linux/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h, branch v5.17

drm/amdgpu: introduce new amdgpu_fence object to indicate the job embedded fence

2021-12-17T17:47:28Z

The job embedded fence donesn't initialize the flags at dma_fence_init(). Then we will go a wrong way in amdgpu_fence_get_timeline_name callback and trigger a null pointer panic once we enabled the trace event here. So introduce new amdgpu_fence object to indicate the job embedded fence. [ 156.131790] BUG: kernel NULL pointer dereference, address: 00000000000002a0 [ 156.131804] #PF: supervisor read access in kernel mode [ 156.131811] #PF: error_code(0x0000) - not-present page [ 156.131817] PGD 0 P4D 0 [ 156.131824] Oops: 0000 [#1] PREEMPT SMP PTI [ 156.131832] CPU: 6 PID: 1404 Comm: sdma0 Tainted: G OE 5.16.0-rc1-custom #1 [ 156.131842] Hardware name: Gigabyte Technology Co., Ltd. Z170XP-SLI/Z170XP-SLI-CF, BIOS F20 11/04/2016 [ 156.131848] RIP: 0010:strlen+0x0/0x20 [ 156.131859] Code: 89 c0 c3 0f 1f 80 00 00 00 00 48 01 fe eb 0f 0f b6 07 38 d0 74 10 48 83 c7 01 84 c0 74 05 48 39 f7 75 ec 31 c0 c3 48 89 f8 c3 <80> 3f 00 74 10 48 89 f8 48 83 c0 01 80 38 00 75 f7 48 29 f8 c3 31 [ 156.131872] RSP: 0018:ffff9bd0018dbcf8 EFLAGS: 00010206 [ 156.131880] RAX: 00000000000002a0 RBX: ffff8d0305ef01b0 RCX: 000000000000000b [ 156.131888] RDX: ffff8d03772ab924 RSI: ffff8d0305ef01b0 RDI: 00000000000002a0 [ 156.131895] RBP: ffff9bd0018dbd60 R08: ffff8d03002094d0 R09: 0000000000000000 [ 156.131901] R10: 000000000000005e R11: 0000000000000065 R12: ffff8d03002094d0 [ 156.131907] R13: 000000000000001f R14: 0000000000070018 R15: 0000000000000007 [ 156.131914] FS: 0000000000000000(0000) GS:ffff8d062ed80000(0000) knlGS:0000000000000000 [ 156.131923] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 156.131929] CR2: 00000000000002a0 CR3: 000000001120a005 CR4: 00000000003706e0 [ 156.131937] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [ 156.131942] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 [ 156.131949] Call Trace: [ 156.131953] [ 156.131957] ? trace_event_raw_event_dma_fence+0xcc/0x200 [ 156.131973] ? ring_buffer_unlock_commit+0x23/0x130 [ 156.131982] dma_fence_init+0x92/0xb0 [ 156.131993] amdgpu_fence_emit+0x10d/0x2b0 [amdgpu] [ 156.132302] amdgpu_ib_schedule+0x2f9/0x580 [amdgpu] [ 156.132586] amdgpu_job_run+0xed/0x220 [amdgpu] v2: fix mismatch warning between the prototype and function name (Ray, kernel test robot) Signed-off-by: Huang Rui Reviewed-by: Christian König Signed-off-by: Alex Deucher

drm/amdgpu: cleanup debugfs for amdgpu rings

2021-09-14T19:35:59Z

Use debugfs_create_file_size API for creating ring debugfs, and as its a NULL returning API, change the return type for amdgpu_debugfs_ring_init API as well. Also cleanup surrounding code. Signed-off-by: Nirmoy Das Reviewed-by: Shashank Sharma Signed-off-by: Alex Deucher

drm/amdgpu: detach ring priority from gfx priority

2021-09-01T20:55:11Z

Currently AMDGPU_RING_PRIO_MAX is redefinition of a max gfx hwip priority, this won't work well when we will have a hwip with different set of priorities than gfx. Also, HW ring priorities are different from ring priorities. Create a global enum for ring priority levels which each HWIP can use to define its own priority levels. Signed-off-by: Nirmoy Das Reviewed-by: Lijo Lazar Signed-off-by: Alex Deucher

