linux-2.6-microblaze.git
6 years agodrm/ttm: consistently use reservation_object_unlock
Christian König [Wed, 8 Nov 2017 13:38:34 +0000 (14:38 +0100)]
drm/ttm: consistently use reservation_object_unlock

Instead of having a confusing wrapper or call the underlying ww_mutex
function directly.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-and-Tested-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/ttm: move unlocking out of ttm_bo_cleanup_memtype_use
Christian König [Wed, 8 Nov 2017 11:37:35 +0000 (12:37 +0100)]
drm/ttm: move unlocking out of ttm_bo_cleanup_memtype_use

Needed for the next patch and makes the code quite a bit easier to
understand.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-and-Tested-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/scheduler: add WARN_ON for s_fence->parent
Chunming Zhou [Tue, 7 Nov 2017 02:27:43 +0000 (10:27 +0800)]
drm/amd/scheduler: add WARN_ON for s_fence->parent

Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/scheduler: fix page protection of cb
Chunming Zhou [Tue, 7 Nov 2017 02:40:00 +0000 (10:40 +0800)]
drm/amd/scheduler: fix page protection of cb

We must remove the fence callback.

Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: revise retry init to fully cleanup driver
Pixel Ding [Wed, 8 Nov 2017 03:03:14 +0000 (11:03 +0800)]
drm/amdgpu: revise retry init to fully cleanup driver

Retry at drm_dev_register instead of amdgpu_device_init.

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Pixel Ding <Pixel.Ding@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agoamdgpu/dm: Remove fb_location form fill_plane_attributes
Harry Wentland [Wed, 8 Nov 2017 14:58:59 +0000 (09:58 -0500)]
amdgpu/dm: Remove fb_location form fill_plane_attributes

We no longer set the framebuffer address here so this is now
dead code.

Signed-off-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu:fix gpu recover missing skipping(v2)
Monk Liu [Wed, 8 Nov 2017 06:35:04 +0000 (14:35 +0800)]
drm/amdgpu:fix gpu recover missing skipping(v2)

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>
6 years agodrm/amdgpu:read VRAMLOST from gim
Monk Liu [Mon, 30 Oct 2017 12:11:54 +0000 (20:11 +0800)]
drm/amdgpu:read VRAMLOST from gim

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: bypass FB resizing for SRIOV VF
pding [Tue, 7 Nov 2017 03:02:00 +0000 (11:02 +0800)]
drm/amdgpu: bypass FB resizing for SRIOV VF

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: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: release exclusive mode after hw_init
pding [Mon, 6 Nov 2017 03:21:55 +0000 (11:21 +0800)]
drm/amdgpu: release exclusive mode after hw_init

Signed-off-by: pding <Pixel.Ding@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdkfd: initialise kfd inside amdgpu_device_init
pding [Mon, 6 Nov 2017 02:21:26 +0000 (10:21 +0800)]
drm/amdkfd: initialise kfd inside amdgpu_device_init

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: pding <Pixel.Ding@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: don't use ttm_bo_move_ttm in amdgpu_ttm_bind v2
Christian König [Thu, 26 Oct 2017 15:54:12 +0000 (17:54 +0200)]
drm/amdgpu: don't use ttm_bo_move_ttm in amdgpu_ttm_bind v2

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 <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: rename amdgpu_ttm_bind to amdgpu_ttm_alloc_gart
Christian König [Fri, 27 Oct 2017 13:43:14 +0000 (15:43 +0200)]
drm/amdgpu: rename amdgpu_ttm_bind to amdgpu_ttm_alloc_gart

We actually don't bind here, but rather allocate GART space if necessary.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: switch to use new SOC15 reg read/write macros for soc15 ih
Hawking Zhang [Thu, 2 Nov 2017 09:53:37 +0000 (17:53 +0800)]
drm/amdgpu: switch to use new SOC15 reg read/write macros for soc15 ih

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: resize VRAM BAR for CPU access v6
Christian König [Tue, 28 Feb 2017 09:36:43 +0000 (10:36 +0100)]
drm/amdgpu: resize VRAM BAR for CPU access v6

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: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: refine SR-IOV firmware VRAM reservation to protect data
Horace Chen [Wed, 1 Nov 2017 11:32:11 +0000 (19:32 +0800)]
drm/amdgpu: refine SR-IOV firmware VRAM reservation to protect data

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: monk liu <monk.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: retry init if exclusive mode request is failed
pding [Mon, 30 Oct 2017 06:07:24 +0000 (14:07 +0800)]
drm/amdgpu: retry init if exclusive mode request is failed

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 <Pixel.Ding@amd.com>
Reviewed-by: Xiangliang Yu <Xiangliang.Yu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: return error when sriov access requests get timeout
pding [Mon, 30 Oct 2017 05:51:54 +0000 (13:51 +0800)]
drm/amdgpu: return error when sriov access requests get timeout

Reported-by: Sun Gary <Gary.Sun@amd.com>
Signed-off-by: pding <Pixel.Ding@amd.com>
Reviewed-by: Xiangliang Yu <Xiangliang.Yu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Remove fb_location parameter from get_fb_info
Michel Dänzer [Thu, 26 Oct 2017 14:56:35 +0000 (16:56 +0200)]
drm/amd/display: Remove fb_location parameter from get_fb_info

It's dead code.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agoamdgpu: Remove AMDGPU_{HPD,CRTC_IRQ,PAGEFLIP_IRQ}_LAST
Michel Dänzer [Tue, 24 Oct 2017 09:15:52 +0000 (11:15 +0200)]
amdgpu: Remove AMDGPU_{HPD,CRTC_IRQ,PAGEFLIP_IRQ}_LAST

Not used anymore.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Use real number of CRTCs and HPDs in set_irq_funcs
Michel Dänzer [Tue, 24 Oct 2017 09:13:34 +0000 (11:13 +0200)]
drm/amd/display: Use real number of CRTCs and HPDs in set_irq_funcs

Corresponding to the previous non-DC change.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agoamdgpu/dce: Use actual number of CRTCs and HPDs in set_irq_funcs
Michel Dänzer [Tue, 24 Oct 2017 09:03:21 +0000 (11:03 +0200)]
amdgpu/dce: Use actual number of CRTCs and HPDs in set_irq_funcs

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: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Move conn_state to header
Harry Wentland [Thu, 19 Oct 2017 18:28:23 +0000 (14:28 -0400)]
drm/amd/display: Move conn_state to header

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: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: dal 3.1.10
Tony Cheng [Wed, 25 Oct 2017 04:15:05 +0000 (00:15 -0400)]
drm/amd/display: dal 3.1.10

Signed-off-by: Tony Cheng <tony.cheng@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: correct DP is always in full range or bt609
Charlene Liu [Tue, 24 Oct 2017 22:19:06 +0000 (18:19 -0400)]
drm/amd/display: correct DP is always in full range or bt609

Signed-off-by: Charlene Liu <charlene.liu@amd.com>
Reviewed-by: Krunoslav Kovac <Krunoslav.Kovac@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: fix bug from last commit for hubbub
Yue Hin Lau [Tue, 24 Oct 2017 15:40:50 +0000 (11:40 -0400)]
drm/amd/display: fix bug from last commit for hubbub

fix memory leak

Signed-off-by: Yue Hin Lau <Yuehin.Lau@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Move hdr_metadata from plane to stream
Anthony Koo [Mon, 23 Oct 2017 21:02:02 +0000 (17:02 -0400)]
drm/amd/display: Move hdr_metadata from plane to stream

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: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Apply VQ adjustments in MPO case
SivapiriyanKumarasamy [Thu, 19 Oct 2017 17:41:30 +0000 (13:41 -0400)]
drm/amd/display: Apply VQ adjustments in MPO case

Signed-off-by: SivapiriyanKumarasamy <sivapiriyan.kumarasamy@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: create new structure for hubbub
Yue Hin Lau [Mon, 23 Oct 2017 20:01:36 +0000 (16:01 -0400)]
drm/amd/display: create new structure for hubbub

instantiating new structure hubbub in resource.c

Signed-off-by: Yue Hin Lau <Yuehin.Lau@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: dal 3.1.09
Tony Cheng [Sat, 21 Oct 2017 22:32:15 +0000 (18:32 -0400)]
drm/amd/display: dal 3.1.09

Signed-off-by: Tony Cheng <tony.cheng@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Added disconnect dchub.
Yongqiang Sun [Sat, 21 Oct 2017 19:52:40 +0000 (15:52 -0400)]
drm/amd/display: Added disconnect dchub.

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 <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: dal 3.1.08
Tony Cheng [Fri, 20 Oct 2017 18:27:47 +0000 (14:27 -0400)]
drm/amd/display: dal 3.1.08

Signed-off-by: Tony Cheng <tony.cheng@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Not reset front end when program back end.
Yongqiang Sun [Tue, 17 Oct 2017 18:24:28 +0000 (14:24 -0400)]
drm/amd/display: Not reset front end when program back end.

Since front end is programmed before back end programming,
no need to reset front end in back end programming.

Signed-off-by: Yongqiang Sun <yongqiang.sun@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Power down front end in init_hw.
Yongqiang Sun [Fri, 20 Oct 2017 18:12:35 +0000 (14:12 -0400)]
drm/amd/display: Power down front end in init_hw.

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: Yongqiang Sun <yongqiang.sun@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Reject PPLib clock values if they are invalid
Andrew Jiang [Thu, 19 Oct 2017 18:43:36 +0000 (14:43 -0400)]
drm/amd/display: Reject PPLib clock values if they are invalid

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: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: create new files for hubbub functions
Yue Hin Lau [Wed, 18 Oct 2017 15:59:42 +0000 (11:59 -0400)]
drm/amd/display: create new files for hubbub functions

moving hubbub functions to new file

Signed-off-by: Yue Hin Lau <Yuehin.Lau@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Complete TODO item: use new DRM iterator
Leo (Sunpeng) Li [Wed, 18 Oct 2017 19:32:41 +0000 (15:32 -0400)]
drm/amd/display: Complete TODO item: use new DRM iterator

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: Harry Wentland <Harry.Wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Fix styling of freesync code in commit_tail
Leo (Sunpeng) Li [Wed, 18 Oct 2017 18:31:08 +0000 (14:31 -0400)]
drm/amd/display: Fix styling of freesync code in commit_tail

For better readability.

Signed-off-by: Leo (Sunpeng) Li <sunpeng.li@amd.com>
Reviewed-by: Bhawanpreet Lakha <Bhawanpreet.Lakha@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: move GART recovery into GTT manager v2
Christian König [Mon, 16 Oct 2017 14:50:32 +0000 (16:50 +0200)]
drm/amdgpu: move GART recovery into GTT manager v2

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: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: nuke amdgpu_ttm_is_bound() v2
Christian König [Fri, 27 Oct 2017 12:17:09 +0000 (14:17 +0200)]
drm/amdgpu: nuke amdgpu_ttm_is_bound() v2

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: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu:fix random missing of FLR NOTIFY
Monk Liu [Tue, 24 Oct 2017 07:10:11 +0000 (15:10 +0800)]
drm/amdgpu:fix random missing of FLR NOTIFY

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu/sriov:fix memory leak in psp_load_fw
Monk Liu [Tue, 19 Sep 2017 07:40:56 +0000 (15:40 +0800)]
drm/amdgpu/sriov:fix memory leak in psp_load_fw

for SR-IOV when doing gpu reset this routine shouldn't do
resource allocating otherwise memory leak

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu:cleanup ucode_init_bo
Monk Liu [Tue, 17 Oct 2017 09:51:41 +0000 (17:51 +0800)]
drm/amdgpu:cleanup ucode_init_bo

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 <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu:cleanup in_sriov_reset and lock_reset
Monk Liu [Tue, 17 Oct 2017 07:11:12 +0000 (15:11 +0800)]
drm/amdgpu:cleanup in_sriov_reset and lock_reset

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 <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu:implement new GPU recover(v3)
Monk Liu [Wed, 25 Oct 2017 08:37:02 +0000 (16:37 +0800)]
drm/amdgpu:implement new GPU recover(v3)

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 <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agoamd/scheduler:imple job skip feature(v3)
Monk Liu [Wed, 25 Oct 2017 08:21:08 +0000 (16:21 +0800)]
amd/scheduler:imple job skip feature(v3)

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: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: fix indentation in amdgpu_display.h
Christian König [Mon, 23 Oct 2017 14:45:49 +0000 (16:45 +0200)]
drm/amdgpu: fix indentation in amdgpu_display.h

That was somehow completely of.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Only add stream to freesync when modeset required
Leo (Sunpeng) Li [Wed, 18 Oct 2017 18:22:23 +0000 (14:22 -0400)]
drm/amd/display: Only add stream to freesync when modeset required

This is a follow-up patch to:

Leo (Sunpeng) Li    Cleanup code that enables freesync

We should only add a stream to freesync if a modeset was requested, so
we don't fill the core freesync map with pointless streams.

Signed-off-by: Leo (Sunpeng) Li <sunpeng.li@amd.com>
Reviewed-by: Harry Wentland <Harry.Wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Cleanup code that enables freesync
Leo (Sunpeng) Li [Tue, 17 Oct 2017 16:42:58 +0000 (12:42 -0400)]
drm/amd/display: Cleanup code that enables freesync

This is a follow-up patch to:

Bhawanpreet Lakha    Atomic freesync ASSERT fix

Changes:
- Combine to use one iterator
- Use new DRM iterators.

Signed-off-by: Leo (Sunpeng) Li <sunpeng.li@amd.com>
Reviewed-by: Bhawanpreet Lakha <Bhawanpreet.Lakha@amd.com>
Acked-by: Harry Wentland <Harry.Wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Atomic freesync ASSERT fix
Bhawanpreet Lakha [Mon, 16 Oct 2017 19:51:00 +0000 (15:51 -0400)]
drm/amd/display: Atomic freesync ASSERT fix

Changes to atomic set property for freesync.

Now In set property, just set the freesync variables and return 0.
Based on the variables call mod_freesync_set_user_enable() inside
commit_tail

Signed-off-by: Bhawanpreet Lakha <Bhawanpreet.Lakha@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <Harry.Wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: add mod_freesync_user_enable to dm_connector_state
Alex Deucher [Mon, 4 Dec 2017 20:49:59 +0000 (15:49 -0500)]
drm/amd/display: add mod_freesync_user_enable to dm_connector_state

We don't currently expose variable refresh rate, but add the state to the
connector state to make it easier to maintain the support for it from a
hw support perspective while we figure out the uapi for drm.

Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: DMCU FW loading from PSP
Anthony Koo [Wed, 18 Oct 2017 20:14:40 +0000 (16:14 -0400)]
drm/amd/display: DMCU FW loading from PSP

Update the programming sequence to allow DMCU firmware to be loaded by
PSP. This code detects whether the firmware is loaded and does a check
to verify the expected interface version and checks for correct response
from micro controller.

Added registry key method to allow force loading of firmware from kernel
mode driver for test purposes. This is old method of firmware loading
without PSP.

Moved some init sequences into dc/dmcu.

Changed loading sequence to initialize IRAM after firmware completely
loaded. Firmware will now disable features that use IRAM
until initialized.

Signed-off-by: Anthony Koo <anthony.koo@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Set cursor position as per address.
Yongqiang Sun [Wed, 18 Oct 2017 15:59:13 +0000 (11:59 -0400)]
drm/amd/display: Set cursor position as per address.

In case of pipe split, cursor position should also be
programmed as per cursor address.

Signed-off-by: Yongqiang Sun <yongqiang.sun@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Explicit casting for grph object ids
Harry Wentland [Thu, 12 Oct 2017 14:23:22 +0000 (10:23 -0400)]
drm/amd/display: Explicit casting for grph object ids

C++ compilers don't like the implicit conversion

Signed-off-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Andrey Grodzovsky <Andrey.Grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: cleaning up hubp for dcn
Yue Hin Lau [Thu, 12 Oct 2017 18:46:38 +0000 (14:46 -0400)]
drm/amd/display: cleaning up hubp for dcn

removing duplicate functions, renaming struct mi_regs, etc.

Signed-off-by: Yue Hin Lau <Yuehin.Lau@amd.com>
Reviewed-by: Andrey Grodzovsky <Andrey.Grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Clean some unneeded defines from ddc_service_types.h
Harry Wentland [Fri, 8 Sep 2017 14:41:46 +0000 (10:41 -0400)]
drm/amd/display: Clean some unneeded defines from ddc_service_types.h

Remove defines we no longer need

Signed-off-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Andrey Grodzovsky <Andrey.Grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Cleanup some fail labels in dcn10_resource
Harry Wentland [Thu, 28 Sep 2017 19:39:41 +0000 (15:39 -0400)]
drm/amd/display: Cleanup some fail labels in dcn10_resource

Signed-off-by: Harry Wentland <harry.wentland@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Adding DCN1 registers
Mikita Lipski [Tue, 17 Oct 2017 14:53:43 +0000 (10:53 -0400)]
drm/amd/display: Adding DCN1 registers

Registers added to definition list that are required
for multi display synchronization

Signed-off-by: Mikita Lipski <mikita.lipski@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <Harry.Wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Check cursor address before program.
Yongqiang Sun [Tue, 17 Oct 2017 18:54:58 +0000 (14:54 -0400)]
drm/amd/display: Check cursor address before program.

Program cursor attributes during set mode is only needed
in case of pipe slipt, bottom pipe should be programmed
same as top pipe. Need to program if address is 0.

Signed-off-by: Yongqiang Sun <yongqiang.sun@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <Harry.Wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Move lock to front end program.
Yongqiang Sun [Tue, 17 Oct 2017 18:48:11 +0000 (14:48 -0400)]
drm/amd/display: Move lock to front end program.

Moved lock and unlock to apply_ctx_to_surface, since all
the front end programming is within apply_ctx_to_surface.

Signed-off-by: Yongqiang Sun <yongqiang.sun@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <Harry.Wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/display: Add APU cap in dc_caps
Anthony Koo [Mon, 16 Oct 2017 14:43:59 +0000 (10:43 -0400)]
drm/amd/display: Add APU cap in dc_caps

Some features should only be enabled on APUs or should not
be enabled on APUs.

Signed-off-by: Anthony Koo <anthony.koo@amd.com>
Reviewed-by: Tony Cheng <Tony.Cheng@amd.com>
Acked-by: Harry Wentland <Harry.Wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: delete duplicated code.
Rex Zhu [Fri, 20 Oct 2017 06:52:43 +0000 (14:52 +0800)]
drm/amdgpu: delete duplicated code.

the variable ref_clock was assigned same
value twice in same function.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/pp: delete an outdated comment in amd_powerplay.c
Rex Zhu [Mon, 9 Oct 2017 03:30:42 +0000 (11:30 +0800)]
drm/amd/pp: delete an outdated comment in amd_powerplay.c

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/pp: implement notify_smu_memory_info on Powerplay
Rex Zhu [Mon, 9 Oct 2017 04:22:21 +0000 (12:22 +0800)]
drm/amd/pp: implement notify_smu_memory_info on Powerplay

Used to set up smu power logging.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: add new pp function point notify_smu_memory_info
Rex Zhu [Fri, 15 Sep 2017 10:34:42 +0000 (18:34 +0800)]
drm/amdgpu: add new pp function point notify_smu_memory_info

Used to set up smu power logging.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: add header kgd_pp_interface.h
Rex Zhu [Fri, 29 Sep 2017 07:58:19 +0000 (15:58 +0800)]
drm/amdgpu: add header kgd_pp_interface.h

move powerplay and amdgpu shared structures
and definitions to kgd_pp_interface.h.  This
is the interface between the base driver
and powerplay.

Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: move struct amd_powerplay to amdgpu.h
Rex Zhu [Fri, 29 Sep 2017 08:07:14 +0000 (16:07 +0800)]
drm/amdgpu: move struct amd_powerplay to amdgpu.h

Clean up the interface.

Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/powerplay: move functions to amd_pm_funcs table
Rex Zhu [Thu, 26 Oct 2017 03:55:23 +0000 (23:55 -0400)]
drm/amd/powerplay: move functions to amd_pm_funcs table

those functions are exported to DC

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: move DC and PP shared data structures to dm_pp_interface.h
Rex Zhu [Thu, 28 Sep 2017 07:20:51 +0000 (15:20 +0800)]
drm/amdgpu: move DC and PP shared data structures to dm_pp_interface.h

Move the display/power interfaces to one place.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: remove extra parameter from amdgpu_ttm_bind() v2
Christian König [Mon, 16 Oct 2017 15:24:21 +0000 (17:24 +0200)]
drm/amdgpu: remove extra parameter from amdgpu_ttm_bind() v2

We always use the BO mem now.

v2: minor rebase

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: don't wait interruptible while binding GART space
Christian König [Wed, 25 Oct 2017 19:37:35 +0000 (21:37 +0200)]
drm/amdgpu: don't wait interruptible while binding GART space

Display can't seem to handle this correctly.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: fix pin domain compatibility check
Christian König [Mon, 23 Oct 2017 15:29:36 +0000 (17:29 +0200)]
drm/amdgpu: fix pin domain compatibility check

We need to test if any domain fits, not all of them.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: always bind pinned BOs
Christian König [Fri, 20 Oct 2017 11:12:12 +0000 (13:12 +0200)]
drm/amdgpu: always bind pinned BOs

We always need to bind pinned BOs, not just when the caller requested the
address.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: use the actual placement for pin accounting
Christian König [Fri, 20 Oct 2017 11:11:00 +0000 (13:11 +0200)]
drm/amdgpu: use the actual placement for pin accounting

This allows us to specify multiple possible placements again.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: retry init if it fails due to exclusive mode timeout (v3)
pding [Mon, 23 Oct 2017 09:22:09 +0000 (17:22 +0800)]
drm/amdgpu: retry init if it fails due to exclusive mode timeout (v3)

The exclusive mode has real-time limitation in reality, such like being
done in 300ms. It's easy observed if running many VF/VMs in single host
with heavy CPU workload.

If we find the init fails due to exclusive mode timeout, try it again.

v2:
 - rewrite the condition for readable value.

v3:
 - fix typo, add comments for sleep

Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: pding <Pixel.Ding@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu/virt: implement wait_reset callbacks for vi/ai
pding [Tue, 24 Oct 2017 01:53:16 +0000 (09:53 +0800)]
drm/amdgpu/virt: implement wait_reset callbacks for vi/ai

Reviewed-by: Monk Liu <monk.liu@amd.com>
Signed-off-by: pding <Pixel.Ding@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/powerplay: describe the PCIE link speed in right GT/s
Evan Quan [Thu, 26 Oct 2017 09:29:34 +0000 (17:29 +0800)]
drm/amd/powerplay: describe the PCIE link speed in right GT/s

Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu/virt: add wait_reset virt ops
pding [Tue, 24 Oct 2017 01:51:04 +0000 (09:51 +0800)]
drm/amdgpu/virt: add wait_reset virt ops

Driver can use this interface to check if there's a function level
reset done in hypervisor. It's helpful when IRQ handler for reset
is not ready, or special handling is required.

Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Monk Liu <monk.liu@amd.com>
Signed-off-by: pding <Pixel.Ding@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu/virt: add function to check MMIO (v2)
pding [Tue, 24 Oct 2017 02:01:13 +0000 (10:01 +0800)]
drm/amdgpu/virt: add function to check MMIO (v2)

MMIO space can be blocked on virtualised device. Add this
function to check if MMIO is blocked or not.

Todo: need a reliable method such like communation
with hypervisor.

v2:
 - add comments inline

Signed-off-by: pding <Pixel.Ding@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: avoid soft lockup when waiting for RLC serdes (v2)
pding [Mon, 23 Oct 2017 08:31:04 +0000 (16:31 +0800)]
drm/amdgpu: avoid soft lockup when waiting for RLC serdes (v2)

Normally all waiting get timeout if there's one.
Release the lock and return immediately when timeout happens.

v2:
 - set the se_sh to broadcase before return

Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: pding <Pixel.Ding@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: change redundant init logs to debug level
pding [Thu, 26 Oct 2017 01:30:38 +0000 (09:30 +0800)]
drm/amdgpu: change redundant init logs to debug level

When this VF stays in exclusive mode for long, other VFs will be
impacted.

The redundant messages causes exclusive mode timeout when they're
redirected. That is a normal use case for cloud service to redirect
guest log to virtual serial port.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: pding <Pixel.Ding@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu:implement ctx query2
Monk Liu [Tue, 17 Oct 2017 06:58:01 +0000 (14:58 +0800)]
drm/amdgpu:implement ctx query2

this query will give flag bits to indicate what happend
on the given context

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu:don't change ctx->reset_couner upon query
Monk Liu [Tue, 17 Oct 2017 06:39:23 +0000 (14:39 +0800)]
drm/amdgpu:don't change ctx->reset_couner upon query

reset_counter marks the reset counter number once the context
is created, shouldn't be changed due to query.

To keep U/K interface on the ctx_query and keep ctx's reset_counter
logic compatible with GPU RESET feature, now use another var named
"reset_counter_query" to replace the original checked & updated in
amdgpu_ctx_query.

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: Remove job->s_entity to avoid keeping reference to stale pointer.
Andrey Grodzovsky [Tue, 24 Oct 2017 17:30:16 +0000 (13:30 -0400)]
drm/amdgpu: Remove job->s_entity to avoid keeping reference to stale pointer.

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: Fix deadlock during GPU reset.
Andrey Grodzovsky [Thu, 12 Oct 2017 20:46:26 +0000 (16:46 -0400)]
drm/amdgpu: Fix deadlock during GPU reset.

Bug:
Kfifo is limited at size, during GPU reset it would fill up to limit
and the pushing thread (producer) would wait for the scheduler worker to
consume the items in the fifo while holding reservation lock
on a BO. The gpu reset thread on the other hand blocks the scheduler
during reset. Before it unblocks the sceduler it might want
to recover VRAM and so will try to reserve the same BO the producer
thread is already holding creating a deadlock.

Fix:
Switch from kfifo to SPSC queue which is unlimited in size.

Signed-off-by: Andrey Grodzovsky <Andrey.Grodzovsky@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: Add SPSC queue to scheduler.
Andrey Grodzovsky [Thu, 12 Oct 2017 20:41:39 +0000 (16:41 -0400)]
drm/amdgpu: Add SPSC queue to scheduler.

It is intended to sabstitute the bounded fifo we are currently
using.

Signed-off-by: Andrey Grodzovsky <Andrey.Grodzovsky@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu:cleanup job reset routine(v2)
Monk Liu [Mon, 16 Oct 2017 11:46:43 +0000 (19:46 +0800)]
drm/amdgpu:cleanup job reset routine(v2)

merge the setting guilty on context into this function
to avoid implement extra routine.

v2:
go through entity list and compare the fence_ctx
before operate on the entity, otherwise the entity
may be just a wild pointer

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Chunming Zhou <David1.Zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu:skip job for guilty ctx in parser_init
Monk Liu [Tue, 17 Oct 2017 04:08:02 +0000 (12:08 +0800)]
drm/amdgpu:skip job for guilty ctx in parser_init

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu:pass ctx->guilty address to entity init
Monk Liu [Mon, 23 Oct 2017 04:25:24 +0000 (12:25 +0800)]
drm/amdgpu:pass ctx->guilty address to entity init

this way the real interested guilty is connected to entity->guilty
pointer, and we can use entity->pointer later in gpu recovery procedure

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Chunming Zhou <David1.Zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/scheduler:introduce guilty pointer member
Monk Liu [Mon, 23 Oct 2017 04:23:29 +0000 (12:23 +0800)]
drm/amd/scheduler:introduce guilty pointer member

this member will be used later, it will points to
the real var inside of context and CS_SUBMIT & gpu schdduler
can decide if skip a job depends on context->guilty or *entity->guilty

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Chunming Zhou <David1.Zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu:add hang_limit for sched(v2)
Monk Liu [Tue, 17 Oct 2017 05:40:54 +0000 (13:40 +0800)]
drm/amdgpu:add hang_limit for sched(v2)

since gpu_scheduler source domain cannot access amdgpu variable
so need create the hang_limit membewr for sched, and it can
refer it for the upcoming GPU RESET patches

v2:
make hang_limit a parameter of sched_init()

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Chunming Zhou <David1.Zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu:cleanup force_completion
Monk Liu [Mon, 16 Oct 2017 06:38:10 +0000 (14:38 +0800)]
drm/amdgpu:cleanup force_completion

cleanups, now only operate on the given ring

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: Avoid accessing job->entity after the job is scheduled.
Andrey Grodzovsky [Thu, 19 Oct 2017 18:29:46 +0000 (14:29 -0400)]
drm/amdgpu: Avoid accessing job->entity after the job is scheduled.

Bug: amdgpu_job_free_cb was accessing s_job->s_entity when the allocated
amdgpu_ctx (and the entity inside it) were already deallocated from
amdgpu_cs_parser_fini.

Fix: Save job's priority on it's creation instead of accessing it from
s_entity later on.

Signed-off-by: Andrey Grodzovsky <Andrey.Grodzovsky@amd.com>
Reviewed-by: Andres Rodriguez <andresx7@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agoMerge tag 'drm-intel-next-2017-11-17-1' of git://anongit.freedesktop.org/drm/drm...
Dave Airlie [Sun, 3 Dec 2017 23:40:35 +0000 (09:40 +1000)]
Merge tag 'drm-intel-next-2017-11-17-1' of git://anongit.freedesktop.org/drm/drm-intel into drm-next

More change sets for 4.16:

- Many improvements for selftests and other igt tests (Chris)
- Forcewake with PUNIT->PMIC bus fixes and robustness (Hans)
- Define an engine class for uABI (Tvrtko)
- Context switch fixes and improvements (Chris)
- GT powersavings and power gating simplification and fixes (Chris)
- Other general driver clean-ups (Chris, Lucas, Ville)
- Removing old, useless and/or bad workarounds (Chris, Oscar, Radhakrishna)
- IPS, pipe config, etc in preparation for another Fast Boot attempt (Maarten)
- OA perf fixes and support to Coffee Lake and Cannonlake (Lionel)
- Fixes around GPU fault registers (Michel)
- GEM Proxy (Tina)
- Refactor of Geminilake and Cannonlake plane color handling (James)
- Generalize transcoder loop (Mika Kahola)
- New HW Workaround for Cannonlake and Geminilake (Rodrigo)
- Resume GuC before using GEM (Chris)
- Stolen Memory handling improvements (Ville)
- Initialize entry in PPAT for older compilers (Chris)
- Other fixes and robustness improvements on execbuf (Chris)
- Improve logs of GEM_BUG_ON (Mika Kuoppala)
- Rework with massive rename of GuC functions and files (Sagar)
- Don't sanitize frame start delay if pipe is off (Ville)
- Cannonlake clock fixes (Rodrigo)
- Cannonlake HDMI 2.0 support (Rodrigo)
- Add a GuC doorbells selftest (Michel)
- Add might_sleep() check to our wait_for() (Chris)

Many GVT changes for 4.16:

- CSB HWSP update support (Weinan)
- GVT debug helpers, dyndbg and debugfs (Chuanxiao, Shuo)
- full virtualized opregion (Xiaolin)
- VM health check for sane fallback (Fred)
- workload submission code refactor for future enabling (Zhi)
- Updated repo URL in MAINTAINERS (Zhenyu)
- other many misc fixes

* tag 'drm-intel-next-2017-11-17-1' of git://anongit.freedesktop.org/drm/drm-intel: (260 commits)
  drm/i915: Update DRIVER_DATE to 20171117
  drm/i915: Add a policy note for removing workarounds
  drm/i915/selftests: Report ENOMEM clearly for an allocation failure
  Revert "drm/i915: Display WA #1133 WaFbcSkipSegments:cnl, glk"
  drm/i915: Calculate g4x intermediate watermarks correctly
  drm/i915: Calculate vlv/chv intermediate watermarks correctly, v3.
  drm/i915: Pass crtc_state to ips toggle functions, v2
  drm/i915: Pass idle crtc_state to intel_dp_sink_crc
  drm/i915: Enable FIFO underrun reporting after initial fastset, v4.
  drm/i915: Mark the userptr invalidate workqueue as WQ_MEM_RECLAIM
  drm/i915: Add might_sleep() check to wait_for()
  drm/i915/selftests: Add a GuC doorbells selftest
  drm/i915/cnl: Extend HDMI 2.0 support to CNL.
  drm/i915/cnl: Simplify dco_fraction calculation.
  drm/i915/cnl: Don't blindly replace qdiv.
  drm/i915/cnl: Fix wrpll math for higher freqs.
  drm/i915/cnl: Fix, simplify and unify wrpll variable sizes.
  drm/i915/cnl: Remove useless conversion.
  drm/i915/cnl: Remove spurious central_freq.
  drm/i915/selftests: exercise_ggtt may have nothing to do
  ...

6 years agoMerge tag 'drm-misc-next-2017-11-30' of git://anongit.freedesktop.org/drm/drm-misc...
Dave Airlie [Sun, 3 Dec 2017 19:38:52 +0000 (05:38 +1000)]
Merge tag 'drm-misc-next-2017-11-30' of git://anongit.freedesktop.org/drm/drm-misc into drm-next

Cross-subsystem Changes:

- device tree doc for the Mitsubishi AA070MC01 and Tianma TM070RVHG71
panels (Lukasz Majewski) and for a 2nd endpoint on stm32 (Philippe Cornu)

Core Changes:

The most important changes are:

- Add drm_driver .last_close and .output_poll_changed helpers to reduce
fbdev emulation footprint in drivers (Noralf)
- Fix plane clipping in core and for vmwgfx (Ville)

Then we have a bunch of of improvement for print and debug such as the
addition of a framebuffer debugfs file. ELD connector, HDMI and
improvements.  And a bunch of misc improvements, clean ups and style
changes and doc updates

[airlied: drop eld bits from amdgpu_dm]

Driver Changes:

- sii8620: filter unsupported modes and add DVI mode support (Maciej Purski)
- rockchip: analogix_dp: Remove unnecessary init code (Jeffy Chen)
- virtio, cirrus: add fb create_handle support to enable screenshots(Lepton Wu)
- virtio: replace reference/unreference with get/put (Aastha Gupta)
- vc4, gma500: Convert timers to use timer_setup() (Kees Cook)
- vc4: Reject HDMI modes with too high of clocks (Eric)
- vc4: Add support for more pixel formats (Dave Stevenson)
- stm: dsi: Rename driver name to "stm32-display-dsi" (Philippe Cornu)
- stm: ltdc: add a 2nd endpoint (Philippe Cornu)
- via: use monotonic time for VIA_WAIT_IRQ (Arnd Bergmann)

* tag 'drm-misc-next-2017-11-30' of git://anongit.freedesktop.org/drm/drm-misc: (96 commits)
  drm/bridge: tc358767: add copyright lines
  MAINTAINERS: change maintainer for Rockchip drm drivers
  drm/vblank: Fix vblank timestamp debugs
  drm/via: use monotonic time for VIA_WAIT_IRQ
  dma-buf: Fix ifnullfree.cocci warnings
  drm/printer: Add drm_vprintf()
  drm/edid: Allow HDMI infoframe without VIC or S3D
  video/hdmi: Allow "empty" HDMI infoframes
  dma-buf/fence: Fix lock inversion within dma-fence-array
  drm/sti: Handle return value of platform_get_irq_byname
  drm/vc4: Add support for NV21 and NV61.
  drm/vc4: Use .pixel_order instead of custom .flip_cbcr
  drm/vc4: Add support for DRM_FORMAT_RGB888 and DRM_FORMAT_BGR888
  drm: Move drm_plane_helper_check_state() into drm_atomic_helper.c
  drm: Check crtc_state->enable rather than crtc->enabled in drm_plane_helper_check_state()
  drm/vmwgfx: Try to fix plane clipping
  drm/vmwgfx: Use drm_plane_helper_check_state()
  drm/vmwgfx: Remove bogus crtc coords vs fb size check
  gpu: gma500: remove unneeded DRIVER_LICENSE #define
  drm: don't link DP aux i2c adapter to the hardware device node
  ...

6 years agoLinux 4.15-rc2
Linus Torvalds [Sun, 3 Dec 2017 16:01:47 +0000 (11:01 -0500)]
Linux 4.15-rc2

6 years agoMerge branch 'fixes' of git://git.armlinux.org.uk/~rmk/linux-arm
Linus Torvalds [Sun, 3 Dec 2017 15:51:08 +0000 (10:51 -0500)]
Merge branch 'fixes' of git://git.armlinux.org.uk/~rmk/linux-arm

Pull ARM fix from Russell King:
 "Just one fix this time around, for the late commit in the merge window
  that triggered a problem with qemu. Qemu is apparently also going to
  receive a fix for the discovered issue"

* 'fixes' of git://git.armlinux.org.uk/~rmk/linux-arm:
  ARM: avoid faulting on qemu

6 years agoMerge branch 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa...
Linus Torvalds [Sun, 3 Dec 2017 15:48:24 +0000 (10:48 -0500)]
Merge branch 'i2c/for-current' of git://git./linux/kernel/git/wsa/linux

Pull i2c fixes from Wolfram Sang:
 "Here are two bugfixes for I2C, fixing a memleak in the core and irq
  allocation for i801.

  Also three bugfixes for the at24 eeprom driver which Bartosz collected
  while taking over maintainership for this driver"

* 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  eeprom: at24: check at24_read/write arguments
  eeprom: at24: fix reading from 24MAC402/24MAC602
  eeprom: at24: correctly set the size for at24mac402
  i2c: i2c-boardinfo: fix memory leaks on devinfo
  i2c: i801: Fix Failed to allocate irq -2147483648 error

6 years agoMerge tag 'hwmon-for-linus-v4.15-rc2' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 3 Dec 2017 15:46:16 +0000 (10:46 -0500)]
Merge tag 'hwmon-for-linus-v4.15-rc2' of git://git./linux/kernel/git/groeck/linux-staging

Pull hwmon fixes from Guenter Roeck:
 "Fixes:

   - Drop reference to obsolete maintainer tree

   - Fix overflow bug in pmbus driver

   - Fix SMBUS timeout problem in jc42 driver

  For the SMBUS timeout handling, we had a brief discussion if this
  should be considered a bug fix or a feature. Peter says "it fixes real
  problems where the application misbehave due to faulty content when
  reading from an eeprom", and he needs the patch in his company's v4.14
  images. This is good enough for me and warrants backport to stable
  kernels"

* tag 'hwmon-for-linus-v4.15-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging:
  hwmon: (jc42) optionally try to disable the SMBUS timeout
  hwmon: (pmbus) Use 64bit math for DIRECT format values
  hwmon: Drop reference to Jean's tree

6 years agoMerge tag 'at24-4.15-fixes-for-wolfram' of git://git.kernel.org/pub/scm/linux/kernel...
Wolfram Sang [Sat, 2 Dec 2017 22:32:13 +0000 (23:32 +0100)]
Merge tag 'at24-4.15-fixes-for-wolfram' of git://git./linux/kernel/git/brgl/linux into i2c/for-current

Please consider pulling the following fixes for v4.15. While it doesn't
fix any regression introduced in the v4.15 merge window, we have a
feature in at24 since linux v4.8 - reading the mac address block from
at24mac series - which turned out to be not working.

This pull request contains changes that fix it together with a patch
that hardens the read and write argument sanitization with
out-of-bounds checks that were missing.