IOMMU harms performance signficantly when we run very fast networking
workloads. It's 40GB networking doing XDP test. Software overhead is
almost unaware, but it's the IOTLB miss (based on our analysis) which
kills the performance. We observed the same performance issue even with
software passthrough (identity mapping), only the hardware passthrough
survives. The pps with iommu (with software passthrough) is only about
~30% of that without it. This is a limitation in hardware based on our
observation, so we'd like to disable the IOMMU force on, but we do want
to use TBOOT and we can sacrifice the DMA security bought by IOMMU. I
must admit I know nothing about TBOOT, but TBOOT guys (cc-ed) think not
eabling IOMMU is totally ok.
So introduce a new boot option to disable the force on. It's kind of
silly we need to run into intel_iommu_init even without force on, but we
need to disable TBOOT PMR registers. For system without the boot option,
nothing is changed.
Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
When booting into a kexec kernel with intel_iommu=off, and
the previous kernel had intel_iommu=on, the IOMMU hardware
is still enabled and gets not disabled by the new kernel.
This causes the boot to fail because DMA is blocked by the
hardware. Disable the IOMMUs when we find it enabled in the
kexec kernel and boot with intel_iommu=off.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The introduction of reserved regions has left a couple of rough edges
which we could do with sorting out sooner rather than later. Since we
are not yet addressing the potential dynamic aspect of software-managed
reservations and presenting them at arbitrary fixed addresses, it is
incongruous that we end up displaying hardware vs. software-managed MSI
regions to userspace differently, especially since ARM-based systems may
actually require one or the other, or even potentially both at once,
(which iommu-dma currently has no hope of dealing with at all). Let's
resolve the former user-visible inconsistency ASAP before the ABI has
been baked into a kernel release, in a way that also lays the groundwork
for the latter shortcoming to be addressed by follow-up patches.
For clarity, rename the software-managed type to IOMMU_RESV_SW_MSI, use
IOMMU_RESV_MSI to describe the hardware type, and document everything a
little bit. Since the x86 MSI remapping hardware falls squarely under
this meaning of IOMMU_RESV_MSI, apply that type to their regions as well,
so that we tell the same story to userspace across all platforms.
Secondly, as the various region types require quite different handling,
and it really makes little sense to ever try combining them, convert the
bitfield-esque #defines to a plain enum in the process before anyone
gets the wrong impression.
Fixes: d30ddcaa7b ("iommu: Add a new type field in iommu_resv_region")
Reviewed-by: Eric Auger <eric.auger@redhat.com>
CC: Alex Williamson <alex.williamson@redhat.com>
CC: David Woodhouse <dwmw2@infradead.org>
CC: kvm@vger.kernel.org
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The link between the iommu sysfs-device and the struct
intel_iommu is no longer stored as driver-data. Update the
code to use the new access method.
Reported-by: Dave Jones <davej@codemonkey.org.uk>
Fixes: 39ab9555c2 ('iommu: Add sysfs bindings for struct iommu_device')
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The callers of the DMA alloc functions already provide the proper
context GFP flags. Make sure to pass them through to the CMA allocator,
to make the CMA compaction context aware.
Link: http://lkml.kernel.org/r/20170127172328.18574-3-l.stach@pengutronix.de
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Radim Krcmar <rkrcmar@redhat.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Alexander Graf <agraf@suse.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There is currently support for iommu sysfs bindings, but
those need to be implemented in the IOMMU drivers. Add a
more generic version of this by adding a struct device to
struct iommu_device and use that for the sysfs bindings.
Also convert the AMD and Intel IOMMU driver to make use of
it.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
This struct represents one hardware iommu in the iommu core
code. For now it only has the iommu-ops associated with it,
but that will be extended soon.
The register/unregister interface is also added, as well as
making use of it in the Intel and AMD IOMMU drivers.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
dma_pte_free_level() recurses down the IOMMU page tables and frees
directory pages that are entirely contained in the given PFN range.
Unfortunately, it incorrectly calculates the starting address covered
by the PTE under consideration, which can lead to it clearing an entry
that is still in use.
This occurs if we have a scatterlist with an entry that has a length
greater than 1026 MB and is aligned to 2 MB for both the IOMMU and
physical addresses. For example, if __domain_mapping() is asked to map a
two-entry scatterlist with 2 MB and 1028 MB segments to PFN 0xffff80000,
it will ask if dma_pte_free_pagetable() is asked to PFNs from
0xffff80200 to 0xffffc05ff, it will also incorrectly clear the PFNs from
0xffff80000 to 0xffff801ff because of this issue. The current code will
set level_pfn to 0xffff80200, and 0xffff80200-0xffffc01ff fits inside
the range being cleared. Properly setting the level_pfn for the current
level under consideration catches that this PTE is outside of the range
being cleared.
