suthikulpanit@amd.com
Signed-off-by: Will Deacon <will@kernel.org>
Merge swiotlb updates from Konrad, as we depend on the updated function
prototype for swiotlb_tbl_map_single(), which dropped the 'tbl_dma_addr'
argument in -rc4.
* 'stable/for-linus-5.10-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/swiotlb:
swiotlb: remove the tbl_dma_addr argument to swiotlb_tbl_map_single
swiotlb: fix "x86: Don't panic if can not alloc buffer for swiotlb"
Pull iommu fixes from Will Deacon:
"Two straightforward vt-d fixes:
- Fix boot when intel iommu initialisation fails under TXT (tboot)
- Fix intel iommu compilation error when DMAR is enabled without ATS
and temporarily update IOMMU MAINTAINERs entry"
* tag 'iommu-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
MAINTAINERS: Temporarily add myself to the IOMMU entry
iommu/vt-d: Fix compile error with CONFIG_PCI_ATS not set
iommu/vt-d: Avoid panic if iommu init fails in tboot system
Pull in x86 fixes from Thomas, as they include a change to the Intel DMAR
code on which we depend:
* tag 'x86-urgent-2020-11-15' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
iommu/vt-d: Cure VF irqdomain hickup
x86/platform/uv: Fix copied UV5 output archtype
x86/platform/uv: Drop last traces of uv_flush_tlb_others
The AMD IOMMU has two modes for generating its own interrupts.
The first is very much based on PCI MSI, and can be configured by Linux
precisely that way. But like legacy unmapped PCI MSI it's limited to
8 bits of APIC ID.
The second method does not use PCI MSI at all in hardawre, and instead
configures the INTCAPXT registers in the IOMMU directly with the APIC ID
and vector.
In the latter case, the IOMMU driver would still use pci_enable_msi(),
read back (through MMIO) the MSI message that Linux wrote to the PCI MSI
table, then swizzle those bits into the appropriate register.
Historically, this worked because__irq_compose_msi_msg() would silently
generate an invalid MSI message with the high bits of the APIC ID in the
high bits of the MSI address. That hack was intended only for the Intel
IOMMU, and I recently enforced that, introducing a warning in
__irq_msi_compose_msg() if it was invoked with an APIC ID above 255.
Fix the AMD IOMMU not to depend on that hack any more, by having its own
irqdomain and directly putting the bits from the irq_cfg into the right
place in its ->activate() method.
Fixes: 47bea873cf "x86/msi: Only use high bits of MSI address for DMAR unit")
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Link: https://lore.kernel.org/r/05e3a5ba317f5ff48d2f8356f19e617f8b9d23a4.camel@infradead.org
"intel_iommu=off" command line is used to disable iommu but iommu is force
enabled in a tboot system for security reason.
However for better performance on high speed network device, a new option
"intel_iommu=tboot_noforce" is introduced to disable the force on.
By default kernel should panic if iommu init fail in tboot for security
reason, but it's unnecessory if we use "intel_iommu=tboot_noforce,off".
Fix the code setting force_on and move intel_iommu_tboot_noforce
from tboot code to intel iommu code.
Fixes: 7304e8f28b ("iommu/vt-d: Correctly disable Intel IOMMU force on")
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@gmail.com>
Tested-by: Lukasz Hawrylko <lukasz.hawrylko@linux.intel.com>
Acked-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20201110071908.3133-1-zhenzhong.duan@gmail.com
Signed-off-by: Will Deacon <will@kernel.org>
When ever an iova alloc request fails we free the iova
ranges present in the percpu iova rcaches and then retry
but the global iova rcache is not freed as a result we could
still see iova alloc failure even after retry as global
rcache is holding the iova's which can cause fragmentation.
So, free the global iova rcache as well and then go for the
retry.
Signed-off-by: Vijayanand Jitta <vjitta@codeaurora.org>
Reviewed-by: Robin Murphy <robin.murphy@arm.com>
Acked-by: John Garry <john.garry@huaqwei.com>
Link: https://lore.kernel.org/r/1601451864-5956-2-git-send-email-vjitta@codeaurora.org
Signed-off-by: Will Deacon <will@kernel.org>
When ever a new iova alloc request comes iova is always searched
from the cached node and the nodes which are previous to cached
node. So, even if there is free iova space available in the nodes
which are next to the cached node iova allocation can still fail
because of this approach.
