This commit applies isolation enforcement to the GFX and Compute rings
in the gfx_v9_4_3 module.
The commit sets `amdgpu_gfx_enforce_isolation_ring_begin_use` and
`amdgpu_gfx_enforce_isolation_ring_end_use` as the functions to be
called when a ring begins and ends its use, respectively.
`amdgpu_gfx_enforce_isolation_ring_begin_use` is called when a ring
begins its use. This function cancels any scheduled
`enforce_isolation_work` and, if necessary, signals the Kernel Fusion
Driver (KFD) to stop the runqueue.
`amdgpu_gfx_enforce_isolation_ring_end_use` is called when a ring ends
its use. This function schedules `enforce_isolation_work` to be run
after a delay.
These functions are part of the Enforce Isolation Handler, which
enforces shader isolation on AMD GPUs to prevent data leakage between
different processes.
The commit also includes a check for the type of the ring. If the type
of the ring is `AMDGPU_RING_TYPE_COMPUTE`, the `xcp_id` of the
`enforce_isolation` structure in the `gfx` structure of the
`amdgpu_device` is set to the `xcp_id` of the ring. This ensures that
the correct `xcp_id` is used when enforcing isolation on compute rings.
The `xcp_id` is an identifier for an XCP partition, and different rings
can be associated with different XCP partitions.
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
This commit applies isolation enforcement to the GFX and Compute rings
in the gfx_v9_0 module.
The commit sets `amdgpu_gfx_enforce_isolation_ring_begin_use` and
`amdgpu_gfx_enforce_isolation_ring_end_use` as the functions to be
called when a ring begins and ends its use, respectively.
`amdgpu_gfx_enforce_isolation_ring_begin_use` is called when a ring
begins its use. This function cancels any scheduled
`enforce_isolation_work` and, if necessary, signals the Kernel Fusion
Driver (KFD) to stop the runqueue.
`amdgpu_gfx_enforce_isolation_ring_end_use` is called when a ring ends
its use. This function schedules `enforce_isolation_work` to be run
after a delay.
These functions are part of the Enforce Isolation Handler, which
enforces shader isolation on AMD GPUs to prevent data leakage between
different processes.
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Suggested-by: Christian König <christian.koenig@amd.com>
This commit introduces the Enforce Isolation Handler designed to enforce
shader isolation on AMD GPUs, which helps to prevent data leakage
between different processes.
The handler counts the number of emitted fences for each GFX and compute
ring. If there are any fences, it schedules the `enforce_isolation_work`
to be run after a delay of `GFX_SLICE_PERIOD`. If there are no fences,
it signals the Kernel Fusion Driver (KFD) to resume the runqueue.
The function is synchronized using the `enforce_isolation_mutex`.
This commit also introduces a reference count mechanism
(kfd_sch_req_count) to keep track of the number of requests to enable
the KFD scheduler. When a request to enable the KFD scheduler is made,
the reference count is decremented. When the reference count reaches
zero, a delayed work is scheduled to enforce isolation after a delay of
GFX_SLICE_PERIOD.
When a request to disable the KFD scheduler is made, the function first
checks if the reference count is zero. If it is, it cancels the delayed
work for enforcing isolation and checks if the KFD scheduler is active.
If the KFD scheduler is active, it sends a request to stop the KFD
scheduler and sets the KFD scheduler state to inactive. Then, it
increments the reference count.
The function is synchronized using the kfd_sch_mutex to ensure that the
KFD scheduler state and reference count are updated atomically.
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Suggested-by: Christian König <christian.koenig@amd.com>
Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Provide amdgpu_amdkfd_stop_sched() for amdgpu to stop KFD scheduling
compute work on HIQ. amdgpu_amdkfd_start_sched() resumes the scheduling.
When amdgpu_amdkfd_stop_sched is called, KFD will unmap queues from
runlist. If users send ioctls to KFD to create queues, they'll be added
but those queues won't be mapped to runlist (so not scheduled) until
amdgpu_amdkfd_start_sched is called.
v2: fix build (Alex)
Signed-off-by: Amber Lin <Amber.Lin@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit extends the cleaner shader feature to support GFX9.4.4
hardware.
