- 21 Apr, 2021 40 commits
-
-
Philip Yang authored
When application explicitly call unmap or unmap from mmput when application exit, driver will receive MMU_NOTIFY_UNMAP event to remove svm range from process svms object tree and list first, unmap from GPUs (in the following patch). Split the svm ranges to handle partial unmapping of svm ranges. To avoid deadlocks, updating MMU notifiers, range lists and interval trees is done in a deferred worker. New child ranges are attached to their parent range's child_list until the worker can update the svm_range_list. svm_range_set_attr flushes deferred work and takes the mmap_write_lock to guarantee that it has an up-to-date svm_range_list. Signed-off-by:
Philip Yang <Philip.Yang@amd.com> Signed-off-by:
Alex Sierra <alex.sierra@amd.com> Reviewed-by:
Felix Kuehling <Felix.Kuehling@amd.com> Signed-off-by:
Alex Deucher <alexander.deucher@amd.com>
-
Philip Yang authored
Use HMM to get system memory pages address, which will be used to map to GPUs or migrate to vram. Signed-off-by:
-
Philip Yang authored
For larger range allocation, if hmm_range_fault return -EBUSY, set retry timeout based on 1 second for every 512MB, this is safe timeout value. Signed-off-by:
-
Philip Yang authored
Move the HMM get pages function from amdgpu_ttm and to amdgpu_mn. This common function will be used by new svm APIs. Signed-off-by:
-
Philip Yang authored
Get the intersection of attributes over all memory in the given range Signed-off-by:
-
Philip Yang authored
svm range structure stores the range start address, size, attributes, flags, prefetch location and gpu bitmap which indicates which GPU this range maps to. Same virtual address is shared by CPU and GPUs. Process has svm range list which uses both interval tree and list to store all svm ranges registered by the process. Interval tree is used by GPU vm fault handler and CPU page fault handler to get svm range structure from the specific address. List is used to scan all ranges in eviction restore work. No overlap range interval [start, last] exist in svms object interval tree. If process registers new range which has overlap with old range, the old range split into 2 ranges depending on the overlap happens at head or tail part of old range. Apply attributes preferred location, prefetch location, mapping flags, migration granularity to svm range, store mapping gpu index into bitmap. Signed-off-by:
-
Philip Yang authored
Add svm (shared virtual memory) ioctl data structure and API definition. The svm ioctl API is designed to be extensible in the future. All operations are provided by a single IOCTL to preserve ioctl number space. The arguments structure ends with a variable size array of attributes that can be used to set or get one or multiple attributes. Signed-off-by:
-
Alex Sierra authored
svm range uses gpu bitmap to store which GPU svm range maps to. Application pass driver gpu id to specify GPU, the helper is needed to convert gpu id to gpu bitmap idx. Access through kfd_process_device pointers array from kfd_process. Signed-off-by:
-
Felix Kuehling authored
This shortcut is no longer needed with access managed properly by KFD. Signed-off-by:
-
Felix Kuehling authored
DRM render node file handles are used for CPU mapping of BOs using mmap by the Thunk. It uses the DRM render node of the GPU where the BO was allocated. DRM allows mmap access automatically when it creates a GEM handle for a BO. KFD BOs don't have GEM handles, so KFD needs to manage access manually. Use drm_vma_node_allow to allow user mode to mmap BOs allocated with kfd_ioctl_alloc_memory_of_gpu through the DRM render node that was used in the kfd_ioctl_acquire_vm call for the same GPU. Signed-off-by:
Christian König <christian.koenig@amd.com> Reviewed-by:
-
Felix Kuehling authored
amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu needs the drm_priv to allow mmap to access the BO through the corresponding file descriptor. The VM can also be extracted from drm_priv, so drm_priv can replace the vm parameter in the kfd2kgd interface. Signed-off-by:
-
Alex Deucher authored
Aldebaran has a hw fix so no longer requires the workaround. Reviewed-by:
Hawking Zhang <Hawking.Zhang@amd.com> Signed-off-by:
-
Jinzhou Su authored
The buffer of SA bo will be used by many cases. So it's better to invalidate the cache of indirect buffer allocated by SA before commit the IB. Signed-off-by:
Jinzhou Su <Jinzhou.Su@amd.com> Reviewed-by:
-
Mukul Joshi authored
Fix the following issues with SDMA RAS error reporting: 1. Read the EDC_COUNTER2 register also to fetch error counts for all sub-blocks in SDMA. 2. SDMA RAS on Aldebaran suports single-bit uncorrectable errors only. So, report error count in UE count instead of CE count. Signed-off-by:
Mukul Joshi <mukul.joshi@amd.com> Reviewed-By:
John Clements <John.Clements@amd.com> Signed-off-by:
-
Mukul Joshi authored
Reset the RAS error count and error status registers after reading to prevent over reporting error counts on Aldebaran. Signed-off-by:
-
Oak Zeng authored
This reverts commit 2f055097. 2f055097 was a driver workaround when PSP firmware was not ready. Now the PSP fw is ready so we revert this driver workaround. Signed-off-by:
Oak Zeng <Oak.Zeng@amd.com> Reviewed-by:
Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com> Signed-off-by:
-
Aric Cyr authored
Signed-off-by:
Aric Cyr <aric.cyr@amd.com> Acked-by:
Aurabindo Pillai <aurabindo.pillai@amd.com> Tested-by:
Daniel Wheeler <daniel.wheeler@amd.com> Signed-off-by:
-
Anthony Koo authored
Signed-off-by:
Anthony Koo <Anthony.Koo@amd.com> Acked-by:
-
Bing Guo authored
[Why] Some MST devices uses different method to enable mst specific stream features. [How] Add dm_helpers_mst_enable_stream features. This can be modified later when we are ready to implement those features. Signed-off-by:
Bing Guo <bing.guo@amd.com> Reviewed-by:
-
Dillon Varone authored
[Why?] When a monitor does not set both QS and QY bits, DC does not set Q0, Q1, QY0 and QY1 bits in AVI infoframe. Setting RGB bits should be separate from setting YCC bits. [How?] Separate logic for setting RGB and YCC quantization range bits in the AVI infoframe. Signed-off-by:
Dillon Varone <dillon.varone@amd.com> Reviewed-by:
Chris Park <Chris.Park@amd.com> Acked-by:
-
Dillon Varone authored
[Why & How?] Call to dc_dsc_compute_bandwidth_range should have min and max bpp in 16ths of a bit. Multiply min and max bpp from policy. Signed-off-by:
Eryk Brol <Eryk.Brol@amd.com> Acked-by:
-
David Galiffi authored
[How & Why] Changed "prsent" to "present". Signed-off-by:
David Galiffi <David.Galiffi@amd.com> Reviewed-by:
-
Nicholas Kazlauskas authored
[Why] Requirement from the spec - we shouldn't be potentially exiting out early based on encryption status. [How] Drop the calls from HDCP1 and HDCP2 execution that exit out early based on link encryption status. Signed-off-by:
Nicholas Kazlauskas <nicholas.kazlauskas@amd.com> Reviewed-by:
Wenjing Liu <Wenjing.Liu@amd.com> Acked-by:
-
Michael Strauss authored
[Why&How] Add logs to verify ILR optimization behaviour on boot Signed-off-by:
Michael Strauss <michael.strauss@amd.com> Reviewed-by:
-
Wesley Chalmers authored
