- 04 Dec, 2017 40 commits
-
-
Monk Liu authored
if app close CTX right after IB submit, gpu recover will fail to find out the entity behind this guilty job thus lead to no job skipping for this guilty job. to fix this corner case just move the increasement of job->karma out of the entity iteration. v2: only do karma increasment if bad->s_priority != KERNEL because we always consider KERNEL job be correct and always want to recover an unfinished kernel job (sometimes kernel job is interrupted by VF FLR or other GPU hang event) Signed-off-by:
Monk Liu <Monk.Liu@amd.com> Reviewed-by:
Christian König <christian.koenig@amd.com> Reviewed-By:
Xiangliang Yu <Xiangliang.Yu@amd.com> Signed-off-by:
Alex Deucher <alexander.deucher@amd.com>
-
Monk Liu authored
Signed-off-by:
-
pding authored
It introduces 900ms latency in exclusive mode which causes failure of driver loading. Host can resize the BAR before guest staring, so the resizing is not necessary here. Signed-off-by:
Pixel Ding <Pixel.Ding@amd.com> Reviewed-by:
-
pding authored
Signed-off-by:
Felix Kuehling <Felix.Kuehling@amd.com> Signed-off-by:
-
pding authored
Also finalize kfd inside amdgpu_device_fini. kfd device_init needs SRIOV exclusive accessing. Try to gather exclusive accessing to reduce time consuming. Signed-off-by:
-
Christian König authored
Just allocate the GART space and fill it. This prevents forcing the BO to be idle. v2: don't unbind/bind at all, just fill the allocated GART space Signed-off-by:
-
Christian König authored
We actually don't bind here, but rather allocate GART space if necessary. Signed-off-by:
-
Hawking Zhang authored
Signed-off-by:
Hawking Zhang <Hawking.Zhang@amd.com> Acked-by:
-
Christian König authored
Try to resize BAR0 to let CPU access all of VRAM. v2: rebased, style cleanups, disable mem decode before resize, handle gmc_v9 as well, round size up to power of two. v3: handle gmc_v6 as well, release and reassign all BARs in the driver. v4: rename new function to amdgpu_device_resize_fb_bar, reenable mem decoding only if all resources are assigned. v5: reorder resource release, return -ENODEV instead of BUG_ON(). v6: squash in rebase fix Signed-off-by: -
Horace Chen authored
The previous solution will create a zero buffer on the system domain and then move the zeroes to the VRAM. This will break the original data on the VRAM. Refine the code to create bo on VRAM domain directly and then remove and re-create mem node to the exact position before bo_pin. This can avoid breaking the data and will not cause eviction. Signed-off-by:
Horace Chen <horace.chen@amd.com> Reviewed-by:
-
pding authored
This is caused of that hypervisor fails to handle request, one known issue is MMIO unblocking timeout. In theory we can retry init here. Signed-off-by:
-
pding authored
Reported-by:
Sun Gary <Gary.Sun@amd.com> Signed-off-by:
-
Michel Dänzer authored
It's dead code. Signed-off-by:
Michel Dänzer <michel.daenzer@amd.com> Reviewed-by:
-
Michel Dänzer authored
Not used anymore. Signed-off-by:
-
Michel Dänzer authored
Corresponding to the previous non-DC change. Signed-off-by:
-
Michel Dänzer authored
Hardcoding the maximum numbers could result in spurious error messages from the IRQ state callbacks, e.g. on Polaris 11/12: [drm:dce_v11_0_set_pageflip_irq_state [amdgpu]] *ERROR* invalid pageflip crtc 5 [drm:amdgpu_irq_disable_all [amdgpu]] *ERROR* error disabling interrupt (-22) Signed-off-by:
-
Harry Wentland authored
We'll need it in amdgpu_dm_mst_types.c as well. Signed-off-by:
Harry Wentland <harry.wentland@amd.com> Reviewed-by:
Tony Cheng <Tony.Cheng@amd.com> Acked-by:
Andrey Grodzovsky <andrey.grodzovsky@amd.com> Signed-off-by:
-
Tony Cheng authored
Signed-off-by:
-
Charlene Liu authored
Signed-off-by:
Charlene Liu <charlene.liu@amd.com> Reviewed-by:
Krunoslav Kovac <Krunoslav.Kovac@amd.com> Acked-by:
-
Yue Hin Lau authored
fix memory leak Signed-off-by:
Yue Hin Lau <Yuehin.Lau@amd.com> Reviewed-by:
-
Anthony Koo authored
Need to move HDR Metadata from Surface to Stream since there is only one infoframe possible per stream. Also cleaning up some duplicate definitions. Signed-off-by:
Anthony Koo <anthony.koo@amd.com> Reviewed-by:
-
SivapiriyanKumarasamy authored
Signed-off-by:
SivapiriyanKumarasamy <sivapiriyan.kumarasamy@amd.com> Reviewed-by:
-
Yue Hin Lau authored
instantiating new structure hubbub in resource.c Signed-off-by:
-
Tony Cheng authored
Signed-off-by:
-
Yongqiang Sun authored
Add disable ttu interface to dcn10, when remove mpc, disable ttu as well. Signed-off-by:
Yongqiang Sun <yongqiang.sun@amd.com> Reviewed-by:
-
Tony Cheng authored
Signed-off-by:
-
Yongqiang Sun authored
Since front end is programmed before back end programming, no need to reset front end in back end programming. Signed-off-by:
-
Yongqiang Sun authored
front end is initialized during init_hw, but not power gated. There are some left over valuse and will cause some diags test failed. Power gated all front end pipes will make sure every test has same starting point. Signed-off-by:
-
Andrew Jiang authored
We should be sticking with the default clock values if the values obtained from PPLib are bogus. Signed-off-by:
Andrew Jiang <Andrew.Jiang@amd.com> Reviewed-by:
-
Yue Hin Lau authored
moving hubbub functions to new file Signed-off-by:
-
Leo (Sunpeng) Li authored
Abandon new_crtcs array and use for_each_new iterator to acquire new crtcs. Signed-off-by:
Leo (Sunpeng) Li <sunpeng.li@amd.com> Reviewed-by:
-
Leo (Sunpeng) Li authored
For better readability. Signed-off-by:
Bhawanpreet Lakha <Bhawanpreet.Lakha@amd.com> Acked-by:
-
Christian König authored
The GTT manager handles the GART address space anyway, so it is completely pointless to keep the same information around twice. v2: rebased Signed-off-by:
Chunming Zhou <david1.zhou@amd.com> Signed-off-by:
-
Christian König authored
Rename amdgpu_gtt_mgr_is_allocated() to amdgpu_gtt_mgr_has_gart_addr() and use that instead. v2: rename the function as well. Signed-off-by:
-
Monk Liu authored
Signed-off-by:
-
Monk Liu authored
for SR-IOV when doing gpu reset this routine shouldn't do resource allocating otherwise memory leak Signed-off-by:
-
Monk Liu authored
1,no sriov check since gpu recover is unified 2,need CPU_ACCESS_REQUIRED flag for VRAM if SRIOV because otherwise after following PIN the first allocated VRAM bo is wasted due to some TTM mgr reason. Signed-off-by:
-
Monk Liu authored
since now gpu reset is unified with gpu_recover for both bare-metal and SR-IOV: 1)rename in_sriov_reset to in_gpu_reset 2)move lock_reset from adev->virt to adev Signed-off-by:
-
Monk Liu authored
1,new imple names amdgpu_gpu_recover which gives more hint on what it does compared with gpu_reset 2,gpu_recover unify bare-metal and SR-IOV, only the asic reset part is implemented differently 3,gpu_recover will increase hang job karma and mark its entity/context as guilty if exceeds limit V2: 4,in scheduler main routine the job from guilty context will be immedialy fake signaled after it poped from queue and its fence be set with "-ECANCELED" error 5,in scheduler recovery routine all jobs from the guilty entity would be dropped 6,in run_job() routine the real IB submission would be skipped if @skip parameter equales true or there was VRAM lost occured. V3: 7,replace deprecated gpu reset, use new gpu recover Signed-off-by:
-
Monk Liu authored
jobs are skipped under two cases 1)when the entity behind this job marked guilty, the job poped from this entity's queue will be dropped in sched_main loop. 2)in job_recovery(), skip the scheduling job if its karma detected above limit, and also skipped as well for other jobs sharing the same fence context. this approach is becuase job_recovery() cannot access job->entity due to entity may already dead. v2: some logic fix v3: when entity detected guilty, don't drop the job in the poping stage, instead set its fence error as -ECANCELED in run_job(), skip the scheduling either:1) fence->error < 0 or 2) there was a VRAM LOST occurred on this job. this way we can unify the job skipping logic. with this feature we can introduce new gpu recover feature. Signed-off-by:
-