This patch also changes the value passed into dma_pte_free_level() when
it recurses. This only affects the first PTE of the range being cleared,
and is handled by the existing code that ensures we start our cursor no
lower than start_pfn.
This was found when using dma_map_sg() to map large chunks of contiguous
memory, which immediatedly led to faults on the first access of the
erroneously-deleted mappings.
Fixes: 3269ee0bd6 ("intel-iommu: Fix leaks in pagetable freeing")
Reviewed-by: Benjamin Serebrin <serebrin@google.com>
Signed-off-by: David Dillow <dillow@google.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The check to set identity map for tylersburg is done too late. It needs
to be done before the check for identity_map domain is done.
To: Joerg Roedel <joro@8bytes.org>
To: David Woodhouse <dwmw2@infradead.org>
Cc: iommu@lists.linux-foundation.org
Cc: linux-kernel@vger.kernel.org
Cc: stable@vger.kernel.org
Cc: Ashok Raj <ashok.raj@intel.com>
Fixes: 86080ccc22 ("iommu/vt-d: Allocate si_domain in init_dmars()")
Signed-off-by: Ashok Raj <ashok.raj@intel.com>
Reported-by: Yunhong Jiang <yunhong.jiang@intel.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
This patch registers the [FEE0_0000h - FEF0_000h] 1MB MSI
range as a reserved region and RMRR regions as direct regions.
This will allow to report those reserved regions in the
iommu-group sysfs.
Signed-off-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Different encodings are used to represent supported PASID bits
and number of PASID table entries.
The current code assigns ecap_pss directly to extended context
table entry PTS which is wrong and could result in writing
non-zero bits to the reserved fields. IOMMU fault reason
11 will be reported when reserved bits are nonzero.
This patch converts ecap_pss to extend context entry pts encoding
based on VT-d spec. Chapter 9.4 as follows:
- number of PASID bits = ecap_pss + 1
- number of PASID table entries = 2^(pts + 5)
Software assigned limit of pasid_max value is also respected to
match the allocation limitation of PASID table.
cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
cc: Ashok Raj <ashok.raj@intel.com>
Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Tested-by: Mika Kuoppala <mika.kuoppala@intel.com>
Fixes: 2f26e0a9c9 ('iommu/vt-d: Add basic SVM PASID support')
Signed-off-by: Joerg Roedel <jroedel@suse.de>
We met the DMAR fault both on hpsa P420i and P421 SmartArray controllers
under kdump, it can be steadily reproduced on several different machines,
the dmesg log is like:
HP HPSA Driver (v 3.4.16-0)
hpsa 0000:02:00.0: using doorbell to reset controller
hpsa 0000:02:00.0: board ready after hard reset.
hpsa 0000:02:00.0: Waiting for controller to respond to no-op
DMAR: Setting identity map for device 0000:02:00.0 [0xe8000 - 0xe8fff]
DMAR: Setting identity map for device 0000:02:00.0 [0xf4000 - 0xf4fff]
DMAR: Setting identity map for device 0000:02:00.0 [0xbdf6e000 - 0xbdf6efff]
DMAR: Setting identity map for device 0000:02:00.0 [0xbdf6f000 - 0xbdf7efff]
DMAR: Setting identity map for device 0000:02:00.0 [0xbdf7f000 - 0xbdf82fff]
DMAR: Setting identity map for device 0000:02:00.0 [0xbdf83000 - 0xbdf84fff]
DMAR: DRHD: handling fault status reg 2
DMAR: [DMA Read] Request device [02:00.0] fault addr fffff000 [fault reason 06] PTE Read access is not set
hpsa 0000:02:00.0: controller message 03:00 timed out
hpsa 0000:02:00.0: no-op failed; re-trying
After some debugging, we found that the fault addr is from DMA initiated at
the driver probe stage after reset(not in-flight DMA), and the corresponding
pte entry value is correct, the fault is likely due to the old iommu caches
of the in-flight DMA before it.
Thus we need to flush the old cache after context mapping is setup for the
device, where the device is supposed to finish reset at its driver probe
stage and no in-flight DMA exists hereafter.
I'm not sure if the hardware is responsible for invalidating all the related
caches allocated in the iommu hardware before, but seems not the case for hpsa,
actually many device drivers have problems in properly resetting the hardware.