[WHY] While Link Training is being performed, and the LTTPRs are in Non-LTTPR or LTTPR Transparent mode, any DPCD registers besides those used for Link Training are not to be accessed. The spec defines the link training registers as DP_TRAINING_PATTERN_SET (102h) to DP_TRAINING_LANE3_SET (106h), and DP_LANE0_1_STATUS (202h) to DP_ADJUST_REQUEST_LANE2_3 (207h). [HOW] Move the current write to DPCD Address DP_LINK_TRAINING_PATTERN_SET out of its conditional block. Signed-off-by:
Wesley Chalmers <Wesley.Chalmers@amd.com> Reviewed-by:
Jun Lei <Jun.Lei@amd.com> Acked-by:
-
Aric Cyr authored
[WHY] We should skip programming manual trigger on non-primary planes when MPO is enabled. [HOW] Implement an explicit mechanism for skipping manual trigger programming for planes that shouldn't cause the frame to end. Signed-off-by:
Krunoslav Kovac <Krunoslav.Kovac@amd.com> Tested-by:
-
Hugo Hu authored
Previous change had been reverted since it caused hang. Remake change to avoid defect. [Why] 1. Driver use umachannelnumber to calculate watermarks for stutter. In asymmetric memory config, the actual bandwidth is less than dual-channel. The bandwidth should be the same as single-channel. 2. We found single rank dimm need additional delay time for stutter. [How] Get information from each DIMM. Treat memory config as a single-channel for asymmetric memory in bandwidth calculating. Add additional delay time for single rank dimm. Fixes: b8720ed0 ("drm/amd/display: System black screen hangs on driver load") Signed-off-by:
Hugo Hu <hugo.hu@amd.com> Reviewed-by:
Sung Lee <Sung.Lee@amd.com> Acked-by:
-
Robin Singh authored
[Why] Found that dc_link_reallocate_mst_payload is not used anymore in any of the use case scenario. [How] removed dc_link_reallocate_mst_payload function definition and declaration. Signed-off-by:
Robin Singh <robin.singh@amd.com> Reviewed-by:
Harry Wentland <Harry.Wentland@amd.com> Acked-by:
-
Anthony Wang authored
[Why] Primary feature use case is with eDP panels. [How] Fail seamless boot validation if display is not an eDP panel. Signed-off-by:
Anthony Wang <anthony1.wang@amd.com> Reviewed-by:
Martin Leung <Martin.Leung@amd.com> Acked-by:
-
Dingchen (David) Zhang authored
[why] the current implementation of hdcp2 rx id list validation does not have handler/checker for invalid message status, e.g. HMAC, the V parameter calculated from PSP not matching the V prime from Rx. [how] return a generic FAILURE for any message status not SUCCESS or REVOKED. Signed-off-by:
Dingchen (David) Zhang <dingchen.zhang@amd.com> Reviewed-by:
Bhawanpreet Lakha <Bhawanpreet.Lakha@amd.com> Acked-by:
-
Dingchen (David) Zhang authored
[why] currently we enforce to update hdcp display using TYPE0, but there is case that connector CP type prop be TYPE1 instead of type0. [how] using the drm prop of CP type of the connector as input argument. Signed-off-by:
-
Anthony Wang authored
[Why & How] We want to immediately fail seamless boot validation if DSC is active, as VBIOS currently does not support DSC timings. Add a check for the relevant flag in dc_validate_seamless_boot_timing. Signed-off-by:
-
Robin Singh authored
[why] During dsc enable, a divide by zero condition triggered the kernel crash. [how] An IGT test, which enable the DSC, was crashing at the time of restore the default dsc status, becaue of h_totals value becoming 0. So add a check before divide condition. If h_total is zero, gracefully ignore and set the default value. kernel panic log: [ 128.758827] divide error: 0000 [#1] PREEMPT SMP NOPTI [ 128.762714] CPU: 5 PID: 4562 Comm: amd_dp_dsc Tainted: G W 5.4.19-android-x86_64 #1 [ 128.769728] Hardware name: ADVANCED MICRO DEVICES, INC. Mauna/Mauna, BIOS WMN0B13N Nov 11 2020 [ 128.777695] RIP: 