The cleaner shader feature is used to clear or initialize certain GPU
resources, such as Local Data Share (LDS), Vector General Purpose
Registers (VGPRs), and Scalar General Purpose Registers (SGPRs). This
operation needs to be performed in isolation, while no other tasks
should be running on the GPU at the same time.
Previously, the cleaner shader feature was implemented for GFX9.4.3
hardware. This commit adds support for GFX9.4.4 hardware by allowing the
cleaner shader to be used with this hardware version.
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit adds the cleaner shader microcode for GFX9.4.3 GPUs. The
cleaner shader is a piece of GPU code that is used to clear or
initialize certain GPU resources, such as Local Data Share (LDS), Vector
General Purpose Registers (VGPRs), and Scalar General Purpose Registers
(SGPRs).
Clearing these resources is important for ensuring data isolation
between different workloads running on the GPU. Without the cleaner
shader, residual data from a previous workload could potentially be
accessed by a subsequent workload, leading to data leaks and incorrect
computation results.
The cleaner shader microcode is represented as an array of 32-bit words
(`gfx_9_4_3_cleaner_shader_hex`). This array is the binary
representation of the cleaner shader code, which is written in a
low-level GPU instruction set.
When the cleaner shader feature is enabled, the AMDGPU driver loads this
array into a specific location in the GPU memory. The GPU then reads
this memory location to fetch and execute the cleaner shader
instructions.
The cleaner shader is executed automatically by the GPU at the end of
each workload, before the next workload starts. This ensures that all
GPU resources are in a clean state before the start of each workload.
This addition is part of the cleaner shader feature implementation. The
cleaner shader feature helps improve GPU performance and resource
utilization by cleaning up GPU resources after they are used. It also
enhances security and reliability by preventing data leaks between
workloads.
v2: fix copyright date (Alex)
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The patch modifies the gfx_v9_4_3_kiq_set_resources function to write
the cleaner shader's memory controller address to the ring buffer. It
also adds a new function, gfx_v9_4_3_ring_emit_cleaner_shader, which
emits the PACKET3_RUN_CLEANER_SHADER packet to the ring buffer.
This patch adds support for the PACKET3_RUN_CLEANER_SHADER packet in the
gfx_v9_4_3 module. This packet is used to emit the cleaner shader, which
is used to clear GPU memory before it's reused, helping to prevent data
leakage between different processes.
Finally, the patch updates the ring function structures to include the
new gfx_v9_4_3_ring_emit_cleaner_shader function. This allows the
cleaner shader to be emitted as part of the ring's operations.
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The patch modifies the gfx_v9_0_kiq_set_resources function to write
the cleaner shader's memory controller address to the ring buffer. It
also adds a new function, gfx_v9_0_ring_emit_cleaner_shader, which
emits the PACKET3_RUN_CLEANER_SHADER packet to the ring buffer.
This patch adds support for the PACKET3_RUN_CLEANER_SHADER packet in the
gfx_v9_0 module. This packet is used to emit the cleaner shader, which
is used to clear GPU memory before it's reused, helping to prevent data
leakage between different processes.
Finally, the patch updates the ring function structures to include the
new gfx_v9_0_ring_emit_cleaner_shader function. This allows the
cleaner shader to be emitted as part of the ring's operations.
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit adds the PACKET3_RUN_CLEANER_SHADER definition. This packet
is a command packet used to instruct the GPU to execute the cleaner
shader.
The cleaner shader is a piece of GPU code that is used to clear or
initialize certain GPU resources, such as Local Data Share (LDS), Vector
General Purpose Registers (VGPRs), and Scalar General Purpose Registers
(SGPRs). Clearing these resources is important for ensuring data
isolation between different workloads running on the GPU.
The PACKET3_RUN_CLEANER_SHADER packet is used to trigger the execution
of the cleaner shader on the GPU. The packet consists of a header
followed by a RESERVED field, which is programmed to zero. When the GPU
receives this packet, it fetches and executes the cleaner shader
instructions from the location specified in the packet.
The cleaner shader feature helps to enhances security and reliability by
preventing data leaks between workloads.
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This patch adds a new sysfs interface for running the cleaner shader on
AMD GPUs. The cleaner shader is used to clear GPU memory before it's
reused, which can help prevent data leakage between different processes.
The new sysfs file is write-only and is named `run_cleaner_shader`.
Write the number of the partition to this file to trigger the cleaner shader
on that partition. There is only one partition on GPUs which do not
support partitioning.
Changes made in this patch:
- Added `amdgpu_set_run_cleaner_shader` function to handle writes to the
`run_cleaner_shader` sysfs file.
- Added `run_cleaner_shader` to the list of device attributes in
`amdgpu_device_attrs`.
- Updated `default_attr_update` to handle `run_cleaner_shader`.
- Added `AMDGPU_DEVICE_ATTR_WO` macro to create write-only device
attributes.
v2: fix error handling (Alex)
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
This commit adds a new sysfs attribute 'enforce_isolation' to control
the 'enforce_isolation' setting per GPU. The attribute can be read and
written, and accepts values 0 (disabled) and 1 (enabled).
When 'enforce_isolation' is enabled, reserved VMIDs are allocated for
each ring. When it's disabled, the reserved VMIDs are freed.
The set function locks a mutex before changing the 'enforce_isolation'
flag and the VMIDs, and unlocks it afterwards. This ensures that these
operations are atomic and prevents race conditions and other concurrency
issues.
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This patch adds a new parameter 'enforce_isolation' to the amdgpu_job
structure. This parameter is used to determine whether shader isolation
should be enforced for a job. The enforce_isolation parameter is then
stored in the amdgpu_job structure and used when flushing the VM.
The enforce_isolation field of the amdgpu_job structure is set directly
after the job is allocated
This change allows more fine-grained control over shader isolation,
making it possible to enforce isolation on a per-job basis rather than
globally. This can be useful in scenarios where only certain jobs
require isolation.
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Suggested-by: Christian König <christian.koenig@amd.com>
The HW fence ctx objects are not ref counted rather tied to the life of
an LRC object. HW fences reference the HW fence ctx, HW fences can
outlive LRCs thus resulting in UAF. Drop the HW fence pointer to HW
fence ctx rather just store what is needed directly in HW fence.
v2:
- Fix typo in commit (Ashutosh)
- Use snprintf (Ashutosh)
Fixes: dd08ebf6c3 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240815193522.16008-1-matthew.brost@intel.com
Wa_15015404425 asks us to perform four "dummy" writes to a
non-existent register offset before every real register read.
Although the specific offset of the writes doesn't directly
matter, the workaround suggests offset 0x130030 as a good target
so that these writes will be easy to recognize and filter out in
debugging traces.
V5(MattR):
- Avoid negating an equality comparison
V4(MattR):
- Use writel and remove xe_reg usage
V3(MattR):
- Define dummy reg local to function
- Avoid tracing dummy writes
- Update commit message
V2:
- Add WA to 8/16/32bit reads also - MattR
- Corrected dummy reg address - MattR
- Use for loop to avoid mental pause - JaniN
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240709155606.2998941-1-tejas.upadhyay@intel.com
(cherry picked from commit 86c5b70a9c)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Move the "exit sleep mode" and "set display on" command from
enable() to init() function.
As mentioned in the patch:
https://lore.kernel.org/all/20240624141926.5250-2-lvzhaoxiong@huaqin.corp-partner.google.com/
The Mediatek Soc DSI host has different modes in prepare() and
enable() functions, prepare() is in LP mode and enable() is in
HS mode. Since the "exit sleep mode" and "set display on"
command must also be sent in LP mode, so we also move "exit
sleep mode" and "set display on" command to the init() function.
We have no other actions in the enable() function after moves
"exit sleep mode" and "set display on", and we checked the call
of the enable() function during the "startup" process. It seems
that only one judgment was made in drm_panel_enabel(). If the
panel does not define enable(), the judgment will skip the
enable() and continue execution. This does not seem to have
any other effect, and we found that some drivers also seem
to have no enable() function added, for example:
panel-asus-z00t-tm5p5-n35596 / panel-boe-himax8279d...
In addition, we briefly tested the kingdisplay_kd101ne3 panel and
melfas_lmfbx101117480 panel, and it seems that there is no garbage
on the panel, so we delete enable() function.
After moving the "exit sleep mode" and "set display on" command
to the init() function, we no longer need additional delay
judgment, so we delete variables "exit_sleep_to_display_on_delay_ms"
and "display_on_delay_ms".
