Pull hmm updates from Jason Gunthorpe:
"This is another round of bug fixing and cleanup. This time the focus
is on the driver pattern to use mmu notifiers to monitor a VA range.
This code is lifted out of many drivers and hmm_mirror directly into
the mmu_notifier core and written using the best ideas from all the
driver implementations.
This removes many bugs from the drivers and has a very pleasing
diffstat. More drivers can still be converted, but that is for another
cycle.
- A shared branch with RDMA reworking the RDMA ODP implementation
- New mmu_interval_notifier API. This is focused on the use case of
monitoring a VA and simplifies the process for drivers
- A common seq-count locking scheme built into the
mmu_interval_notifier API usable by drivers that call
get_user_pages() or hmm_range_fault() with the VA range
- Conversion of mlx5 ODP, hfi1, radeon, nouveau, AMD GPU, and Xen
GntDev drivers to the new API. This deletes a lot of wonky driver
code.
- Two improvements for hmm_range_fault(), from testing done by Ralph"
* tag 'for-linus-hmm' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
mm/hmm: remove hmm_range_dma_map and hmm_range_dma_unmap
mm/hmm: make full use of walk_page_range()
xen/gntdev: use mmu_interval_notifier_insert
mm/hmm: remove hmm_mirror and related
drm/amdgpu: Use mmu_interval_notifier instead of hmm_mirror
drm/amdgpu: Use mmu_interval_insert instead of hmm_mirror
drm/amdgpu: Call find_vma under mmap_sem
nouveau: use mmu_interval_notifier instead of hmm_mirror
nouveau: use mmu_notifier directly for invalidate_range_start
drm/radeon: use mmu_interval_notifier_insert
RDMA/hfi1: Use mmu_interval_notifier_insert for user_exp_rcv
RDMA/odp: Use mmu_interval_notifier_insert()
mm/hmm: define the pre-processor related parts of hmm.h even if disabled
mm/hmm: allow hmm_range to be used with a mmu_interval_notifier or hmm_mirror
mm/mmu_notifier: add an interval tree notifier
mm/mmu_notifier: define the header pre-processor parts even if disabled
mm/hmm: allow snapshot of the special zero page
Be less pessimistic about estimated page table use for KFD. Most
allocations use 2MB pages and therefore need less VRAM for page
tables. This allows more VRAM to be used for applications especially
on large systems with many GPUs and hundreds of GB of system memory.
Example: 8 GPUs with 32GB VRAM each + 256GB system memory = 512GB
Old page table reservation per GPU: 1GB
New page table reservation per GPU: 32MB
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: xinhui pan <xinhui.pan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Convert the collision-retry lock around hmm_range_fault to use the one now
provided by the mmu_interval notifier.
Although this driver does not seem to use the collision retry lock that
hmm provides correctly, it can still be converted over to use the
mmu_interval_notifier api instead of hmm_mirror without too much trouble.
This also deletes another place where a driver is associating additional
data (struct amdgpu_mn) with a mmu_struct.
Link: https://lore.kernel.org/r/20191112202231.3856-13-jgg@ziepe.ca
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Philip Yang <Philip.Yang@amd.com>
Tested-by: Philip Yang <Philip.Yang@amd.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Remove the interval tree in the driver and rely on the tree maintained by
the mmu_notifier for delivering mmu_notifier invalidation callbacks.
For some reason amdgpu has a very complicated arrangement where it tries
to prevent duplicate entries in the interval_tree, this is not necessary,
each amdgpu_bo can be its own stand alone entry. interval_tree already
allows duplicates and overlaps in the tree.
Also, there is no need to remove entries upon a release callback, the
mmu_interval API safely allows objects to remain registered beyond the
lifetime of the mm. The driver only has to stop touching the pages during
release.
Link: https://lore.kernel.org/r/20191112202231.3856-12-jgg@ziepe.ca
Reviewed-by: Philip Yang <Philip.Yang@amd.com>
Tested-by: Philip Yang <Philip.Yang@amd.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
find_vma() must be called under the mmap_sem, reorganize this code to
do the vma check after entering the lock.
Further, fix the unlocked use of struct task_struct's mm, instead use
the mm from hmm_mirror which has an active mm_grab. Also the mm_grab
must be converted to a mm_get before acquiring mmap_sem or calling
find_vma().
