On Tue, Feb 07, 2023 at 07:21:48PM -0500, Lyude Paul wrote:
On Tue, 2023-02-07 at 14:11 +0200, Imre Deak wrote:
On Mon, Feb 06, 2023 at 07:42:32PM -0500, Lyude Paul wrote:
On Wed, 2023-02-01 at 17:04 +0200, Imre Deak wrote:
Yes, this patch should have no functional change, so please check what would apply to other drivers as well.
Could you also check Ville's comment about storing start_slot elsewhere than the atomic state (leaving only time_slots there). I wonder if that would work, at least it would simplify things I think.
Ideally it'd be nice if we could have all this info in the atomic commit, but yeah - you're not the first person to find this a bit confusing. FWIW though, the way we store start_slot right now is intended to follow the same kind of behavior we have with atomic CRTC commit dependencies, I think I also made it that way so all of the state would be in a single place but there's no hard requirement for it.
As I understood the atomic state should be precomputed during atomic check and not changed later during the commit phase. In case of start_slot this would mean knowing in advance the actual (driver specific) disabling / enabling sequence of the payloads, not sure how feasible it is to figure that out. start_slot also changes dynamically
It isn't feasible afaict, which was the main motivation for having the strange update procedure - this is pretty much something that needs to be determined at commit time.
as each payload is disabled, so these intermediate versions of the field would need to be tracked somehow separately from the final (precomputed) versions.
FWIW, the current way this works is that we call drm_dp_mst_atomic_wait_for_dependencies() in order to ensure that no one's currently writing to the start_slot field (remember, at this point we may not have atomic locks held anymore). Then at that point, start_slot in the new atomic state should hold the current in-hw start_slot value. start_slot isn't actually set to the new starting slot until drm_dp_add_payload_part1(). That of course as you pointed out, doesn't help us unless we read all of the start_slot values before we start changing payloads - since disabling payloads inevitably changes the start slot of payloads that come afterwards.
I'll admit I'm a bit surprised this logic was wrong - if I recall the whole reason I assumed this was OK when writing this code was that since we're updating one payload at a time, ACT, repeat, that the start slots of each payload in hardware _would_ actually change in the same way we modify them in helpers. e.g., my expectation was if we had a bunch of payloads like this:
Payload #1: 15 slots, start_slot=0 Payload #2: 15 slots, start_slot=15 Payload #3: 15 slots, start_slot=30
And then disabled say, payload #1, that immediately after we get the ACT that the payload table in hardware would look like this:
Payload #2: 15 slots, start_slot=0 Payload #3: 15 slots, start_slot=15
The above is the actual and expected HW state state yes.
But it sounds like it actually would look like this in the real world?
Payload #2: 15 slots, start_slot=15 Payload #3: 15 slots, start_slot=30
No, the problem currently is not that start_slot of the subsequent payloads are not shifted towards the beginning. Rather the atomic state doesn't get updated properly, becoming out of sync with the HW. For instance in a commit resizing payload #1, in the commit phase (intel_atomic_commit_tail()) will begin by removing payload #1. The initial state is
old payload state new payload state Payload #1: 15 slots, start_slot=0 20 slots, start_slot=0 Payload #2: 15 slots, start_slot=15 15 slots, start_slot=15 Payload #3: 15 slots, start_slot=30 15 slots, start_slot=30
mgr->next_start_slot = 45
intel_mst_disable_dp() will pass the old MST and payload state to drm_dp_remove_payload(): The MST state was added during atomic check, since payload #1 changed, then intel_atomic_commit() -> drm_atomic_helper_swap_state() sets the MST current state (returned by drm_atomic_get_mst_topology_state()) to point to the old state. So at the point drm_dp_remove_payload() returns we have:
old payload state new payload state Payload #1: 15 slots, start_slot=-1 20 slots, start_slot=0 Payload #2: 15 slots, start_slot=0 15 slots, start_slot=15 Payload #3: 15 slots, start_slot=15 15 slots, start_slot=30
mgr->next_start_slot = 30
then after re-enabling payload #1, after drm_dp_add_payload_part1() returns (passing to it the new MST and payload state) we have:
old payload state new payload state Payload #1: 15 slots, start_slot=-1 20 slots, start_slot=30 Payload #2: 15 slots, start_slot=0 15 slots, start_slot=15 Payload #3: 15 slots, start_slot=15 15 slots, start_slot=30
mgr->next_start_slot = 50
So in the new SW state payload #1 and #3 incorrectly overlap, with the actual HW state being:
Payload #1: 20 slots, start_slot=30 Payload #2: 15 slots, start_slot=0 Payload #3: 15 slots, start_slot=15
A subsequent commit will see the wrong start_slot in the SW state for payload #2 (15) and #3 (30).
