From: Ivan Lipski <[email protected]>

[Why]
Multiple paths issue DMUB commands without managing IPS state, causing
dc_wake_and_execute_gpint/dmub_cmd to internally wake from IPS and
reallow idle. This flips idle_allowed back to true while
idle_optimizations_allowed remains false during in-flight commits,
desynchronizing the two flags.

Affected paths:

- amdgpu_dm_psr_set_event() and amdgpu_dm_replay_set_event() calls from
  amdgpu_dm_handle_vrr_transition(), amdgpu_dm_commit_planes() and
  amdgpu_dm_mod_power_update_streams(), that are invoked on atomic commits.
- debugfs psr_get(), psr_read_residency(), replay_get_state(),
  replay_set_residency() access hardware without holding dc_lock or
  disabling IPS.

[How]
- Explicitly exit IPS before PSR/Replay set_event w/ hw_programming,
  called within atomic commit.
- Wrap debugfs PSR/Replay state getters and setters with IPS exit/entry +
  dc_lock.

Reviewed-by: Sunpeng Li <[email protected]>
Signed-off-by: Ivan Lipski <[email protected]>
Signed-off-by: James Lin <[email protected]>
---
 .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c |  4 +
 .../amd/display/amdgpu_dm/amdgpu_dm_debugfs.c | 77 +++++++++++++++++++
 2 files changed, 81 insertions(+)

diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c 
b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
index 7ff1af3528dd..4e9b4fd505c2 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
@@ -9937,6 +9937,7 @@ static void amdgpu_dm_handle_vrr_transition(struct 
amdgpu_display_manager *dm,
                                 __func__, new_state->base.crtc->base.id);
 
                scoped_guard(mutex, &dm->dc_lock) {
+                       dc_exit_ips_for_hw_access(dm->dc);
                        amdgpu_dm_psr_set_event(dm, new_state->stream, true,
                                psr_event_vrr_transition, true);
                        amdgpu_dm_replay_set_event(dm, new_state->stream, true,
@@ -9952,6 +9953,7 @@ static void amdgpu_dm_handle_vrr_transition(struct 
amdgpu_display_manager *dm,
                                 __func__, new_state->base.crtc->base.id);
 
                scoped_guard(mutex, &dm->dc_lock) {
+                       dc_exit_ips_for_hw_access(dm->dc);
                        amdgpu_dm_psr_set_event(dm, new_state->stream, false,
                                psr_event_vrr_transition, false);
                        amdgpu_dm_replay_set_event(dm, new_state->stream, false,
@@ -10253,6 +10255,7 @@ static void amdgpu_dm_commit_planes(struct 
drm_atomic_state *state,
                                mutex_lock(&dm->dc_lock);
                                
acrtc_state->stream->link->psr_settings.psr_dirty_rects_change_timestamp_ns =
                                timestamp_ns;
+                               dc_exit_ips_for_hw_access(dm->dc);
                                amdgpu_dm_psr_set_event(dm, 
acrtc_state->stream, true,
                                        psr_event_hw_programming, true);
                                mutex_unlock(&dm->dc_lock);
@@ -10610,6 +10613,7 @@ static void amdgpu_dm_mod_power_update_streams(struct 
drm_atomic_state *state,
                 */
                if (old_crtc_state->active) {
                        scoped_guard(mutex, &dm->dc_lock) {
+                               dc_exit_ips_for_hw_access(dm->dc);
                                amdgpu_dm_psr_set_event(dm, 
dm_old_crtc_state->stream, true,
                                        psr_event_hw_programming, true);
                                amdgpu_dm_replay_set_event(dm, 
dm_old_crtc_state->stream, true,
diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_debugfs.c 
b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_debugfs.c
index 49226d6d0311..4e68a3541639 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_debugfs.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_debugfs.c
@@ -3163,10 +3163,25 @@ static int replay_get_state(void *data, u64 *val)
 {
        struct amdgpu_dm_connector *connector = data;
        struct dc_link *link = connector->dc_link;
+       struct amdgpu_device *adev = drm_to_adev(connector->base.dev);
+       struct dc *dc = adev->dm.dc;
        uint64_t state = REPLAY_STATE_INVALID;
+       bool reallow_idle = false;
+
+       mutex_lock(&adev->dm.dc_lock);
+
+       if (dc->idle_optimizations_allowed) {
+               dc_allow_idle_optimizations(dc, false);
+               reallow_idle = true;
+       }
 
        dc_link_get_replay_state(link, &state);
 