Fixes: 66c45500bf ("drm/amdgpu: use new HMM APIs and helpers")
Fixes: 0919195f2b ("drm/amdgpu: Enable amdgpu_ttm_tt_get_user_pages in worker threads")
Link: https://lore.kernel.org/r/20191112202231.3856-11-jgg@ziepe.ca
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Philip Yang <Philip.Yang@amd.com>
Tested-by: Philip Yang <Philip.Yang@amd.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
It may lose gpuvm invalidate acknowldege state across power-gating off
cycle. To avoid this issue in gmc9/gmc10 invalidation, add semaphore acquire
before invalidation and semaphore release after invalidation.
After adding semaphore acquire before invalidation, the semaphore
register become read-only if another process try to acquire semaphore.
Then it will not be able to release this semaphore. Then it may cause
deadlock problem. If this deadlock problem happens, it needs a semaphore
firmware fix.
Signed-off-by: changzhu <Changfeng.Zhu@amd.com>
Acked-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
SW must acquire/release one of the vm_invalidate_eng*_sem around the
invalidation req/ack. Through this way,it can avoid losing invalidate
acknowledge state across power-gating off cycle.
To use vm_invalidate_eng*_sem, it needs to initialize
vm_invalidate_eng*_sem firstly.
Signed-off-by: changzhu <Changfeng.Zhu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
After rlcg fw 2.1, kmd driver starts to load extra fw for
LIST_CNTL,GPM_MEM,SRM_MEM. We needs to skip the three fw
because all rlcg related fw have been loaded by host driver.
Guest driver would load the three fw fail without this change.
Signed-off-by: Jack Zhang <Jack.Zhang1@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Temporarily skip ras,dtm,hdcp initialize and terminate for arcturus VF
Currently the three features haven't been enabled at SRIOV, it would
trigger guest driver load fail with the bare-metal path of the three
features.
Signed-off-by: Jack Zhang <Jack.Zhang1@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This patch fixes 2nd baco reset failure with gfxoff enabled on navi1x.
clear state buffer (resides in vram) is corrupted after 1st baco reset,
upon gfxoff exit, CPF gets garbage header in CSIB and hangs.
Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
The INTERRUPT_CNTL2 register expects a valid DMA address, but is
currently set with a GPU MC address. This can cause problems on
systems that detect the resulting DMA read from an invalid address
(found on a Power8 guest).
Instead, use the DMA address of the dummy page because it will always
be safe.
Fixes: 27ae10641e ("drm/amdgpu: add interupt handler implementation for si v3")
Signed-off-by: Sam Bobroff <sbobroff@linux.ibm.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There is one regression from 042f3d7b745cd76aa
To put flush_delayed_work after adev->shutdown = true
which will make amdgpu_ih_process not response the irq
At last, all ib ring tests will be failed just like below
[drm] amdgpu: finishing device.
[drm] Fence fallback timer expired on ring gfx
[drm] Fence fallback timer expired on ring comp_1.0.0
[drm] Fence fallback timer expired on ring comp_1.1.0
[drm] Fence fallback timer expired on ring comp_1.2.0
[drm] Fence fallback timer expired on ring comp_1.3.0
[drm] Fence fallback timer expired on ring comp_1.0.1
amdgpu 0000:00:07.0: [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR* IB test failed on comp_1.1.1 (-110).
amdgpu 0000:00:07.0: [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR* IB test failed on comp_1.2.1 (-110).
amdgpu 0000:00:07.0: [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR* IB test failed on comp_1.3.1 (-110).
amdgpu 0000:00:07.0: [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR* IB test failed on sdma0 (-110).
amdgpu 0000:00:07.0: [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR* IB test failed on sdma1 (-110).
amdgpu 0000:00:07.0: [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR* IB test failed on uvd_enc_0.0 (-110).
amdgpu 0000:00:07.0: [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR* IB test failed on vce0 (-110).
