aboutsummaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
AgeCommit message (Collapse)AuthorFilesLines
2016-10-21drm/amdgpu: avoid drm error log during S3 on RHEL7.3jimqu1-1/+2
Signed-off-by: JimQu <[email protected]> Reviewed-by: Christian König <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-09-28drm/amdgpu: add a custom GTT memory manager v2Christian König1-2/+2
Only allocate address space when we really need it. v2: fix a typo, add correct function description, stop leaking the node in the error case. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Reviewed-by: Felix Kuehling <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-09-14drm/amdgpu: validate size and offset of user fence BOChristian König1-0/+6
We need to validate the offset to make sure that we don't write after the BO. Additional to that a page should be enough and can make address space handling much easier. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-09-14drm/amdgpu: mark symbols static where possibleBaoyou Xie1-1/+1
We get a few warnings when building kernel with W=1: drivers/gpu/drm/amd/amdgpu/cz_smc.c:51:5: warning: no previous prototype for 'cz_send_msg_to_smc_async' [-Wmissing-prototypes] drivers/gpu/drm/amd/amdgpu/cz_smc.c:143:5: warning: no previous prototype for 'cz_write_smc_sram_dword' [-Wmissing-prototypes] drivers/gpu/drm/amd/amdgpu/iceland_smc.c:124:6: warning: no previous prototype for 'iceland_start_smc' [-Wmissing-prototypes] drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c:3926:6: warning: no previous prototype for 'gfx_v8_0_rlc_stop' [-Wmissing-prototypes] drivers/gpu/drm/amd/amdgpu/amdgpu_job.c:94:6: warning: no previous prototype for 'amdgpu_job_free_cb' [-Wmissing-prototypes] .... In fact, these functions are only used in the file in which they are declared and don't need a declaration, but can be made static. So this patch marks these functions with 'static'. Reviewed-by: Christian König <[email protected]> Acked-by: Huang Rui <[email protected]> Reviewed-by: Edward O'Callaghan <[email protected]> Signed-off-by: Baoyou Xie <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-09-14drm/amdgpu: bind GTT on demandChristian König1-2/+32
We don't really need the GTT table any more most of the time. So bind it only on demand. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Acked-by: Felix Kuehling <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-09-14drm/amdgpu:implement CONTEXT_CONTROL (v5)Monk Liu1-0/+8
v1: for gfx8, use CONTEXT_CONTROL package to dynamically skip preamble CEIB and other load_xxx command in sequence. v2: support GFX7 as well. remove cntxcntl in compute ring funcs because CPC doesn't support this packet. v3: fix reduntant judgement in cntxcntl. v4: some cleanups, don't change cs_submit() v5: keep old MESA supported & bump up KMS version. Signed-off-by: Monk Liu <[email protected]> Ack-by: Chunming Zhou <[email protected]> Reviewed-by: Christian König <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-09-12drm/amdgpu: change job->ctx field nameMonk Liu1-1/+1
job->ctx actually is a fence_context of the entity it belongs to, naming it as ctx is too vague, and we'll need add amdgpu_ctx into the job structure later. Signed-off-by: Monk Liu <[email protected]> Reviewed-by: Christian König <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-09-02drm/amdgpu: prevent command submission failures under memory pressure v2Christian König1-1/+62
As last resort try to evict BOs from the current working set into other memory domains. This effectively prevents command submission failures when VM page tables have been swapped out. v2: fix typos Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-09-02drm/amdgpu: only try again if we actually run into -ENOMEMChristian König1-5/+3
All other errors can't be fixed by using a different memory domain. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-08-30drm/amdgpu: throttle buffer migrations at CS using a fixed MBps limit (v2)Marek Olšák1-48/+104
The old mechanism used a per-submission limit that didn't take previous submissions within the same time frame into account. It also filled VRAM slowly when VRAM usage dropped due to a big eviction or buffer deallocation. This new method establishes a configurable MBps limit that is obeyed when VRAM usage is very high. When VRAM usage is not very high, it gives the driver the freedom to fill it quickly. The result is more consistent performance. It can't keep the BO move rate low if lots of evictions are happening due to VRAM fragmentation, or if a big buffer is being migrated. The amdgpu.moverate parameter can be used to set a non-default limit. Measurements can be done to find out which amdgpu.moverate setting gives the best results. Mainly APUs and cards with small VRAM will benefit from this. For F1 2015, anything with 2 GB VRAM or less will benefit. Some benchmark results - F1 2015 (Tonga 2GB): Limit MinFPS AvgFPS Old code: 14 32.6 128 MB/s: 28 41 64 MB/s: 15.5 43 32 MB/s: 28.7 43.4 8 MB/s: 27.8 44.4 8 MB/s: 21.9 42.8 (different run) Random drops in Min FPS can still occur (due to fragmented VRAM?), but the average FPS is much better. 8 MB/s is probably a good limit for this game & the current VRAM management. The random FPS drops are still to be tackled. v2: use a spinlock Signed-off-by: Marek Olšák <[email protected]> Acked-by: Christian König <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-08-22drm/amdgpu: cleanup amdgpu_vm_bo_update paramsChristian König1-1/+1