Anyway flushing (again) by software in kdump kernel when the device gets context
mapped which is a quite infrequent operation does little harm.
With this patch, the problematic machine can survive the kdump tests.
CC: Myron Stowe <myron.stowe@gmail.com>
CC: Joseph Szczypek <jszczype@redhat.com>
CC: Don Brace <don.brace@microsemi.com>
CC: Baoquan He <bhe@redhat.com>
CC: Dave Young <dyoung@redhat.com>
Fixes: 091d42e43d ("iommu/vt-d: Copy translation tables from old kernel")
Fixes: dbcd861f25 ("iommu/vt-d: Do not re-use domain-ids from the old kernel")
Fixes: cf484d0e69 ("iommu/vt-d: Mark copied context entries")
Signed-off-by: Xunlei Pang <xlpang@redhat.com>
Tested-by: Don Brace <don.brace@microsemi.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Pull smp hotplug updates from Thomas Gleixner:
"This is the final round of converting the notifier mess to the state
machine. The removal of the notifiers and the related infrastructure
will happen around rc1, as there are conversions outstanding in other
trees.
The whole exercise removed about 2000 lines of code in total and in
course of the conversion several dozen bugs got fixed. The new
mechanism allows to test almost every hotplug step standalone, so
usage sites can exercise all transitions extensively.
There is more room for improvement, like integrating all the
pointlessly different architecture mechanisms of synchronizing,
setting cpus online etc into the core code"
* 'smp-hotplug-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (60 commits)
tracing/rb: Init the CPU mask on allocation
soc/fsl/qbman: Convert to hotplug state machine
soc/fsl/qbman: Convert to hotplug state machine
zram: Convert to hotplug state machine
KVM/PPC/Book3S HV: Convert to hotplug state machine
arm64/cpuinfo: Convert to hotplug state machine
arm64/cpuinfo: Make hotplug notifier symmetric
mm/compaction: Convert to hotplug state machine
iommu/vt-d: Convert to hotplug state machine
mm/zswap: Convert pool to hotplug state machine
mm/zswap: Convert dst-mem to hotplug state machine
mm/zsmalloc: Convert to hotplug state machine
mm/vmstat: Convert to hotplug state machine
mm/vmstat: Avoid on each online CPU loops
mm/vmstat: Drop get_online_cpus() from init_cpu_node_state/vmstat_cpu_dead()
tracing/rb: Convert to hotplug state machine
oprofile/nmi timer: Convert to hotplug state machine
net/iucv: Use explicit clean up labels in iucv_init()
x86/pci/amd-bus: Convert to hotplug state machine
x86/oprofile/nmi: Convert to hotplug state machine
...
Pull IOMMU fixes from David Woodhouse:
"Two minor fixes.
The first fixes the assignment of SR-IOV virtual functions to the
correct IOMMU unit, and the second fixes the excessively large (and
physically contiguous) PASID tables used with SVM"
* git://git.infradead.org/intel-iommu:
iommu/vt-d: Fix PASID table allocation
iommu/vt-d: Fix IOMMU lookup for SR-IOV Virtual Functions
It turns out that the disable_dmar_iommu() code-path tried
to get the device_domain_lock recursivly, which will
dead-lock when this code runs on dmar removal. Fix both
code-paths that could lead to the dead-lock.
Fixes: 55d940430a ('iommu/vt-d: Get rid of domain->iommu_lock')
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The VT-d specification (§8.3.3) says:
‘Virtual Functions’ of a ‘Physical Function’ are under the scope
of the same remapping unit as the ‘Physical Function’.
The BIOS is not required to list all the possible VFs in the scope
tables, and arguably *shouldn't* make any attempt to do so, since there
could be a huge number of them.
This has been broken basically for ever — the VF is never going to match
against a specific unit's scope, so it ends up being assigned to the
INCLUDE_ALL IOMMU. Which was always actually correct by coincidence, but
now we're looking at Root-Complex integrated devices with SR-IOV support
it's going to start being wrong.
Fix it to simply use pci_physfn() before doing the lookup for PCI devices.
Cc: stable@vger.kernel.org
Signed-off-by: Sainath Grandhi <sainath.grandhi@intel.com>
Signed-off-by: Ashok Raj <ashok.raj@intel.com>
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
When a domain is allocated through the get_valid_domain_for_dev
path, it will be context-mapped before the RMRR regions are
mapped in the page-table. This opens a short time window
where device-accesses to these regions fail and causing DMAR
faults.
Fix this by mapping the RMRR regions before the domain is
context-mapped.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Split out the search for an already existing domain and the
context mapping of the device to the new domain.
This allows to map possible RMRR regions into the domain
before it is context mapped.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The dma-mapping core and the implementations do not change the DMA
attributes passed by pointer. Thus the pointer can point to const data.
However the attributes do not have to be a bitfield. Instead unsigned
long will do fine:
1. This is just simpler. Both in terms of reading the code and setting
attributes. Instead of initializing local attributes on the stack
and passing pointer to it to dma_set_attr(), just set the bits.
2. It brings safeness and checking for const correctness because the
attributes are passed by value.
Semantic patches for this change (at least most of them):
virtual patch
virtual context
@r@
identifier f, attrs;
@@
f(...,
- struct dma_attrs *attrs
+ unsigned long attrs
, ...)
{
...
}
@@
identifier r.f;
@@
f(...,
- NULL
+ 0
)
and
// Options: --all-includes
virtual patch
virtual context
@r@
identifier f, attrs;
type t;
@@
t f(..., struct dma_attrs *attrs);
@@
identifier r.f;
@@
f(...,
- NULL
+ 0
)
Link: http://lkml.kernel.org/r/1468399300-5399-2-git-send-email-k.kozlowski@samsung.com
Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Acked-by: Vineet Gupta <vgupta@synopsys.com>
Acked-by: Robin Murphy <robin.murphy@arm.com>
Acked-by: Hans-Christian Noren Egtvedt <egtvedt@samfundet.no>
Acked-by: Mark Salter <msalter@redhat.com> [c6x]
Acked-by: Jesper Nilsson <jesper.nilsson@axis.com> [cris]
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch> [drm]
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Acked-by: Joerg Roedel <jroedel@suse.de> [iommu]
Acked-by: Fabien Dessenne <fabien.dessenne@st.com> [bdisp]
Reviewed-by: Marek Szyprowski <m.szyprowski@samsung.com> [vb2-core]
Acked-by: David Vrabel <david.vrabel@citrix.com> [xen]
Acked-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> [xen swiotlb]
Acked-by: Joerg Roedel <jroedel@suse.de> [iommu]
Acked-by: Richard Kuo <rkuo@codeaurora.org> [hexagon]
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org> [m68k]
Acked-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> [s390]
Acked-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Acked-by: Hans-Christian Noren Egtvedt <egtvedt@samfundet.no> [avr32]
Acked-by: Vineet Gupta <vgupta@synopsys.com> [arc]
Acked-by: Robin Murphy <robin.murphy@arm.com> [arm64 and dma-iommu]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In the updates:
* Big endian support and preparation for defered probing for the
Exynos IOMMU driver
* Simplifications in iommu-group id handling
* Support for Mediatek generation one IOMMU hardware
* Conversion of the AMD IOMMU driver to use the generic IOVA
allocator. This driver now also benefits from the recent
scalability improvements in the IOVA code.
* Preparations to use generic DMA mapping code in the Rockchip
IOMMU driver
* Device tree adaption and conversion to use generic page-table
code for the MSM IOMMU driver
* An iova_to_phys optimization in the ARM-SMMU driver to greatly
improve page-table teardown performance with VFIO
* Various other small fixes and conversions
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2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=xiS2
-----END PGP SIGNATURE-----
Merge tag 'iommu-updates-v4.8' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu
Pull IOMMU updates from Joerg Roedel:
- big-endian support and preparation for defered probing for the Exynos
IOMMU driver
- simplifications in iommu-group id handling
- support for Mediatek generation one IOMMU hardware
- conversion of the AMD IOMMU driver to use the generic IOVA allocator.
This driver now also benefits from the recent scalability
improvements in the IOVA code.
- preparations to use generic DMA mapping code in the Rockchip IOMMU
driver
- device tree adaption and conversion to use generic page-table code
for the MSM IOMMU driver
- an iova_to_phys optimization in the ARM-SMMU driver to greatly
improve page-table teardown performance with VFIO
- various other small fixes and conversions
* tag 'iommu-updates-v4.8' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu: (59 commits)
iommu/amd: Initialize dma-ops domains with 3-level page-table
iommu/amd: Update Alias-DTE in update_device_table()
iommu/vt-d: Return error code in domain_context_mapping_one()
iommu/amd: Use container_of to get dma_ops_domain
iommu/amd: Flush iova queue before releasing dma_ops_domain
iommu/amd: Handle IOMMU_DOMAIN_DMA in ops->domain_free call-back
iommu/amd: Use dev_data->domain in get_domain()
iommu/amd: Optimize map_sg and unmap_sg
iommu/amd: Introduce dir2prot() helper
iommu/amd: Implement timeout to flush unmap queues
iommu/amd: Implement flush queue
iommu/amd: Allow NULL pointer parameter for domain_flush_complete()
iommu/amd: Set up data structures for flush queue
iommu/amd: Remove align-parameter from __map_single()
iommu/amd: Remove other remains of old address allocator
iommu/amd: Make use of the generic IOVA allocator
iommu/amd: Remove special mapping code for dma_ops path
iommu/amd: Pass gfp-flags to iommu_map_page()
iommu/amd: Implement apply_dm_region call-back
iommu/amd: Create a list of reserved iova addresses
...
Some of our "for_each_xyz()" macro constructs make gcc unhappy about
lack of braces around if-statements inside or outside the loop, because
the loop construct itself has a "if-then-else" statement inside of it.
The resulting warnings look something like this:
drivers/gpu/drm/i915/i915_debugfs.c: In function ‘i915_dump_lrc’:
drivers/gpu/drm/i915/i915_debugfs.c:2103:6: warning: suggest explicit braces to avoid ambiguous ‘else’ [-Wparentheses]
if (ctx != dev_priv->kernel_context)
^
even if the code itself is fine.
Since the warning is fairly easy to avoid by adding a braces around the
if-statement near the for_each_xyz() construct, do so, rather than
disabling the otherwise potentially useful warning.
(The if-then-else statements used in the "for_each_xyz()" constructs are
designed to be inherently safe even with no braces, but in this case
it's quite understandable that gcc isn't really able to tell that).
This finally leaves the standard "allmodconfig" build with just a
handful of remaining warnings, so new and valid warnings hopefully will
stand out.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In 'commit <55d940430ab9> ("iommu/vt-d: Get rid of domain->iommu_lock")',
the error handling path is changed a little, which makes the function
always return 0.
This path fixes this.
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Fixes: 55d940430a ('iommu/vt-d: Get rid of domain->iommu_lock')
Cc: stable@vger.kernel.org # v4.3+
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Per VT-d spec Section 10.4.2 ("Capability Register"), the maximum
number of possible domains is 64K; indeed this is the maximum value
that the cap_ndoms() macro will expand to. Since the value 65536
will not fix in a u16, the 'did' variable must be promoted to an
int, otherwise the test for < 65536 will always be true and the
loop will never end.
The symptom, in my case, was a hung machine during suspend.
Fixes: 3bd4f9112f ("iommu/vt-d: Fix overflow of iommu->domains array")
Signed-off-by: Aaron Campbell <aaron@monkey.org>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The valid range of 'did' in get_iommu_domain(*iommu, did)
is 0..cap_ndoms(iommu->cap), so don't exceed that
range in free_all_cpu_cached_iovas().
The user-visible impact of the out-of-bounds access is the machine
hanging on suspend-to-ram. It is, in fact, a kernel panic, but due
to already suspended devices, that's often not visible to the user.
Fixes: 22e2f9fa63 ("iommu/vt-d: Use per-cpu IOVA caching")
Signed-off-by: Jan Niehusmann <jan@gondor.com>
Tested-By: Marius Vlad <marius.c.vlad@intel.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
This seems to be required on some X58 chipsets on systems
with more than one IOMMU. QI does not work until it is
enabled on all IOMMUs in the system.
Reported-by: Dheeraj CVR <cvr.dheeraj@gmail.com>
Tested-by: Dheeraj CVR <cvr.dheeraj@gmail.com>
Fixes: 5f0a7f7614 ('iommu/vt-d: Make root entry visible for hardware right after allocation')
Cc: stable@vger.kernel.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
In commit <8bf478163e69> ("iommu/vt-d: Split up iommu->domains array"), it
it splits iommu->domains in two levels. Each first level contains 256
entries of second level. In case of the ndomains is exact a multiple of
256, it would have one more extra first level entry for current
implementation.
This patch refines this calculation to reduce the extra first level entry.
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Pull intel IOMMU updates from David Woodhouse:
"This patchset improves the scalability of the Intel IOMMU code by
resolving two spinlock bottlenecks and eliminating the linearity of
the IOVA allocator, yielding up to ~5x performance improvement and
approaching 'iommu=off' performance"
* git://git.infradead.org/intel-iommu:
iommu/vt-d: Use per-cpu IOVA caching
iommu/iova: introduce per-cpu caching to iova allocation
iommu/vt-d: change intel-iommu to use IOVA frame numbers
iommu/vt-d: avoid dev iotlb logic for domains with no dev iotlbs
iommu/vt-d: only unmap mapped entries
iommu/vt-d: correct flush_unmaps pfn usage
iommu/vt-d: per-cpu deferred invalidation queues
iommu/vt-d: refactoring of deferred flush entries
Commit 9257b4a2 ('iommu/iova: introduce per-cpu caching to iova allocation')
introduced per-CPU IOVA caches to massively improve scalability. Use them.