Consider the following sequence of iova alloc and frees on
1GB of iova space
1) alloc - 500MB
2) alloc - 12MB
3) alloc - 499MB
4) free - 12MB which was allocated in step 2
5) alloc - 13MB
After the above sequence we will have 12MB of free iova space and
cached node will be pointing to the iova pfn of last alloc of 13MB
which will be the lowest iova pfn of that iova space. Now if we get an
alloc request of 2MB we just search from cached node and then look
for lower iova pfn's for free iova and as they aren't any, iova alloc
fails though there is 12MB of free iova space.
To avoid such iova search failures do a retry from the last rb tree node
when iova search fails, this will search the entire tree and get an iova
if its available.
Signed-off-by: Vijayanand Jitta <vjitta@codeaurora.org>
Reviewed-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/1601451864-5956-1-git-send-email-vjitta@codeaurora.org
Signed-off-by: Will Deacon <will@kernel.org>
Commit 6ee1b77ba3 ("iommu/vt-d: Add svm/sva invalidate function")
introduced intel_iommu_sva_invalidate() when CONFIG_INTEL_IOMMU_SVM.
This function uses the dedicated static variable inv_type_granu_table
and functions to_vtd_granularity() and to_vtd_size().
These parts are unused when !CONFIG_INTEL_IOMMU_SVM, and hence,
make CC=clang W=1 warns with an -Wunused-function warning.
Include these parts conditionally on CONFIG_INTEL_IOMMU_SVM.
Fixes: 6ee1b77ba3 ("iommu/vt-d: Add svm/sva invalidate function")
Signed-off-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Acked-by: Lu Baolu <baolu.lu@linux.intel.com>
Link: https://lore.kernel.org/r/20201115205951.20698-1-lukas.bulwahn@gmail.com
Signed-off-by: Will Deacon <will@kernel.org>
Pull x86 fixes from Thomas Gleixner:
"A small set of fixes for x86:
- Cure the fallout from the MSI irqdomain overhaul which missed that
the Intel IOMMU does not register virtual function devices and
therefore never reaches the point where the MSI interrupt domain is
assigned. This made the VF devices use the non-remapped MSI domain
which is trapped by the IOMMU/remap unit
- Remove an extra space in the SGI_UV architecture type procfs output
for UV5
- Remove a unused function which was missed when removing the UV BAU
TLB shootdown handler"
* tag 'x86-urgent-2020-11-15' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
iommu/vt-d: Cure VF irqdomain hickup
x86/platform/uv: Fix copied UV5 output archtype
x86/platform/uv: Drop last traces of uv_flush_tlb_others
The recent changes to store the MSI irqdomain pointer in struct device
missed that Intel DMAR does not register virtual function devices. Due to
that a VF device gets the plain PCI-MSI domain assigned and then issues
compat MSI messages which get caught by the interrupt remapping unit.
Cure that by inheriting the irq domain from the physical function
device.
Ideally the irqdomain would be associated to the bus, but DMAR can have
multiple units and therefore irqdomains on a single bus. The VF 'bus' could
of course inherit the domain from the PF, but that'd be yet another x86
oddity.
Fixes: 85a8dfc57a ("iommm/vt-d: Store irq domain in struct device")
Reported-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Lu Baolu <baolu.lu@linux.intel.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Bjorn Helgaas <bhelgaas@google.com>
Cc: Marc Zyngier <maz@kernel.org>
Cc: David Woodhouse <dwmw2@infradead.org>
Link: https://lore.kernel.org/r/draft-87eekymlpz.fsf@nanos.tec.linutronix.de
Pull swiotlb fixes from Konrad Rzeszutek Wilk:
"Two tiny fixes for issues that make drivers under Xen unhappy under
certain conditions"
* 'stable/for-linus-5.10-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/swiotlb:
swiotlb: remove the tbl_dma_addr argument to swiotlb_tbl_map_single
swiotlb: fix "x86: Don't panic if can not alloc buffer for swiotlb"
We need commit f8f6ae5d07 ("mm: always have io_remap_pfn_range() set
pgprot_decrypted()") to be able to merge Jason's cleanup patch.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The "data->flags" variable is a u64 so if one of the high 32 bits is
set the original code will allow it, but it should be rejected. The
fix is to declare "mask" as a u64 instead of a u32.
