Nouveau - Nov 2018 - [PATCH v6 3/6] drm/dp_mst: Start tracking per-port VCPI allocations

On Thu, Nov 15, 2018 at 07:50:05PM -0500, Lyude Paul
wrote:
> There has been a TODO waiting for quite a long time in
> drm_dp_mst_topology.c:
> 
> 	/* We cannot rely on port->vcpi.num_slots to update
> 	 * topology_state->avail_slots as the port may not exist if the parent
> 	 * branch device was unplugged. This should be fixed by tracking
> 	 * per-port slot allocation in drm_dp_mst_topology_state instead of
> 	 * depending on the caller to tell us how many slots to release.
> 	 */
> 
> That's not the only reason we should fix this: forcing the driver to
> track the VCPI allocations throughout a state's atomic check is
> error prone, because it means that extra care has to be taken with the
> order that drm_dp_atomic_find_vcpi_slots() and
> drm_dp_atomic_release_vcpi_slots() are called in in order to ensure
> idempotency. Currently the only driver actually using these helpers,
> i915, doesn't even do this correctly: multiple ->best_encoder()
checks
> with i915's current implementation would not be idempotent and would
> over-allocate VCPI slots, something I learned trying to implement
> fallback retraining in MST.
> 
> So: simplify this whole mess, and teach drm_dp_atomic_find_vcpi_slots()
> and drm_dp_atomic_release_vcpi_slots() to track the VCPI allocations for
> each port. This allows us to ensure idempotency without having to rely
> on the driver as much. Additionally: the driver doesn't need to do any
> kind of VCPI slot tracking anymore if it doesn't need it for it's
own
> internal state.
> 
> Additionally; this adds a new drm_dp_mst_atomic_check() helper which
> must be used by atomic drivers to perform validity checks for the new
> VCPI allocations incurred by a state.
> 
> Also: update the documentation and make it more obvious that these
> /must/ be called by /all/ atomic drivers supporting MST.
> 
> Changes since v6:
>  - Keep a kref to all of the ports we have allocations on. This required
>    a good bit of changing to when we call drm_dp_find_vcpi_slots(),
>    mainly that we need to ensure that we only redo VCPI allocations on
>    actual mode or CRTC changes, not crtc_state->active changes.
>    Additionally, we no longer take the registration of the DRM connector
>    for each port into account because so long as we have a kref to the
>    port in the new or previous atomic state, the connector will stay
>    registered.
I write an entire pile of small nitpits (still included most of them
below), until I realized this here wont work. Delaying the call to destroy
the connector (well, unregister it really) wreaks the design we've come up
with thus far, resulting in most of my comments here.

Instead, all we need to do is delay the kfree(port) at the bottom of
drm_dp_destroy_port(). The vcpi allocation structure _only_ needs the
pointer value to stay valid, as a lookup key. It doesn't care at all about
anything actually stored in there. So the only thing we need to delay is
the kfree. I think the simplest way to achieve this is to add a 2nd kref
(port->kfree_ref or something like that), with on reference held by the
port itself (released in drm_dp_destroy_port), and the other one held
as-needed by the vcpi allocation lists.

I think if we go with this design instead of retrofitting a semantic
change of the port lifetime itself, all the complications I complain about
below should disappear.

Piles of comments below.

Cheers, Daniel
>  - Use the small changes to drm_dp_put_port() to add even more error
>    checking to make misusage of the helpers more obvious. I added this
>    after having to chase down various use-after-free conditions that
>    started popping up from the new helpers so no one else has to
>    troubleshoot that.
>  - Move some accidental DRM_DEBUG_KMS() calls to DRM_DEBUG_ATOMIC()
>  - Update documentation again, note that find/release() should both not be
>    called on the same port in a single atomic check phase (but multiple
>    calls to one or the other is OK)
> 
> Changes since v4:
>  - Don't skip the atomic checks for VCPI allocations if no new VCPI
>    allocations happen in a state. This makes the next change I'm about
>    to list here a lot easier to implement.
>  - Don't ignore VCPI allocations on destroyed ports, instead ensure
that
>    when ports are destroyed and still have VCPI allocations in the
>    topology state, the only state changes allowed are releasing said
>    ports' VCPI. This prevents a state with a mix of VCPI allocations
>    from destroyed ports, and allocations from valid ports.
> 
> Changes since v3:
>  - Don't release VCPI allocations in the topology state immediately in
>    drm_dp_atomic_release_vcpi_slots(), instead mark them as 0 and skip
>    over them in drm_dp_mst_duplicate_state(). This makes it so
>    drm_dp_atomic_release_vcpi_slots() is still idempotent while also
>    throwing warnings if the driver messes up it's book keeping and
tries
>    to release VCPI slots on a port that doesn't have any pre-existing
>    VCPI allocation - danvet
>  - Change mst_state/state in some debugging messages to "mst
state"
> 
> Changes since v2:
>  - Use kmemdup() for duplicating MST state - danvet
>  - Move port validation out of duplicate state callback - danvet
>  - Handle looping through MST topology states in
>    drm_dp_mst_atomic_check() so the driver doesn't have to do it
>  - Fix documentation in drm_dp_atomic_find_vcpi_slots()
>  - Move the atomic check for each individual topology state into it's
>    own function, reduces indenting
>  - Don't consider "stale" MST ports when calculating the
bandwidth
>    requirements. This is needed because originally we relied on the
>    state duplication functions to prune any stale ports from the new
>    state, which would prevent us from incorrectly considering their
>    bandwidth requirements alongside legitimate new payloads.
>  - Add function references in drm_dp_atomic_release_vcpi_slots() - danvet
>  - Annotate atomic VCPI and atomic check functions with __must_check
>    - danvet
> 
> Changes since v1:
>  - Don't use the now-removed ->atomic_check() for private objects
hook,
>    just give drivers a function to call themselves
> 
> Signed-off-by: Lyude Paul <lyude at redhat.com>
> Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
> ---
>  drivers/gpu/drm/drm_dp_mst_topology.c | 294 +++++++++++++++++++++++---
>  drivers/gpu/drm/i915/intel_display.c  |   4 +
>  drivers/gpu/drm/i915/intel_dp_mst.c   |  64 +++---
>  include/drm/drm_dp_mst_helper.h       |  24 ++-
>  4 files changed, 319 insertions(+), 67 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_dp_mst_topology.c
b/drivers/gpu/drm/drm_dp_mst_topology.c
> index 00fbe7a2699d..183d0e832ced 100644
> --- a/drivers/gpu/drm/drm_dp_mst_topology.c
> +++ b/drivers/gpu/drm/drm_dp_mst_topology.c
> @@ -2617,21 +2617,42 @@ static int drm_dp_init_vcpi(struct
drm_dp_mst_topology_mgr *mgr,
>  }
>  
>  /**
> - * drm_dp_atomic_find_vcpi_slots() - Find and add vcpi slots to the state
> + * drm_dp_atomic_find_vcpi_slots() - Find and add VCPI slots to the state
>   * @state: global atomic state
>   * @mgr: MST topology manager for the port
>   * @port: port to find vcpi slots for
>   * @pbn: bandwidth required for the mode in PBN
>   *
> - * RETURNS:
> - * Total slots in the atomic state assigned for this port or error
> + * Allocates VCPI slots to @port, replacing any previous VCPI allocations
it
> + * may have had. Any atomic drivers which support MST must call this
function
> + * in their &drm_encoder_helper_funcs.atomic_check() callback to
change the
> + * current VCPI allocation for the new state, but only when
> + * &drm_crtc_state.mode_changed or
&drm_crtc_state.connectors_changed is set
> + * to ensure compatibility with userspace applications that still use the
> + * legacy modesetting UAPI.
> + *
> + * Allocations set by this function are not checked against the bandwidth
> + * restraints of @mgr until the driver calls drm_dp_mst_atomic_check().
> + *
> + * Additionally, it is OK to call this function multiple times on the same
> + * @port as needed. It is not OK however, to call this function and
> + * drm_dp_atomic_release_vcpi_slots() in the same atomic check phase.
> + *
> + * See also:
> + * drm_dp_atomic_release_vcpi_slots()
> + * drm_dp_mst_atomic_check()
> + *
> + * Returns:
> + * Total slots in the atomic state assigned for this port, or a negative
error
> + * code if the port no longer exists
>   */
>  int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
>  				  struct drm_dp_mst_topology_mgr *mgr,
>  				  struct drm_dp_mst_port *port, int pbn)
>  {
>  	struct drm_dp_mst_topology_state *topology_state;
> -	int req_slots;
> +	struct drm_dp_vcpi_allocation *pos, *vcpi = NULL;
> +	int prev_slots, req_slots, ret;
>  
>  	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
>  	if (IS_ERR(topology_state))
> @@ -2640,20 +2661,60 @@ int drm_dp_atomic_find_vcpi_slots(struct
drm_atomic_state *state,
>  	port = drm_dp_get_validated_port_ref(mgr, port);
>  	if (port == NULL)
>  		return -EINVAL;
> -	req_slots = DIV_ROUND_UP(pbn, mgr->pbn_div);
> -	DRM_DEBUG_KMS("vcpi slots req=%d, avail=%d\n",
> -			req_slots, topology_state->avail_slots);
>  
> -	if (req_slots > topology_state->avail_slots) {
> -		drm_dp_put_port(port);
> -		return -ENOSPC;
> +	topology_state->vcpi_allocated = true;
> +
> +	/* Find the current allocation for this port, if any */
> +	list_for_each_entry(pos, &topology_state->vcpis, next) {
> +		if (pos->port == port) {
> +			vcpi = pos;
> +			prev_slots = vcpi->vcpi;
> +
> +			/*
> +			 * This should never happen, unless the driver tries
> +			 * releasing and allocating the same VCPI allocation,
> +			 * which is an error
> +			 */
> +			if (WARN_ON(!prev_slots)) {
> +				DRM_ERROR("cannot allocate and release VCPI on [MST PORT:%p] in
the same state\n",
> +					  port);
> +				return -EINVAL;
> +			}
Allowing atomic changes to output state (like disabling it on one crtc,
and then enabling on the other, thus both releasing and re-allocating the
vcpi in one go) is kinda the point of atomic. So definitely can't EINVAL
here.

Sounds like we need more testcases in igt?
> +
> +			break;
> +		}
>  	}
> +	if (!vcpi)
> +		prev_slots = 0;
>  
> -	topology_state->avail_slots -= req_slots;
> -	DRM_DEBUG_KMS("vcpi slots avail=%d",
topology_state->avail_slots);
> +	req_slots = DIV_ROUND_UP(pbn, mgr->pbn_div);
> +
> +	DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] [MST PORT:%p] VCPI %d ->
%d\n",
> +			 port->connector->base.id, port->connector->name,
> +			 port, prev_slots, req_slots);
> +
> +	/* Add the new allocation to the state */
> +	if (!vcpi) {
> +		vcpi = kzalloc(sizeof(*vcpi), GFP_KERNEL);
> +		if (!vcpi) {
> +			ret = -ENOMEM;
> +			goto out;
> +		}
>  
> +		vcpi->port = drm_dp_get_validated_port_ref(mgr, port);
> +		if (!vcpi->port) {
> +			ret = -EINVAL;
> +			kfree(vcpi);
> +			goto out;
> +		}
> +		list_add(&vcpi->next, &topology_state->vcpis);
> +	}
> +	vcpi->vcpi = req_slots;
> +
> +	ret = req_slots;
> +out:
>  	drm_dp_put_port(port);
> -	return req_slots;
> +	return ret;
>  }
>  EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
>  
> @@ -2661,31 +2722,66 @@ EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
>   * drm_dp_atomic_release_vcpi_slots() - Release allocated vcpi slots
>   * @state: global atomic state
>   * @mgr: MST topology manager for the port
> - * @slots: number of vcpi slots to release
> + * @port: The port to release the VCPI slots from
>   *
> - * RETURNS:
> - * 0 if @slots were added back to
&drm_dp_mst_topology_state->avail_slots or
> - * negative error code
> + * Releases any VCPI slots that have been allocated to a port in the
atomic
> + * state. Any atomic drivers which support MST must call this function in
> + * their &drm_connector_helper_funcs.atomic_check() callback when the
> + * connector will no longer have VCPI allocated (e.g. because it's
CRTC was
> + * removed) when it had VCPI allocated in the previous atomic state.
> + *
> + * It is OK to call this even if @port has been removed from the system.
> + * Additionally, it is OK to call this function multiple times on the same
> + * @port as needed. It is not OK however, to call this function and
> + * drm_dp_atomic_find_vcpi_slots() on the same @port in a single atomic
check
> + * phase.
> + *
> + * See also:
> + * drm_dp_atomic_find_vcpi_slots()
> + * drm_dp_mst_atomic_check()
> + *
> + * Returns:
> + * 0 if all slots for this port were added back to
> + * &drm_dp_mst_topology_state.avail_slots or negative error code
>   */
>  int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
>  				     struct drm_dp_mst_topology_mgr *mgr,
> -				     int slots)
> +				     struct drm_dp_mst_port *port)
>  {
>  	struct drm_dp_mst_topology_state *topology_state;
> +	struct drm_dp_vcpi_allocation *pos;
> +	bool found = false;
>  
>  	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
>  	if (IS_ERR(topology_state))
>  		return PTR_ERR(topology_state);
>  
> -	/* We cannot rely on port->vcpi.num_slots to update
> -	 * topology_state->avail_slots as the port may not exist if the parent
> -	 * branch device was unplugged. This should be fixed by tracking
> -	 * per-port slot allocation in drm_dp_mst_topology_state instead of
> -	 * depending on the caller to tell us how many slots to release.
> -	 */
> -	topology_state->avail_slots += slots;
> -	DRM_DEBUG_KMS("vcpi slots released=%d, avail=%d\n",
> -			slots, topology_state->avail_slots);
> +	list_for_each_entry(pos, &topology_state->vcpis, next) {
> +		if (pos->port == port) {
> +			found = true;
> +			break;
> +		}
> +	}
> +	if (WARN_ON(!found)) {
> +		DRM_ERROR("no VCPI for [MST PORT:%p] found in mst state %p\n",
> +			  port, &topology_state->base);
> +		return -EINVAL;
> +	}
> +
> +	DRM_DEBUG_ATOMIC("[MST PORT:%p] VCPI %d -> 0\n", port,
pos->vcpi);
> +	if (pos->vcpi) {
> +		/*
> +		 * This should be safe-as the previous state will still hold a
> +		 * reference to this port until we swap states. If it's not,
> +		 * the driver has a bookkeeping error
> +		 */
> +		if (WARN_ON(drm_dp_put_port(port))) {
The usual way to do this (instead of adding a return value to the normal
_put function) is something like

	kref_put(port, function_that_with_just_a_WARNING_DRM_ERROR)