Make it more obvious what we are doing here. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-08-16drm/amdgpu: validate shadow as well when validating boChunming Zhou1-30/+46
Signed-off-by: Chunming Zhou <[email protected]> Reviewed-by: Christian König <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-08-08drm/amdgpu: print more accurate error messages on IB submission failureMarek Olšák1-4/+13
It's useful for debugging. Signed-off-by: Marek Olšák <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Reviewed-by: Christian König <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-07-07drm/amdgpu: remove fence parameter from amd_sched_job_initChristian König1-5/+3
We return the fence as part of the job structur anyway, no need to do this twice. Signed-off-by: Christian König <[email protected]> Reviewed-by: Chunming Zhou <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-07-07drm/amdgpu: earlier free SA resourcesChristian König1-0/+1
Keep the time we don't have a fence associated with the resource smaller. Signed-off-by: Christian König <[email protected]> Reviewed-by: Chunming Zhou <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-07-07drm/amdgpu: fix user fence handling once moreChristian König1-6/+6
Same problem as with the VM page tables. The user fence address must be determined before the job is scheduled, not when the IB is executed. This fixes a security problem where user fences could be used to overwrite any part of VRAM. Signed-off-by: Christian König <[email protected]> Reviewed-by: Chunming Zhou <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-07-07drm/amdgpu: don't update page tables for VM emulationChristian König1-7/+8
It's just overhead to do so and allocating a VMID when we don't need one is actually a bit dangerous. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-07-07drm/amdgpu: validate VM PTs only on evictionChristian König1-1/+4
We don't need to validate them again if the eviction counter didn't changed. Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Christian König <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-07-07drm/amdgpu: save the PD addr before scheduling the jobChristian König1-0/+2
When we pipeline evictions the page directory could already be moving somewhere else when grab_id is called. Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Christian König <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-07-07drm/amdgpu: fix and cleanup job destructionChristian König1-2/+1
Remove the job reference counting and just properly destroy it from a work item which blocks on any potential running timeout handler. Signed-off-by: Christian König <[email protected]> Reviewed-by: Monk.Liu <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-07-07drm/amdgpu: properly abstract scheduler timeout handlingChristian König1-2/+1
The driver shouldn't mess with the scheduler internals. Signed-off-by: Christian König <[email protected]> Reviewed-by: Monk.Liu <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-05-17drm: Remove unused drm_device from drm_gem_object_lookup()Chris Wilson1-2/+1
drm_gem_object_lookup() has never required the drm_device for its file local translation of the user handle to the GEM object. Let's remove the unused parameter and save some space. Signed-off-by: Chris Wilson <[email protected]> Cc: [email protected] Cc: Dave Airlie <[email protected]> Cc: Daniel Vetter <[email protected]> [danvet: Fixup kerneldoc too.] Signed-off-by: Daniel Vetter <[email protected]>
2016-05-11drm/amdgpu: fix and cleanup user fence handling v2Christian König1-30/+25
We leaked the BO in the error pass, additional to that we only have one user fence for all IBs in a job. v2: remove white space changes Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-05-11drm/amdgpu: move VM fields into jobChristian König1-20/+16
They are the same for all IBs. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-05-11drm/amdgpu: move the context from the IBs into the jobChristian König1-5/+5
We only have one context for all IBs. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-05-11drm/amdgpu: use fence_context to judge ctx switch v2Christian König1-1/+1
Use of the ctx pointer is not safe, because they are likely already be assigned to another ctx when doing comparing. v2: recreate from scratch, avoid all unnecessary changes. Signed-off-by: Christian König <[email protected]> Reviewed-by: Monk.Liu <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-05-11drm/amdgpu: keep vm in job instead of ib (v2)Monk Liu1-1/+2
ib.vm is a legacy way to get vm, after scheduler implemented vm should be get from job, and all ibs from one job share the same vm, no need to keep ib.vm just move vm field to job. this patch as well add job as paramter to ib_schedule so it can get vm from job->vm. v2: agd: sqaush in: drm/amdgpu: check if ring emit_vm_flush exists in vm flush No vm flush on engines that don't support VM. bug: https://bugs.freedesktop.org/show_bug.cgi?id=95195 Signed-off-by: Monk Liu <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-05-04drm/amdgpu: remove sorting of CS BOsChristian König1-23/+0
Not needed any more. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-05-02drm/amdgpu: use ref to keep job aliveMonk Liu1-0/+1
this is to fix fatal page fault error that occured if: job is signaled/released after its timeout work is already put to the global queue (in this case the cancel_delayed_work will return false), which will lead to NX-protection error page fault during job_timeout_func. Signed-off-by: Monk Liu <[email protected]> Reviewed-by: Chunming Zhou <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-05-02drm/amdgpu: rework TDR in scheduler (v2)Monk Liu1-0/+1