Fixes: d90573812e ("iommu/uapi: Handle data and argsz filled by users")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Link: https://lore.kernel.org/r/20201103101623.GA1127762@mwanda
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Certain device drivers allocate IO queues on a per-cpu basis.
On AMD EPYC platform, which can support up-to 256 cpu threads,
this can exceed the current MAX_IRQ_PER_TABLE limit of 256,
and result in the error message:
AMD-Vi: Failed to allocate IRTE
This has been observed with certain NVME devices.
AMD IOMMU hardware can actually support upto 512 interrupt
remapping table entries. Therefore, update the driver to
match the hardware limit.
Please note that this also increases the size of interrupt remapping
table to 8KB per device when using the 128-bit IRTE format.
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Link: https://lore.kernel.org/r/20201015025002.87997-1-suravee.suthikulpanit@amd.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The tbl_dma_addr argument is used to check the DMA boundary for the
allocations, and thus needs to be a dma_addr_t. swiotlb-xen instead
passed a physical address, which could lead to incorrect results for
strange offsets. Fix this by removing the parameter entirely and hard
code the DMA address for io_tlb_start instead.
Fixes: 91ffe4ad53 ("swiotlb-xen: introduce phys_to_dma/dma_to_phys translations")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Stefano Stabellini <sstabellini@kernel.org>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
The firmware found in some Qualcomm platforms intercepts writes to S2CR
in order to replace bypass type streams with fault; and ignore S2CR
updates of type fault.
Detect this behavior and implement a custom write_s2cr function in order
to trick the firmware into supporting bypass streams by the means of
configuring the stream for translation using a reserved and disabled
context bank.
Also circumvent the problem of configuring faulting streams by
configuring the stream as bypass.
Cc: <stable@vger.kernel.org>
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Tested-by: Steev Klimaszewski <steev@kali.org>
Acked-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/20201019182323.3162386-4-bjorn.andersson@linaro.org
Signed-off-by: Will Deacon <will@kernel.org>
The I/O-APIC generates an MSI cycle with address/data bits taken from its
Redirection Table Entry in some combination which used to make sense, but
now is just a bunch of bits which get passed through in some seemingly
arbitrary order.
Instead of making IRQ remapping drivers directly frob the I/OA-PIC RTE, let
them just do their job and generate an MSI message. The bit swizzling to
turn that MSI message into the I/O-APIC's RTE is the same in all cases,
since it's a function of the I/O-APIC hardware. The IRQ remappers have no
real need to get involved with that.
The only slight caveat is that the I/OAPIC is interpreting some of those
fields too, and it does want the 'vector' field to be unique to make EOI
work. The AMD IOMMU happens to put its IRTE index in the bits that the
I/O-APIC thinks are the vector field, and accommodates this requirement by
reserving the first 32 indices for the I/O-APIC. The Intel IOMMU doesn't
actually use the bits that the I/O-APIC thinks are the vector field, so it
fills in the 'pin' value there instead.
[ tglx: Replaced the unreadably macro maze with the cleaned up RTE/msi_msg
bitfields and added commentry to explain the mapping magic ]
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/20201024213535.443185-22-dwmw2@infradead.org
Having two seperate structs for the I/O-APIC RTE entries (non-remapped and
DMAR remapped) requires type casts and makes it hard to map.
Combine them in IO_APIC_routing_entry by defining a union of two 64bit
bitfields. Use naming which reflects which bits are shared and which bits
are actually different for the operating modes.
[dwmw2: Fix it up and finish the job, pulling the 32-bit w1,w2 words for
register access into the same union and eliminating a few more
places where bits were accessed through masks and shifts.]
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/20201024213535.443185-21-dwmw2@infradead.org
'trigger' and 'polarity' are used throughout the I/O-APIC code for handling
the trigger type (edge/level) and the active low/high configuration. While
there are defines for initializing these variables and struct members, they
are not used consequently and the meaning of 'trigger' and 'polarity' is
opaque and confusing at best.
Rename them to 'is_level' and 'active_low' and make them boolean in various
structs so it's entirely clear what the meaning is.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/20201024213535.443185-20-dwmw2@infradead.org