drm/amd/amdgpu embed hw_fence into amdgpu_job

2021-08-16T19:16:58Z

Why: Previously hw fence is alloced separately with job. It caused historical lifetime issues and corner cases. The ideal situation is to take fence to manage both job and fence's lifetime, and simplify the design of gpu-scheduler. How: We propose to embed hw_fence into amdgpu_job. 1. We cover the normal job submission by this method. 2. For ib_test, and submit without a parent job keep the legacy way to create a hw fence separately. v2: use AMDGPU_FENCE_FLAG_EMBED_IN_JOB_BIT to show that the fence is embedded in a job. v3: remove redundant variable ring in amdgpu_job v4: add tdr sequence support for this feature. Add a job_run_counter to indicate whether this job is a resubmit job. v5 add missing handling in amdgpu_fence_enable_signaling Signed-off-by: Jingwen Chen Signed-off-by: Jack Zhang Reviewed-by: Andrey Grodzovsky Reviewed by: Monk Liu Signed-off-by: Alex Deucher

drm/amdgpu: avoid over-handle of fence driver fini in s3 test (v2)

2021-08-06T01:17:58Z

In amdgpu_fence_driver_hw_fini, no need to call drm_sched_fini to stop scheduler in s3 test, otherwise, fence related failure will arrive after resume. To fix this and for a better clean up, move drm_sched_fini from fence_hw_fini to fence_sw_fini, as it's part of driver shutdown, and should never be called in hw_fini. v2: rename amdgpu_fence_driver_init to amdgpu_fence_driver_sw_init, to keep sw_init and sw_fini paired. Bug: https://gitlab.freedesktop.org/drm/amd/-/issues/1668 Fixes: 8d35a2596164c1 ("drm/amdgpu: adjust fence driver enable sequence") Suggested-by: Christian König Tested-by: Mike Lothian Signed-off-by: Guchun Chen Reviewed-by: Christian König Signed-off-by: Alex Deucher

drm/amdgpu: adjust fence driver enable sequence

2021-07-29T02:15:44Z

Fence driver was enabled per ring when sw init on per IP block before. Change to enable all the fence driver at the same time after amdgpu_device_ip_init finished. Rename some function related to fence to make it reasonable for read. Signed-off-by: Likun Gao Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher

drm/amdgpu: Split amdgpu_device_fini into early and late

2021-05-20T03:45:49Z

Some of the stuff in amdgpu_device_fini such as HW interrupts disable and pending fences finilization must be done right away on pci_remove while most of the stuff which relates to finilizing and releasing driver data structures can be kept until drm_driver.release hook is called, i.e. when the last device reference is dropped. v4: Change functions prefix early->hw and late->sw Signed-off-by: Andrey Grodzovsky Acked-by: Christian König Reviewed-by: Alex Deucher Link: https://patchwork.freedesktop.org/patch/msgid/20210512142648.666476-3-andrey.grodzovsky@amd.com

drm/amdgpu: add the sched_score to amdgpu_ring_init

2021-04-09T20:44:56Z

Allow separate ring to share the same scheduler score. No functional change. Signed-off-by: Christian König Reviewed-and-Tested-by: Leo Liu Signed-off-by: Alex Deucher

drm/amdgpu: add wave limit functionality for gfx8,9

2021-02-09T20:27:04Z

Wave limiting can be use to load balance high priority compute jobs along with gfx jobs. When enabled, this will reserve ~75% of waves for compute jobs. We do not need this from gfx10 onwards because >=gfx10 has asynchronous compute tunneling to replace wave limit requirement. Signed-off-by: Nirmoy Das Reviewed-by: Christian König Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher

drm/amdgpu: cleanup struct amdgpu_ring

2021-02-09T20:26:41Z

This patch consist of below related changes: 1 Rename ring->priority to ring->hw_prio. 2 Assign correct hardware ring priority. 3 Remove ring->priority_mutex as ring priority remains unchanged after initialization. 4 Remove unused ring->num_jobs. v3: remove ring->num_jobs. v2: remove ring->priority_mutex. Fixes: 33abcb1f5a17 ("drm/amdgpu: set compute queue priority at mqd_init") Signed-off-by: Nirmoy Das Reviewed-by: Christian König Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher