Am 19.07.2016 um 10:13 schrieb Chunming Zhou:
> Check gpu status first, if MC/VMC/DISPLAY hang, directly triger full reset.
> If engine hangs, then triger engine soft reset, if soft reset fails, will
> fallback to full reset.
>
> Change-Id: I6f946db3624cd950e11e669f5dc80be58dad4711
> Signed-off-by: Chunming Zhou <David1.Zhou at amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 42 +++++++++++++++++++++++++++++-
> 1 file changed, 41 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index 97957ab..4521725 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -1976,7 +1976,8 @@ int amdgpu_pre_soft_reset(struct amdgpu_device *adev)
> for (i = 0; i < adev->num_ip_blocks; i++) {
> if (!adev->ip_block_status[i].valid)
> continue;
> - if (adev->ip_blocks[i].funcs->pre_soft_reset)
> + if (adev->ip_block_status[i].hang &&
> + adev->ip_blocks[i].funcs->pre_soft_reset)
> r = adev->ip_blocks[i].funcs->pre_soft_reset(adev);
> if (r)
> return r;
> @@ -1985,6 +1986,38 @@ int amdgpu_pre_soft_reset(struct amdgpu_device *adev)
> return 0;
> }
>
> +static bool amdgpu_need_full_reset(struct amdgpu_device *adev)
> +{
> + if (adev->ip_block_status[AMD_IP_BLOCK_TYPE_GMC].hang ||
> + adev->ip_block_status[AMD_IP_BLOCK_TYPE_IH].hang ||
> + adev->ip_block_status[AMD_IP_BLOCK_TYPE_SMC].hang ||
> + adev->ip_block_status[AMD_IP_BLOCK_TYPE_GFX].hang ||
> + adev->ip_block_status[AMD_IP_BLOCK_TYPE_SDMA].hang ||
> + adev->ip_block_status[AMD_IP_BLOCK_TYPE_UVD].hang ||
> + adev->ip_block_status[AMD_IP_BLOCK_TYPE_VCE].hang ||
> + adev->ip_block_status[AMD_IP_BLOCK_TYPE_ACP].hang ||
> + adev->ip_block_status[AMD_IP_BLOCK_TYPE_DCE].hang)
> + return true;
> + return false;
> +}
> +
> +static int amdgpu_soft_reset(struct amdgpu_device *adev)
> +{
> + int i, r = 0;
> +
> + for (i = 0; i < adev->num_ip_blocks; i++) {
> + if (!adev->ip_block_status[i].valid)
> + continue;
> + if (adev->ip_block_status[i].hang &&
> + adev->ip_blocks[i].funcs->soft_reset)
> + r = adev->ip_blocks[i].funcs->soft_reset(adev);
> + if (r)
> + return r;
Same as with patch #3, please move the return code checking into the if.
> + }
> +
> + return 0;
> +}
> +
> /**
> * amdgpu_gpu_reset - reset the asic
> *
> @@ -2024,6 +2057,12 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
> if (amdgpu_device_has_dal_support(adev))
> state = drm_atomic_helper_suspend(adev->ddev);
>
> + if (!amdgpu_need_full_reset(adev)) {
> + amdgpu_pre_soft_reset(adev);
> + r = amdgpu_soft_reset(adev);
> + if (!amdgpu_check_soft_reset(adev))
> + goto out;
This should probably be "if (r || !amdgpu_check_soft_reset(adev))...".
Christian.
> + }
> /* save scratch */
> amdgpu_atombios_scratch_regs_save(adev);
> r = amdgpu_suspend(adev);
> @@ -2045,6 +2084,7 @@ retry:
> }
> /* restore scratch */
> amdgpu_atombios_scratch_regs_restore(adev);
> +out:
> if (!r) {
> r = amdgpu_ib_ring_tests(adev);
> if (r) {