Signed-off-by: Omer Peleg <omer@cs.technion.ac.il>
[mad@cs.technion.ac.il: rebased, cleaned up and reworded the commit message]
Signed-off-by: Adam Morrison <mad@cs.technion.ac.il>
Reviewed-by: Shaohua Li <shli@fb.com>
Reviewed-by: Ben Serebrin <serebrin@google.com>
[dwmw2: split out VT-d part into a separate patch]
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Make intel-iommu map/unmap/invalidate work with IOVA pfns instead of
pointers to "struct iova". This avoids using the iova struct from the IOVA
red-black tree and the resulting explicit find_iova() on unmap.
This patch will allow us to cache IOVAs in the next patch, in order to
avoid rbtree operations for the majority of map/unmap operations.
Note: In eliminating the find_iova() operation, we have also eliminated
the sanity check previously done in the unmap flow. Arguably, this was
overhead that is better avoided in production code, but it could be
brought back as a debug option for driver development.
Signed-off-by: Omer Peleg <omer@cs.technion.ac.il>
[mad@cs.technion.ac.il: rebased, fixed to not break iova api, and reworded
the commit message]
Signed-off-by: Adam Morrison <mad@cs.technion.ac.il>
Reviewed-by: Shaohua Li <shli@fb.com>
Reviewed-by: Ben Serebrin <serebrin@google.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
This patch avoids taking the device_domain_lock in iommu_flush_dev_iotlb()
for domains with no dev iotlb devices.
Signed-off-by: Omer Peleg <omer@cs.technion.ac.il>
[gvdl@google.com: fixed locking issues]
Signed-off-by: Godfrey van der Linden <gvdl@google.com>
[mad@cs.technion.ac.il: rebased and reworded the commit message]
Signed-off-by: Adam Morrison <mad@cs.technion.ac.il>
Reviewed-by: Shaohua Li <shli@fb.com>
Reviewed-by: Ben Serebrin <serebrin@google.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Current unmap implementation unmaps the entire area covered by the IOVA
range, which is a power-of-2 aligned region. The corresponding map,
however, only maps those pages originally mapped by the user. This
discrepancy can lead to unmapping of already unmapped entries, which is
unneeded work.
With this patch, only mapped pages are unmapped. This is also a baseline
for a map/unmap implementation based on IOVAs and not iova structures,
which will allow caching.
Signed-off-by: Omer Peleg <omer@cs.technion.ac.il>
[mad@cs.technion.ac.il: rebased and reworded the commit message]
Signed-off-by: Adam Morrison <mad@cs.technion.ac.il>
Reviewed-by: Shaohua Li <shli@fb.com>
Reviewed-by: Ben Serebrin <serebrin@google.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Change flush_unmaps() to correctly pass iommu_flush_iotlb_psi()
dma addresses. (x86_64 mm and dma have the same size for pages
at the moment, but this usage improves consistency.)
Signed-off-by: Omer Peleg <omer@cs.technion.ac.il>
[mad@cs.technion.ac.il: rebased and reworded the commit message]
Signed-off-by: Adam Morrison <mad@cs.technion.ac.il>
Reviewed-by: Shaohua Li <shli@fb.com>
Reviewed-by: Ben Serebrin <serebrin@google.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
The IOMMU's IOTLB invalidation is a costly process. When iommu mode
is not set to "strict", it is done asynchronously. Current code
amortizes the cost of invalidating IOTLB entries by batching all the
invalidations in the system and performing a single global invalidation
instead. The code queues pending invalidations in a global queue that
is accessed under the global "async_umap_flush_lock" spinlock, which
can result is significant spinlock contention.
This patch splits this deferred queue into multiple per-cpu deferred
queues, and thus gets rid of the "async_umap_flush_lock" and its
contention. To keep existing deferred invalidation behavior, it still
invalidates the pending invalidations of all CPUs whenever a CPU
reaches its watermark or a timeout occurs.