Reviewed-by: Douglas Anderson <dianders@chromium.org>
Signed-off-by: Zhaoxiong Lv <lvzhaoxiong@huaqin.corp-partner.google.com>
Acked-by: Jessica Zhang <quic_jesszhan@quicinc.com>
Link: https://lore.kernel.org/r/20240807100429.13260-2-lvzhaoxiong@huaqin.corp-partner.google.com
Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20240807100429.13260-2-lvzhaoxiong@huaqin.corp-partner.google.com
On LNL because of flat CCS, driver creates migrates job to clear
CCS meta data. Extend that to also clear system pages using GPU.
Inform TTM to allocate pages without __GFP_ZERO to avoid double page
clearing by clearing out TTM_TT_FLAG_ZERO_ALLOC flag and set
TTM_TT_FLAG_CLEARED_ON_FREE while freeing to skip ttm pool's clear
on free as XE now takes care of clearing pages. If a bo is in system
placement such as BO created with DRM_XE_GEM_CREATE_FLAG_DEFER_BACKING
and there is a cpu map then for such BO gpu clear will be avoided as
there is no dma mapping for such BO at that moment to create migration
jobs.
Tested this patch api_overhead_benchmark_l0 from
https://github.com/intel/compute-benchmarks
Without the patch:
api_overhead_benchmark_l0 --testFilter=UsmMemoryAllocation:
UsmMemoryAllocation(api=l0 type=Host size=4KB) 84.206 us
UsmMemoryAllocation(api=l0 type=Host size=1GB) 105775.56 us
erf tool top 5 entries:
71.44% api_overhead_be [kernel.kallsyms] [k] clear_page_erms
6.34% api_overhead_be [kernel.kallsyms] [k] __pageblock_pfn_to_page
2.24% api_overhead_be [kernel.kallsyms] [k] cpa_flush
2.15% api_overhead_be [kernel.kallsyms] [k] pages_are_mergeable
1.94% api_overhead_be [kernel.kallsyms] [k] find_next_iomem_res
With the patch:
api_overhead_benchmark_l0 --testFilter=UsmMemoryAllocation:
UsmMemoryAllocation(api=l0 type=Host size=4KB) 79.439 us
UsmMemoryAllocation(api=l0 type=Host size=1GB) 98677.75 us
Perf tool top 5 entries:
11.16% api_overhead_be [kernel.kallsyms] [k] __pageblock_pfn_to_page
7.85% api_overhead_be [kernel.kallsyms] [k] cpa_flush
7.59% api_overhead_be [kernel.kallsyms] [k] find_next_iomem_res
7.24% api_overhead_be [kernel.kallsyms] [k] pages_are_mergeable
5.53% api_overhead_be [kernel.kallsyms] [k] lookup_address_in_pgd_attr
Without this patch clear_page_erms() dominates execution time which is
also not pipelined with migration jobs. With this patch page clearing
will get pipelined with migration job and will free CPU for more work.
v2: Handle regression on dgfx(Himal)
Update commit message as no ttm API changes needed.
v3: Fix Kunit test.
v4: handle data leak on cpu mmap(Thomas)
v5: s/gpu_page_clear/gpu_page_clear_sys and move setting
it to xe_ttm_sys_mgr_init() and other nits (Matt Auld)
v6: Disable it when init_on_alloc and/or init_on_free is active(Matt)
Use compute-benchmarks as reporter used it to report this
allocation latency issue also a proper test application than mime.
In v5, the test showed significant reduction in alloc latency but
that is not the case any more, I think this was mostly because
previous test was done on IFWI which had low mem BW from CPU.
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240816135154.19678-2-nirmoy.das@intel.com
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
There's no need to get the length of an DT array property before
parsing the array. of_property_read_variable_u8_array() takes a minimum
and maximum length and returns the actual length (or error code).
This is part of a larger effort to remove callers of of_get_property()
and similar functions. of_get_property() leaks the DT property data
pointer which is a problem for dynamically allocated nodes which may
be freed.
Signed-off-by: Rob Herring (Arm) <robh@kernel.org>
Signed-off-by: Robert Foss <rfoss@kernel.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20240731201407.1838385-4-robh@kernel.org