Add two callbacks to scheduler to maintain jobs, and invoked for job timeout calculations. Now TDR measures time gap from job is processed by hw. v2: fix typo Signed-off-by: Monk Liu <[email protected]> Reviewed-by: Chunming Zhou <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-05-02drm/amdgpu: use sched_job_init to initialize sched_jobMonk Liu1-13/+10
Consolidate job initialization in one place rather than duplicating it in multiple places. Signed-off-by: Monk Liu <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-03-21drm/amdgpu: release_pages requires linux/pagemap.hStephen Rothwell1-0/+1
Signed-off-by: Stephen Rothwell <[email protected]> Reviewed-by: Christian König <[email protected]. Signed-off-by: Dave Airlie <[email protected]>
2016-03-08drm/amdgpu: move get_user_pages out of amdgpu_ttm_tt_pin_userptr v6Christian König1-4/+110
That avoids lock inversion between the BO reservation lock and the anon_vma lock. v2: * Changed amdgpu_bo_list_entry.user_pages to an array of pointers * Lock mmap_sem only for get_user_pages * Added invalidation of unbound userpointer BOs * Fixed memory leak and page reference leak v3 (chk): * Revert locking mmap_sem only for_get user_pages * Revert adding invalidation of unbound userpointer BOs * Sanitize and fix error handling v4 (chk): * Init userpages pointer everywhere. * Fix error handling when get_user_pages() fails. * Add invalidation of unbound userpointer BOs again. v5 (chk): * Add maximum number of tries. v6 (chk): * Fix error handling when we run out of tries. Signed-off-by: Christian König <[email protected]> Signed-off-by: Felix Kuehling <[email protected]> Reviewed-by: Felix Kuehling <[email protected]> (v4) Acked-by: Alex Deucher <[email protected]>
2016-03-08drm/amdgpu: group userptr in the BO list v2Christian König1-1/+2
We need them together with the next patch. v2: Don't take bo reference twice Signed-off-by: Christian König <[email protected]> Signed-off-by: Felix Kuehling <[email protected]> Reviewed-by: Felix Kuehling <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: move sync into job objectChristian König1-8/+6
No need to keep that for every IB. Signed-off-by: Christian König <[email protected]> Acked-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: cleanup in kernel job submissionChristian König1-9/+1
Add a job_alloc_with_ib helper and proper job submission. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: move ring from IBs into jobChristian König1-15/+18
We can't submit to multiple rings at the same time anyway. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: cleanup user fence handling in the CSChristian König1-16/+13
Don't keep that around twice. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: add proper job alloc/free functionsChristian König1-43/+22
And use them in the CS instead of allocating IBs and jobs separately. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: fix num_ibs checkChristian König1-11/+5
Specifying no IBs on command submission is invalid, stop crashing badly when somebody tries it. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Cc: [email protected] Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: check userptrs mm earlierChristian König1-1/+6
Instead of when we try to bind it check the usermm when we try to use it in the IOCTLs. Signed-off-by: Christian König <[email protected]> Acked-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: separate pushing CS to schedulerChristian König1-46/+50
Move that out of the main IOCTL function. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: clean up non-scheduler code path (v2)Chunming Zhou1-10/+1
Non-scheduler code is longer supported. v2: agd: rebased on upstream Signed-off-by: Chunming Zhou <[email protected]> Reviewed-by: Ken Wang <[email protected]> Reviewed-by: Monk Liu <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: try to find BO VAs only for the BOs in the listChristian König1-3/+13
The other ones don't have any VAs assigned anyway or are uninteresting to us. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: search only the BO list for VM mappingsChristian König1-7/+13
Make UVD/VCE VM emulation more efficient. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: keep the prefered/allowed domains in the BOChristian König1-6/+4
Stop copying that to the bo list entry, it doesn't change anyway. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: cleanup bo list bucket handlingChristian König1-49/+2
Move that into the BO list. No functional change. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: cleanup amdgpu_cs_parser_relocsChristian König1-11/+10
Rename it to amdgpu_cs_parser_bos and move validation and bo list init there. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: cleanup amdgpu_cs_parser structureChristian König1-1/+0
Remove unused user_ptr field, group fields by usage. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
2016-02-10drm/amdgpu: cleanup amdgpu_cs_list_validateChristian König1-35/+31
No need to actually check the current placement. Just use the allowed domains when the threshold is reached. Signed-off-by: Christian König <[email protected]> Reviewed-by: Alex Deucher <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
Powered by cgit, Gruvboxed by Blaster4385.