Signed-off-by: Omer Peleg <omer@cs.technion.ac.il>
[mad@cs.technion.ac.il: rebased, cleaned up and reworded the commit message]
Signed-off-by: Adam Morrison <mad@cs.technion.ac.il>
Reviewed-by: Shaohua Li <shli@fb.com>
Reviewed-by: Ben Serebrin <serebrin@google.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Currently, deferred flushes' info is striped between several lists in
the flush tables. Instead, move all information about a specific flush
to a single entry in this table.
This patch does not introduce any functional change.
Signed-off-by: Omer Peleg <omer@cs.technion.ac.il>
[mad@cs.technion.ac.il: rebased and reworded the commit message]
Signed-off-by: Adam Morrison <mad@cs.technion.ac.il>
Reviewed-by: Shaohua Li <shli@fb.com>
Reviewed-by: Ben Serebrin <serebrin@google.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
My static checker complains that "dma_alias" is uninitialized unless we
are dealing with a pci device. This is true but harmless. Anyway, we
can flip the condition around to silence the warning.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
dma_pte_free_pagetable no longer depends on last level ptes
being clear, it clears them itself. Fix up the comment to
match.
Cc: Jiang Liu <jiang.liu@linux.intel.com>
Suggested-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
In the PCI hotplug path of the Intel IOMMU driver, replace
the usage of the BUS_NOTIFY_DEL_DEVICE notifier, which is
executed before the driver is unbound from the device, with
BUS_NOTIFY_REMOVED_DEVICE, which runs after that.
This fixes a kernel BUG being triggered in the VT-d code
when the device driver tries to unmap DMA buffers and the
VT-d driver already destroyed all mappings.
Reported-by: Stefani Seibold <stefani@seibold.net>
Cc: stable@vger.kernel.org # v4.3+
Signed-off-by: Joerg Roedel <jroedel@suse.de>
commit db0fa0cb01 "scatterlist: use sg_phys()" did replacements of
the form:
phys_addr_t phys = page_to_phys(sg_page(s));
phys_addr_t phys = sg_phys(s) & PAGE_MASK;
However, this breaks platforms where sizeof(phys_addr_t) >
sizeof(unsigned long). Revert for 4.3 and 4.4 to make room for a
combined helper in 4.5.
Cc: <stable@vger.kernel.org>
Cc: Jens Axboe <axboe@fb.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Fixes: db0fa0cb01 ("scatterlist: use sg_phys()")
Suggested-by: Joerg Roedel <joro@8bytes.org>
Reported-by: Vitaly Lavrov <vel21ripn@gmail.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
__GFP_WAIT has been used to identify atomic context in callers that hold
spinlocks or are in interrupts. They are expected to be high priority and
have access one of two watermarks lower than "min" which can be referred
to as the "atomic reserve". __GFP_HIGH users get access to the first
lower watermark and can be called the "high priority reserve".
Over time, callers had a requirement to not block when fallback options
were available. Some have abused __GFP_WAIT leading to a situation where
an optimisitic allocation with a fallback option can access atomic
reserves.
This patch uses __GFP_ATOMIC to identify callers that are truely atomic,
cannot sleep and have no alternative. High priority users continue to use
__GFP_HIGH. __GFP_DIRECT_RECLAIM identifies callers that can sleep and
are willing to enter direct reclaim. __GFP_KSWAPD_RECLAIM to identify
callers that want to wake kswapd for background reclaim. __GFP_WAIT is
redefined as a caller that is willing to enter direct reclaim and wake
kswapd for background reclaim.
This patch then converts a number of sites
o __GFP_ATOMIC is used by callers that are high priority and have memory
pools for those requests. GFP_ATOMIC uses this flag.
o Callers that have a limited mempool to guarantee forward progress clear
__GFP_DIRECT_RECLAIM but keep __GFP_KSWAPD_RECLAIM. bio allocations fall
into this category where kswapd will still be woken but atomic reserves
are not used as there is a one-entry mempool to guarantee progress.
o Callers that are checking if they are non-blocking should use the
helper gfpflags_allow_blocking() where possible. This is because
checking for __GFP_WAIT as was done historically now can trigger false
positives. Some exceptions like dm-crypt.c exist where the code intent
is clearer if __GFP_DIRECT_RECLAIM is used instead of the helper due to
flag manipulations.
o Callers that built their own GFP flags instead of starting with GFP_KERNEL
and friends now also need to specify __GFP_KSWAPD_RECLAIM.
The first key hazard to watch out for is callers that removed __GFP_WAIT
and was depending on access to atomic reserves for inconspicuous reasons.
In some cases it may be appropriate for them to use __GFP_HIGH.
The second key hazard is callers that assembled their own combination of
GFP flags instead of starting with something like GFP_KERNEL. They may
now wish to specify __GFP_KSWAPD_RECLAIM. It's almost certainly harmless
if it's missed in most cases as other activity will wake kswapd.
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Vitaly Wool <vitalywool@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This time including:
* A new IOMMU driver for s390 pci devices
* Common dma-ops support based on iommu-api for ARM64. The plan is to
use this as a basis for ARM32 and hopefully other architectures as
well in the future.
* MSI support for ARM-SMMUv3
* Cleanups and dead code removal in the AMD IOMMU driver
* Better RMRR handling for the Intel VT-d driver
* Various other cleanups and small fixes
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)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=jLlB
-----END PGP SIGNATURE-----
Merge tag 'iommu-updates-v4.4' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu
Pull iommu updates from Joerg Roedel:
"This time including:
- A new IOMMU driver for s390 pci devices
- Common dma-ops support based on iommu-api for ARM64. The plan is
to use this as a basis for ARM32 and hopefully other architectures
as well in the future.
- MSI support for ARM-SMMUv3
- Cleanups and dead code removal in the AMD IOMMU driver
- Better RMRR handling for the Intel VT-d driver
- Various other cleanups and small fixes"
* tag 'iommu-updates-v4.4' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu: (41 commits)
iommu/vt-d: Fix return value check of parse_ioapics_under_ir()
iommu/vt-d: Propagate error-value from ir_parse_ioapic_hpet_scope()
iommu/vt-d: Adjust the return value of the parse_ioapics_under_ir
iommu: Move default domain allocation to iommu_group_get_for_dev()
iommu: Remove is_pci_dev() fall-back from iommu_group_get_for_dev
iommu/arm-smmu: Switch to device_group call-back
iommu/fsl: Convert to device_group call-back
iommu: Add device_group call-back to x86 iommu drivers
iommu: Add generic_device_group() function
iommu: Export and rename iommu_group_get_for_pci_dev()
iommu: Revive device_group iommu-ops call-back
iommu/amd: Remove find_last_devid_on_pci()
iommu/amd: Remove first/last_device handling
iommu/amd: Initialize amd_iommu_last_bdf for DEV_ALL
iommu/amd: Cleanup buffer allocation
iommu/amd: Remove cmd_buf_size and evt_buf_size from struct amd_iommu
iommu/amd: Align DTE flag definitions
iommu/amd: Remove old alias handling code
iommu/amd: Set alias DTE in do_attach/do_detach
iommu/amd: WARN when __[attach|detach]_device are called with irqs enabled
...
Pull intel iommu updates from David Woodhouse:
"This adds "Shared Virtual Memory" (aka PASID support) for the Intel
IOMMU. This allows devices to do DMA using process address space,
translated through the normal CPU page tables for the relevant mm.
With corresponding support added to the i915 driver, this has been
tested with the graphics device on Skylake. We don't have the
required TLP support in our PCIe root ports for supporting discrete
devices yet, so it's only integrated devices that can do it so far"
* git://git.infradead.org/intel-iommu: (23 commits)
iommu/vt-d: Fix rwxp flags in SVM device fault callback
iommu/vt-d: Expose struct svm_dev_ops without CONFIG_INTEL_IOMMU_SVM
iommu/vt-d: Clean up pasid_enabled() and ecs_enabled() dependencies
iommu/vt-d: Handle Caching Mode implementations of SVM
iommu/vt-d: Fix SVM IOTLB flush handling
iommu/vt-d: Use dev_err(..) in intel_svm_device_to_iommu(..)
iommu/vt-d: fix a loop in prq_event_thread()
iommu/vt-d: Fix IOTLB flushing for global pages
iommu/vt-d: Fix address shifting in page request handler
iommu/vt-d: shift wrapping bug in prq_event_thread()
iommu/vt-d: Fix NULL pointer dereference in page request error case
iommu/vt-d: Implement SVM_FLAG_SUPERVISOR_MODE for kernel access
iommu/vt-d: Implement SVM_FLAG_PRIVATE_PASID to allocate unique PASIDs
iommu/vt-d: Add callback to device driver on page faults
iommu/vt-d: Implement page request handling
iommu/vt-d: Generalise DMAR MSI setup to allow for page request events
iommu/vt-d: Implement deferred invalidate for SVM
iommu/vt-d: Add basic SVM PASID support
iommu/vt-d: Always enable PASID/PRI PCI capabilities before ATS
iommu/vt-d: Add initial support for PASID tables
...
Joerg Roedel