avoids the need for patch 2.
> +			DRM_ERROR("cannot allocate and release VCPI on [MST PORT:%p] in
the same state\n",
> +				  port);
> +			return -EINVAL;
> +		}
> +		pos->vcpi = 0;
> +	}
>  
>  	return 0;
>  }
> @@ -3115,15 +3211,45 @@ static void drm_dp_destroy_connector_work(struct
work_struct *work)
>  static struct drm_private_state *
>  drm_dp_mst_duplicate_state(struct drm_private_obj *obj)
>  {
> -	struct drm_dp_mst_topology_state *state;
> +	struct drm_dp_mst_topology_state *state, *old_state > +	
to_dp_mst_topology_state(obj->state);
> +	struct drm_dp_vcpi_allocation *pos, *vcpi;
>  
> -	state = kmemdup(obj->state, sizeof(*state), GFP_KERNEL);
> +	state = kmemdup(old_state, sizeof(*state), GFP_KERNEL);
>  	if (!state)
>  		return NULL;
>  
>  	__drm_atomic_helper_private_obj_duplicate_state(obj,
&state->base);
>  
> +	INIT_LIST_HEAD(&state->vcpis);
> +	state->vcpi_allocated = false;
> +
> +	list_for_each_entry(pos, &old_state->vcpis, next) {
> +		/* Prune leftover freed VCPI allocations */
> +		if (!pos->vcpi)
> +			continue;
> +
> +		vcpi = kmemdup(pos, sizeof(*vcpi), GFP_KERNEL);
> +		if (!vcpi)
> +			goto fail;
> +
> +		/* Take a plain old kref here and don't validate the port, as
> +		 * it's topology might not even exist anymore
> +		 */
> +		kref_get(&vcpi->port->kref);
> +		list_add(&vcpi->next, &state->vcpis);
> +	}
> +
>  	return &state->base;
> +
> +fail:
> +	list_for_each_entry_safe(pos, vcpi, &state->vcpis, next) {
> +		drm_dp_put_port(pos->port);
> +		kfree(pos);
> +	}
> +	kfree(state);
> +
> +	return NULL;
>  }
>  
>  static void drm_dp_mst_destroy_state(struct drm_private_obj *obj,
> @@ -3131,14 +3257,116 @@ static void drm_dp_mst_destroy_state(struct
drm_private_obj *obj,
>  {
>  	struct drm_dp_mst_topology_state *mst_state >  	
to_dp_mst_topology_state(state);
> +	struct drm_dp_vcpi_allocation *pos, *tmp;
> +
> +	list_for_each_entry_safe(pos, tmp, &mst_state->vcpis, next) {
> +		/* We only keep references to ports with non-zero VCPIs */
> +		if (pos->vcpi)
> +			drm_dp_put_port(pos->port);
> +		kfree(pos);
> +	}
>  
>  	kfree(mst_state);
>  }
>  
> -static const struct drm_private_state_funcs mst_state_funcs = {
> +static inline int
> +drm_dp_mst_atomic_check_topology_state(struct drm_dp_mst_topology_mgr
*mgr,
> +				       struct drm_dp_mst_topology_state *mst_state)
> +{
> +	struct drm_dp_vcpi_allocation *vcpi;
> +	struct drm_dp_mst_port *port;
> +	int avail_slots = 63, ret;
> +
> +	/* There's no possible scenario where releasing VCPI or keeping it
the
> +	 * same would make the state invalid
> +	 */
> +	if (!mst_state->vcpi_allocated) {
Do we need this? If not I'm always in favour of removing code if it's
not
needed :-)
> +		DRM_DEBUG_ATOMIC("[MST MGR:%p] state %p allocates no VCPI, check
skipped\n",
> +				 mgr, &mst_state->base);
> +		return 0;
> +	}
> +
> +	list_for_each_entry(vcpi, &mst_state->vcpis, next) {
> +		/* Releasing VCPI is always OK-even if the port is gone */
> +		if (!vcpi->vcpi) {
> +			DRM_DEBUG_ATOMIC("[MST PORT:%p] releases all VCPI slots\n",
> +					 vcpi->port);
> +			continue;
> +		}
> +
> +		port = drm_dp_get_validated_port_ref(mgr, vcpi->port);
> +		if (!port) {
> +			DRM_DEBUG_ATOMIC("[MST PORT:%p] is gone but still has VCPI, cannot
add new VCPI\n",
> +					 vcpi->port);
> +			return -EINVAL;
> +		}
For find_vcpi we've already done this. For unplugged vcpi, we can't do
this here I think - otherwise if you enable 2 mst outputs on the same
branch, and unplug both, old non-atomic userspace can never ever again
disable them because it can't disable both. The check (to skip already
deallocated ports) above isn't enough to handle that.
> +
> +		DRM_DEBUG_ATOMIC("[MST PORT:%p] requires %d vcpi slots\n",
> +				 vcpi->port, vcpi->vcpi);
> +
> +		avail_slots -= vcpi->vcpi;
> +		if (avail_slots < 0) {
> +			DRM_DEBUG_ATOMIC("[MST PORT:%p] not enough VCPI slots in mst state
%p (avail=%d)\n",
> +					 vcpi->port, mst_state,
> +					 avail_slots + vcpi->vcpi);
> +			ret = -ENOSPC;
> +			goto port_fail;
> +		}
> +
> +		drm_dp_put_port(port);
> +	}
> +	DRM_DEBUG_ATOMIC("[MST MGR:%p] mst state %p VCPI avail=%d
used=%d\n",
> +			 mgr, mst_state, avail_slots,
> +			 63 - avail_slots);
> +
> +	return 0;
> +port_fail:
> +	drm_dp_put_port(port);
> +	return ret;
> +}
> +
> +/**
> + * drm_dp_mst_atomic_check - Check that the new state of an MST topology
in an
> + * atomic update is valid
> + * @state: Pointer to the new &struct drm_dp_mst_topology_state
> + *
> + * Checks the given topology state for an atomic update to ensure that
it's
> + * valid. This includes checking whether there's enough bandwidth to
support
> + * the new VCPI allocations in the atomic update.
> + *
> + * Any atomic drivers supporting DP MST must make sure to call this after
> + * checking the rest of their state in their
> + * &drm_mode_config_funcs.atomic_check() callback.
> + *
> + * See also:
> + * drm_dp_atomic_find_vcpi_slots()
> + * drm_dp_atomic_release_vcpi_slots()
> + *
> + * Returns:
> + *
> + * 0 if the new state is valid, negative error code otherwise.
> + */
> +int drm_dp_mst_atomic_check(struct drm_atomic_state *state)
> +{
> +	struct drm_dp_mst_topology_mgr *mgr;
> +	struct drm_dp_mst_topology_state *mst_state;
> +	int i, ret = 0;
> +
> +	for_each_new_mst_mgr_in_state(state, mgr, mst_state, i) {
> +		ret = drm_dp_mst_atomic_check_topology_state(mgr, mst_state);
> +		if (ret)
> +			break;
> +	}
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL(drm_dp_mst_atomic_check);
> +
> +const struct drm_private_state_funcs drm_dp_mst_topology_state_funcs = {
>  	.atomic_duplicate_state = drm_dp_mst_duplicate_state,
>  	.atomic_destroy_state = drm_dp_mst_destroy_state,
>  };
> +EXPORT_SYMBOL(drm_dp_mst_topology_state_funcs);
>  
>  /**
>   * drm_atomic_get_mst_topology_state: get MST topology state
> @@ -3216,13 +3444,11 @@ int drm_dp_mst_topology_mgr_init(struct
drm_dp_mst_topology_mgr *mgr,
>  		return -ENOMEM;
>  
>  	mst_state->mgr = mgr;
> -
> -	/* max. time slots - one slot for MTP header */
> -	mst_state->avail_slots = 63;
> +	INIT_LIST_HEAD(&mst_state->vcpis);
>  
>  	drm_atomic_private_obj_init(&mgr->base,
>  				    &mst_state->base,
> -				    &mst_state_funcs);
> +				    &drm_dp_mst_topology_state_funcs);
>  
>  	return 0;
>  }
> diff --git a/drivers/gpu/drm/i915/intel_display.c
b/drivers/gpu/drm/i915/intel_display.c
> index 132e978227fb..8e85e04a5948 100644
> --- a/drivers/gpu/drm/i915/intel_display.c
> +++ b/drivers/gpu/drm/i915/intel_display.c
> @@ -12592,6 +12592,10 @@ static int intel_atomic_check(struct drm_device
*dev,
>  				       "[modeset]" : "[fastset]");
>  	}
>  
> +	ret = drm_dp_mst_atomic_check(state);
> +	if (ret)
> +		return ret;
> +
>  	if (any_ms) {
>  		ret = intel_modeset_checks(state);
>  
> diff --git a/drivers/gpu/drm/i915/intel_dp_mst.c
b/drivers/gpu/drm/i915/intel_dp_mst.c
> index 4de247ddf05f..be2d1ac7d87d 100644
> --- a/drivers/gpu/drm/i915/intel_dp_mst.c
> +++ b/drivers/gpu/drm/i915/intel_dp_mst.c
> @@ -41,8 +41,13 @@ static bool intel_dp_mst_compute_config(struct
intel_encoder *encoder,
>  	struct drm_connector *connector = conn_state->connector;
>  	void *port = to_intel_connector(connector)->port;
>  	struct drm_atomic_state *state = pipe_config->base.state;
> +	struct drm_crtc *crtc = pipe_config->base.crtc;
> +	struct drm_crtc_state *old_crtc_state > +	
drm_atomic_get_old_crtc_state(state, crtc);
> +	struct drm_crtc_state *new_crtc_state = &pipe_config->base;
>  	int bpp;
> -	int lane_count, slots = 0;
> +	int lane_count, slots > +	
to_intel_crtc_state(old_crtc_state)->dp_m_n.tu;
>  	const struct drm_display_mode *adjusted_mode =
&pipe_config->base.adjusted_mode;
>  	int mst_pbn;
>  	bool constant_n = drm_dp_has_quirk(&intel_dp->desc,
> @@ -77,8 +82,12 @@ static bool intel_dp_mst_compute_config(struct
intel_encoder *encoder,
>  	mst_pbn = drm_dp_calc_pbn_mode(adjusted_mode->crtc_clock, bpp);
>  	pipe_config->pbn = mst_pbn;
>  
> -	/* Zombie connectors can't have VCPI slots */
> -	if (!drm_connector_is_unregistered(connector)) {
> +	/* Only change VCPI allocation on actual mode changes, to prevent us
> +	 * from trying to allocate VCPI to ports that no longer exist when we
> +	 * may just be trying to disable DPMS on them
> +	 */
> +	if (new_crtc_state->mode_changed ||
> +	    new_crtc_state->connectors_changed) {
Why exactly is this needed? We only call compute_config on the enocoder
when enabling an output. Not when disabling one. Similar for the atomic
helpers and the various encoder callbacks (you have the same change in the
nouveau code). Did this actually blow up while testing?
>  		slots = drm_dp_atomic_find_vcpi_slots(state,
>  						      &intel_dp->mst_mgr,
>  						      port,
> @@ -107,36 +116,39 @@ static bool intel_dp_mst_compute_config(struct
intel_encoder *encoder,
>  	return true;
>  }
>  
> -static int intel_dp_mst_atomic_check(struct drm_connector *connector,
> -		struct drm_connector_state *new_conn_state)
> +static int
> +intel_dp_mst_atomic_check(struct drm_connector *connector,
> +			  struct drm_connector_state *new_conn_state)
>  {
> +	struct intel_connector *intel_connector > +	
to_intel_connector(connector);
> +	struct drm_dp_mst_topology_mgr *mgr > +	
&intel_connector->mst_port->mst_mgr;
> +	struct drm_dp_mst_port *port = intel_connector->port;
>  	struct drm_atomic_state *state = new_conn_state->state;
> -	struct drm_connector_state *old_conn_state;
> -	struct drm_crtc *old_crtc;
> -	struct drm_crtc_state *crtc_state;
> -	int slots, ret = 0;
> +	struct drm_connector_state *old_conn_state > +	
drm_atomic_get_old_connector_state(state, connector);
> +	struct drm_crtc *old_crtc = old_conn_state->crtc,
> +			*new_crtc = new_conn_state->crtc;
> +	struct drm_crtc_state *old_crtc_state, *new_crtc_state;
>  
> -	old_conn_state = drm_atomic_get_old_connector_state(state, connector);
> -	old_crtc = old_conn_state->crtc;
Bikeshed: Reworking all the above hides the real changes quite a bit.
>  	if (!old_crtc)
> -		return ret;
> +		return 0;
>  
> -	crtc_state = drm_atomic_get_new_crtc_state(state, old_crtc);
> -	slots = to_intel_crtc_state(crtc_state)->dp_m_n.tu;
> -	if (drm_atomic_crtc_needs_modeset(crtc_state) && slots > 0) {
> -		struct drm_dp_mst_topology_mgr *mgr;
> -		struct drm_encoder *old_encoder;
> +	old_crtc_state = drm_atomic_get_old_crtc_state(state, old_crtc);
> +	if (!old_crtc_state ||
!old_crtc_state would be a framework bug. Just remove that check.
> +	    !to_intel_crtc_state(old_crtc_state)->dp_m_n.tu)
> +		return 0;
Hm, why do we need this? Having a DP port previously enabled with a 0 tu
sounds like a pretty big driver bug.
>  
> -		old_encoder = old_conn_state->best_encoder;
> -		mgr = &enc_to_mst(old_encoder)->primary->dp.mst_mgr;
> +	/* If we switched CRTCs, clear the previous one's allocation */
> +	new_crtc_state = drm_atomic_get_new_crtc_state(state, old_crtc);
> +	if (new_crtc_state->connectors_changed &&
!new_crtc_state->enable)
> +		to_intel_crtc_state(new_crtc_state)->dp_m_n.tu = 0;
No idea why we need to clear this here? Shouldn't we just clear this
unconditionally if we release the allocation.

Also naming convention, I'd call this old_crtc_new_state, and maybe
old_crtc_old_state. To make it clear these are the old/new states for
old_crtc. Instead of old/new state for the new crtc.
>  
> -		ret = drm_dp_atomic_release_vcpi_slots(state, mgr, slots);
> -		if (ret)
> -			DRM_DEBUG_KMS("failed releasing %d vcpi slots:%d\n", slots,
ret);
> -		else
> -			to_intel_crtc_state(crtc_state)->dp_m_n.tu = 0;
> -	}
> -	return ret;
> +	if (new_crtc)
> +		return 0;
Not sure why we shouldn't release this? If we handle this in helpers, it
would remove piles of special casing code here I think. Plus remove
special code in the hlpers.

> +
> +	return drm_dp_atomic_release_vcpi_slots(state, mgr, port);
>  }
>  
>  static void intel_mst_disable_dp(struct intel_encoder *encoder,
> diff --git a/include/drm/drm_dp_mst_helper.h
b/include/drm/drm_dp_mst_helper.h
> index 3faceb66f5cb..2392e12b6bce 100644
> --- a/include/drm/drm_dp_mst_helper.h
> +++ b/include/drm/drm_dp_mst_helper.h
> @@ -406,9 +406,16 @@ struct drm_dp_payload {
>  
>  #define to_dp_mst_topology_state(x) container_of(x, struct
drm_dp_mst_topology_state, base)
>  
> +struct drm_dp_vcpi_allocation {
I think a comment here that this pointer holds a full reference (even past
when the connector is unplugged) would be good. Otherwise the loop in
release_vcpi would be buggy.
> +	struct drm_dp_mst_port *port;
> +	int vcpi;
> +	struct list_head next;
> +};
> +
>  struct drm_dp_mst_topology_state {
>  	struct drm_private_state base;
> -	int avail_slots;
> +	struct list_head vcpis;
> +	u8 vcpi_allocated;
>  	struct drm_dp_mst_topology_mgr *mgr;
>  };
>  
> @@ -619,14 +626,17 @@ void drm_dp_mst_topology_mgr_suspend(struct
drm_dp_mst_topology_mgr *mgr);
>  int drm_dp_mst_topology_mgr_resume(struct drm_dp_mst_topology_mgr *mgr);
>  struct drm_dp_mst_topology_state *drm_atomic_get_mst_topology_state(struct
drm_atomic_state *state,
>  								    struct drm_dp_mst_topology_mgr *mgr);
> -int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
> -				  struct drm_dp_mst_topology_mgr *mgr,
> -				  struct drm_dp_mst_port *port, int pbn);
> -int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
> -				     struct drm_dp_mst_topology_mgr *mgr,
> -				     int slots);
> +int __must_check
> +drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
> +			      struct drm_dp_mst_topology_mgr *mgr,
> +			      struct drm_dp_mst_port *port, int pbn);
> +int __must_check
> +drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
> +				 struct drm_dp_mst_topology_mgr *mgr,
> +				 struct drm_dp_mst_port *port);
>  int drm_dp_send_power_updown_phy(struct drm_dp_mst_topology_mgr *mgr,
>  				 struct drm_dp_mst_port *port, bool power_up);
> +int __must_check drm_dp_mst_atomic_check(struct drm_atomic_state *state);
>  
>  extern const struct drm_private_state_funcs
drm_dp_mst_topology_state_funcs;
>  
> -- 
> 2.19.1
> 
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

On Mon, Nov 26, 2018 at 10:04:21PM +0100, Daniel Vetter
wrote:
> On Thu, Nov 15, 2018 at 07:50:05PM -0500, Lyude Paul wrote:
> > There has been a TODO waiting for quite a long time in
> > drm_dp_mst_topology.c:
> > 
> > 	/* We cannot rely on port->vcpi.num_slots to update
> > 	 * topology_state->avail_slots as the port may not exist if the
parent
> > 	 * branch device was unplugged. This should be fixed by tracking
> > 	 * per-port slot allocation in drm_dp_mst_topology_state instead of
> > 	 * depending on the caller to tell us how many slots to release.
> > 	 */
> > 
> > That's not the only reason we should fix this: forcing the driver
to
> > track the VCPI allocations throughout a state's atomic check is
> > error prone, because it means that extra care has to be taken with the
> > order that drm_dp_atomic_find_vcpi_slots() and
> > drm_dp_atomic_release_vcpi_slots() are called in in order to ensure
> > idempotency. Currently the only driver actually using these helpers,
> > i915, doesn't even do this correctly: multiple ->best_encoder()
checks
> > with i915's current implementation would not be idempotent and
would
> > over-allocate VCPI slots, something I learned trying to implement
> > fallback retraining in MST.
> > 
> > So: simplify this whole mess, and teach
drm_dp_atomic_find_vcpi_slots()
> > and drm_dp_atomic_release_vcpi_slots() to track the VCPI allocations
for
> > each port. This allows us to ensure idempotency without having to rely
> > on the driver as much. Additionally: the driver doesn't need to do
any
> > kind of VCPI slot tracking anymore if it doesn't need it for
it's own
> > internal state.
> > 
> > Additionally; this adds a new drm_dp_mst_atomic_check() helper which
> > must be used by atomic drivers to perform validity checks for the new
> > VCPI allocations incurred by a state.
> > 
> > Also: update the documentation and make it more obvious that these
> > /must/ be called by /all/ atomic drivers supporting MST.
> > 
> > Changes since v6:
> >  - Keep a kref to all of the ports we have allocations on. This
required
> >    a good bit of changing to when we call drm_dp_find_vcpi_slots(),
> >    mainly that we need to ensure that we only redo VCPI allocations on
> >    actual mode or CRTC changes, not crtc_state->active changes.
> >    Additionally, we no longer take the registration of the DRM
connector
> >    for each port into account because so long as we have a kref to the
> >    port in the new or previous atomic state, the connector will stay
> >    registered.
> 
> I write an entire pile of small nitpits (still included most of them
> below), until I realized this here wont work. Delaying the call to destroy
> the connector (well, unregister it really) wreaks the design we've come
up
> with thus far, resulting in most of my comments here.
> 
> Instead, all we need to do is delay the kfree(port) at the bottom of
> drm_dp_destroy_port(). The vcpi allocation structure _only_ needs the
> pointer value to stay valid, as a lookup key. It doesn't care at all
about
> anything actually stored in there. So the only thing we need to delay is
> the kfree. I think the simplest way to achieve this is to add a 2nd kref
> (port->kfree_ref or something like that), with on reference held by the
> port itself (released in drm_dp_destroy_port), and the other one held
> as-needed by the vcpi allocation lists.
> 
> I think if we go with this design instead of retrofitting a semantic
> change of the port lifetime itself, all the complications I complain about
> below should disappear.
In the above I meant drm_dp_destroy_port or drm_dp_destroy_connector_work.

Aside: I think creating a kref for the final kfree would also solve a
bunch of other issues in a much neater way: In

commit f038c5b99fc1332f558b495d136d4f433ee65caa
Author: Lyude Paul <lyude at redhat.com>
Date:   Tue Nov 13 17:46:14 2018 -0500

    drm/dp_mst: Skip validating ports during destruction, just ref

we could use that kfree reference to make sure the port pointer is alive.
This of course means that drm_dp_update_payload_part1() would also need to
use the kfree reference for the vcpi allocations. And probably everywhere
else (e.g. in your nouveau series for payload information).