So I'm curious, is there something I missed here? At what point does the MST hub at the other end decide that it's time to move the start slots back?
The hub shifts back payloads after the DPCD write command to 0x1c0 - 0x1c2 to remove a payload. (The HW OTOH does the corresponding shift at the point of disabling the stream, in intel_mst_post_disable_dp() -> intel_disable_transcoder() for i915).
(keep in mind, the MST specification does explicitly mention that there should never be holes in the payload table - so something has to be shifting the payloads back).
Right, the hubs I checked conform to this.
So if you guys think it'd be better design-wise to store this something else, I've got no strong feelings either way
For 0-2:
Reviewed-by: Lyude Paul lyude@redhat.com
Thanks.
Cc: Lyude Paul lyude@redhat.com Cc: Ville Syrjälä ville.syrjala@linux.intel.com Cc: Ben Skeggs bskeggs@redhat.com Cc: Karol Herbst kherbst@redhat.com Cc: Harry Wentland harry.wentland@amd.com Cc: Alex Deucher alexander.deucher@amd.com Cc: Wayne Lin Wayne.Lin@amd.com Cc: stable@vger.kernel.org # 6.1 Cc: dri-devel@lists.freedesktop.org Reviewed-by: Ville Syrjälä ville.syrjala@linux.intel.com Signed-off-by: Imre Deak imre.deak@intel.com
.../amd/display/amdgpu_dm/amdgpu_dm_helpers.c | 2 +- drivers/gpu/drm/display/drm_dp_mst_topology.c | 26 ++++++++++--------- drivers/gpu/drm/i915/display/intel_dp_mst.c | 4 ++- drivers/gpu/drm/nouveau/dispnv50/disp.c | 2 +- include/drm/display/drm_dp_mst_helper.h | 3 ++- 5 files changed, 21 insertions(+), 16 deletions(-)
diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c index a50319fc42b11..180d3893b68da 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_helpers.c @@ -208,7 +208,7 @@ bool dm_helpers_dp_mst_write_payload_allocation_table( if (enable) drm_dp_add_payload_part1(mst_mgr, mst_state, payload); else
drm_dp_remove_payload(mst_mgr, mst_state, payload);
drm_dp_remove_payload(mst_mgr, mst_state, payload, payload);/* mst_mgr->->payloads are VC payload notify MST branch using DPCD or * AUX message. The sequence is slot 1-63 allocated sequence for each diff --git a/drivers/gpu/drm/display/drm_dp_mst_topology.c b/drivers/gpu/drm/display/drm_dp_mst_topology.c index 847c10aa2098c..1990ff5dc7ddd 100644 --- a/drivers/gpu/drm/display/drm_dp_mst_topology.c +++ b/drivers/gpu/drm/display/drm_dp_mst_topology.c @@ -3342,7 +3342,8 @@ EXPORT_SYMBOL(drm_dp_add_payload_part1);
- drm_dp_remove_payload() - Remove an MST payload
- @mgr: Manager to use.
- @mst_state: The MST atomic state
- @payload: The payload to write
- @old_payload: The payload with its old state
- @new_payload: The payload to write
- Removes a payload from an MST topology if it was successfully assigned a start slot. Also updates
- the starting time slots of all other payloads which would have been shifted towards the start of
@@ -3350,36 +3351,37 @@ EXPORT_SYMBOL(drm_dp_add_payload_part1); */ void drm_dp_remove_payload(struct drm_dp_mst_topology_mgr *mgr, struct drm_dp_mst_topology_state *mst_state,
struct drm_dp_mst_atomic_payload *payload)
const struct drm_dp_mst_atomic_payload *old_payload,struct drm_dp_mst_atomic_payload *new_payload){ struct drm_dp_mst_atomic_payload *pos; bool send_remove = false; /* We failed to make the payload, so nothing to do */
- if (payload->vc_start_slot == -1)
- if (new_payload->vc_start_slot == -1) return;
mutex_lock(&mgr->lock);
- send_remove = drm_dp_mst_port_downstream_of_branch(payload->port, mgr->mst_primary);
- send_remove = drm_dp_mst_port_downstream_of_branch(new_payload->port, mgr->mst_primary); mutex_unlock(&mgr->lock);
if (send_remove)
drm_dp_destroy_payload_step1(mgr, mst_state, payload);
drm_dp_destroy_payload_step1(mgr, mst_state, new_payload);
payload->vcpi);
new_payload->vcpi);list_for_each_entry(pos, &mst_state->payloads, next) {
if (pos != payload && pos->vc_start_slot > payload->vc_start_slot)pos->vc_start_slot -= payload->time_slots;
if (pos != new_payload && pos->vc_start_slot > new_payload->vc_start_slot)pos->vc_start_slot -= old_payload->time_slots;
- payload->vc_start_slot = -1;
- new_payload->vc_start_slot = -1;
mgr->payload_count--;
- mgr->next_start_slot -= payload->time_slots;
- mgr->next_start_slot -= old_payload->time_slots;
- if (payload->delete)
drm_dp_mst_put_port_malloc(payload->port);
- if (new_payload->delete)
drm_dp_mst_put_port_malloc(new_payload->port);} EXPORT_SYMBOL(drm_dp_remove_payload); diff --git a/drivers/gpu/drm/i915/display/intel_dp_mst.c b/drivers/gpu/drm/i915/display/intel_dp_mst.c index f3cb12dcfe0a7..dc4e5ff1dbb31 100644 --- a/drivers/gpu/drm/i915/display/intel_dp_mst.c +++ b/drivers/gpu/drm/i915/display/intel_dp_mst.c @@ -526,6 +526,8 @@ static void intel_mst_disable_dp(struct intel_atomic_state *state, to_intel_connector(old_conn_state->connector); struct drm_dp_mst_topology_state *mst_state = drm_atomic_get_mst_topology_state(&state->base, &intel_dp->mst_mgr);
- struct drm_dp_mst_atomic_payload *payload =
drm_atomic_get_mst_payload_state(mst_state, connector->port);drm_dbg_kms(&i915->drm, "active links %d\n", @@ -534,7 +536,7 @@ static void intel_mst_disable_dp(struct intel_atomic_state *state, intel_hdcp_disable(intel_mst->connector); drm_dp_remove_payload(&intel_dp->mst_mgr, mst_state,
drm_atomic_get_mst_payload_state(mst_state, connector->port));
payload, payload);intel_audio_codec_disable(encoder, old_crtc_state, old_conn_state); } diff --git a/drivers/gpu/drm/nouveau/dispnv50/disp.c b/drivers/gpu/drm/nouveau/dispnv50/disp.c index edcb2529b4025..ed9d374147b8d 100644 --- a/drivers/gpu/drm/nouveau/dispnv50/disp.c +++ b/drivers/gpu/drm/nouveau/dispnv50/disp.c @@ -885,7 +885,7 @@ nv50_msto_prepare(struct drm_atomic_state *state, // TODO: Figure out if we want to do a better job of handling VCPI allocation failures here? if (msto->disabled) {
drm_dp_remove_payload(mgr, mst_state, payload);
drm_dp_remove_payload(mgr, mst_state, payload, payload);nvif_outp_dp_mst_vcpi(&mstm->outp->outp, msto->head->base.index, 0, 0, 0, 0); } else { diff --git a/include/drm/display/drm_dp_mst_helper.h b/include/drm/display/drm_dp_mst_helper.h index 41fd8352ab656..f5eb9aa152b14 100644 --- a/include/drm/display/drm_dp_mst_helper.h +++ b/include/drm/display/drm_dp_mst_helper.h @@ -841,7 +841,8 @@ int drm_dp_add_payload_part2(struct drm_dp_mst_topology_mgr *mgr, struct drm_dp_mst_atomic_payload *payload); void drm_dp_remove_payload(struct drm_dp_mst_topology_mgr *mgr, struct drm_dp_mst_topology_state *mst_state,
struct drm_dp_mst_atomic_payload *payload);
const struct drm_dp_mst_atomic_payload *old_payload,struct drm_dp_mst_atomic_payload *new_payload);int drm_dp_check_act_status(struct drm_dp_mst_topology_mgr *mgr);
-- Cheers, Lyude Paul (she/her) Software Engineer at Red Hat
-- Cheers, Lyude Paul (she/her) Software Engineer at Red Hat
-- Cheers, Lyude Paul (she/her) Software Engineer at Red Hat