0010:hubp2_vready_at_or_After_vsync+0x37/0x7a [amdgpu] [ 128.785707] Code: 80 02 00 00 48 89 f3 48 8b 7f 08 b ...... [ 128.805696] RSP: 0018:ffffad8f82d43628 EFLAGS: 00010246 ...... [ 128.857707] CR2: 00007106d8465000 CR3: 0000000426530000 CR4: 0000000000140ee0 [ 128.865695] Call Trace: [ 128.869712] hubp3_setup+0x1f/0x7f [amdgpu] [ 128.873705] dcn20_update_dchubp_dpp+0xc8/0x54a [amdgpu] [ 128.877706] dcn20_program_front_end_for_ctx+0x31d/0x463 [amdgpu] [ 128.885706] dc_commit_state+0x3d2/0x658 [amdgpu] [ 128.889707] amdgpu_dm_atomic_commit_tail+0x4b3/0x1e7c [amdgpu] [ 128.897699] ? dm_read_reg_func+0x41/0xb5 [amdgpu] [ 128.901707] ? dm_read_reg_func+0x41/0xb5 [amdgpu] [ 128.905706] ? __is_insn_slot_addr+0x43/0x48 [ 128.909706] ? fill_plane_buffer_attributes+0x29e/0x3dc [amdgpu] [ 128.917705] ? dm_plane_helper_prepare_fb+0x255/0x284 [amdgpu] [ 128.921700] ? usleep_range+0x7c/0x7c [ 128.925705] ? preempt_count_sub+0xf/0x18 [ 128.929706] ? _raw_spin_unlock_irq+0x13/0x24 [ 128.933732] ? __wait_for_common+0x11e/0x18f [ 128.937705] ? _raw_spin_unlock_irq+0x13/0x24 [ 128.941706] ? __wait_for_common+0x11e/0x18f [ 128.945705] commit_tail+0x8b/0xd2 [drm_kms_helper] [ 128.949707] drm_atomic_helper_commit+0xd8/0xf5 [drm_kms_helper] [ 128.957706] amdgpu_dm_atomic_commit+0x337/0x360 [amdgpu] [ 128.961705] ? drm_atomic_check_only+0x543/0x68d [drm] [ 128.969705] ? drm_atomic_set_property+0x760/0x7af [drm] [ 128.973704] ? drm_mode_atomic_ioctl+0x6f3/0x85a [drm] [ 128.977705] drm_mode_atomic_ioctl+0x6f3/0x85a [drm] [ 128.985705] ? drm_atomic_set_property+0x7af/0x7af [drm] [ 128.989706] drm_ioctl_kernel+0x82/0xda [drm] [ 128.993706] drm_ioctl+0x225/0x319 [drm] [ 128.997707] ? drm_atomic_set_property+0x7af/0x7af [drm] [ 129.001706] ? preempt_count_sub+0xf/0x18 [ 129.005713] amdgpu_drm_ioctl+0x4b/0x76 [amdgpu] [ 129.009705] vfs_ioctl+0x1d/0x2a [ 129.013705] do_vfs_ioctl+0x419/0x43d [ 129.017707] ksys_ioctl+0x52/0x71 [ 129.021707] __x64_sys_ioctl+0x16/0x19 [ 129.025706] do_syscall_64+0x78/0x85 [ 129.029705] entry_SYSCALL_64_after_hwframe+0x44/0xa9 Signed-off-by:
-
Jiansong Chen authored
dimgrey_cavefish has similar gc_10_3 ip with sienna_cichlid, so follow its registers offset setting. Signed-off-by:
Jiansong Chen <Jiansong.Chen@amd.com> Reviewed-by:
Tao Zhou <tao.zhou1@amd.com> Signed-off-by:
-
John Clements authored
resolve bug on aldebaran where gfx error counts will print on driver load when there are no errors present Reviewed-by:
-
Dennis Li authored
because "sscanf(str, "retire_page")" always return 0, if application use the raw data for error injection, it always wrongly falls into "op == 3". Change to use strstr instead. Signed-off-by:
Dennis Li <Dennis.Li@amd.com> Reviewed-by:
-
Hawking Zhang authored
In aldebaran, driver only needs to harvest SDP RdRspStatus, WrRspStatus and first parity error on RdRsp data. Check error type before harvest error information. Signed-off-by:
Stanley Yang <Stanley.Yang@amd.com> Signed-off-by:
-
Hawking Zhang authored
SDP RdRspStatus/WrRspStatus or first parity error on RdRsp data can cause system fatal error in arcturus. GPU will be freezed in such case. Driver needs to harvest these error information before reset the GPU. Check error type to avoid harvest normal gcea/mmea information. Signed-off-by:
-
Huang Rui authored
The tmz functions are verified on renoir chips as well. So enable it by default. Signed-off-by:
Huang Rui <ray.huang@amd.com> Tested-by:
Lang Yu <Lang.Yu@amd.com> Reviewed-by:
-
Hawking Zhang authored
When gfx wdt was configured to fatal_disable, the timeout period should be configured to 0x0 (timeout disabled) Signed-off-by:
-