That would give us a very clear separation overall between "port can still
be used because it's not yet unplugged" vs. "port only hangs
around
because a bunch of vcpi allocations and other payload things upstream of
the port might still need the port lookup key to free resources". Instead
of again sprinkling complicated conditions and magic tricks all over the
code to figure out whethe we should allow the get_validate_port to succeed
or not, or the modeset to succeed or not.
-Daniel
> 
> Piles of comments below.
> 
> Cheers, Daniel
> 
> >  - Use the small changes to drm_dp_put_port() to add even more error
> >    checking to make misusage of the helpers more obvious. I added this
> >    after having to chase down various use-after-free conditions that
> >    started popping up from the new helpers so no one else has to
> >    troubleshoot that.
> >  - Move some accidental DRM_DEBUG_KMS() calls to DRM_DEBUG_ATOMIC()
> >  - Update documentation again, note that find/release() should both
not be
> >    called on the same port in a single atomic check phase (but
multiple
> >    calls to one or the other is OK)
> > 
> > Changes since v4:
> >  - Don't skip the atomic checks for VCPI allocations if no new
VCPI
> >    allocations happen in a state. This makes the next change I'm
about
> >    to list here a lot easier to implement.
> >  - Don't ignore VCPI allocations on destroyed ports, instead
ensure that
> >    when ports are destroyed and still have VCPI allocations in the
> >    topology state, the only state changes allowed are releasing said
> >    ports' VCPI. This prevents a state with a mix of VCPI
allocations
> >    from destroyed ports, and allocations from valid ports.
> > 
> > Changes since v3:
> >  - Don't release VCPI allocations in the topology state
immediately in
> >    drm_dp_atomic_release_vcpi_slots(), instead mark them as 0 and skip
> >    over them in drm_dp_mst_duplicate_state(). This makes it so
> >    drm_dp_atomic_release_vcpi_slots() is still idempotent while also
> >    throwing warnings if the driver messes up it's book keeping and
tries
> >    to release VCPI slots on a port that doesn't have any
pre-existing
> >    VCPI allocation - danvet
> >  - Change mst_state/state in some debugging messages to "mst
state"
> > 
> > Changes since v2:
> >  - Use kmemdup() for duplicating MST state - danvet
> >  - Move port validation out of duplicate state callback - danvet
> >  - Handle looping through MST topology states in
> >    drm_dp_mst_atomic_check() so the driver doesn't have to do it
> >  - Fix documentation in drm_dp_atomic_find_vcpi_slots()
> >  - Move the atomic check for each individual topology state into
it's
> >    own function, reduces indenting
> >  - Don't consider "stale" MST ports when calculating the
bandwidth
> >    requirements. This is needed because originally we relied on the
> >    state duplication functions to prune any stale ports from the new
> >    state, which would prevent us from incorrectly considering their
> >    bandwidth requirements alongside legitimate new payloads.
> >  - Add function references in drm_dp_atomic_release_vcpi_slots() -
danvet
> >  - Annotate atomic VCPI and atomic check functions with __must_check
> >    - danvet
> > 
> > Changes since v1:
> >  - Don't use the now-removed ->atomic_check() for private
objects hook,
> >    just give drivers a function to call themselves
> > 
> > Signed-off-by: Lyude Paul <lyude at redhat.com>
> > Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
> > ---
> >  drivers/gpu/drm/drm_dp_mst_topology.c | 294
+++++++++++++++++++++++---
> >  drivers/gpu/drm/i915/intel_display.c  |   4 +
> >  drivers/gpu/drm/i915/intel_dp_mst.c   |  64 +++---
> >  include/drm/drm_dp_mst_helper.h       |  24 ++-
> >  4 files changed, 319 insertions(+), 67 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/drm_dp_mst_topology.c
b/drivers/gpu/drm/drm_dp_mst_topology.c
> > index 00fbe7a2699d..183d0e832ced 100644
> > --- a/drivers/gpu/drm/drm_dp_mst_topology.c
> > +++ b/drivers/gpu/drm/drm_dp_mst_topology.c
> > @@ -2617,21 +2617,42 @@ static int drm_dp_init_vcpi(struct
drm_dp_mst_topology_mgr *mgr,
> >  }
> >  
> >  /**
> > - * drm_dp_atomic_find_vcpi_slots() - Find and add vcpi slots to the
state
> > + * drm_dp_atomic_find_vcpi_slots() - Find and add VCPI slots to the
state
> >   * @state: global atomic state
> >   * @mgr: MST topology manager for the port
> >   * @port: port to find vcpi slots for
> >   * @pbn: bandwidth required for the mode in PBN
> >   *
> > - * RETURNS:
> > - * Total slots in the atomic state assigned for this port or error
> > + * Allocates VCPI slots to @port, replacing any previous VCPI
allocations it
> > + * may have had. Any atomic drivers which support MST must call this
function
> > + * in their &drm_encoder_helper_funcs.atomic_check() callback to
change the
> > + * current VCPI allocation for the new state, but only when
> > + * &drm_crtc_state.mode_changed or
&drm_crtc_state.connectors_changed is set
> > + * to ensure compatibility with userspace applications that still use
the
> > + * legacy modesetting UAPI.
> > + *
> > + * Allocations set by this function are not checked against the
bandwidth
> > + * restraints of @mgr until the driver calls
drm_dp_mst_atomic_check().
> > + *
> > + * Additionally, it is OK to call this function multiple times on the
same
> > + * @port as needed. It is not OK however, to call this function and
> > + * drm_dp_atomic_release_vcpi_slots() in the same atomic check phase.
> > + *
> > + * See also:
> > + * drm_dp_atomic_release_vcpi_slots()
> > + * drm_dp_mst_atomic_check()
> > + *
> > + * Returns:
> > + * Total slots in the atomic state assigned for this port, or a
negative error
> > + * code if the port no longer exists
> >   */
> >  int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
> >  				  struct drm_dp_mst_topology_mgr *mgr,
> >  				  struct drm_dp_mst_port *port, int pbn)
> >  {
> >  	struct drm_dp_mst_topology_state *topology_state;
> > -	int req_slots;
> > +	struct drm_dp_vcpi_allocation *pos, *vcpi = NULL;
> > +	int prev_slots, req_slots, ret;
> >  
> >  	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
> >  	if (IS_ERR(topology_state))
> > @@ -2640,20 +2661,60 @@ int drm_dp_atomic_find_vcpi_slots(struct
drm_atomic_state *state,
> >  	port = drm_dp_get_validated_port_ref(mgr, port);
> >  	if (port == NULL)
> >  		return -EINVAL;
> > -	req_slots = DIV_ROUND_UP(pbn, mgr->pbn_div);
> > -	DRM_DEBUG_KMS("vcpi slots req=%d, avail=%d\n",
> > -			req_slots, topology_state->avail_slots);
> >  
> > -	if (req_slots > topology_state->avail_slots) {
> > -		drm_dp_put_port(port);
> > -		return -ENOSPC;
> > +	topology_state->vcpi_allocated = true;
> > +
> > +	/* Find the current allocation for this port, if any */
> > +	list_for_each_entry(pos, &topology_state->vcpis, next) {
> > +		if (pos->port == port) {
> > +			vcpi = pos;
> > +			prev_slots = vcpi->vcpi;
> > +
> > +			/*
> > +			 * This should never happen, unless the driver tries
> > +			 * releasing and allocating the same VCPI allocation,
> > +			 * which is an error
> > +			 */
> > +			if (WARN_ON(!prev_slots)) {
> 
> 
> > +				DRM_ERROR("cannot allocate and release VCPI on [MST PORT:%p]
in the same state\n",
> > +					  port);
> > +				return -EINVAL;
> > +			}
> 
> Allowing atomic changes to output state (like disabling it on one crtc,
> and then enabling on the other, thus both releasing and re-allocating the
> vcpi in one go) is kinda the point of atomic. So definitely can't
EINVAL
> here.
> 
> Sounds like we need more testcases in igt?
> 
> > +
> > +			break;
> > +		}
> >  	}
> > +	if (!vcpi)
> > +		prev_slots = 0;
> >  
> > -	topology_state->avail_slots -= req_slots;
> > -	DRM_DEBUG_KMS("vcpi slots avail=%d",
topology_state->avail_slots);
> > +	req_slots = DIV_ROUND_UP(pbn, mgr->pbn_div);
> > +
> > +	DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] [MST PORT:%p] VCPI %d ->
%d\n",
> > +			 port->connector->base.id, port->connector->name,
> > +			 port, prev_slots, req_slots);
> > +
> > +	/* Add the new allocation to the state */
> > +	if (!vcpi) {
> > +		vcpi = kzalloc(sizeof(*vcpi), GFP_KERNEL);
> > +		if (!vcpi) {
> > +			ret = -ENOMEM;
> > +			goto out;
> > +		}
> >  
> > +		vcpi->port = drm_dp_get_validated_port_ref(mgr, port);
> > +		if (!vcpi->port) {
> > +			ret = -EINVAL;
> > +			kfree(vcpi);
> > +			goto out;
> > +		}
> > +		list_add(&vcpi->next, &topology_state->vcpis);
> > +	}
> > +	vcpi->vcpi = req_slots;
> > +
> > +	ret = req_slots;
> > +out:
> >  	drm_dp_put_port(port);
> > -	return req_slots;
> > +	return ret;
> >  }
> >  EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
> >  
> > @@ -2661,31 +2722,66 @@ EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
> >   * drm_dp_atomic_release_vcpi_slots() - Release allocated vcpi slots
> >   * @state: global atomic state
> >   * @mgr: MST topology manager for the port
> > - * @slots: number of vcpi slots to release
> > + * @port: The port to release the VCPI slots from
> >   *
> > - * RETURNS:
> > - * 0 if @slots were added back to
&drm_dp_mst_topology_state->avail_slots or
> > - * negative error code
> > + * Releases any VCPI slots that have been allocated to a port in the
atomic
> > + * state. Any atomic drivers which support MST must call this
function in
> > + * their &drm_connector_helper_funcs.atomic_check() callback when
the
> > + * connector will no longer have VCPI allocated (e.g. because
it's CRTC was
> > + * removed) when it had VCPI allocated in the previous atomic state.
> > + *
> > + * It is OK to call this even if @port has been removed from the
system.
> > + * Additionally, it is OK to call this function multiple times on the
same
> > + * @port as needed. It is not OK however, to call this function and
> > + * drm_dp_atomic_find_vcpi_slots() on the same @port in a single
atomic check
> > + * phase.
> > + *
> > + * See also:
> > + * drm_dp_atomic_find_vcpi_slots()
> > + * drm_dp_mst_atomic_check()
> > + *
> > + * Returns:
> > + * 0 if all slots for this port were added back to
> > + * &drm_dp_mst_topology_state.avail_slots or negative error code
> >   */
> >  int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
> >  				     struct drm_dp_mst_topology_mgr *mgr,
> > -				     int slots)
> > +				     struct drm_dp_mst_port *port)
> >  {
> >  	struct drm_dp_mst_topology_state *topology_state;
> > +	struct drm_dp_vcpi_allocation *pos;
> > +	bool found = false;
> >  
> >  	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
> >  	if (IS_ERR(topology_state))
> >  		return PTR_ERR(topology_state);
> >  
> > -	/* We cannot rely on port->vcpi.num_slots to update
> > -	 * topology_state->avail_slots as the port may not exist if the
parent
> > -	 * branch device was unplugged. This should be fixed by tracking
> > -	 * per-port slot allocation in drm_dp_mst_topology_state instead of
> > -	 * depending on the caller to tell us how many slots to release.
> > -	 */
> > -	topology_state->avail_slots += slots;
> > -	DRM_DEBUG_KMS("vcpi slots released=%d, avail=%d\n",
> > -			slots, topology_state->avail_slots);
> > +	list_for_each_entry(pos, &topology_state->vcpis, next) {
> 
> > +		if (pos->port == port) {
> > +			found = true;
> > +			break;
> > +		}
> > +	}
> > +	if (WARN_ON(!found)) {
> > +		DRM_ERROR("no VCPI for [MST PORT:%p] found in mst state
%p\n",
> > +			  port, &topology_state->base);
> > +		return -EINVAL;
> > +	}
> > +
> > +	DRM_DEBUG_ATOMIC("[MST PORT:%p] VCPI %d -> 0\n", port,
pos->vcpi);
> > +	if (pos->vcpi) {
> > +		/*
> > +		 * This should be safe-as the previous state will still hold a
> > +		 * reference to this port until we swap states. If it's not,
> > +		 * the driver has a bookkeeping error
> > +		 */
> > +		if (WARN_ON(drm_dp_put_port(port))) {
> 
> The usual way to do this (instead of adding a return value to the normal
> _put function) is something like
> 
> 	kref_put(port, function_that_with_just_a_WARNING_DRM_ERROR)
> 
> avoids the need for patch 2.
> 
> > +			DRM_ERROR("cannot allocate and release VCPI on [MST PORT:%p]
in the same state\n",
> > +				  port);
> > +			return -EINVAL;
> > +		}
> > +		pos->vcpi = 0;
> > +	}
> >  
> >  	return 0;
> >  }
> > @@ -3115,15 +3211,45 @@ static void
drm_dp_destroy_connector_work(struct work_struct *work)
> >  static struct drm_private_state *
> >  drm_dp_mst_duplicate_state(struct drm_private_obj *obj)
> >  {
> > -	struct drm_dp_mst_topology_state *state;
> > +	struct drm_dp_mst_topology_state *state, *old_state > > +	
to_dp_mst_topology_state(obj->state);
> > +	struct drm_dp_vcpi_allocation *pos, *vcpi;
> >  
> > -	state = kmemdup(obj->state, sizeof(*state), GFP_KERNEL);
> > +	state = kmemdup(old_state, sizeof(*state), GFP_KERNEL);
> >  	if (!state)
> >  		return NULL;
> >  
> >  	__drm_atomic_helper_private_obj_duplicate_state(obj,
&state->base);
> >  
> > +	INIT_LIST_HEAD(&state->vcpis);
> > +	state->vcpi_allocated = false;
> > +
> > +	list_for_each_entry(pos, &old_state->vcpis, next) {
> > +		/* Prune leftover freed VCPI allocations */
> > +		if (!pos->vcpi)
> > +			continue;
> > +
> > +		vcpi = kmemdup(pos, sizeof(*vcpi), GFP_KERNEL);
> > +		if (!vcpi)
> > +			goto fail;
> > +
> > +		/* Take a plain old kref here and don't validate the port, as
> > +		 * it's topology might not even exist anymore
> > +		 */
> > +		kref_get(&vcpi->port->kref);
> > +		list_add(&vcpi->next, &state->vcpis);
> > +	}
> > +
> >  	return &state->base;
> > +
> > +fail:
> > +	list_for_each_entry_safe(pos, vcpi, &state->vcpis, next) {
> > +		drm_dp_put_port(pos->port);
> > +		kfree(pos);
> > +	}
> > +	kfree(state);
> > +
> > +	return NULL;
> >  }
> >  
> >  static void drm_dp_mst_destroy_state(struct drm_private_obj *obj,
> > @@ -3131,14 +3257,116 @@ static void drm_dp_mst_destroy_state(struct
drm_private_obj *obj,
> >  {
> >  	struct drm_dp_mst_topology_state *mst_state > >  	
to_dp_mst_topology_state(state);
> > +	struct drm_dp_vcpi_allocation *pos, *tmp;
> > +
> > +	list_for_each_entry_safe(pos, tmp, &mst_state->vcpis, next) {
> > +		/* We only keep references to ports with non-zero VCPIs */
> > +		if (pos->vcpi)
> > +			drm_dp_put_port(pos->port);
> > +		kfree(pos);
> > +	}
> >  
> >  	kfree(mst_state);
> >  }
> >  
> > -static const struct drm_private_state_funcs mst_state_funcs = {
> > +static inline int
> > +drm_dp_mst_atomic_check_topology_state(struct drm_dp_mst_topology_mgr
*mgr,
> > +				       struct drm_dp_mst_topology_state *mst_state)
> > +{
> > +	struct drm_dp_vcpi_allocation *vcpi;
> > +	struct drm_dp_mst_port *port;
> > +	int avail_slots = 63, ret;
> > +
> > +	/* There's no possible scenario where releasing VCPI or keeping
it the
> > +	 * same would make the state invalid
> > +	 */
> > +	if (!mst_state->vcpi_allocated) {
> 
> Do we need this? If not I'm always in favour of removing code if
it's not
> needed :-)
> 
> > +		DRM_DEBUG_ATOMIC("[MST MGR:%p] state %p allocates no VCPI,
check skipped\n",
> > +				 mgr, &mst_state->base);
> > +		return 0;
> > +	}
> > +
> > +	list_for_each_entry(vcpi, &mst_state->vcpis, next) {
> > +		/* Releasing VCPI is always OK-even if the port is gone */
> > +		if (!vcpi->vcpi) {
> > +			DRM_DEBUG_ATOMIC("[MST PORT:%p] releases all VCPI
slots\n",
> > +					 vcpi->port);
> > +			continue;
> > +		}
> > +
> > +		port = drm_dp_get_validated_port_ref(mgr, vcpi->port);
> > +		if (!port) {
> > +			DRM_DEBUG_ATOMIC("[MST PORT:%p] is gone but still has VCPI,
cannot add new VCPI\n",
> > +					 vcpi->port);
> > +			return -EINVAL;
> > +		}
> 
> For find_vcpi we've already done this. For unplugged vcpi, we can't
do
> this here I think - otherwise if you enable 2 mst outputs on the same
> branch, and unplug both, old non-atomic userspace can never ever again
> disable them because it can't disable both. The check (to skip already
> deallocated ports) above isn't enough to handle that.
> 
> > +
> > +		DRM_DEBUG_ATOMIC("[MST PORT:%p] requires %d vcpi
slots\n",
> > +				 vcpi->port, vcpi->vcpi);
> > +
> > +		avail_slots -= vcpi->vcpi;
> > +		if (avail_slots < 0) {
> > +			DRM_DEBUG_ATOMIC("[MST PORT:%p] not enough VCPI slots in mst
state %p (avail=%d)\n",
> > +					 vcpi->port, mst_state,
> > +					 avail_slots + vcpi->vcpi);
> > +			ret = -ENOSPC;
> > +			goto port_fail;
> > +		}
> > +
> > +		drm_dp_put_port(port);
> > +	}
> > +	DRM_DEBUG_ATOMIC("[MST MGR:%p] mst state %p VCPI avail=%d
used=%d\n",
> > +			 mgr, mst_state, avail_slots,
> > +			 63 - avail_slots);
> > +
> > +	return 0;
> > +port_fail:
> > +	drm_dp_put_port(port);
> > +	return ret;
> > +}
> > +
> > +/**
> > + * drm_dp_mst_atomic_check - Check that the new state of an MST
topology in an
> > + * atomic update is valid
> > + * @state: Pointer to the new &struct drm_dp_mst_topology_state
> > + *
> > + * Checks the given topology state for an atomic update to ensure
that it's
> > + * valid. This includes checking whether there's enough bandwidth
to support
> > + * the new VCPI allocations in the atomic update.
> > + *
> > + * Any atomic drivers supporting DP MST must make sure to call this
after
> > + * checking the rest of their state in their
> > + * &drm_mode_config_funcs.atomic_check() callback.
> > + *
> > + * See also:
> > + * drm_dp_atomic_find_vcpi_slots()
> > + * drm_dp_atomic_release_vcpi_slots()
> > + *
> > + * Returns:
> > + *
> > + * 0 if the new state is valid, negative error code otherwise.
> > + */
> > +int drm_dp_mst_atomic_check(struct drm_atomic_state *state)
> > +{
> > +	struct drm_dp_mst_topology_mgr *mgr;
> > +	struct drm_dp_mst_topology_state *mst_state;
> > +	int i, ret = 0;
> > +
> > +	for_each_new_mst_mgr_in_state(state, mgr, mst_state, i) {
> > +		ret = drm_dp_mst_atomic_check_topology_state(mgr, mst_state);
> > +		if (ret)
> > +			break;
> > +	}
> > +
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL(drm_dp_mst_atomic_check);
> > +
> > +const struct drm_private_state_funcs drm_dp_mst_topology_state_funcs
= {
> >  	.atomic_duplicate_state = drm_dp_mst_duplicate_state,
> >  	.atomic_destroy_state = drm_dp_mst_destroy_state,
> >  };
> > +EXPORT_SYMBOL(drm_dp_mst_topology_state_funcs);
> >  
> >  /**
> >   * drm_atomic_get_mst_topology_state: get MST topology state
> > @@ -3216,13 +3444,11 @@ int drm_dp_mst_topology_mgr_init(struct
drm_dp_mst_topology_mgr *mgr,
> >  		return -ENOMEM;
> >  
> >  	mst_state->mgr = mgr;
> > -
> > -	/* max. time slots - one slot for MTP header */
> > -	mst_state->avail_slots = 63;
> > +	INIT_LIST_HEAD(&mst_state->vcpis);
> >  
> >  	drm_atomic_private_obj_init(&mgr->base,
> >  				    &mst_state->base,
> > -				    &mst_state_funcs);
> > +				    &drm_dp_mst_topology_state_funcs);
> >  
> >  	return 0;
> >  }
> > diff --git a/drivers/gpu/drm/i915/intel_display.c
b/drivers/gpu/drm/i915/intel_display.c
> > index 132e978227fb..8e85e04a5948 100644
> > --- a/drivers/gpu/drm/i915/intel_display.c
> > +++ b/drivers/gpu/drm/i915/intel_display.c
> > @@ -12592,6 +12592,10 @@ static int intel_atomic_check(struct
drm_device *dev,
> >  				       "[modeset]" : "[fastset]");
> >  	}
> >  
> > +	ret = drm_dp_mst_atomic_check(state);
> > +	if (ret)
> > +		return ret;
> > +
> >  	if (any_ms) {
> >  		ret = intel_modeset_checks(state);
> >  
> > diff --git a/drivers/gpu/drm/i915/intel_dp_mst.c
b/drivers/gpu/drm/i915/intel_dp_mst.c
> > index 4de247ddf05f..be2d1ac7d87d 100644
> > --- a/drivers/gpu/drm/i915/intel_dp_mst.c
> > +++ b/drivers/gpu/drm/i915/intel_dp_mst.c
> > @@ -41,8 +41,13 @@ static bool intel_dp_mst_compute_config(struct
intel_encoder *encoder,
> >  	struct drm_connector *connector = conn_state->connector;
> >  	void *port = to_intel_connector(connector)->port;
> >  	struct drm_atomic_state *state = pipe_config->base.state;
> > +	struct drm_crtc *crtc = pipe_config->base.crtc;
> > +	struct drm_crtc_state *old_crtc_state > > +	
drm_atomic_get_old_crtc_state(state, crtc);
> > +	struct drm_crtc_state *new_crtc_state = &pipe_config->base;
> >  	int bpp;
> > -	int lane_count, slots = 0;
> > +	int lane_count, slots > > +	
to_intel_crtc_state(old_crtc_state)->dp_m_n.tu;
> >  	const struct drm_display_mode *adjusted_mode =
&pipe_config->base.adjusted_mode;
> >  	int mst_pbn;
> >  	bool constant_n = drm_dp_has_quirk(&intel_dp->desc,
> > @@ -77,8 +82,12 @@ static bool intel_dp_mst_compute_config(struct
intel_encoder *encoder,
> >  	mst_pbn = drm_dp_calc_pbn_mode(adjusted_mode->crtc_clock, bpp);
> >  	pipe_config->pbn = mst_pbn;
> >  
> > -	/* Zombie connectors can't have VCPI slots */
> > -	if (!drm_connector_is_unregistered(connector)) {
> > +	/* Only change VCPI allocation on actual mode changes, to prevent us
> > +	 * from trying to allocate VCPI to ports that no longer exist when
we
> > +	 * may just be trying to disable DPMS on them
> > +	 */
> > +	if (new_crtc_state->mode_changed ||
> > +	    new_crtc_state->connectors_changed) {
> 
> Why exactly is this needed? We only call compute_config on the enocoder
> when enabling an output. Not when disabling one. Similar for the atomic
> helpers and the various encoder callbacks (you have the same change in the
> nouveau code). Did this actually blow up while testing?
> 
> >  		slots = drm_dp_atomic_find_vcpi_slots(state,
> >  						      &intel_dp->mst_mgr,
> >  						      port,
> > @@ -107,36 +116,39 @@ static bool intel_dp_mst_compute_config(struct
intel_encoder *encoder,
> >  	return true;
> >  }
> >  
> > -static int intel_dp_mst_atomic_check(struct drm_connector *connector,
> > -		struct drm_connector_state *new_conn_state)
> > +static int
> > +intel_dp_mst_atomic_check(struct drm_connector *connector,
> > +			  struct drm_connector_state *new_conn_state)
> >  {
> > +	struct intel_connector *intel_connector > > +	
to_intel_connector(connector);
> > +	struct drm_dp_mst_topology_mgr *mgr > > +	
&intel_connector->mst_port->mst_mgr;
> > +	struct drm_dp_mst_port *port = intel_connector->port;
> >  	struct drm_atomic_state *state = new_conn_state->state;
> > -	struct drm_connector_state *old_conn_state;
> > -	struct drm_crtc *old_crtc;
> > -	struct drm_crtc_state *crtc_state;
> > -	int slots, ret = 0;
> > +	struct drm_connector_state *old_conn_state > > +	
drm_atomic_get_old_connector_state(state, connector);
> > +	struct drm_crtc *old_crtc = old_conn_state->crtc,
> > +			*new_crtc = new_conn_state->crtc;
> > +	struct drm_crtc_state *old_crtc_state, *new_crtc_state;
> >  
> > -	old_conn_state = drm_atomic_get_old_connector_state(state,
connector);
> > -	old_crtc = old_conn_state->crtc;
> 
> Bikeshed: Reworking all the above hides the real changes quite a bit.
> 
> >  	if (!old_crtc)
> > -		return ret;
> > +		return 0;
> >  
> > -	crtc_state = drm_atomic_get_new_crtc_state(state, old_crtc);
> > -	slots = to_intel_crtc_state(crtc_state)->dp_m_n.tu;
> > -	if (drm_atomic_crtc_needs_modeset(crtc_state) && slots >
0) {
> > -		struct drm_dp_mst_topology_mgr *mgr;
> > -		struct drm_encoder *old_encoder;
> > +	old_crtc_state = drm_atomic_get_old_crtc_state(state, old_crtc);
> > +	if (!old_crtc_state ||
> 
> !old_crtc_state would be a framework bug. Just remove that check.
> 
> > +	    !to_intel_crtc_state(old_crtc_state)->dp_m_n.tu)
> > +		return 0;
> 
> Hm, why do we need this? Having a DP port previously enabled with a 0 tu
> sounds like a pretty big driver bug.
> >  
> > -		old_encoder = old_conn_state->best_encoder;
> > -		mgr = &enc_to_mst(old_encoder)->primary->dp.mst_mgr;
> > +	/* If we switched CRTCs, clear the previous one's allocation */
> > +	new_crtc_state = drm_atomic_get_new_crtc_state(state, old_crtc);
> > +	if (new_crtc_state->connectors_changed &&
!new_crtc_state->enable)
> > +		to_intel_crtc_state(new_crtc_state)->dp_m_n.tu = 0;
> 
> No idea why we need to clear this here? Shouldn't we just clear this
> unconditionally if we release the allocation.
> 
> Also naming convention, I'd call this old_crtc_new_state, and maybe
> old_crtc_old_state. To make it clear these are the old/new states for
> old_crtc. Instead of old/new state for the new crtc.
> 
> >  
> > -		ret = drm_dp_atomic_release_vcpi_slots(state, mgr, slots);
> > -		if (ret)
> > -			DRM_DEBUG_KMS("failed releasing %d vcpi slots:%d\n",
slots, ret);
> > -		else
> > -			to_intel_crtc_state(crtc_state)->dp_m_n.tu = 0;
> > -	}
> > -	return ret;
> > +	if (new_crtc)
> > +		return 0;
> 
> Not sure why we shouldn't release this? If we handle this in helpers,
it
> would remove piles of special casing code here I think. Plus remove
> special code in the hlpers.
> 
> 
> > +
> > +	return drm_dp_atomic_release_vcpi_slots(state, mgr, port);
> >  }
> >  
> >  static void intel_mst_disable_dp(struct intel_encoder *encoder,
> > diff --git a/include/drm/drm_dp_mst_helper.h
b/include/drm/drm_dp_mst_helper.h
> > index 3faceb66f5cb..2392e12b6bce 100644
> > --- a/include/drm/drm_dp_mst_helper.h
> > +++ b/include/drm/drm_dp_mst_helper.h
> > @@ -406,9 +406,16 @@ struct drm_dp_payload {
> >  
> >  #define to_dp_mst_topology_state(x) container_of(x, struct
drm_dp_mst_topology_state, base)
> >  
> > +struct drm_dp_vcpi_allocation {
> 
> I think a comment here that this pointer holds a full reference (even past
> when the connector is unplugged) would be good. Otherwise the loop in
> release_vcpi would be buggy.
> 
> > +	struct drm_dp_mst_port *port;
> > +	int vcpi;
> > +	struct list_head next;
> > +};
> > +
> >  struct drm_dp_mst_topology_state {
> >  	struct drm_private_state base;
> > -	int avail_slots;
> > +	struct list_head vcpis;
> > +	u8 vcpi_allocated;
> >  	struct drm_dp_mst_topology_mgr *mgr;
> >  };
> >  
> > @@ -619,14 +626,17 @@ void drm_dp_mst_topology_mgr_suspend(struct
drm_dp_mst_topology_mgr *mgr);
> >  int drm_dp_mst_topology_mgr_resume(struct drm_dp_mst_topology_mgr
*mgr);
> >  struct drm_dp_mst_topology_state
*drm_atomic_get_mst_topology_state(struct drm_atomic_state *state,
> >  								    struct drm_dp_mst_topology_mgr *mgr);
> > -int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
> > -				  struct drm_dp_mst_topology_mgr *mgr,
> > -				  struct drm_dp_mst_port *port, int pbn);
> > -int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
> > -				     struct drm_dp_mst_topology_mgr *mgr,
> > -				     int slots);
> > +int __must_check
> > +drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
> > +			      struct drm_dp_mst_topology_mgr *mgr,
> > +			      struct drm_dp_mst_port *port, int pbn);
> > +int __must_check
> > +drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
> > +				 struct drm_dp_mst_topology_mgr *mgr,
> > +				 struct drm_dp_mst_port *port);
> >  int drm_dp_send_power_updown_phy(struct drm_dp_mst_topology_mgr *mgr,
> >  				 struct drm_dp_mst_port *port, bool power_up);
> > +int __must_check drm_dp_mst_atomic_check(struct drm_atomic_state
*state);
> >  
> >  extern const struct drm_private_state_funcs
drm_dp_mst_topology_state_funcs;
> >  
> > -- 
> > 2.19.1
> > 
> 
> -- 
> Daniel Vetter
> Software Engineer, Intel Corporation
> http://blog.ffwll.ch
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

