[ Upstream commit 121660bba631104154b7c15e88f208c48c8c3297 ]
Previously the AMD IOMMU would only enable SWIOTLB in certain
circumstances:
* IOMMU in passthrough mode
* SME enabled
This logic however doesn't work when an untrusted device is plugged in
that doesn't do page aligned DMA transactions. The expectation is
that a bounce buffer is used for those transactions.
This fails like this:
swiotlb buffer is full (sz: 4096 bytes), total 0 (slots), used 0 (slots)
That happens because the bounce buffers have been allocated, followed by
freed during startup but the bounce buffering code expects that all IOMMUs
have left it enabled.
Remove the criteria to set up bounce buffers on AMD systems to ensure
they're always available for supporting untrusted devices.
Fixes: 82612d66d5 ("iommu: Allow the dma-iommu api to use bounce buffers")
Suggested-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Reviewed-by: Robin Murphy <robin.murphy@arm.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220404204723.9767-2-mario.limonciello@amd.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 5ce97f4ec5e0f8726a5dda1710727b1ee9badcac upstream.
The AMD IOMMU logs I/O page faults and such to a ring buffer in
system memory, and this ring buffer can overflow. The AMD IOMMU
spec has the following to say about the interrupt status bit that
signals this overflow condition:
EventOverflow: Event log overflow. RW1C. Reset 0b. 1 = IOMMU
event log overflow has occurred. This bit is set when a new
event is to be written to the event log and there is no usable
entry in the event log, causing the new event information to
be discarded. An interrupt is generated when EventOverflow = 1b
and MMIO Offset 0018h[EventIntEn] = 1b. No new event log
entries are written while this bit is set. Software Note: To
resume logging, clear EventOverflow (W1C), and write a 1 to
MMIO Offset 0018h[EventLogEn].
The AMD IOMMU driver doesn't currently implement this recovery
sequence, meaning that if a ring buffer overflow occurs, logging
of EVT/PPR/GA events will cease entirely.
This patch implements the spec-mandated reset sequence, with the
minor tweak that the hardware seems to want to have a 0 written to
MMIO Offset 0018h[EventLogEn] first, before writing an 1 into this
field, or the IOMMU won't actually resume logging events.
Signed-off-by: Lennert Buytenhek <buytenh@arista.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/YVrSXEdW2rzEfOvk@wantstofly.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When running on an AMD vIOMMU, it is better to avoid TLB flushes
of unmodified PTEs. vIOMMUs require the hypervisor to synchronize the
virtualized IOMMU's PTEs with the physical ones. This process induce
overheads.
AMD IOMMU allows us to flush any range that is aligned to the power of
2. So when running on top of a vIOMMU, break the range into sub-ranges
that are naturally aligned, and flush each one separately. This apporach
is better when running with a vIOMMU, but on physical IOMMUs, the
penalty of IOTLB misses due to unnecessary flushed entries is likely to
be low.
Repurpose (i.e., keeping the name, changing the logic)
domain_flush_pages() so it is used to choose whether to perform one
flush of the whole range or multiple ones to avoid flushing unnecessary
ranges. Use NpCache, as usual, to infer whether the IOMMU is physical or
virtual.
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Will Deacon <will@kernel.org>
Cc: Jiajun Cao <caojiajun@vmware.com>
Cc: Lu Baolu <baolu.lu@linux.intel.com>
Cc: iommu@lists.linux-foundation.org
Cc: linux-kernel@vger.kernel.org
Suggested-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Nadav Amit <namit@vmware.com>
Link: https://lore.kernel.org/r/20210723093209.714328-8-namit@vmware.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
AMD's IOMMU can flush efficiently (i.e., in a single flush) any range.
This is in contrast, for instnace, to Intel IOMMUs that have a limit on
the number of pages that can be flushed in a single flush. In addition,
AMD's IOMMU do not care about the page-size, so changes of the page size
do not need to trigger a TLB flush.
So in most cases, a TLB flush due to disjoint range is not needed for
AMD. Yet, vIOMMUs require the hypervisor to synchronize the virtualized
IOMMU's PTEs with the physical ones. This process induce overheads, so
it is better not to cause unnecessary flushes, i.e., flushes of PTEs
that were not modified.
Implement and use amd_iommu_iotlb_gather_add_page() and use it instead
of the generic iommu_iotlb_gather_add_page(). Ignore disjoint regions
unless "non-present cache" feature is reported by the IOMMU
capabilities, as this is an indication we are running on a physical
IOMMU. A similar indication is used by VT-d (see "caching mode"). The
new logic retains the same flushing behavior that we had before the
introduction of page-selective IOTLB flushes for AMD.
On virtualized environments, check if the newly flushed region and the
gathered one are disjoint and flush if it is.
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Will Deacon <will@kernel.org>
Cc: Jiajun Cao <caojiajun@vmware.com>
Cc: Lu Baolu <baolu.lu@linux.intel.com>
Cc: iommu@lists.linux-foundation.org
Cc: linux-kernel@vger.kernel.org>
Reviewed-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Nadav Amit <namit@vmware.com>
Link: https://lore.kernel.org/r/20210723093209.714328-6-namit@vmware.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
For the printing of RMP_HW_ERROR / RMP_PAGE_FAULT / IO_PAGE_FAULT
events, the AMD IOMMU code uses such logic:
if (pdev)
dev_data = dev_iommu_priv_get(&pdev->dev);
if (dev_data && __ratelimit(&dev_data->rs)) {
pci_err(pdev, ...
} else {
printk_ratelimit() / pr_err{,_ratelimited}(...
}
This means that if we receive an event for a PCI devid which actually
does have a struct pci_dev and an attached struct iommu_dev_data, but
rate limiting kicks in, we'll fall back to the non-PCI branch of the
test, and print the event in a different format.
Fix this by changing the logic to:
if (dev_data) {
if (__ratelimit(&dev_data->rs)) {
pci_err(pdev, ...
}
} else {
pr_err_ratelimited(...
}
Suggested-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Lennert Buytenhek <buytenh@wantstofly.org>
Reviewed-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Link: https://lore.kernel.org/r/YPgk1dD1gPMhJXgY@wantstofly.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Make IOMMU_DEFAULT_LAZY default for when AMD_IOMMU config is set, which
matches current behaviour.
For "fullflush" param, just call iommu_set_dma_strict(true) directly.
Since we get a strict vs lazy mode print already in iommu_subsys_init(),
and maintain a deprecation print when "fullflush" param is passed, drop the
prints in amd_iommu_init_dma_ops().
Finally drop global flag amd_iommu_unmap_flush, as it has no longer has any
purpose.
[jpg: Rebase for relocated file and drop amd_iommu_unmap_flush]
Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Link: https://lore.kernel.org/r/1626088340-5838-6-git-send-email-john.garry@huawei.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Passing a 64-bit address width to iommu_setup_dma_ops() is valid on
virtual platforms, but isn't currently possible. The overflow check in
iommu_dma_init_domain() prevents this even when @dma_base isn't 0. Pass
a limit address instead of a size, so callers don't have to fake a size
to work around the check.
The base and limit parameters are being phased out, because:
* they are redundant for x86 callers. dma-iommu already reserves the
first page, and the upper limit is already in domain->geometry.
* they can now be obtained from dev->dma_range_map on Arm.
But removing them on Arm isn't completely straightforward so is left for
future work. As an intermediate step, simplify the x86 callers by
passing dummy limits.
Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Reviewed-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/20210618152059.1194210-5-jean-philippe@linaro.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Now that DMA ops are part of the core API via iommu-dma, fold the
vestigial remains of the IOMMU_DMA_OPS init state into the IOMMU API
phase, and clean up a few other leftovers. This should also close the
race window wherein bus_set_iommu() effectively makes the DMA ops state
visible before its nominal initialisation - it seems this was previously
fairly benign, but since commit a250c23f15 ("iommu: remove
DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE") it can now lead to the strict flush
queue policy inadvertently being picked for default domains allocated
during that window, with a corresponding unexpected perfomance impact.
Reported-by: Jussi Maki <joamaki@gmail.com>
Tested-by: Jussi Maki <joamaki@gmail.com>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Fixes: a250c23f15 ("iommu: remove DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE")
Link: https://lore.kernel.org/r/665db61e23ff8d54ac5eb391bef520b3a803fcb9.1622727974.git.robin.murphy@arm.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Since commit 08a27c1c3e ("iommu: Add support to change default domain
of an iommu group") a user can switch a device between IOMMU and direct
DMA through sysfs. This doesn't work for AMD IOMMU at the moment because
dev->dma_ops is not cleared when switching from a DMA to an identity
IOMMU domain. The DMA layer thus attempts to use the dma-iommu ops on an
identity domain, causing an oops:
# echo 0000:00:05.0 > /sys/sys/bus/pci/drivers/e1000e/unbind
# echo identity > /sys/bus/pci/devices/0000:00:05.0/iommu_group/type
# echo 0000:00:05.0 > /sys/sys/bus/pci/drivers/e1000e/bind
...
BUG: kernel NULL pointer dereference, address: 0000000000000028
...
Call Trace:
iommu_dma_alloc
e1000e_setup_tx_resources
e1000e_open
Since iommu_change_dev_def_domain() calls probe_finalize() again, clear
the dma_ops there like Vt-d does.
