From: Alex Deucher alexander.deucher@amd.com
stable inclusion from stable-v6.6.8 commit 3aae4ef4d799fb3d0381157640fdb251008cf0ae category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/I8ST59
Reference: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id=...
--------------------------------
commit ab4750332dbe535243def5dcebc24ca00c1f98ac upstream.
Add begin/end_use ring callbacks to disallow GFXOFF when SDMA work is submitted and allow it again afterward.
This should avoid corner cases where GFXOFF is erroneously entered when SDMA is still active. For now just allow/disallow GFXOFF in the begin and end helpers until we root cause the issue. This should not impact power as SDMA usage is pretty minimal and GFXOSS should not be active when SDMA is active anyway, this just makes it explicit.
v2: move everything into sdma5.2 code. No reason for this to be generic at this point. v3: Add comments in new code
Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2220 Reviewed-by: Mario Limonciello mario.limonciello@amd.com (v1) Tested-by: Mario Limonciello mario.limonciello@amd.com (v1) Reviewed-by: Christian König christian.koenig@amd.com Signed-off-by: Alex Deucher alexander.deucher@amd.com Cc: stable@vger.kernel.org # 5.15+ Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org Signed-off-by: Zheng Zengkai zhengzengkai@huawei.com --- drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c | 28 ++++++++++++++++++++++++++ 1 file changed, 28 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c b/drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c index 2b3ebebc4299..0f930fd8a383 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c @@ -1651,6 +1651,32 @@ static void sdma_v5_2_get_clockgating_state(void *handle, u64 *flags) *flags |= AMD_CG_SUPPORT_SDMA_LS; }
+static void sdma_v5_2_ring_begin_use(struct amdgpu_ring *ring) +{ + struct amdgpu_device *adev = ring->adev; + + /* SDMA 5.2.3 (RMB) FW doesn't seem to properly + * disallow GFXOFF in some cases leading to + * hangs in SDMA. Disallow GFXOFF while SDMA is active. + * We can probably just limit this to 5.2.3, + * but it shouldn't hurt for other parts since + * this GFXOFF will be disallowed anyway when SDMA is + * active, this just makes it explicit. + */ + amdgpu_gfx_off_ctrl(adev, false); +} + +static void sdma_v5_2_ring_end_use(struct amdgpu_ring *ring) +{ + struct amdgpu_device *adev = ring->adev; + + /* SDMA 5.2.3 (RMB) FW doesn't seem to properly + * disallow GFXOFF in some cases leading to + * hangs in SDMA. Allow GFXOFF when SDMA is complete. + */ + amdgpu_gfx_off_ctrl(adev, true); +} + const struct amd_ip_funcs sdma_v5_2_ip_funcs = { .name = "sdma_v5_2", .early_init = sdma_v5_2_early_init, @@ -1698,6 +1724,8 @@ static const struct amdgpu_ring_funcs sdma_v5_2_ring_funcs = { .test_ib = sdma_v5_2_ring_test_ib, .insert_nop = sdma_v5_2_ring_insert_nop, .pad_ib = sdma_v5_2_ring_pad_ib, + .begin_use = sdma_v5_2_ring_begin_use, + .end_use = sdma_v5_2_ring_end_use, .emit_wreg = sdma_v5_2_ring_emit_wreg, .emit_reg_wait = sdma_v5_2_ring_emit_reg_wait, .emit_reg_write_reg_wait = sdma_v5_2_ring_emit_reg_write_reg_wait,