On Mon, 2018-11-26 at 22:04 +0100, Daniel Vetter wrote:
> On Thu, Nov 15, 2018 at 07:50:05PM -0500, Lyude Paul wrote:
> > There has been a TODO waiting for quite a long time in
> > drm_dp_mst_topology.c:
> > 
> > 	/* We cannot rely on port->vcpi.num_slots to update
> > 	 * topology_state->avail_slots as the port may not exist if the
parent
> > 	 * branch device was unplugged. This should be fixed by tracking
> > 	 * per-port slot allocation in drm_dp_mst_topology_state instead of
> > 	 * depending on the caller to tell us how many slots to release.
> > 	 */
> > 
> > That's not the only reason we should fix this: forcing the driver
to
> > track the VCPI allocations throughout a state's atomic check is
> > error prone, because it means that extra care has to be taken with the
> > order that drm_dp_atomic_find_vcpi_slots() and
> > drm_dp_atomic_release_vcpi_slots() are called in in order to ensure
> > idempotency. Currently the only driver actually using these helpers,
> > i915, doesn't even do this correctly: multiple ->best_encoder()
checks
> > with i915's current implementation would not be idempotent and
would
> > over-allocate VCPI slots, something I learned trying to implement
> > fallback retraining in MST.
> > 
> > So: simplify this whole mess, and teach
drm_dp_atomic_find_vcpi_slots()
> > and drm_dp_atomic_release_vcpi_slots() to track the VCPI allocations
for
> > each port. This allows us to ensure idempotency without having to rely
> > on the driver as much. Additionally: the driver doesn't need to do
any
> > kind of VCPI slot tracking anymore if it doesn't need it for
it's own
> > internal state.
> > 
> > Additionally; this adds a new drm_dp_mst_atomic_check() helper which
> > must be used by atomic drivers to perform validity checks for the new
> > VCPI allocations incurred by a state.
> > 
> > Also: update the documentation and make it more obvious that these
> > /must/ be called by /all/ atomic drivers supporting MST.
> > 
> > Changes since v6:
> >  - Keep a kref to all of the ports we have allocations on. This
required
> >    a good bit of changing to when we call drm_dp_find_vcpi_slots(),
> >    mainly that we need to ensure that we only redo VCPI allocations on
> >    actual mode or CRTC changes, not crtc_state->active changes.
> >    Additionally, we no longer take the registration of the DRM
connector
> >    for each port into account because so long as we have a kref to the
> >    port in the new or previous atomic state, the connector will stay
> >    registered.
> 
> I write an entire pile of small nitpits (still included most of them
> below), until I realized this here wont work. Delaying the call to destroy
> the connector (well, unregister it really) wreaks the design we've come
up
> with thus far, resulting in most of my comments here.
> 
> Instead, all we need to do is delay the kfree(port) at the bottom of
> drm_dp_destroy_port(). The vcpi allocation structure _only_ needs the
> pointer value to stay valid, as a lookup key. It doesn't care at all
about
> anything actually stored in there. So the only thing we need to delay is
> the kfree. I think the simplest way to achieve this is to add a 2nd kref
> (port->kfree_ref or something like that), with on reference held by the
> port itself (released in drm_dp_destroy_port), and the other one held
> as-needed by the vcpi allocation lists.
> 
> I think if we go with this design instead of retrofitting a semantic
> change of the port lifetime itself, all the complications I complain about
> below should disappear.
> 
> Piles of comments below.
> 
> Cheers, Daniel
> 
> >  - Use the small changes to drm_dp_put_port() to add even more error
> >    checking to make misusage of the helpers more obvious. I added this
> >    after having to chase down various use-after-free conditions that
> >    started popping up from the new helpers so no one else has to
> >    troubleshoot that.
> >  - Move some accidental DRM_DEBUG_KMS() calls to DRM_DEBUG_ATOMIC()
> >  - Update documentation again, note that find/release() should both
not be
> >    called on the same port in a single atomic check phase (but
multiple
> >    calls to one or the other is OK)
> > 
> > Changes since v4:
> >  - Don't skip the atomic checks for VCPI allocations if no new
VCPI
> >    allocations happen in a state. This makes the next change I'm
about
> >    to list here a lot easier to implement.
> >  - Don't ignore VCPI allocations on destroyed ports, instead
ensure that
> >    when ports are destroyed and still have VCPI allocations in the
> >    topology state, the only state changes allowed are releasing said
> >    ports' VCPI. This prevents a state with a mix of VCPI
allocations
> >    from destroyed ports, and allocations from valid ports.
> > 
> > Changes since v3:
> >  - Don't release VCPI allocations in the topology state
immediately in
> >    drm_dp_atomic_release_vcpi_slots(), instead mark them as 0 and skip
> >    over them in drm_dp_mst_duplicate_state(). This makes it so
> >    drm_dp_atomic_release_vcpi_slots() is still idempotent while also
> >    throwing warnings if the driver messes up it's book keeping and
tries
> >    to release VCPI slots on a port that doesn't have any
pre-existing
> >    VCPI allocation - danvet
> >  - Change mst_state/state in some debugging messages to "mst
state"
> > 
> > Changes since v2:
> >  - Use kmemdup() for duplicating MST state - danvet
> >  - Move port validation out of duplicate state callback - danvet
> >  - Handle looping through MST topology states in
> >    drm_dp_mst_atomic_check() so the driver doesn't have to do it
> >  - Fix documentation in drm_dp_atomic_find_vcpi_slots()
> >  - Move the atomic check for each individual topology state into
it's
> >    own function, reduces indenting
> >  - Don't consider "stale" MST ports when calculating the
bandwidth
> >    requirements. This is needed because originally we relied on the
> >    state duplication functions to prune any stale ports from the new
> >    state, which would prevent us from incorrectly considering their
> >    bandwidth requirements alongside legitimate new payloads.
> >  - Add function references in drm_dp_atomic_release_vcpi_slots() -
danvet
> >  - Annotate atomic VCPI and atomic check functions with __must_check
> >    - danvet
> > 
> > Changes since v1:
> >  - Don't use the now-removed ->atomic_check() for private
objects hook,
> >    just give drivers a function to call themselves
> > 
> > Signed-off-by: Lyude Paul <lyude at redhat.com>
> > Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
> > ---
> >  drivers/gpu/drm/drm_dp_mst_topology.c | 294
+++++++++++++++++++++++---
> >  drivers/gpu/drm/i915/intel_display.c  |   4 +
> >  drivers/gpu/drm/i915/intel_dp_mst.c   |  64 +++---
> >  include/drm/drm_dp_mst_helper.h       |  24 ++-
> >  4 files changed, 319 insertions(+), 67 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/drm_dp_mst_topology.c
> > b/drivers/gpu/drm/drm_dp_mst_topology.c
> > index 00fbe7a2699d..183d0e832ced 100644
> > --- a/drivers/gpu/drm/drm_dp_mst_topology.c
> > +++ b/drivers/gpu/drm/drm_dp_mst_topology.c
> > @@ -2617,21 +2617,42 @@ static int drm_dp_init_vcpi(struct
> > drm_dp_mst_topology_mgr *mgr,
> >  }
> >  
> >  /**
> > - * drm_dp_atomic_find_vcpi_slots() - Find and add vcpi slots to the
state
> > + * drm_dp_atomic_find_vcpi_slots() - Find and add VCPI slots to the
state
> >   * @state: global atomic state
> >   * @mgr: MST topology manager for the port
> >   * @port: port to find vcpi slots for
> >   * @pbn: bandwidth required for the mode in PBN
> >   *
> > - * RETURNS:
> > - * Total slots in the atomic state assigned for this port or error
> > + * Allocates VCPI slots to @port, replacing any previous VCPI
allocations
> > it
> > + * may have had. Any atomic drivers which support MST must call this
> > function
> > + * in their &drm_encoder_helper_funcs.atomic_check() callback to
change
> > the
> > + * current VCPI allocation for the new state, but only when
> > + * &drm_crtc_state.mode_changed or
&drm_crtc_state.connectors_changed is
> > set
> > + * to ensure compatibility with userspace applications that still use
the
> > + * legacy modesetting UAPI.
> > + *
> > + * Allocations set by this function are not checked against the
bandwidth
> > + * restraints of @mgr until the driver calls
drm_dp_mst_atomic_check().
> > + *
> > + * Additionally, it is OK to call this function multiple times on the
> > same
> > + * @port as needed. It is not OK however, to call this function and
> > + * drm_dp_atomic_release_vcpi_slots() in the same atomic check phase.
> > + *
> > + * See also:
> > + * drm_dp_atomic_release_vcpi_slots()
> > + * drm_dp_mst_atomic_check()
> > + *
> > + * Returns:
> > + * Total slots in the atomic state assigned for this port, or a
negative
> > error
> > + * code if the port no longer exists
> >   */
> >  int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
> >  				  struct drm_dp_mst_topology_mgr *mgr,
> >  				  struct drm_dp_mst_port *port, int pbn)
> >  {
> >  	struct drm_dp_mst_topology_state *topology_state;
> > -	int req_slots;
> > +	struct drm_dp_vcpi_allocation *pos, *vcpi = NULL;
> > +	int prev_slots, req_slots, ret;
> >  
> >  	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
> >  	if (IS_ERR(topology_state))
> > @@ -2640,20 +2661,60 @@ int drm_dp_atomic_find_vcpi_slots(struct
> > drm_atomic_state *state,
> >  	port = drm_dp_get_validated_port_ref(mgr, port);
> >  	if (port == NULL)
> >  		return -EINVAL;
> > -	req_slots = DIV_ROUND_UP(pbn, mgr->pbn_div);
> > -	DRM_DEBUG_KMS("vcpi slots req=%d, avail=%d\n",
> > -			req_slots, topology_state->avail_slots);
> >  
> > -	if (req_slots > topology_state->avail_slots) {
> > -		drm_dp_put_port(port);
> > -		return -ENOSPC;
> > +	topology_state->vcpi_allocated = true;
> > +
> > +	/* Find the current allocation for this port, if any */
> > +	list_for_each_entry(pos, &topology_state->vcpis, next) {
> > +		if (pos->port == port) {
> > +			vcpi = pos;
> > +			prev_slots = vcpi->vcpi;
> > +
> > +			/*
> > +			 * This should never happen, unless the driver tries
> > +			 * releasing and allocating the same VCPI allocation,
> > +			 * which is an error
> > +			 */
> > +			if (WARN_ON(!prev_slots)) {
> > +				DRM_ERROR("cannot allocate and release VCPI on
> > [MST PORT:%p] in the same state\n",
> > +					  port);
> > +				return -EINVAL;
> > +			}
> 
> Allowing atomic changes to output state (like disabling it on one crtc,
> and then enabling on the other, thus both releasing and re-allocating the
> vcpi in one go) is kinda the point of atomic. So definitely can't
EINVAL
> here.
Actually-so long the original assertion we had of not allowing "stale"
(e.g.
VCPIs from now-destroyed topologies) VCPI allocations and real VCPI
allocations in the same atomic state is held true, this should be fine.

So: (and this answers the question below as well), the way this is supposed to
work is that we just skip the whole VCPI validation sequence unless part of
the state actually tried to add VCPI. This means userland can still disable
displays one-by-one or set them as inactive, so long as it doesn't try to
enable new MST displays until the point at which it's disabled everything
leftover from the previous topology - since disabling displays won't
allocate
VCPI and thus skips the port validation checks here.
> 
> Sounds like we need more testcases in igt?
> 
> > +
> > +			break;
> > +		}
> >  	}
> > +	if (!vcpi)
> > +		prev_slots = 0;
> >  
> > -	topology_state->avail_slots -= req_slots;
> > -	DRM_DEBUG_KMS("vcpi slots avail=%d",
topology_state->avail_slots);
> > +	req_slots = DIV_ROUND_UP(pbn, mgr->pbn_div);
> > +
> > +	DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] [MST PORT:%p] VCPI %d ->
%d\n",
> > +			 port->connector->base.id, port->connector->name,
> > +			 port, prev_slots, req_slots);
> > +
> > +	/* Add the new allocation to the state */
> > +	if (!vcpi) {
> > +		vcpi = kzalloc(sizeof(*vcpi), GFP_KERNEL);
> > +		if (!vcpi) {
> > +			ret = -ENOMEM;
> > +			goto out;
> > +		}
> >  
> > +		vcpi->port = drm_dp_get_validated_port_ref(mgr, port);
> > +		if (!vcpi->port) {
> > +			ret = -EINVAL;
> > +			kfree(vcpi);
> > +			goto out;
> > +		}
> > +		list_add(&vcpi->next, &topology_state->vcpis);
> > +	}
> > +	vcpi->vcpi = req_slots;
> > +
> > +	ret = req_slots;
> > +out:
> >  	drm_dp_put_port(port);
> > -	return req_slots;
> > +	return ret;
> >  }
> >  EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
> >  
> > @@ -2661,31 +2722,66 @@ EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
> >   * drm_dp_atomic_release_vcpi_slots() - Release allocated vcpi slots
> >   * @state: global atomic state
> >   * @mgr: MST topology manager for the port
> > - * @slots: number of vcpi slots to release
> > + * @port: The port to release the VCPI slots from
> >   *
> > - * RETURNS:
> > - * 0 if @slots were added back to
&drm_dp_mst_topology_state->avail_slots
> > or
> > - * negative error code
> > + * Releases any VCPI slots that have been allocated to a port in the
> > atomic
> > + * state. Any atomic drivers which support MST must call this
function in
> > + * their &drm_connector_helper_funcs.atomic_check() callback when
the
> > + * connector will no longer have VCPI allocated (e.g. because
it's CRTC
> > was
> > + * removed) when it had VCPI allocated in the previous atomic state.
> > + *
> > + * It is OK to call this even if @port has been removed from the
system.
> > + * Additionally, it is OK to call this function multiple times on the
> > same
> > + * @port as needed. It is not OK however, to call this function and
> > + * drm_dp_atomic_find_vcpi_slots() on the same @port in a single
atomic
> > check
> > + * phase.
> > + *
> > + * See also:
> > + * drm_dp_atomic_find_vcpi_slots()
> > + * drm_dp_mst_atomic_check()
> > + *
> > + * Returns:
> > + * 0 if all slots for this port were added back to
> > + * &drm_dp_mst_topology_state.avail_slots or negative error code
> >   */
> >  int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
> >  				     struct drm_dp_mst_topology_mgr *mgr,
> > -				     int slots)
> > +				     struct drm_dp_mst_port *port)
> >  {
> >  	struct drm_dp_mst_topology_state *topology_state;
> > +	struct drm_dp_vcpi_allocation *pos;
> > +	bool found = false;
> >  
> >  	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
> >  	if (IS_ERR(topology_state))
> >  		return PTR_ERR(topology_state);
> >  
> > -	/* We cannot rely on port->vcpi.num_slots to update
> > -	 * topology_state->avail_slots as the port may not exist if the
parent
> > -	 * branch device was unplugged. This should be fixed by tracking
> > -	 * per-port slot allocation in drm_dp_mst_topology_state instead of
> > -	 * depending on the caller to tell us how many slots to release.
> > -	 */
> > -	topology_state->avail_slots += slots;
> > -	DRM_DEBUG_KMS("vcpi slots released=%d, avail=%d\n",
> > -			slots, topology_state->avail_slots);
> > +	list_for_each_entry(pos, &topology_state->vcpis, next) {
> > +		if (pos->port == port) {
> > +			found = true;
> > +			break;
> > +		}
> > +	}
> > +	if (WARN_ON(!found)) {
> > +		DRM_ERROR("no VCPI for [MST PORT:%p] found in mst state
%p\n",
> > +			  port, &topology_state->base);
> > +		return -EINVAL;
> > +	}
> > +
> > +	DRM_DEBUG_ATOMIC("[MST PORT:%p] VCPI %d -> 0\n", port,
pos->vcpi);
> > +	if (pos->vcpi) {
> > +		/*
> > +		 * This should be safe-as the previous state will still hold a
> > +		 * reference to this port until we swap states. If it's not,
> > +		 * the driver has a bookkeeping error
> > +		 */
> > +		if (WARN_ON(drm_dp_put_port(port))) {
> 
> The usual way to do this (instead of adding a return value to the normal
> _put function) is something like
> 
> 	kref_put(port, function_that_with_just_a_WARNING_DRM_ERROR)
> 
> avoids the need for patch 2.
> 
> > +			DRM_ERROR("cannot allocate and release VCPI on [MST
> > PORT:%p] in the same state\n",
> > +				  port);
> > +			return -EINVAL;
> > +		}
> > +		pos->vcpi = 0;
> > +	}
> >  
> >  	return 0;
> >  }
> > @@ -3115,15 +3211,45 @@ static void
drm_dp_destroy_connector_work(struct
> > work_struct *work)
> >  static struct drm_private_state *
> >  drm_dp_mst_duplicate_state(struct drm_private_obj *obj)
> >  {
> > -	struct drm_dp_mst_topology_state *state;
> > +	struct drm_dp_mst_topology_state *state, *old_state > > +	
to_dp_mst_topology_state(obj->state);
> > +	struct drm_dp_vcpi_allocation *pos, *vcpi;
> >  
> > -	state = kmemdup(obj->state, sizeof(*state), GFP_KERNEL);
> > +	state = kmemdup(old_state, sizeof(*state), GFP_KERNEL);
> >  	if (!state)
> >  		return NULL;
> >  
> >  	__drm_atomic_helper_private_obj_duplicate_state(obj,
&state->base);
> >  
> > +	INIT_LIST_HEAD(&state->vcpis);
> > +	state->vcpi_allocated = false;
> > +
> > +	list_for_each_entry(pos, &old_state->vcpis, next) {
> > +		/* Prune leftover freed VCPI allocations */
> > +		if (!pos->vcpi)
> > +			continue;
> > +
> > +		vcpi = kmemdup(pos, sizeof(*vcpi), GFP_KERNEL);
> > +		if (!vcpi)
> > +			goto fail;
> > +
> > +		/* Take a plain old kref here and don't validate the port, as
> > +		 * it's topology might not even exist anymore
> > +		 */
> > +		kref_get(&vcpi->port->kref);
> > +		list_add(&vcpi->next, &state->vcpis);
> > +	}
> > +
> >  	return &state->base;
> > +
> > +fail:
> > +	list_for_each_entry_safe(pos, vcpi, &state->vcpis, next) {
> > +		drm_dp_put_port(pos->port);
> > +		kfree(pos);
> > +	}
> > +	kfree(state);
> > +
> > +	return NULL;
> >  }
> >  
> >  static void drm_dp_mst_destroy_state(struct drm_private_obj *obj,
> > @@ -3131,14 +3257,116 @@ static void drm_dp_mst_destroy_state(struct
> > drm_private_obj *obj,
> >  {
> >  	struct drm_dp_mst_topology_state *mst_state > >  	
to_dp_mst_topology_state(state);
> > +	struct drm_dp_vcpi_allocation *pos, *tmp;
> > +
> > +	list_for_each_entry_safe(pos, tmp, &mst_state->vcpis, next) {
> > +		/* We only keep references to ports with non-zero VCPIs */
> > +		if (pos->vcpi)
> > +			drm_dp_put_port(pos->port);
> > +		kfree(pos);
> > +	}
> >  
> >  	kfree(mst_state);
> >  }
> >  
> > -static const struct drm_private_state_funcs mst_state_funcs = {
> > +static inline int
> > +drm_dp_mst_atomic_check_topology_state(struct drm_dp_mst_topology_mgr
> > *mgr,
> > +				       struct drm_dp_mst_topology_state
> > *mst_state)
> > +{
> > +	struct drm_dp_vcpi_allocation *vcpi;
> > +	struct drm_dp_mst_port *port;
> > +	int avail_slots = 63, ret;
> > +
> > +	/* There's no possible scenario where releasing VCPI or keeping
it the
> > +	 * same would make the state invalid
> > +	 */
> > +	if (!mst_state->vcpi_allocated) {
> 
> Do we need this? If not I'm always in favour of removing code if
it's not
> needed :-)
answered by my explanation above ^
> 
> > +		DRM_DEBUG_ATOMIC("[MST MGR:%p] state %p allocates no VCPI,
> > check skipped\n",
> > +				 mgr, &mst_state->base);
> > +		return 0;
> > +	}
> > +
> > +	list_for_each_entry(vcpi, &mst_state->vcpis, next) {
> > +		/* Releasing VCPI is always OK-even if the port is gone */
> > +		if (!vcpi->vcpi) {
> > +			DRM_DEBUG_ATOMIC("[MST PORT:%p] releases all VCPI
> > slots\n",
> > +					 vcpi->port);
> > +			continue;
> > +		}
> > +
> > +		port = drm_dp_get_validated_port_ref(mgr, vcpi->port);
> > +		if (!port) {
> > +			DRM_DEBUG_ATOMIC("[MST PORT:%p] is gone but still has
> > VCPI, cannot add new VCPI\n",
> > +					 vcpi->port);
> > +			return -EINVAL;
> > +		}
> 
> For find_vcpi we've already done this. For unplugged vcpi, we can't
do
> this here I think - otherwise if you enable 2 mst outputs on the same
> branch, and unplug both, old non-atomic userspace can never ever again
> disable them because it can't disable both. The check (to skip already
> deallocated ports) above isn't enough to handle that.
answered by my explanation above ^
> 
> > +
> > +		DRM_DEBUG_ATOMIC("[MST PORT:%p] requires %d vcpi
slots\n",
> > +				 vcpi->port, vcpi->vcpi);
> > +
> > +		avail_slots -= vcpi->vcpi;
> > +		if (avail_slots < 0) {
> > +			DRM_DEBUG_ATOMIC("[MST PORT:%p] not enough VCPI slots
> > in mst state %p (avail=%d)\n",
> > +					 vcpi->port, mst_state,
> > +					 avail_slots + vcpi->vcpi);
> > +			ret = -ENOSPC;
> > +			goto port_fail;
> > +		}
> > +
> > +		drm_dp_put_port(port);
> > +	}
> > +	DRM_DEBUG_ATOMIC("[MST MGR:%p] mst state %p VCPI avail=%d
used=%d\n",
> > +			 mgr, mst_state, avail_slots,
> > +			 63 - avail_slots);
> > +
> > +	return 0;
> > +port_fail:
> > +	drm_dp_put_port(port);
> > +	return ret;
> > +}
> > +
> > +/**
> > + * drm_dp_mst_atomic_check - Check that the new state of an MST
topology
> > in an
> > + * atomic update is valid
> > + * @state: Pointer to the new &struct drm_dp_mst_topology_state
> > + *
> > + * Checks the given topology state for an atomic update to ensure
that
> > it's
> > + * valid. This includes checking whether there's enough bandwidth
to
> > support
> > + * the new VCPI allocations in the atomic update.
> > + *
> > + * Any atomic drivers supporting DP MST must make sure to call this
after
> > + * checking the rest of their state in their
> > + * &drm_mode_config_funcs.atomic_check() callback.
> > + *
> > + * See also:
> > + * drm_dp_atomic_find_vcpi_slots()
> > + * drm_dp_atomic_release_vcpi_slots()
> > + *
> > + * Returns:
> > + *
> > + * 0 if the new state is valid, negative error code otherwise.
> > + */
> > +int drm_dp_mst_atomic_check(struct drm_atomic_state *state)
> > +{
> > +	struct drm_dp_mst_topology_mgr *mgr;
> > +	struct drm_dp_mst_topology_state *mst_state;
> > +	int i, ret = 0;
> > +
> > +	for_each_new_mst_mgr_in_state(state, mgr, mst_state, i) {
> > +		ret = drm_dp_mst_atomic_check_topology_state(mgr, mst_state);
> > +		if (ret)
> > +			break;
> > +	}
> > +
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL(drm_dp_mst_atomic_check);
> > +
> > +const struct drm_private_state_funcs drm_dp_mst_topology_state_funcs
= {
> >  	.atomic_duplicate_state = drm_dp_mst_duplicate_state,
> >  	.atomic_destroy_state = drm_dp_mst_destroy_state,
> >  };
> > +EXPORT_SYMBOL(drm_dp_mst_topology_state_funcs);
> >  
> >  /**
> >   * drm_atomic_get_mst_topology_state: get MST topology state
> > @@ -3216,13 +3444,11 @@ int drm_dp_mst_topology_mgr_init(struct
> > drm_dp_mst_topology_mgr *mgr,
> >  		return -ENOMEM;
> >  
> >  	mst_state->mgr = mgr;
> > -
> > -	/* max. time slots - one slot for MTP header */
> > -	mst_state->avail_slots = 63;
> > +	INIT_LIST_HEAD(&mst_state->vcpis);
> >  
> >  	drm_atomic_private_obj_init(&mgr->base,
> >  				    &mst_state->base,
> > -				    &mst_state_funcs);
> > +				    &drm_dp_mst_topology_state_funcs);
> >  
> >  	return 0;
> >  }
> > diff --git a/drivers/gpu/drm/i915/intel_display.c
> > b/drivers/gpu/drm/i915/intel_display.c
> > index 132e978227fb..8e85e04a5948 100644
> > --- a/drivers/gpu/drm/i915/intel_display.c
> > +++ b/drivers/gpu/drm/i915/intel_display.c
> > @@ -12592,6 +12592,10 @@ static int intel_atomic_check(struct
drm_device
> > *dev,
> >  				       "[modeset]" : "[fastset]");
> >  	}
> >  
> > +	ret = drm_dp_mst_atomic_check(state);
> > +	if (ret)
> > +		return ret;
> > +
> >  	if (any_ms) {
> >  		ret = intel_modeset_checks(state);
> >  
> > diff --git a/drivers/gpu/drm/i915/intel_dp_mst.c
> > b/drivers/gpu/drm/i915/intel_dp_mst.c
> > index 4de247ddf05f..be2d1ac7d87d 100644
> > --- a/drivers/gpu/drm/i915/intel_dp_mst.c
> > +++ b/drivers/gpu/drm/i915/intel_dp_mst.c
> > @@ -41,8 +41,13 @@ static bool intel_dp_mst_compute_config(struct
> > intel_encoder *encoder,
> >  	struct drm_connector *connector = conn_state->connector;
> >  	void *port = to_intel_connector(connector)->port;
> >  	struct drm_atomic_state *state = pipe_config->base.state;
> > +	struct drm_crtc *crtc = pipe_config->base.crtc;
> > +	struct drm_crtc_state *old_crtc_state > > +	
drm_atomic_get_old_crtc_state(state, crtc);
> > +	struct drm_crtc_state *new_crtc_state = &pipe_config->base;
> >  	int bpp;
> > -	int lane_count, slots = 0;
> > +	int lane_count, slots > > +	
to_intel_crtc_state(old_crtc_state)->dp_m_n.tu;
> >  	const struct drm_display_mode *adjusted_mode = &pipe_config-
> > >base.adjusted_mode;
> >  	int mst_pbn;
> >  	bool constant_n = drm_dp_has_quirk(&intel_dp->desc,
> > @@ -77,8 +82,12 @@ static bool intel_dp_mst_compute_config(struct
> > intel_encoder *encoder,
> >  	mst_pbn = drm_dp_calc_pbn_mode(adjusted_mode->crtc_clock, bpp);
> >  	pipe_config->pbn = mst_pbn;
> >  
> > -	/* Zombie connectors can't have VCPI slots */
> > -	if (!drm_connector_is_unregistered(connector)) {
> > +	/* Only change VCPI allocation on actual mode changes, to prevent us
> > +	 * from trying to allocate VCPI to ports that no longer exist when
we
> > +	 * may just be trying to disable DPMS on them
> > +	 */
> > +	if (new_crtc_state->mode_changed ||
> > +	    new_crtc_state->connectors_changed) {
> 
> Why exactly is this needed? We only call compute_config on the enocoder
> when enabling an output. Not when disabling one. Similar for the atomic
> helpers and the various encoder callbacks (you have the same change in the
> nouveau code). Did this actually blow up while testing?
yes-it looked like we were calling compute_config on simple DPMS on->off
chnges as well, this ended up making it so we wouldn't try changing the VCPI
allocation on such commits.

ACK on all of the other comments here.
> 
> >  		slots = drm_dp_atomic_find_vcpi_slots(state,
> >  						      &intel_dp->mst_mgr,
> >  						      port,
> > @@ -107,36 +116,39 @@ static bool intel_dp_mst_compute_config(struct
> > intel_encoder *encoder,
> >  	return true;
> >  }
> >  
> > -static int intel_dp_mst_atomic_check(struct drm_connector *connector,
> > -		struct drm_connector_state *new_conn_state)
> > +static int
> > +intel_dp_mst_atomic_check(struct drm_connector *connector,
> > +			  struct drm_connector_state *new_conn_state)
> >  {
> > +	struct intel_connector *intel_connector > > +	
to_intel_connector(connector);
> > +	struct drm_dp_mst_topology_mgr *mgr > > +	
&intel_connector->mst_port->mst_mgr;
> > +	struct drm_dp_mst_port *port = intel_connector->port;
> >  	struct drm_atomic_state *state = new_conn_state->state;
> > -	struct drm_connector_state *old_conn_state;
> > -	struct drm_crtc *old_crtc;
> > -	struct drm_crtc_state *crtc_state;
> > -	int slots, ret = 0;
> > +	struct drm_connector_state *old_conn_state > > +	
drm_atomic_get_old_connector_state(state, connector);
> > +	struct drm_crtc *old_crtc = old_conn_state->crtc,
> > +			*new_crtc = new_conn_state->crtc;
> > +	struct drm_crtc_state *old_crtc_state, *new_crtc_state;
> >  
> > -	old_conn_state = drm_atomic_get_old_connector_state(state,
connector);
> > -	old_crtc = old_conn_state->crtc;
> 
> Bikeshed: Reworking all the above hides the real changes quite a bit.
> 
> >  	if (!old_crtc)
> > -		return ret;
> > +		return 0;
> >  
> > -	crtc_state = drm_atomic_get_new_crtc_state(state, old_crtc);
> > -	slots = to_intel_crtc_state(crtc_state)->dp_m_n.tu;
> > -	if (drm_atomic_crtc_needs_modeset(crtc_state) && slots >
0) {
> > -		struct drm_dp_mst_topology_mgr *mgr;
> > -		struct drm_encoder *old_encoder;
> > +	old_crtc_state = drm_atomic_get_old_crtc_state(state, old_crtc);
> > +	if (!old_crtc_state ||
> 
> !old_crtc_state would be a framework bug. Just remove that check.
> 
> > +	    !to_intel_crtc_state(old_crtc_state)->dp_m_n.tu)
> > +		return 0;
> 
> Hm, why do we need this? Having a DP port previously enabled with a 0 tu
> sounds like a pretty big driver bug.
> >  
> > -		old_encoder = old_conn_state->best_encoder;
> > -		mgr = &enc_to_mst(old_encoder)->primary->dp.mst_mgr;
> > +	/* If we switched CRTCs, clear the previous one's allocation */
> > +	new_crtc_state = drm_atomic_get_new_crtc_state(state, old_crtc);
> > +	if (new_crtc_state->connectors_changed &&
!new_crtc_state->enable)
> > +		to_intel_crtc_state(new_crtc_state)->dp_m_n.tu = 0;
> 
> No idea why we need to clear this here? Shouldn't we just clear this
> unconditionally if we release the allocation.
> 
> Also naming convention, I'd call this old_crtc_new_state, and maybe
> old_crtc_old_state. To make it clear these are the old/new states for
> old_crtc. Instead of old/new state for the new crtc.
> 
> >  
> > -		ret = drm_dp_atomic_release_vcpi_slots(state, mgr, slots);
> > -		if (ret)
> > -			DRM_DEBUG_KMS("failed releasing %d vcpi slots:%d\n",
> > slots, ret);
> > -		else
> > -			to_intel_crtc_state(crtc_state)->dp_m_n.tu = 0;
> > -	}
> > -	return ret;
> > +	if (new_crtc)
> > +		return 0;
> 
> Not sure why we shouldn't release this? If we handle this in helpers,
it
> would remove piles of special casing code here I think. Plus remove
> special code in the hlpers.
> 
> 
> > +
> > +	return drm_dp_atomic_release_vcpi_slots(state, mgr, port);
> >  }
> >  
> >  static void intel_mst_disable_dp(struct intel_encoder *encoder,
> > diff --git a/include/drm/drm_dp_mst_helper.h
> > b/include/drm/drm_dp_mst_helper.h
> > index 3faceb66f5cb..2392e12b6bce 100644
> > --- a/include/drm/drm_dp_mst_helper.h
> > +++ b/include/drm/drm_dp_mst_helper.h
> > @@ -406,9 +406,16 @@ struct drm_dp_payload {
> >  
> >  #define to_dp_mst_topology_state(x) container_of(x, struct
> > drm_dp_mst_topology_state, base)
> >  
> > +struct drm_dp_vcpi_allocation {
> 
> I think a comment here that this pointer holds a full reference (even past
> when the connector is unplugged) would be good. Otherwise the loop in
> release_vcpi would be buggy.
> 
> > +	struct drm_dp_mst_port *port;
> > +	int vcpi;
> > +	struct list_head next;
> > +};
> > +
> >  struct drm_dp_mst_topology_state {
> >  	struct drm_private_state base;
> > -	int avail_slots;
> > +	struct list_head vcpis;
> > +	u8 vcpi_allocated;
> >  	struct drm_dp_mst_topology_mgr *mgr;
> >  };
> >  
> > @@ -619,14 +626,17 @@ void drm_dp_mst_topology_mgr_suspend(struct
> > drm_dp_mst_topology_mgr *mgr);
> >  int drm_dp_mst_topology_mgr_resume(struct drm_dp_mst_topology_mgr
*mgr);
> >  struct drm_dp_mst_topology_state
> > *drm_atomic_get_mst_topology_state(struct drm_atomic_state *state,
> >  								    struct
> > drm_dp_mst_topology_mgr *mgr);
> > -int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
> > -				  struct drm_dp_mst_topology_mgr *mgr,
> > -				  struct drm_dp_mst_port *port, int pbn);
> > -int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
> > -				     struct drm_dp_mst_topology_mgr *mgr,
> > -				     int slots);
> > +int __must_check
> > +drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
> > +			      struct drm_dp_mst_topology_mgr *mgr,
> > +			      struct drm_dp_mst_port *port, int pbn);
> > +int __must_check
> > +drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
> > +				 struct drm_dp_mst_topology_mgr *mgr,
> > +				 struct drm_dp_mst_port *port);
> >  int drm_dp_send_power_updown_phy(struct drm_dp_mst_topology_mgr *mgr,
> >  				 struct drm_dp_mst_port *port, bool power_up);
> > +int __must_check drm_dp_mst_atomic_check(struct drm_atomic_state
*state);
> >  
> >  extern const struct drm_private_state_funcs
> > drm_dp_mst_topology_state_funcs;
> >  
> > -- 
> > 2.19.1
> > 
-- 
Cheers,
	Lyude Paul

On Mon, Nov 26, 2018 at 04:36:46PM -0500, Lyude Paul
wrote:
> On Mon, 2018-11-26 at 22:04 +0100, Daniel Vetter wrote:
> > On Thu, Nov 15, 2018 at 07:50:05PM -0500, Lyude Paul wrote:
> > > There has been a TODO waiting for quite a long time in
> > > drm_dp_mst_topology.c:
> > > 
> > > 	/* We cannot rely on port->vcpi.num_slots to update
> > > 	 * topology_state->avail_slots as the port may not exist if
the parent
> > > 	 * branch device was unplugged. This should be fixed by tracking
> > > 	 * per-port slot allocation in drm_dp_mst_topology_state instead
of
> > > 	 * depending on the caller to tell us how many slots to release.
> > > 	 */
> > > 
> > > That's not the only reason we should fix this: forcing the
driver to
> > > track the VCPI allocations throughout a state's atomic check
is
> > > error prone, because it means that extra care has to be taken
with the
> > > order that drm_dp_atomic_find_vcpi_slots() and
> > > drm_dp_atomic_release_vcpi_slots() are called in in order to
ensure
> > > idempotency. Currently the only driver actually using these
helpers,
> > > i915, doesn't even do this correctly: multiple
->best_encoder() checks
> > > with i915's current implementation would not be idempotent
and would
> > > over-allocate VCPI slots, something I learned trying to implement
> > > fallback retraining in MST.
> > > 
> > > So: simplify this whole mess, and teach
drm_dp_atomic_find_vcpi_slots()
> > > and drm_dp_atomic_release_vcpi_slots() to track the VCPI
allocations for
> > > each port. This allows us to ensure idempotency without having to
rely
> > > on the driver as much. Additionally: the driver doesn't need
to do any
> > > kind of VCPI slot tracking anymore if it doesn't need it for
it's own
> > > internal state.
> > > 
> > > Additionally; this adds a new drm_dp_mst_atomic_check() helper
which
> > > must be used by atomic drivers to perform validity checks for the
new
> > > VCPI allocations incurred by a state.
> > > 
> > > Also: update the documentation and make it more obvious that
these
> > > /must/ be called by /all/ atomic drivers supporting MST.
> > > 
> > > Changes since v6:
> > >  - Keep a kref to all of the ports we have allocations on. This
required
> > >    a good bit of changing to when we call
drm_dp_find_vcpi_slots(),
> > >    mainly that we need to ensure that we only redo VCPI
allocations on
> > >    actual mode or CRTC changes, not crtc_state->active
changes.
> > >    Additionally, we no longer take the registration of the DRM
connector
> > >    for each port into account because so long as we have a kref
to the
> > >    port in the new or previous atomic state, the connector will
stay
> > >    registered.
> > 
> > I write an entire pile of small nitpits (still included most of them
> > below), until I realized this here wont work. Delaying the call to
destroy
> > the connector (well, unregister it really) wreaks the design we've
come up
> > with thus far, resulting in most of my comments here.
> > 
> > Instead, all we need to do is delay the kfree(port) at the bottom of
> > drm_dp_destroy_port(). The vcpi allocation structure _only_ needs the
> > pointer value to stay valid, as a lookup key. It doesn't care at
all about
> > anything actually stored in there. So the only thing we need to delay
is
> > the kfree. I think the simplest way to achieve this is to add a 2nd
kref
> > (port->kfree_ref or something like that), with on reference held by
the
> > port itself (released in drm_dp_destroy_port), and the other one held
> > as-needed by the vcpi allocation lists.
> > 
> > I think if we go with this design instead of retrofitting a semantic
> > change of the port lifetime itself, all the complications I complain
about
> > below should disappear.
> > 
> > Piles of comments below.
> > 
> > Cheers, Daniel
> > 
> > >  - Use the small changes to drm_dp_put_port() to add even more
error
> > >    checking to make misusage of the helpers more obvious. I added
this
> > >    after having to chase down various use-after-free conditions
that
> > >    started popping up from the new helpers so no one else has to
> > >    troubleshoot that.
> > >  - Move some accidental DRM_DEBUG_KMS() calls to
DRM_DEBUG_ATOMIC()
> > >  - Update documentation again, note that find/release() should
both not be
> > >    called on the same port in a single atomic check phase (but
multiple
> > >    calls to one or the other is OK)
> > > 
> > > Changes since v4:
> > >  - Don't skip the atomic checks for VCPI allocations if no
new VCPI
> > >    allocations happen in a state. This makes the next change
I'm about
> > >    to list here a lot easier to implement.
> > >  - Don't ignore VCPI allocations on destroyed ports, instead
ensure that
> > >    when ports are destroyed and still have VCPI allocations in
the
> > >    topology state, the only state changes allowed are releasing
said
> > >    ports' VCPI. This prevents a state with a mix of VCPI
allocations
> > >    from destroyed ports, and allocations from valid ports.
> > > 
> > > Changes since v3:
> > >  - Don't release VCPI allocations in the topology state
immediately in
> > >    drm_dp_atomic_release_vcpi_slots(), instead mark them as 0 and
skip
> > >    over them in drm_dp_mst_duplicate_state(). This makes it so
> > >    drm_dp_atomic_release_vcpi_slots() is still idempotent while
also
> > >    throwing warnings if the driver messes up it's book
keeping and tries
> > >    to release VCPI slots on a port that doesn't have any
pre-existing
> > >    VCPI allocation - danvet
> > >  - Change mst_state/state in some debugging messages to "mst
state"
> > > 
> > > Changes since v2:
> > >  - Use kmemdup() for duplicating MST state - danvet
> > >  - Move port validation out of duplicate state callback - danvet
> > >  - Handle looping through MST topology states in
> > >    drm_dp_mst_atomic_check() so the driver doesn't have to do
it
> > >  - Fix documentation in drm_dp_atomic_find_vcpi_slots()
> > >  - Move the atomic check for each individual topology state into
it's
> > >    own function, reduces indenting
> > >  - Don't consider "stale" MST ports when
calculating the bandwidth
> > >    requirements. This is needed because originally we relied on
the
> > >    state duplication functions to prune any stale ports from the
new
> > >    state, which would prevent us from incorrectly considering
their
> > >    bandwidth requirements alongside legitimate new payloads.
> > >  - Add function references in drm_dp_atomic_release_vcpi_slots()
- danvet
> > >  - Annotate atomic VCPI and atomic check functions with
__must_check
> > >    - danvet
> > > 
> > > Changes since v1:
> > >  - Don't use the now-removed ->atomic_check() for private
objects hook,
> > >    just give drivers a function to call themselves
> > > 
> > > Signed-off-by: Lyude Paul <lyude at redhat.com>
> > > Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
> > > ---
> > >  drivers/gpu/drm/drm_dp_mst_topology.c | 294
+++++++++++++++++++++++---
> > >  drivers/gpu/drm/i915/intel_display.c  |   4 +
> > >  drivers/gpu/drm/i915/intel_dp_mst.c   |  64 +++---
> > >  include/drm/drm_dp_mst_helper.h       |  24 ++-
> > >  4 files changed, 319 insertions(+), 67 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/drm_dp_mst_topology.c
> > > b/drivers/gpu/drm/drm_dp_mst_topology.c
> > > index 00fbe7a2699d..183d0e832ced 100644
> > > --- a/drivers/gpu/drm/drm_dp_mst_topology.c
> > > +++ b/drivers/gpu/drm/drm_dp_mst_topology.c
> > > @@ -2617,21 +2617,42 @@ static int drm_dp_init_vcpi(struct
> > > drm_dp_mst_topology_mgr *mgr,
> > >  }
> > >  
> > >  /**
> > > - * drm_dp_atomic_find_vcpi_slots() - Find and add vcpi slots to
the state
> > > + * drm_dp_atomic_find_vcpi_slots() - Find and add VCPI slots to
the state
> > >   * @state: global atomic state
> > >   * @mgr: MST topology manager for the port
> > >   * @port: port to find vcpi slots for
> > >   * @pbn: bandwidth required for the mode in PBN
> > >   *
> > > - * RETURNS:
> > > - * Total slots in the atomic state assigned for this port or
error
> > > + * Allocates VCPI slots to @port, replacing any previous VCPI
allocations
> > > it
> > > + * may have had. Any atomic drivers which support MST must call
this
> > > function
> > > + * in their &drm_encoder_helper_funcs.atomic_check()
callback to change
> > > the
> > > + * current VCPI allocation for the new state, but only when
> > > + * &drm_crtc_state.mode_changed or
&drm_crtc_state.connectors_changed is
> > > set
> > > + * to ensure compatibility with userspace applications that
still use the
> > > + * legacy modesetting UAPI.
> > > + *
> > > + * Allocations set by this function are not checked against the
bandwidth
> > > + * restraints of @mgr until the driver calls
drm_dp_mst_atomic_check().
> > > + *
> > > + * Additionally, it is OK to call this function multiple times
on the
> > > same
> > > + * @port as needed. It is not OK however, to call this function
and
> > > + * drm_dp_atomic_release_vcpi_slots() in the same atomic check
phase.
> > > + *
> > > + * See also:
> > > + * drm_dp_atomic_release_vcpi_slots()
> > > + * drm_dp_mst_atomic_check()
> > > + *
> > > + * Returns:
> > > + * Total slots in the atomic state assigned for this port, or a
negative
> > > error
> > > + * code if the port no longer exists
> > >   */
> > >  int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state
*state,
> > >  				  struct drm_dp_mst_topology_mgr *mgr,
> > >  				  struct drm_dp_mst_port *port, int pbn)
> > >  {
> > >  	struct drm_dp_mst_topology_state *topology_state;
> > > -	int req_slots;
> > > +	struct drm_dp_vcpi_allocation *pos, *vcpi = NULL;
> > > +	int prev_slots, req_slots, ret;
> > >  
> > >  	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
> > >  	if (IS_ERR(topology_state))
> > > @@ -2640,20 +2661,60 @@ int drm_dp_atomic_find_vcpi_slots(struct
> > > drm_atomic_state *state,
> > >  	port = drm_dp_get_validated_port_ref(mgr, port);
> > >  	if (port == NULL)
> > >  		return -EINVAL;
> > > -	req_slots = DIV_ROUND_UP(pbn, mgr->pbn_div);
> > > -	DRM_DEBUG_KMS("vcpi slots req=%d, avail=%d\n",
> > > -			req_slots, topology_state->avail_slots);
> > >  
> > > -	if (req_slots > topology_state->avail_slots) {
> > > -		drm_dp_put_port(port);
> > > -		return -ENOSPC;
> > > +	topology_state->vcpi_allocated = true;
> > > +
> > > +	/* Find the current allocation for this port, if any */
> > > +	list_for_each_entry(pos, &topology_state->vcpis, next) {
> > > +		if (pos->port == port) {
> > > +			vcpi = pos;
> > > +			prev_slots = vcpi->vcpi;
> > > +
> > > +			/*
> > > +			 * This should never happen, unless the driver tries
> > > +			 * releasing and allocating the same VCPI allocation,
> > > +			 * which is an error
> > > +			 */
> > > +			if (WARN_ON(!prev_slots)) {
> > > +				DRM_ERROR("cannot allocate and release VCPI on
> > > [MST PORT:%p] in the same state\n",
> > > +					  port);
> > > +				return -EINVAL;
> > > +			}
> > 
> > Allowing atomic changes to output state (like disabling it on one
crtc,
> > and then enabling on the other, thus both releasing and re-allocating
the
> > vcpi in one go) is kinda the point of atomic. So definitely can't
EINVAL
> > here.
> Actually-so long the original assertion we had of not allowing
"stale" (e.g.
> VCPIs from now-destroyed topologies) VCPI allocations and real VCPI
> allocations in the same atomic state is held true, this should be fine.
> 
> So: (and this answers the question below as well), the way this is supposed
to
> work is that we just skip the whole VCPI validation sequence unless part of
> the state actually tried to add VCPI. This means userland can still disable
> displays one-by-one or set them as inactive, so long as it doesn't try
to
> enable new MST displays until the point at which it's disabled
everything
> leftover from the previous topology - since disabling displays won't
allocate
> VCPI and thus skips the port validation checks here.
This doesn't sound robust if there's a race, e.g. userspace tries to
light
up a display right at the time when we unplug a different (already enabled
output). I think to fix this properly we need to replace all the
dp_get_validated_port calls in all the vcpi/payload places with the
lower-level kref that only prevents the kfree.

Fundamentally hotunplug is still racy, but this way I think there's at
least no surprises. And we could probably remove a bunch of the hacks
around cleanup paths.
> > Sounds like we need more testcases in igt?
> > 
> > > +
> > > +			break;
> > > +		}
> > >  	}
> > > +	if (!vcpi)
> > > +		prev_slots = 0;
> > >  
> > > -	topology_state->avail_slots -= req_slots;
> > > -	DRM_DEBUG_KMS("vcpi slots avail=%d",
topology_state->avail_slots);
> > > +	req_slots = DIV_ROUND_UP(pbn, mgr->pbn_div);
> > > +
> > > +	DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] [MST PORT:%p] VCPI %d
-> %d\n",
> > > +			 port->connector->base.id, port->connector->name,
> > > +			 port, prev_slots, req_slots);
> > > +
> > > +	/* Add the new allocation to the state */
> > > +	if (!vcpi) {
> > > +		vcpi = kzalloc(sizeof(*vcpi), GFP_KERNEL);
> > > +		if (!vcpi) {
> > > +			ret = -ENOMEM;
> > > +			goto out;
> > > +		}
> > >  
> > > +		vcpi->port = drm_dp_get_validated_port_ref(mgr, port);
> > > +		if (!vcpi->port) {
> > > +			ret = -EINVAL;
> > > +			kfree(vcpi);
> > > +			goto out;
> > > +		}
> > > +		list_add(&vcpi->next, &topology_state->vcpis);
> > > +	}
> > > +	vcpi->vcpi = req_slots;
> > > +
> > > +	ret = req_slots;
> > > +out:
> > >  	drm_dp_put_port(port);
> > > -	return req_slots;
> > > +	return ret;
> > >  }
> > >  EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
> > >  
> > > @@ -2661,31 +2722,66 @@
EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
> > >   * drm_dp_atomic_release_vcpi_slots() - Release allocated vcpi
slots
> > >   * @state: global atomic state
> > >   * @mgr: MST topology manager for the port
> > > - * @slots: number of vcpi slots to release
> > > + * @port: The port to release the VCPI slots from
> > >   *
> > > - * RETURNS:
> > > - * 0 if @slots were added back to
&drm_dp_mst_topology_state->avail_slots
> > > or
> > > - * negative error code
> > > + * Releases any VCPI slots that have been allocated to a port in
the
> > > atomic
> > > + * state. Any atomic drivers which support MST must call this
function in
> > > + * their &drm_connector_helper_funcs.atomic_check() callback
when the
> > > + * connector will no longer have VCPI allocated (e.g. because
it's CRTC
> > > was
> > > + * removed) when it had VCPI allocated in the previous atomic
state.
> > > + *
> > > + * It is OK to call this even if @port has been removed from the
system.
> > > + * Additionally, it is OK to call this function multiple times
on the
> > > same
> > > + * @port as needed. It is not OK however, to call this function
and
> > > + * drm_dp_atomic_find_vcpi_slots() on the same @port in a single
atomic
> > > check
> > > + * phase.
> > > + *
> > > + * See also:
> > > + * drm_dp_atomic_find_vcpi_slots()
> > > + * drm_dp_mst_atomic_check()
> > > + *
> > > + * Returns:
> > > + * 0 if all slots for this port were added back to
> > > + * &drm_dp_mst_topology_state.avail_slots or negative error
code
> > >   */
> > >  int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state
*state,
> > >  				     struct drm_dp_mst_topology_mgr *mgr,
> > > -				     int slots)
> > > +				     struct drm_dp_mst_port *port)
> > >  {
> > >  	struct drm_dp_mst_topology_state *topology_state;
> > > +	struct drm_dp_vcpi_allocation *pos;
> > > +	bool found = false;
> > >  
> > >  	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
> > >  	if (IS_ERR(topology_state))
> > >  		return PTR_ERR(topology_state);
> > >  
> > > -	/* We cannot rely on port->vcpi.num_slots to update
> > > -	 * topology_state->avail_slots as the port may not exist if
the parent
> > > -	 * branch device was unplugged. This should be fixed by
tracking
> > > -	 * per-port slot allocation in drm_dp_mst_topology_state
instead of
> > > -	 * depending on the caller to tell us how many slots to
release.
> > > -	 */
> > > -	topology_state->avail_slots += slots;
> > > -	DRM_DEBUG_KMS("vcpi slots released=%d, avail=%d\n",
> > > -			slots, topology_state->avail_slots);
> > > +	list_for_each_entry(pos, &topology_state->vcpis, next) {
> > > +		if (pos->port == port) {
> > > +			found = true;
> > > +			break;
> > > +		}
> > > +	}
> > > +	if (WARN_ON(!found)) {
> > > +		DRM_ERROR("no VCPI for [MST PORT:%p] found in mst state
%p\n",
> > > +			  port, &topology_state->base);
> > > +		return -EINVAL;
> > > +	}
> > > +
> > > +	DRM_DEBUG_ATOMIC("[MST PORT:%p] VCPI %d -> 0\n",
port, pos->vcpi);
> > > +	if (pos->vcpi) {
> > > +		/*
> > > +		 * This should be safe-as the previous state will still hold a
> > > +		 * reference to this port until we swap states. If it's
not,
> > > +		 * the driver has a bookkeeping error
> > > +		 */
> > > +		if (WARN_ON(drm_dp_put_port(port))) {
> > 
> > The usual way to do this (instead of adding a return value to the
normal
> > _put function) is something like
> > 
> > 	kref_put(port, function_that_with_just_a_WARNING_DRM_ERROR)
> > 
> > avoids the need for patch 2.
> > 
> > > +			DRM_ERROR("cannot allocate and release VCPI on [MST
> > > PORT:%p] in the same state\n",
> > > +				  port);
> > > +			return -EINVAL;
> > > +		}
> > > +		pos->vcpi = 0;
> > > +	}
> > >  
> > >  	return 0;
> > >  }
> > > @@ -3115,15 +3211,45 @@ static void
drm_dp_destroy_connector_work(struct
> > > work_struct *work)
> > >  static struct drm_private_state *
> > >  drm_dp_mst_duplicate_state(struct drm_private_obj *obj)
> > >  {
> > > -	struct drm_dp_mst_topology_state *state;
> > > +	struct drm_dp_mst_topology_state *state, *old_state > >
> +		to_dp_mst_topology_state(obj->state);
> > > +	struct drm_dp_vcpi_allocation *pos, *vcpi;
> > >  
> > > -	state = kmemdup(obj->state, sizeof(*state), GFP_KERNEL);
> > > +	state = kmemdup(old_state, sizeof(*state), GFP_KERNEL);
> > >  	if (!state)
> > >  		return NULL;
> > >  
> > >  	__drm_atomic_helper_private_obj_duplicate_state(obj,
&state->base);
> > >  
> > > +	INIT_LIST_HEAD(&state->vcpis);
> > > +	state->vcpi_allocated = false;
> > > +
> > > +	list_for_each_entry(pos, &old_state->vcpis, next) {
> > > +		/* Prune leftover freed VCPI allocations */
> > > +		if (!pos->vcpi)
> > > +			continue;
> > > +
> > > +		vcpi = kmemdup(pos, sizeof(*vcpi), GFP_KERNEL);
> > > +		if (!vcpi)
> > > +			goto fail;
> > > +
> > > +		/* Take a plain old kref here and don't validate the port,
as
> > > +		 * it's topology might not even exist anymore
> > > +		 */
> > > +		kref_get(&vcpi->port->kref);
> > > +		list_add(&vcpi->next, &state->vcpis);
> > > +	}
> > > +
> > >  	return &state->base;
> > > +
> > > +fail:
> > > +	list_for_each_entry_safe(pos, vcpi, &state->vcpis, next)
{
> > > +		drm_dp_put_port(pos->port);
> > > +		kfree(pos);
> > > +	}
> > > +	kfree(state);
> > > +
> > > +	return NULL;
> > >  }
> > >  
> > >  static void drm_dp_mst_destroy_state(struct drm_private_obj
*obj,
> > > @@ -3131,14 +3257,116 @@ static void
drm_dp_mst_destroy_state(struct
> > > drm_private_obj *obj,
> > >  {
> > >  	struct drm_dp_mst_topology_state *mst_state > > >  	
to_dp_mst_topology_state(state);
> > > +	struct drm_dp_vcpi_allocation *pos, *tmp;
> > > +
> > > +	list_for_each_entry_safe(pos, tmp, &mst_state->vcpis,
next) {
> > > +		/* We only keep references to ports with non-zero VCPIs */
> > > +		if (pos->vcpi)
> > > +			drm_dp_put_port(pos->port);
> > > +		kfree(pos);
> > > +	}
> > >  
> > >  	kfree(mst_state);
> > >  }
> > >  
> > > -static const struct drm_private_state_funcs mst_state_funcs = {
> > > +static inline int
> > > +drm_dp_mst_atomic_check_topology_state(struct
drm_dp_mst_topology_mgr
> > > *mgr,
> > > +				       struct drm_dp_mst_topology_state
> > > *mst_state)
> > > +{
> > > +	struct drm_dp_vcpi_allocation *vcpi;
> > > +	struct drm_dp_mst_port *port;
> > > +	int avail_slots = 63, ret;
> > > +
> > > +	/* There's no possible scenario where releasing VCPI or
keeping it the
> > > +	 * same would make the state invalid
> > > +	 */
> > > +	if (!mst_state->vcpi_allocated) {
> > 
> > Do we need this? If not I'm always in favour of removing code if
it's not
> > needed :-)
> answered by my explanation above ^
If removing this breaks the code then I think it's a bit fragile. And see
above, open to races (yes unlikely ones, but then all of mst is full of
unlikely races it that still seem to happen eventually).

I think it's better if we can remove all these special cases (which should
be possible if we drop the get_validated_port below).
> > 
> > > +		DRM_DEBUG_ATOMIC("[MST MGR:%p] state %p allocates no
VCPI,
> > > check skipped\n",
> > > +				 mgr, &mst_state->base);
> > > +		return 0;
> > > +	}
> > > +
> > > +	list_for_each_entry(vcpi, &mst_state->vcpis, next) {
> > > +		/* Releasing VCPI is always OK-even if the port is gone */
> > > +		if (!vcpi->vcpi) {
> > > +			DRM_DEBUG_ATOMIC("[MST PORT:%p] releases all VCPI
> > > slots\n",
> > > +					 vcpi->port);
> > > +			continue;
> > > +		}
> > > +
> > > +		port = drm_dp_get_validated_port_ref(mgr, vcpi->port);
> > > +		if (!port) {
> > > +			DRM_DEBUG_ATOMIC("[MST PORT:%p] is gone but still has
> > > VCPI, cannot add new VCPI\n",
> > > +					 vcpi->port);
> > > +			return -EINVAL;
> > > +		}
> > 
> > For find_vcpi we've already done this. For unplugged vcpi, we
can't do
> > this here I think - otherwise if you enable 2 mst outputs on the same
> > branch, and unplug both, old non-atomic userspace can never ever again
> > disable them because it can't disable both. The check (to skip
already
> > deallocated ports) above isn't enough to handle that.
> 
> answered by my explanation above ^
> > 
> > > +
> > > +		DRM_DEBUG_ATOMIC("[MST PORT:%p] requires %d vcpi
slots\n",
> > > +				 vcpi->port, vcpi->vcpi);
> > > +
> > > +		avail_slots -= vcpi->vcpi;
> > > +		if (avail_slots < 0) {
> > > +			DRM_DEBUG_ATOMIC("[MST PORT:%p] not enough VCPI slots
> > > in mst state %p (avail=%d)\n",
> > > +					 vcpi->port, mst_state,
> > > +					 avail_slots + vcpi->vcpi);
> > > +			ret = -ENOSPC;
> > > +			goto port_fail;
> > > +		}
> > > +
> > > +		drm_dp_put_port(port);
> > > +	}
> > > +	DRM_DEBUG_ATOMIC("[MST MGR:%p] mst state %p VCPI avail=%d
used=%d\n",
> > > +			 mgr, mst_state, avail_slots,
> > > +			 63 - avail_slots);
> > > +
> > > +	return 0;
> > > +port_fail:
> > > +	drm_dp_put_port(port);
> > > +	return ret;
> > > +}
> > > +
> > > +/**
> > > + * drm_dp_mst_atomic_check - Check that the new state of an MST
topology
> > > in an
> > > + * atomic update is valid
> > > + * @state: Pointer to the new &struct
drm_dp_mst_topology_state
> > > + *
> > > + * Checks the given topology state for an atomic update to
ensure that
> > > it's
> > > + * valid. This includes checking whether there's enough
bandwidth to
> > > support
> > > + * the new VCPI allocations in the atomic update.
> > > + *
> > > + * Any atomic drivers supporting DP MST must make sure to call
this after
> > > + * checking the rest of their state in their
> > > + * &drm_mode_config_funcs.atomic_check() callback.
> > > + *
> > > + * See also:
> > > + * drm_dp_atomic_find_vcpi_slots()
> > > + * drm_dp_atomic_release_vcpi_slots()
> > > + *
> > > + * Returns:
> > > + *
> > > + * 0 if the new state is valid, negative error code otherwise.
> > > + */
> > > +int drm_dp_mst_atomic_check(struct drm_atomic_state *state)
> > > +{
> > > +	struct drm_dp_mst_topology_mgr *mgr;
> > > +	struct drm_dp_mst_topology_state *mst_state;
> > > +	int i, ret = 0;
> > > +
> > > +	for_each_new_mst_mgr_in_state(state, mgr, mst_state, i) {
> > > +		ret = drm_dp_mst_atomic_check_topology_state(mgr, mst_state);
> > > +		if (ret)
> > > +			break;
> > > +	}
> > > +
> > > +	return ret;
> > > +}
> > > +EXPORT_SYMBOL(drm_dp_mst_atomic_check);
> > > +
> > > +const struct drm_private_state_funcs
drm_dp_mst_topology_state_funcs = {
> > >  	.atomic_duplicate_state = drm_dp_mst_duplicate_state,
> > >  	.atomic_destroy_state = drm_dp_mst_destroy_state,
> > >  };
> > > +EXPORT_SYMBOL(drm_dp_mst_topology_state_funcs);
> > >  
> > >  /**
> > >   * drm_atomic_get_mst_topology_state: get MST topology state
> > > @@ -3216,13 +3444,11 @@ int drm_dp_mst_topology_mgr_init(struct
> > > drm_dp_mst_topology_mgr *mgr,
> > >  		return -ENOMEM;
> > >  
> > >  	mst_state->mgr = mgr;
> > > -
> > > -	/* max. time slots - one slot for MTP header */
> > > -	mst_state->avail_slots = 63;
> > > +	INIT_LIST_HEAD(&mst_state->vcpis);
> > >  
> > >  	drm_atomic_private_obj_init(&mgr->base,
> > >  				    &mst_state->base,
> > > -				    &mst_state_funcs);
> > > +				    &drm_dp_mst_topology_state_funcs);
> > >  
> > >  	return 0;
> > >  }
> > > diff --git a/drivers/gpu/drm/i915/intel_display.c
> > > b/drivers/gpu/drm/i915/intel_display.c
> > > index 132e978227fb..8e85e04a5948 100644
> > > --- a/drivers/gpu/drm/i915/intel_display.c
> > > +++ b/drivers/gpu/drm/i915/intel_display.c
> > > @@ -12592,6 +12592,10 @@ static int intel_atomic_check(struct
drm_device
> > > *dev,
> > >  				       "[modeset]" : "[fastset]");
> > >  	}
> > >  
> > > +	ret = drm_dp_mst_atomic_check(state);
> > > +	if (ret)
> > > +		return ret;
> > > +
> > >  	if (any_ms) {
> > >  		ret = intel_modeset_checks(state);
> > >  
> > > diff --git a/drivers/gpu/drm/i915/intel_dp_mst.c
> > > b/drivers/gpu/drm/i915/intel_dp_mst.c
> > > index 4de247ddf05f..be2d1ac7d87d 100644
> > > --- a/drivers/gpu/drm/i915/intel_dp_mst.c
> > > +++ b/drivers/gpu/drm/i915/intel_dp_mst.c
> > > @@ -41,8 +41,13 @@ static bool intel_dp_mst_compute_config(struct
> > > intel_encoder *encoder,
> > >  	struct drm_connector *connector = conn_state->connector;
> > >  	void *port = to_intel_connector(connector)->port;
> > >  	struct drm_atomic_state *state = pipe_config->base.state;
> > > +	struct drm_crtc *crtc = pipe_config->base.crtc;
> > > +	struct drm_crtc_state *old_crtc_state > > > +	
drm_atomic_get_old_crtc_state(state, crtc);
> > > +	struct drm_crtc_state *new_crtc_state =
&pipe_config->base;
> > >  	int bpp;
> > > -	int lane_count, slots = 0;
> > > +	int lane_count, slots > > > +	
to_intel_crtc_state(old_crtc_state)->dp_m_n.tu;
> > >  	const struct drm_display_mode *adjusted_mode =
&pipe_config-
> > > >base.adjusted_mode;
> > >  	int mst_pbn;
> > >  	bool constant_n = drm_dp_has_quirk(&intel_dp->desc,
> > > @@ -77,8 +82,12 @@ static bool intel_dp_mst_compute_config(struct
> > > intel_encoder *encoder,
> > >  	mst_pbn = drm_dp_calc_pbn_mode(adjusted_mode->crtc_clock,
bpp);
> > >  	pipe_config->pbn = mst_pbn;
> > >  
> > > -	/* Zombie connectors can't have VCPI slots */
> > > -	if (!drm_connector_is_unregistered(connector)) {
> > > +	/* Only change VCPI allocation on actual mode changes, to
prevent us
> > > +	 * from trying to allocate VCPI to ports that no longer exist
when we
> > > +	 * may just be trying to disable DPMS on them
> > > +	 */
> > > +	if (new_crtc_state->mode_changed ||
> > > +	    new_crtc_state->connectors_changed) {
> > 
> > Why exactly is this needed? We only call compute_config on the
enocoder
> > when enabling an output. Not when disabling one. Similar for the
atomic
> > helpers and the various encoder callbacks (you have the same change in
the
> > nouveau code). Did this actually blow up while testing?
> 
> yes-it looked like we were calling compute_config on simple DPMS on->off
> chnges as well, this ended up making it so we wouldn't try changing the
VCPI
> allocation on such commits.
Ah right, we discussed all this when adding the is_unregistered() check.
If we don't change the port lifetime (but only when kfree is called) then
there's no need to change anything here. Allows us to keep the nice
consistency of the is_unregistered() check in driver code.

I should have rewritten this when I realized that you've changed the
lifetime rules in a different way than what I had in mind, as-is my
comment didn't make much sense.
-Daniel
> ACK on all of the other comments here.
> > 
> > >  		slots = drm_dp_atomic_find_vcpi_slots(state,
> > >  						      &intel_dp->mst_mgr,
> > >  						      port,
> > > @@ -107,36 +116,39 @@ static bool
intel_dp_mst_compute_config(struct
> > > intel_encoder *encoder,
> > >  	return true;
> > >  }
> > >  
> > > -static int intel_dp_mst_atomic_check(struct drm_connector
*connector,
> > > -		struct drm_connector_state *new_conn_state)
> > > +static int
> > > +intel_dp_mst_atomic_check(struct drm_connector *connector,
> > > +			  struct drm_connector_state *new_conn_state)
> > >  {
> > > +	struct intel_connector *intel_connector > > > +	
to_intel_connector(connector);
> > > +	struct drm_dp_mst_topology_mgr *mgr > > > +	
&intel_connector->mst_port->mst_mgr;
> > > +	struct drm_dp_mst_port *port = intel_connector->port;
> > >  	struct drm_atomic_state *state = new_conn_state->state;
> > > -	struct drm_connector_state *old_conn_state;
> > > -	struct drm_crtc *old_crtc;
> > > -	struct drm_crtc_state *crtc_state;
> > > -	int slots, ret = 0;
> > > +	struct drm_connector_state *old_conn_state > > > +	
drm_atomic_get_old_connector_state(state, connector);
> > > +	struct drm_crtc *old_crtc = old_conn_state->crtc,
> > > +			*new_crtc = new_conn_state->crtc;
> > > +	struct drm_crtc_state *old_crtc_state, *new_crtc_state;
> > >  
> > > -	old_conn_state = drm_atomic_get_old_connector_state(state,
connector);
> > > -	old_crtc = old_conn_state->crtc;
> > 
> > Bikeshed: Reworking all the above hides the real changes quite a bit.
> > 
> > >  	if (!old_crtc)
> > > -		return ret;
> > > +		return 0;
> > >  
> > > -	crtc_state = drm_atomic_get_new_crtc_state(state, old_crtc);
> > > -	slots = to_intel_crtc_state(crtc_state)->dp_m_n.tu;
> > > -	if (drm_atomic_crtc_needs_modeset(crtc_state) && slots
> 0) {
> > > -		struct drm_dp_mst_topology_mgr *mgr;
> > > -		struct drm_encoder *old_encoder;
> > > +	old_crtc_state = drm_atomic_get_old_crtc_state(state,
old_crtc);
> > > +	if (!old_crtc_state ||
> > 
> > !old_crtc_state would be a framework bug. Just remove that check.
> > 
> > > +	    !to_intel_crtc_state(old_crtc_state)->dp_m_n.tu)
> > > +		return 0;
> > 
> > Hm, why do we need this? Having a DP port previously enabled with a 0
tu
> > sounds like a pretty big driver bug.
> > >  
> > > -		old_encoder = old_conn_state->best_encoder;
> > > -		mgr = &enc_to_mst(old_encoder)->primary->dp.mst_mgr;
> > > +	/* If we switched CRTCs, clear the previous one's
allocation */
> > > +	new_crtc_state = drm_atomic_get_new_crtc_state(state,
old_crtc);
> > > +	if (new_crtc_state->connectors_changed &&
!new_crtc_state->enable)
> > > +		to_intel_crtc_state(new_crtc_state)->dp_m_n.tu = 0;
> > 
> > No idea why we need to clear this here? Shouldn't we just clear
this
> > unconditionally if we release the allocation.
> > 
> > Also naming convention, I'd call this old_crtc_new_state, and
maybe
> > old_crtc_old_state. To make it clear these are the old/new states for
> > old_crtc. Instead of old/new state for the new crtc.
> > 
> > >  
> > > -		ret = drm_dp_atomic_release_vcpi_slots(state, mgr, slots);
> > > -		if (ret)
> > > -			DRM_DEBUG_KMS("failed releasing %d vcpi
slots:%d\n",
> > > slots, ret);
> > > -		else
> > > -			to_intel_crtc_state(crtc_state)->dp_m_n.tu = 0;
> > > -	}
> > > -	return ret;
> > > +	if (new_crtc)
> > > +		return 0;
> > 
> > Not sure why we shouldn't release this? If we handle this in
helpers, it
> > would remove piles of special casing code here I think. Plus remove
> > special code in the hlpers.
> > 
> > 
> > > +
> > > +	return drm_dp_atomic_release_vcpi_slots(state, mgr, port);
> > >  }
> > >  
> > >  static void intel_mst_disable_dp(struct intel_encoder *encoder,
> > > diff --git a/include/drm/drm_dp_mst_helper.h
> > > b/include/drm/drm_dp_mst_helper.h
> > > index 3faceb66f5cb..2392e12b6bce 100644
> > > --- a/include/drm/drm_dp_mst_helper.h
> > > +++ b/include/drm/drm_dp_mst_helper.h
> > > @@ -406,9 +406,16 @@ struct drm_dp_payload {
> > >  
> > >  #define to_dp_mst_topology_state(x) container_of(x, struct
> > > drm_dp_mst_topology_state, base)
> > >  
> > > +struct drm_dp_vcpi_allocation {
> > 
> > I think a comment here that this pointer holds a full reference (even
past
> > when the connector is unplugged) would be good. Otherwise the loop in
> > release_vcpi would be buggy.
> > 
> > > +	struct drm_dp_mst_port *port;
> > > +	int vcpi;
> > > +	struct list_head next;
> > > +};
> > > +
> > >  struct drm_dp_mst_topology_state {
> > >  	struct drm_private_state base;
> > > -	int avail_slots;
> > > +	struct list_head vcpis;
> > > +	u8 vcpi_allocated;
> > >  	struct drm_dp_mst_topology_mgr *mgr;
> > >  };
> > >  
> > > @@ -619,14 +626,17 @@ void drm_dp_mst_topology_mgr_suspend(struct
> > > drm_dp_mst_topology_mgr *mgr);
> > >  int drm_dp_mst_topology_mgr_resume(struct
drm_dp_mst_topology_mgr *mgr);
> > >  struct drm_dp_mst_topology_state
> > > *drm_atomic_get_mst_topology_state(struct drm_atomic_state
*state,
> > >  								    struct
> > > drm_dp_mst_topology_mgr *mgr);
> > > -int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state
*state,
> > > -				  struct drm_dp_mst_topology_mgr *mgr,
> > > -				  struct drm_dp_mst_port *port, int pbn);
> > > -int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state
*state,
> > > -				     struct drm_dp_mst_topology_mgr *mgr,
> > > -				     int slots);
> > > +int __must_check
> > > +drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
> > > +			      struct drm_dp_mst_topology_mgr *mgr,
> > > +			      struct drm_dp_mst_port *port, int pbn);
> > > +int __must_check
> > > +drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
> > > +				 struct drm_dp_mst_topology_mgr *mgr,
> > > +				 struct drm_dp_mst_port *port);
> > >  int drm_dp_send_power_updown_phy(struct drm_dp_mst_topology_mgr
*mgr,
> > >  				 struct drm_dp_mst_port *port, bool power_up);
> > > +int __must_check drm_dp_mst_atomic_check(struct drm_atomic_state
*state);
> > >  
> > >  extern const struct drm_private_state_funcs
> > > drm_dp_mst_topology_state_funcs;
> > >  
> > > -- 
> > > 2.19.1
> > > 
> -- 
> Cheers,
> 	Lyude Paul
> 
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