[drm:amdgpu_device_delayed_init_work_handler [amdgpu]] *ERROR* ib ring test failed (-110).
v2: replace cancel_delayed_work_sync() with flush_delayed_work()
Signed-off-by: Yintian Tao <yttao@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
1. no need to allocate an extra member for 'mqd_backup' array
2. backup/restore mqd to/from the correct 'mqd_backup' array slot
v2: warning fix (Alex)
Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Commit 8c0d3a02c1 ("PCI: Add accessors for PCI Express Capability")
added accessors for the PCI Express Capability so that drivers didn't
need to be aware of differences between v1 and v2 of the PCI
Express Capability.
Replace pci_read_config_word() and pci_write_config_word() calls with
pcie_capability_read_word() and pcie_capability_write_word().
[bhelgaas: fix a couple remaining instances in cik.c]
Link: https://lore.kernel.org/r/20191118003513.10852-1-fred@fredlawl.com
Signed-off-by: Frederick Lawler <fred@fredlawl.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Previously we masked PCIe Link Control 2 register values with "7 << 9",
which was apparently intended to be the Transmit Margin field, but instead
was the high order bit of Transmit Margin, the Enter Modified Compliance
bit, and the Compliance SOS bit.
Correct the mask to "7 << 7", which is the Transmit Margin field.
Link: https://lore.kernel.org/r/20191112173503.176611-3-helgaas@kernel.org
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
flush/cancel delayed works before doing finalization
to avoid concurrently requests.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
We have the i915 security fixes to backmerge, but first
let's clear the decks for other drivers to avoid a bigger
mess.
Signed-off-by: Dave Airlie <airlied@redhat.com>
xgmi, ras, hdcp and dtm ta are actually separated ucode and
need to handled case by case to upload to psp.
We support the case that ta binary have one or multiple of
them built-in. As a result, the driver should check each ta
binariy's availablity before decide to upload them to psp.
In the terminate (unload) case, the driver will check the
context readiness before perform unload activity. It's fine
to keep it as is.
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Le Ma <Le.Ma@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
It will cause modprobe atombios stuck problem in raven2 if it doesn't
allow direct upload save restore list from gfx driver.
So it needs to allow direct upload save restore list for raven2
temporarily.
Signed-off-by: changzhu <Changfeng.Zhu@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Problem:
During GPU reset we call the GPU scheduler to suspend it's
thread, those two functions in amdgpu also suspend and resume
the sceduler for their needs but this can collide with GPU
reset in progress and accidently restart a suspended thread
before time.
Fix:
Serialize with GPU reset.
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
the asic callback function of get_pcie_replay_count is not implement on navi asic,
it will cause null pinter error when read this interface.
Signed-off-by: Kevin Wang <kevin1.wang@amd.com>
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Explain fields like aper_base, agp_start etc. The definition
of those fields are confusing as they are from different view
(CPU or GPU). Add comments for easier understand.
Signed-off-by: Oak Zeng <Oak.Zeng@amd.com>
Reviewed-by: Alex Deucher <Alex.Deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
It needs to add warning to update firmware in gfx9
in case that firmware is too old to have function to
realize dummy read in cp firmware.
Signed-off-by: changzhu <Changfeng.Zhu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The GRBM register interface is now capable of bursting 1 cycle per
register wr->wr, wr->rd much faster than previous muticycle per
transaction done interface. This has caused a problem where
status registers requiring HW to update have a 1 cycle delay, due
to the register update having to go through GRBM.
For cp ucode, it has realized dummy read in cp firmware.It covers
the use of WAIT_REG_MEM operation 1 case only.So it needs to call
gfx_v10_0_wait_reg_mem in gfx10. Besides it also needs to add warning to
update firmware in case firmware is too old to have function to realize
dummy read in cp firmware.
For sdma ucode, it hasn't realized dummy read in sdma firmware. sdma is
moved to gfxhub in gfx10. So it needs to add dummy read in driver
between amdgpu_ring_emit_wreg and amdgpu_ring_emit_reg_wait for sdma_v5_0.
Signed-off-by: changzhu <Changfeng.Zhu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Clarify some areas, clean up formatting, add section for
unrecoverable error handling.
v2: fix grammatical errors
Reviewed-by: Yong Zhao <yong.zhao@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The reference to object fence is dropped at the end of the loop.
However, it is dropped again outside the loop. The reference can be
dropped immediately after calling dma_fence_wait() in the loop and
thus the dropping operation outside the loop can be removed.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Pan Bian <bianpan2016@163.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