Fixes: 08a27c1c3e ("iommu: Add support to change default domain of an iommu group")
Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Link: https://lore.kernel.org/r/20210422094216.2282097-1-jean-philippe@linaro.org
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Currently, IOMMU invalidations and device-IOTLB invalidations using
AMD IOMMU fall back to full address-space invalidation if more than a
single page need to be flushed.
Full flushes are especially inefficient when the IOMMU is virtualized by
a hypervisor, since it requires the hypervisor to synchronize the entire
address-space.
AMD IOMMUs allow to provide a mask to perform page-specific
invalidations for multiple pages that match the address. The mask is
encoded as part of the address, and the first zero bit in the address
(in bits [51:12]) indicates the mask size.
Use this hardware feature to perform selective IOMMU and IOTLB flushes.
Combine the logic between both for better code reuse.
The IOMMU invalidations passed a smoke-test. The device IOTLB
invalidations are untested.
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Will Deacon <will@kernel.org>
Cc: Jiajun Cao <caojiajun@vmware.com>
Cc: iommu@lists.linux-foundation.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Nadav Amit <namit@vmware.com>
Link: https://lore.kernel.org/r/20210323210619.513069-1-namit@vmware.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Instead make the global iommu_dma_strict paramete in iommu.c canonical by
exporting helpers to get and set it and use those directly in the drivers.
This make sure that the iommu.strict parameter also works for the AMD and
Intel IOMMU drivers on x86. As those default to lazy flushing a new
IOMMU_CMD_LINE_STRICT is used to turn the value into a tristate to
represent the default if not overriden by an explicit parameter.
[ported on top of the other iommu_attr changes and added a few small
missing bits]
Signed-off-by: Robin Murphy <robin.murphy@arm.com>.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20210401155256.298656-19-hch@lst.de
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The AMD IOMMU initialisation registers the IRQ remapping domain for
each IOMMU before doing the final sanity check that every I/OAPIC is
covered.
This means that the AMD irq_remapping_select() function gets invoked
even when IRQ remapping has been disabled, eventually leading to a NULL
pointer dereference in alloc_irq_table().
Unfortunately, the IVRS isn't fully parsed early enough that the sanity
check can be done in time to registering the IRQ domain altogether.
Doing that would be nice, but is a larger and more error-prone task. The
simple fix is just for irq_remapping_select() to refuse to report a
match when IRQ remapping has disabled.
Link: https://lore.kernel.org/lkml/ed4be9b4-24ac-7128-c522-7ef359e8185d@gmx.at
Fixes: a1a785b572 ("iommu/amd: Implement select() method on remapping irqdomain")
Reported-by: Johnathan Smithinovic <johnathan.smithinovic@gmx.at>
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Link: https://lore.kernel.org/r/04bbe8bca87f81a3cfa93ec4299e53f47e00e5b3.camel@infradead.org
Signed-off-by: Will Deacon <will@kernel.org>
Pull IOMMU updates from Will Deacon:
"There's a good mixture of improvements to the core code and driver
changes across the board.
One thing worth pointing out is that this includes a quirk to work
around behaviour in the i915 driver (see 65f746e828 ("iommu: Add
quirk for Intel graphic devices in map_sg")), which otherwise
interacts badly with the conversion of the intel IOMMU driver over to
the DMA-IOMMU APU but has being fixed properly in the DRM tree.
We'll revert the quirk later this cycle once we've confirmed that
things don't fall apart without it.
Summary:
- IOVA allocation optimisations and removal of unused code
- Introduction of DOMAIN_ATTR_IO_PGTABLE_CFG for parameterising the
page-table of an IOMMU domain
- Support for changing the default domain type in sysfs
- Optimisation to the way in which identity-mapped regions are
created
- Driver updates:
* Arm SMMU updates, including continued work on Shared Virtual
Memory
* Tegra SMMU updates, including support for PCI devices
* Intel VT-D updates, including conversion to the IOMMU-DMA API
- Cleanup, kerneldoc and minor refactoring"
* tag 'iommu-updates-v5.11' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux: (50 commits)
iommu/amd: Add sanity check for interrupt remapping table length macros
dma-iommu: remove __iommu_dma_mmap
iommu/io-pgtable: Remove tlb_flush_leaf
iommu: Stop exporting free_iova_mem()
iommu: Stop exporting alloc_iova_mem()
iommu: Delete split_and_remove_iova()
iommu/io-pgtable-arm: Remove unused 'level' parameter from iopte_type() macro
iommu: Defer the early return in arm_(v7s/lpae)_map
iommu: Improve the performance for direct_mapping
iommu: avoid taking iova_rbtree_lock twice
iommu/vt-d: Avoid GFP_ATOMIC where it is not needed
iommu/vt-d: Remove set but not used variable
iommu: return error code when it can't get group
iommu: Fix htmldocs warnings in sysfs-kernel-iommu_groups
iommu: arm-smmu-impl: Add a space before open parenthesis
iommu: arm-smmu-impl: Use table to list QCOM implementations
iommu/arm-smmu: Move non-strict mode to use io_pgtable_domain_attr
iommu/arm-smmu: Add support for pagetable config domain attribute
iommu: Document usage of "/sys/kernel/iommu_groups/<grp_id>/type" file
iommu: Take lock before reading iommu group default domain type
...
