This patch limits the ref_div_max value to 100, during the
calculation of PLL feedback reference divider. With current
value (128), the produced fb_ref_div value generates unstable
output at particular frequencies. Radeon driver limits this
value at 100.
On Oland, when we try to setup mode 2048x1280@60 (a bit weird,
I know), it demands a clock of 221270 Khz. It's been observed
that the PLL calculations using values 128 and 100 are vastly
different, and look like this:
+------------------------------------------+
|Parameter |AMDGPU |Radeon |
| | | |
+-------------+----------------------------+
|Clock feedback | |
|divider max | 128 | 100 |
|cap value | | |
| | | |
| | | |
+------------------------------------------+
|ref_div_max | | |
| | 42 | 20 |
| | | |
| | | |
+------------------------------------------+
|ref_div | 42 | 20 |
| | | |
+------------------------------------------+
|fb_div | 10326 | 8195 |
+------------------------------------------+
|fb_div | 1024 | 163 |
+------------------------------------------+
|fb_dev_p | 4 | 9 |
|frac fb_de^_p| | |
+----------------------------+-------------+
With ref_div_max value clipped at 100, AMDGPU driver can also
drive videmode 2048x1280@60 (221Mhz) and produce proper output
without any blanking and distortion on the screen.
PS: This value was changed from 128 to 100 in Radeon driver also, here:
4b21ce1b4b
V1:
Got acks from:
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
V2:
- Restricting the changes only for OLAND, just to avoid any regression
for other cards.
- Changed unsigned -> unsigned int to make checkpatch quiet.
V3: Apply the change on SI family (not only oland) (Christian)
Cc: Alex Deucher <Alexander.Deucher@amd.com>
Cc: Christian König <christian.koenig@amd.com>
Cc: Eddy Qin <Eddy.Qin@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Shashank Sharma <shashank.sharma@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
schedule_delayed_work does not push back the work if it was already
scheduled before, so amdgpu_device_delay_enable_gfx_off ran ~100 ms
after the first time GFXOFF was disabled and re-enabled, even if GFXOFF
was disabled and re-enabled again during those 100 ms.
This resulted in frame drops / stutter with the upcoming mutter 41
release on Navi 14, due to constantly enabling GFXOFF in the HW and
disabling it again (for getting the GPU clock counter).
To fix this, call cancel_delayed_work_sync when the disable count
transitions from 0 to 1, and only schedule the delayed work on the
reverse transition, not if the disable count was already 0. This makes
sure the delayed work doesn't run at unexpected times, and allows it to
be lock-free.
v2:
* Use cancel_delayed_work_sync & mutex_trylock instead of
mod_delayed_work.
v3:
* Make amdgpu_device_delay_enable_gfx_off lock-free (Christian König)
v4:
* Fix race condition between amdgpu_gfx_off_ctrl incrementing
adev->gfx.gfx_off_req_count and amdgpu_device_delay_enable_gfx_off
checking for it to be 0 (Evan Quan)
Cc: stable@vger.kernel.org
Reviewed-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com> # v3
Acked-by: Christian König <christian.koenig@amd.com> # v3
Signed-off-by: Michel Dänzer <mdaenzer@redhat.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
In some cases when we unload driver, warning call trace
will show up in vram_mgr_fini which claims that LRU is not empty, caused
by the ttm bo inside delay deleted queue.
[How]
We should flush delayed work to make sure the delay deleting is done.
Signed-off-by: YuBiao Wang <YuBiao.Wang@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The TA has a limit to the amount of data that can be retrieved from
GET_TOPOLOGY. For setups that exceed this limit, the xGMI topology
needs to be re-initialized and data needs to be re-fetched from the
extended link records by setting a flag in the shared command buffer.
The number of hops and the number of links must be accumulated by the
driver. Other data points are all fetched from the first request.
Because the TA has already exceeded its link record limit, it
cannot hold bidirectional information. Otherwise the driver would
have to do more than two fetches so the driver has to reflect the
topology information in the opposite direction.
v2: squashed with internal reviewed fix
Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Hawking Zhang <hawking.zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Currently, the readout of fan speed pwm is transited into percent-based
and then pwm-based. However, the transition into percent-based is totally
unnecessary and make the final output less accurate.
Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When guest received FLR notification from host, it would
lock adapter into reset state. There will be no more
job submission and hardware access after that.
Then it should send a response to host that it has prepared
for host reset.
Signed-off-by: Jiange Zhao <Jiange.Zhao@amd.com>
Signed-off-by: Peng Ju Zhou <PengJu.Zhou@amd.com>
Reviewed-by: Emily.Deng <Emily.Deng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
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 <Jingwen.Chen2@amd.com>
Signed-off-by: Jack Zhang <Jack.Zhang7@hotmail.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed by: Monk Liu <monk.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The variable val is declared without initialization, and its address is
passed to amdgpu_i2c_get_byte(). In this function, the value of val is
accessed in:
DRM_DEBUG("i2c 0x%02x 0x%02x read failed\n",
addr, *val);
Also, when amdgpu_i2c_get_byte() returns, val may remain uninitialized,
but it is accessed in:
val &= ~amdgpu_connector->router.ddc_mux_control_pin;
To fix this possible uninitialized-variable access, initialize val to 0 in
amdgpu_i2c_router_select_ddc_port().
Reported-by: TOTE Robot <oslab@tsinghua.edu.cn>
Signed-off-by: Tuo Li <islituo@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There was an "if" statement that did nothing so it was removed.
Signed-off-by: Sergio Miguéns Iglesias <sergio@lony.xyz>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Don't use "begin kernel-doc notation" (/**) for comments that are
not kernel-doc. This eliminates warnings reported by the 0day bot.
drivers/gpu/drm/amd/amdgpu/gfx_v9_4_2.c:89: warning: This comment starts with '/**', but isn't a kernel-doc comment. Refer Documentation/doc-guide/kernel-doc.rst
* This shader is used to clear VGPRS and LDS, and also write the input
drivers/gpu/drm/amd/amdgpu/gfx_v9_4_2.c:209: warning: This comment starts with '/**', but isn't a kernel-doc comment. Refer Documentation/doc-guide/kernel-doc.rst
* The below shaders are used to clear SGPRS, and also write the input
drivers/gpu/drm/amd/amdgpu/gfx_v9_4_2.c:301: warning: This comment starts with '/**', but isn't a kernel-doc comment. Refer Documentation/doc-guide/kernel-doc.rst
* This shader is used to clear the uninitiated sgprs after the above
Fixes: 0e0036c7d1 ("drm/amdgpu: fix no full coverage issue for gprs initialization")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Reported-by: kernel test robot <lkp@intel.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: Christian König <christian.koenig@amd.com>
Cc: "Pan, Xinhui" <Xinhui.Pan@amd.com>
Cc: Dennis Li <Dennis.Li@amd.com>
Cc: amd-gfx@lists.freedesktop.org
Cc: dri-devel@lists.freedesktop.org
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When init failed in early init stage, amdgpu_object has
not been initialized, so hasn't the ttm delayed queue functions.
Signed-off-by: YuBiao Wang <YuBiao.Wang@amd.com>
Reviewed-by: Emily.Deng <Emily.Deng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The varialbe gtt in the function amdgpu_ttm_tt_populate() and
amdgpu_ttm_tt_unpopulate() is guaranteed to be not NULL in the context.
Thus the null-pointer checks are redundant and can be dropped.
Reported-by: TOTE Robot <oslab@tsinghua.edu.cn>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Tuo Li <islituo@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add the MODE register into the per-wave debug information.
This register holds state such as FP rounding and denorm
modes, which exceptions are enabled, and active clamping
modes.
Signed-off-by: Joseph Greathouse <Joseph.Greathouse@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The previous logic is recording the amount of valid vcn instances
to use them on SRIOV, it is a hard task due to the vcn accessment is
based on the index of the vcn instance.
Check if the vcn instance enabled before do instance init.
Signed-off-by: Peng Ju Zhou <PengJu.Zhou@amd.com>
Reviewed-by: Monk Liu <monk.liu@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Move dce_virtual into amdgpu_vkms and update all references to
dce_virtual with amdgpu_vkms.
v2: Removed more references to dce_virtual.
v3: Restored display modes from previous implementation.
Signed-off-by: Ryan Taylor <Ryan.Taylor@amd.com>
Reported-by: kernel test robot <lkp@intel.com>
Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Modify the VKMS driver into an api that dce_virtual can use to create
virtual displays that obey drm's atomic modesetting api.
v2: Made local functions static.
v3: Switched vkms_output kzalloc for kcalloc.
Cleanup patches by moving display mode fixes to this patch.
v4: Update atomic_check and atomic_update to comply with new kms api.
Signed-off-by: Ryan Taylor <Ryan.Taylor@amd.com>
Reported-by: kernel test robot <lkp@intel.com>
Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Duplicate include header file <drm/drm_drv.h>
line 28: #include <drm/drm_drv.h>
line 44: #include <drm/drm_drv.h>
Signed-off-by: zhouchuangao <zhouchuangao@vivo.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
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: 8d35a25961 ("drm/amdgpu: adjust fence driver enable sequence")
Suggested-by: Christian König <christian.koenig@amd.com>
Tested-by: Mike Lothian <mike@fireburn.co.uk>
Signed-off-by: Guchun Chen <guchun.chen@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Fix the channel_index table layout to fetch the correct
channel_index when calculating physical address from
normalized address during page retirement.
Also, fix the number of UMC instances and number of channels
within each UMC instance for Aldebaran.
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Reviewed-By: John Clements <john.clements@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
'pm_suspend_target_state' is only available when CONFIG_PM_SLEEP
is set/enabled. OTOH, when both SUSPEND and HIBERNATION are not set,
PM_SLEEP is not set, so this variable cannot be used.
../drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c: In function ‘amdgpu_acpi_is_s0ix_active’:
../drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c:1046:11: error: ‘pm_suspend_target_state’ undeclared (first use in this function); did you mean ‘__KSYM_pm_suspend_target_state’?
return pm_suspend_target_state == PM_SUSPEND_TO_IDLE;
^~~~~~~~~~~~~~~~~~~~~~~
__KSYM_pm_suspend_target_state
Also use shorter IS_ENABLED(CONFIG_foo) notation for checking the
2 config symbols.
Fixes: 91e273712a ("drm/amdgpu: Check pmops for desired suspend state")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: Christian König <christian.koenig@amd.com>
Cc: "Pan, Xinhui" <Xinhui.Pan@amd.com>
Cc: amd-gfx@lists.freedesktop.org
Cc: dri-devel@lists.freedesktop.org
Cc: linux-next@vger.kernel.org
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If GC has entered CGPG, ringing doorbell > first page doesn't wakeup GC.
Enlarge CP_MEC_DOORBELL_RANGE_UPPER to workaround this issue.
Signed-off-by: Yifan Zhang <yifan1.zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If one GTT BO has been evicted/swapped out, it should sit in CPU domain.
TTM only alloc struct ttm_resource instead of struct ttm_range_mgr_node
for sysMem.
Now when we update mapping for such invalidated BOs, we might walk out
of bounds of struct ttm_resource.
Three possible fix:
1) Let sysMem manager alloc struct ttm_range_mgr_node, like
ttm_range_manager does.
2) Pass pages_addr to update_mapping function too, but need memset
pages_addr[] to zero when unpopulate.
3) Init amdgpu_res_cursor directly.
bug is detected by kfence.
==================================================================
BUG: KFENCE: out-of-bounds read in amdgpu_vm_bo_update_mapping+0x564/0x6e0
Out-of-bounds read at 0x000000008ea93fe9 (64B right of kfence-#167):
amdgpu_vm_bo_update_mapping+0x564/0x6e0 [amdgpu]
amdgpu_vm_bo_update+0x282/0xa40 [amdgpu]
amdgpu_vm_handle_moved+0x19e/0x1f0 [amdgpu]
amdgpu_cs_vm_handling+0x4e4/0x640 [amdgpu]
amdgpu_cs_ioctl+0x19e7/0x23c0 [amdgpu]
drm_ioctl_kernel+0xf3/0x180 [drm]
drm_ioctl+0x2cb/0x550 [drm]
amdgpu_drm_ioctl+0x5e/0xb0 [amdgpu]
kfence-#167 [0x000000008e11c055-0x000000001f676b3e
ttm_sys_man_alloc+0x35/0x80 [ttm]
ttm_resource_alloc+0x39/0x50 [ttm]
ttm_bo_swapout+0x252/0x5a0 [ttm]
ttm_device_swapout+0x107/0x180 [ttm]
ttm_global_swapout+0x6f/0x130 [ttm]
ttm_tt_populate+0xb1/0x2a0 [ttm]
ttm_bo_handle_move_mem+0x17e/0x1d0 [ttm]
ttm_mem_evict_first+0x59d/0x9c0 [ttm]
ttm_bo_mem_space+0x39f/0x400 [ttm]
ttm_bo_validate+0x13c/0x340 [ttm]
ttm_bo_init_reserved+0x269/0x540 [ttm]
amdgpu_bo_create+0x1d1/0xa30 [amdgpu]
amdgpu_bo_create_user+0x40/0x80 [amdgpu]
amdgpu_gem_object_create+0x71/0xc0 [amdgpu]
amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu+0x2f2/0xcd0 [amdgpu]
kfd_ioctl_alloc_memory_of_gpu+0xe2/0x330 [amdgpu]
kfd_ioctl+0x461/0x690 [amdgpu]
Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
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 <Likun.Gao@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