+       if (reallow_idle)
+               dc_allow_idle_optimizations(dc, true);
+
+       mutex_unlock(&adev->dm.dc_lock);
+
        *val = state;
 
        return 0;
@@ -3179,10 +3194,26 @@ static int replay_set_residency(void *data, u64 val)
 {
        struct amdgpu_dm_connector *connector = data;
        struct dc_link *link = connector->dc_link;
+       struct amdgpu_device *adev = drm_to_adev(connector->base.dev);
+       struct dc *dc = adev->dm.dc;
        bool is_start = (val != 0);
        u32 residency = 0;
+       bool reallow_idle = false;
+
+       mutex_lock(&adev->dm.dc_lock);
+
+       if (dc->idle_optimizations_allowed) {
+               dc_allow_idle_optimizations(dc, false);
+               reallow_idle = true;
+       }
 
        link->dc->link_srv->edp_replay_residency(link, &residency, is_start, 
PR_RESIDENCY_MODE_PHY);
+
+       if (reallow_idle)
+               dc_allow_idle_optimizations(dc, true);
+
+       mutex_unlock(&adev->dm.dc_lock);
+
        return 0;
 }
 
@@ -3193,9 +3224,25 @@ static int replay_get_residency(void *data, u64 *val)
 {
        struct amdgpu_dm_connector *connector = data;
        struct dc_link *link = connector->dc_link;
+       struct amdgpu_device *adev = drm_to_adev(connector->base.dev);
+       struct dc *dc = adev->dm.dc;
        u32 residency = 0;
+       bool reallow_idle = false;
+
+       mutex_lock(&adev->dm.dc_lock);
+
+       if (dc->idle_optimizations_allowed) {
+               dc_allow_idle_optimizations(dc, false);
+               reallow_idle = true;
+       }
 
        link->dc->link_srv->edp_replay_residency(link, &residency, false, 
PR_RESIDENCY_MODE_PHY);
+
+       if (reallow_idle)
+               dc_allow_idle_optimizations(dc, true);
+
+       mutex_unlock(&adev->dm.dc_lock);
+
        *val = (u64)residency;
 
        return 0;
@@ -3208,10 +3255,25 @@ static int psr_get(void *data, u64 *val)
 {
        struct amdgpu_dm_connector *connector = data;
        struct dc_link *link = connector->dc_link;
+       struct amdgpu_device *adev = drm_to_adev(connector->base.dev);
+       struct dc *dc = adev->dm.dc;
        enum dc_psr_state state = PSR_STATE0;
+       bool reallow_idle = false;
+
+       mutex_lock(&adev->dm.dc_lock);
+
+       if (dc->idle_optimizations_allowed) {
+               dc_allow_idle_optimizations(dc, false);
+               reallow_idle = true;
+       }
 
        dc_link_get_psr_state(link, &state);
 
+       if (reallow_idle)
+               dc_allow_idle_optimizations(dc, true);
+
+       mutex_unlock(&adev->dm.dc_lock);
+
        *val = state;
 
        return 0;
@@ -3224,10 +3286,25 @@ static int psr_read_residency(void *data, u64 *val)
 {
        struct amdgpu_dm_connector *connector = data;
        struct dc_link *link = connector->dc_link;
+       struct amdgpu_device *adev = drm_to_adev(connector->base.dev);
+       struct dc *dc = adev->dm.dc;
        u32 residency = 0;
+       bool reallow_idle = false;
+
+       mutex_lock(&adev->dm.dc_lock);
+
+       if (dc->idle_optimizations_allowed) {
+               dc_allow_idle_optimizations(dc, false);
+               reallow_idle = true;
+       }
 
        link->dc->link_srv->edp_get_psr_residency(link, &residency, 
PSR_RESIDENCY_MODE_PHY);
 
+       if (reallow_idle)
+               dc_allow_idle_optimizations(dc, true);
+
+       mutex_unlock(&adev->dm.dc_lock);
+
        *val = (u64)residency;
 
        return 0;
-- 
2.43.0

Reply via email to