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
Pull dma-mapping updates from Christoph Hellwig:
- rework the non-coherent DMA allocator
- move private definitions out of <linux/dma-mapping.h>
- lower CMA_ALIGNMENT (Paul Cercueil)
- remove the omap1 dma address translation in favor of the common code
- make dma-direct aware of multiple dma offset ranges (Jim Quinlan)
- support per-node DMA CMA areas (Barry Song)
- increase the default seg boundary limit (Nicolin Chen)
- misc fixes (Robin Murphy, Thomas Tai, Xu Wang)
- various cleanups
* tag 'dma-mapping-5.10' of git://git.infradead.org/users/hch/dma-mapping: (63 commits)
ARM/ixp4xx: add a missing include of dma-map-ops.h
dma-direct: simplify the DMA_ATTR_NO_KERNEL_MAPPING handling
dma-direct: factor out a dma_direct_alloc_from_pool helper
dma-direct check for highmem pages in dma_direct_alloc_pages
dma-mapping: merge <linux/dma-noncoherent.h> into <linux/dma-map-ops.h>
dma-mapping: move large parts of <linux/dma-direct.h> to kernel/dma
dma-mapping: move dma-debug.h to kernel/dma/
dma-mapping: remove <asm/dma-contiguous.h>
dma-mapping: merge <linux/dma-contiguous.h> into <linux/dma-map-ops.h>
dma-contiguous: remove dma_contiguous_set_default
dma-contiguous: remove dev_set_cma_area
dma-contiguous: remove dma_declare_contiguous
dma-mapping: split <linux/dma-mapping.h>
cma: decrease CMA_ALIGNMENT lower limit to 2
firewire-ohci: use dma_alloc_pages
dma-iommu: implement ->alloc_noncoherent
dma-mapping: add new {alloc,free}_noncoherent dma_map_ops methods
dma-mapping: add a new dma_alloc_pages API
dma-mapping: remove dma_cache_sync
53c700: convert to dma_alloc_noncoherent
...
Pull iommu updates from Joerg Roedel:
- ARM-SMMU Updates from Will:
- Continued SVM enablement, where page-table is shared with CPU
- Groundwork to support integrated SMMU with Adreno GPU
- Allow disabling of MSI-based polling on the kernel command-line
- Minor driver fixes and cleanups (octal permissions, error
messages, ...)
- Secure Nested Paging Support for AMD IOMMU. The IOMMU will fault when
a device tries DMA on memory owned by a guest. This needs new
fault-types as well as a rewrite of the IOMMU memory semaphore for
command completions.
- Allow broken Intel IOMMUs (wrong address widths reported) to still be
used for interrupt remapping.
- IOMMU UAPI updates for supporting vSVA, where the IOMMU can access
address spaces of processes running in a VM.
- Support for the MT8167 IOMMU in the Mediatek IOMMU driver.
- Device-tree updates for the Renesas driver to support r8a7742.
- Several smaller fixes and cleanups all over the place.
* tag 'iommu-updates-v5.10' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu: (57 commits)
iommu/vt-d: Gracefully handle DMAR units with no supported address widths
iommu/vt-d: Check UAPI data processed by IOMMU core
iommu/uapi: Handle data and argsz filled by users
iommu/uapi: Rename uapi functions
iommu/uapi: Use named union for user data
iommu/uapi: Add argsz for user filled data
docs: IOMMU user API
iommu/qcom: add missing put_device() call in qcom_iommu_of_xlate()
iommu/arm-smmu-v3: Add SVA device feature
iommu/arm-smmu-v3: Check for SVA features
iommu/arm-smmu-v3: Seize private ASID
iommu/arm-smmu-v3: Share process page tables
iommu/arm-smmu-v3: Move definitions to a header
iommu/io-pgtable-arm: Move some definitions to a header
iommu/arm-smmu-v3: Ensure queue is read after updating prod pointer
iommu/amd: Re-purpose Exclusion range registers to support SNP CWWB
iommu/amd: Add support for RMP_PAGE_FAULT and RMP_HW_ERR
iommu/amd: Use 4K page for completion wait write-back semaphore
iommu/tegra-smmu: Allow to group clients in same swgroup
iommu/tegra-smmu: Fix iova->phys translation
...