On Mon, 2018-11-26 at 22:22 +0100, Daniel Vetter wrote:
> On Mon, Nov 26, 2018 at 10:04:21PM +0100, Daniel Vetter wrote:
> > On Thu, Nov 15, 2018 at 07:50:05PM -0500, Lyude Paul wrote:
> > > There has been a TODO waiting for quite a long time in
> > > drm_dp_mst_topology.c:
> > > 
> > > 	/* We cannot rely on port->vcpi.num_slots to update
> > > 	 * topology_state->avail_slots as the port may not exist if
the parent
> > > 	 * branch device was unplugged. This should be fixed by tracking
> > > 	 * per-port slot allocation in drm_dp_mst_topology_state instead
of
> > > 	 * depending on the caller to tell us how many slots to release.
> > > 	 */
> > > 
> > > That's not the only reason we should fix this: forcing the
driver to
> > > track the VCPI allocations throughout a state's atomic check
is
> > > error prone, because it means that extra care has to be taken
with the
> > > order that drm_dp_atomic_find_vcpi_slots() and
> > > drm_dp_atomic_release_vcpi_slots() are called in in order to
ensure
> > > idempotency. Currently the only driver actually using these
helpers,
> > > i915, doesn't even do this correctly: multiple
->best_encoder() checks
> > > with i915's current implementation would not be idempotent
and would
> > > over-allocate VCPI slots, something I learned trying to implement
> > > fallback retraining in MST.
> > > 
> > > So: simplify this whole mess, and teach
drm_dp_atomic_find_vcpi_slots()
> > > and drm_dp_atomic_release_vcpi_slots() to track the VCPI
allocations for
> > > each port. This allows us to ensure idempotency without having to
rely
> > > on the driver as much. Additionally: the driver doesn't need
to do any
> > > kind of VCPI slot tracking anymore if it doesn't need it for
it's own
> > > internal state.
> > > 
> > > Additionally; this adds a new drm_dp_mst_atomic_check() helper
which
> > > must be used by atomic drivers to perform validity checks for the
new
> > > VCPI allocations incurred by a state.
> > > 
> > > Also: update the documentation and make it more obvious that
these
> > > /must/ be called by /all/ atomic drivers supporting MST.
> > > 
> > > Changes since v6:
> > >  - Keep a kref to all of the ports we have allocations on. This
required
> > >    a good bit of changing to when we call
drm_dp_find_vcpi_slots(),
> > >    mainly that we need to ensure that we only redo VCPI
allocations on
> > >    actual mode or CRTC changes, not crtc_state->active
changes.
> > >    Additionally, we no longer take the registration of the DRM
connector
> > >    for each port into account because so long as we have a kref
to the
> > >    port in the new or previous atomic state, the connector will
stay
> > >    registered.
> > 
> > I write an entire pile of small nitpits (still included most of them
> > below), until I realized this here wont work. Delaying the call to
destroy
> > the connector (well, unregister it really) wreaks the design we've
come up
> > with thus far, resulting in most of my comments here.
> > 
> > Instead, all we need to do is delay the kfree(port) at the bottom of
> > drm_dp_destroy_port(). The vcpi allocation structure _only_ needs the
> > pointer value to stay valid, as a lookup key. It doesn't care at
all about
> > anything actually stored in there. So the only thing we need to delay
is
> > the kfree. I think the simplest way to achieve this is to add a 2nd
kref
> > (port->kfree_ref or something like that), with on reference held by
the
> > port itself (released in drm_dp_destroy_port), and the other one held
> > as-needed by the vcpi allocation lists.
> > 
> > I think if we go with this design instead of retrofitting a semantic
> > change of the port lifetime itself, all the complications I complain
about
> > below should disappear.
> 
> In the above I meant drm_dp_destroy_port or drm_dp_destroy_connector_work.
> 
> Aside: I think creating a kref for the final kfree would also solve a
> bunch of other issues in a much neater way: In
> 
> commit f038c5b99fc1332f558b495d136d4f433ee65caa
> Author: Lyude Paul <lyude at redhat.com>
> Date:   Tue Nov 13 17:46:14 2018 -0500
> 
>     drm/dp_mst: Skip validating ports during destruction, just ref
> 
> we could use that kfree reference to make sure the port pointer is alive.
> This of course means that drm_dp_update_payload_part1() would also need to
> use the kfree reference for the vcpi allocations. And probably everywhere
> else (e.g. in your nouveau series for payload information).
> 
> That would give us a very clear separation overall between "port can
still
> be used because it's not yet unplugged" vs. "port only hangs
around
> because a bunch of vcpi allocations and other payload things upstream of
> the port might still need the port lookup key to free resources".
Instead
> of again sprinkling complicated conditions and magic tricks all over the
> code to figure out whethe we should allow the get_validate_port to succeed
> or not, or the modeset to succeed or not.
mhm, I've been experimenting with this and I think I agree this is
definitely
the right solution to go with. It'll end up decomplicating a lot of this
:)
> -Daniel
> 
> > Piles of comments below.
> > 
> > Cheers, Daniel
> > 
> > >  - Use the small changes to drm_dp_put_port() to add even more
error
> > >    checking to make misusage of the helpers more obvious. I added
this
> > >    after having to chase down various use-after-free conditions
that
> > >    started popping up from the new helpers so no one else has to
> > >    troubleshoot that.
> > >  - Move some accidental DRM_DEBUG_KMS() calls to
DRM_DEBUG_ATOMIC()
> > >  - Update documentation again, note that find/release() should
both not
> > > be
> > >    called on the same port in a single atomic check phase (but
multiple
> > >    calls to one or the other is OK)
> > > 
> > > Changes since v4:
> > >  - Don't skip the atomic checks for VCPI allocations if no
new VCPI
> > >    allocations happen in a state. This makes the next change
I'm about
> > >    to list here a lot easier to implement.
> > >  - Don't ignore VCPI allocations on destroyed ports, instead
ensure that
> > >    when ports are destroyed and still have VCPI allocations in
the
> > >    topology state, the only state changes allowed are releasing
said
> > >    ports' VCPI. This prevents a state with a mix of VCPI
allocations
> > >    from destroyed ports, and allocations from valid ports.
> > > 
> > > Changes since v3:
> > >  - Don't release VCPI allocations in the topology state
immediately in
> > >    drm_dp_atomic_release_vcpi_slots(), instead mark them as 0 and
skip
> > >    over them in drm_dp_mst_duplicate_state(). This makes it so
> > >    drm_dp_atomic_release_vcpi_slots() is still idempotent while
also
> > >    throwing warnings if the driver messes up it's book
keeping and tries
> > >    to release VCPI slots on a port that doesn't have any
pre-existing
> > >    VCPI allocation - danvet
> > >  - Change mst_state/state in some debugging messages to "mst
state"
> > > 
> > > Changes since v2:
> > >  - Use kmemdup() for duplicating MST state - danvet
> > >  - Move port validation out of duplicate state callback - danvet
> > >  - Handle looping through MST topology states in
> > >    drm_dp_mst_atomic_check() so the driver doesn't have to do
it
> > >  - Fix documentation in drm_dp_atomic_find_vcpi_slots()
> > >  - Move the atomic check for each individual topology state into
it's
> > >    own function, reduces indenting
> > >  - Don't consider "stale" MST ports when
calculating the bandwidth
> > >    requirements. This is needed because originally we relied on
the
> > >    state duplication functions to prune any stale ports from the
new
> > >    state, which would prevent us from incorrectly considering
their
> > >    bandwidth requirements alongside legitimate new payloads.
> > >  - Add function references in drm_dp_atomic_release_vcpi_slots()
-
> > > danvet
> > >  - Annotate atomic VCPI and atomic check functions with
__must_check
> > >    - danvet
> > > 
> > > Changes since v1:
> > >  - Don't use the now-removed ->atomic_check() for private
objects hook,
> > >    just give drivers a function to call themselves
> > > 
> > > Signed-off-by: Lyude Paul <lyude at redhat.com>
> > > Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
> > > ---
> > >  drivers/gpu/drm/drm_dp_mst_topology.c | 294
+++++++++++++++++++++++---
> > >  drivers/gpu/drm/i915/intel_display.c  |   4 +
> > >  drivers/gpu/drm/i915/intel_dp_mst.c   |  64 +++---
> > >  include/drm/drm_dp_mst_helper.h       |  24 ++-
> > >  4 files changed, 319 insertions(+), 67 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/drm_dp_mst_topology.c
> > > b/drivers/gpu/drm/drm_dp_mst_topology.c
> > > index 00fbe7a2699d..183d0e832ced 100644
> > > --- a/drivers/gpu/drm/drm_dp_mst_topology.c
> > > +++ b/drivers/gpu/drm/drm_dp_mst_topology.c
> > > @@ -2617,21 +2617,42 @@ static int drm_dp_init_vcpi(struct
> > > drm_dp_mst_topology_mgr *mgr,
> > >  }
> > >  
> > >  /**
> > > - * drm_dp_atomic_find_vcpi_slots() - Find and add vcpi slots to
the
> > > state
> > > + * drm_dp_atomic_find_vcpi_slots() - Find and add VCPI slots to
the
> > > state
> > >   * @state: global atomic state
> > >   * @mgr: MST topology manager for the port
> > >   * @port: port to find vcpi slots for
> > >   * @pbn: bandwidth required for the mode in PBN
> > >   *
> > > - * RETURNS:
> > > - * Total slots in the atomic state assigned for this port or
error
> > > + * Allocates VCPI slots to @port, replacing any previous VCPI
> > > allocations it
> > > + * may have had. Any atomic drivers which support MST must call
this
> > > function
> > > + * in their &drm_encoder_helper_funcs.atomic_check()
callback to change
> > > the
> > > + * current VCPI allocation for the new state, but only when
> > > + * &drm_crtc_state.mode_changed or
&drm_crtc_state.connectors_changed
> > > is set
> > > + * to ensure compatibility with userspace applications that
still use
> > > the
> > > + * legacy modesetting UAPI.
> > > + *
> > > + * Allocations set by this function are not checked against the
> > > bandwidth
> > > + * restraints of @mgr until the driver calls
drm_dp_mst_atomic_check().
> > > + *
> > > + * Additionally, it is OK to call this function multiple times
on the
> > > same
> > > + * @port as needed. It is not OK however, to call this function
and
> > > + * drm_dp_atomic_release_vcpi_slots() in the same atomic check
phase.
> > > + *
> > > + * See also:
> > > + * drm_dp_atomic_release_vcpi_slots()
> > > + * drm_dp_mst_atomic_check()
> > > + *
> > > + * Returns:
> > > + * Total slots in the atomic state assigned for this port, or a
> > > negative error
> > > + * code if the port no longer exists
> > >   */
> > >  int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state
*state,
> > >  				  struct drm_dp_mst_topology_mgr *mgr,
> > >  				  struct drm_dp_mst_port *port, int pbn)
> > >  {
> > >  	struct drm_dp_mst_topology_state *topology_state;
> > > -	int req_slots;
> > > +	struct drm_dp_vcpi_allocation *pos, *vcpi = NULL;
> > > +	int prev_slots, req_slots, ret;
> > >  
> > >  	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
> > >  	if (IS_ERR(topology_state))
> > > @@ -2640,20 +2661,60 @@ int drm_dp_atomic_find_vcpi_slots(struct
> > > drm_atomic_state *state,
> > >  	port = drm_dp_get_validated_port_ref(mgr, port);
> > >  	if (port == NULL)
> > >  		return -EINVAL;
> > > -	req_slots = DIV_ROUND_UP(pbn, mgr->pbn_div);
> > > -	DRM_DEBUG_KMS("vcpi slots req=%d, avail=%d\n",
> > > -			req_slots, topology_state->avail_slots);
> > >  
> > > -	if (req_slots > topology_state->avail_slots) {
> > > -		drm_dp_put_port(port);
> > > -		return -ENOSPC;
> > > +	topology_state->vcpi_allocated = true;
> > > +
> > > +	/* Find the current allocation for this port, if any */
> > > +	list_for_each_entry(pos, &topology_state->vcpis, next) {
> > > +		if (pos->port == port) {
> > > +			vcpi = pos;
> > > +			prev_slots = vcpi->vcpi;
> > > +
> > > +			/*
> > > +			 * This should never happen, unless the driver tries
> > > +			 * releasing and allocating the same VCPI allocation,
> > > +			 * which is an error
> > > +			 */
> > > +			if (WARN_ON(!prev_slots)) {
> > > +				DRM_ERROR("cannot allocate and release VCPI on
> > > [MST PORT:%p] in the same state\n",
> > > +					  port);
> > > +				return -EINVAL;
> > > +			}
> > 
> > Allowing atomic changes to output state (like disabling it on one
crtc,
> > and then enabling on the other, thus both releasing and re-allocating
the
> > vcpi in one go) is kinda the point of atomic. So definitely can't
EINVAL
> > here.
> > 
> > Sounds like we need more testcases in igt?
> > 
> > > +
> > > +			break;
> > > +		}
> > >  	}
> > > +	if (!vcpi)
> > > +		prev_slots = 0;
> > >  
> > > -	topology_state->avail_slots -= req_slots;
> > > -	DRM_DEBUG_KMS("vcpi slots avail=%d",
topology_state->avail_slots);
> > > +	req_slots = DIV_ROUND_UP(pbn, mgr->pbn_div);
> > > +
> > > +	DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] [MST PORT:%p] VCPI %d
-> %d\n",
> > > +			 port->connector->base.id, port->connector->name,
> > > +			 port, prev_slots, req_slots);
> > > +
> > > +	/* Add the new allocation to the state */
> > > +	if (!vcpi) {
> > > +		vcpi = kzalloc(sizeof(*vcpi), GFP_KERNEL);
> > > +		if (!vcpi) {
> > > +			ret = -ENOMEM;
> > > +			goto out;
> > > +		}
> > >  
> > > +		vcpi->port = drm_dp_get_validated_port_ref(mgr, port);
> > > +		if (!vcpi->port) {
> > > +			ret = -EINVAL;
> > > +			kfree(vcpi);
> > > +			goto out;
> > > +		}
> > > +		list_add(&vcpi->next, &topology_state->vcpis);
> > > +	}
> > > +	vcpi->vcpi = req_slots;
> > > +
> > > +	ret = req_slots;
> > > +out:
> > >  	drm_dp_put_port(port);
> > > -	return req_slots;
> > > +	return ret;
> > >  }
> > >  EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
> > >  
> > > @@ -2661,31 +2722,66 @@
EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
> > >   * drm_dp_atomic_release_vcpi_slots() - Release allocated vcpi
slots
> > >   * @state: global atomic state
> > >   * @mgr: MST topology manager for the port
> > > - * @slots: number of vcpi slots to release
> > > + * @port: The port to release the VCPI slots from
> > >   *
> > > - * RETURNS:
> > > - * 0 if @slots were added back to
&drm_dp_mst_topology_state-
> > > >avail_slots or
> > > - * negative error code
> > > + * Releases any VCPI slots that have been allocated to a port in
the
> > > atomic
> > > + * state. Any atomic drivers which support MST must call this
function
> > > in
> > > + * their &drm_connector_helper_funcs.atomic_check() callback
when the
> > > + * connector will no longer have VCPI allocated (e.g. because
it's CRTC
> > > was
> > > + * removed) when it had VCPI allocated in the previous atomic
state.
> > > + *
> > > + * It is OK to call this even if @port has been removed from the
> > > system.
> > > + * Additionally, it is OK to call this function multiple times
on the
> > > same
> > > + * @port as needed. It is not OK however, to call this function
and
> > > + * drm_dp_atomic_find_vcpi_slots() on the same @port in a single
atomic
> > > check
> > > + * phase.
> > > + *
> > > + * See also:
> > > + * drm_dp_atomic_find_vcpi_slots()
> > > + * drm_dp_mst_atomic_check()
> > > + *
> > > + * Returns:
> > > + * 0 if all slots for this port were added back to
> > > + * &drm_dp_mst_topology_state.avail_slots or negative error
code
> > >   */
> > >  int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state
*state,
> > >  				     struct drm_dp_mst_topology_mgr *mgr,
> > > -				     int slots)
> > > +				     struct drm_dp_mst_port *port)
> > >  {
> > >  	struct drm_dp_mst_topology_state *topology_state;
> > > +	struct drm_dp_vcpi_allocation *pos;
> > > +	bool found = false;
> > >  
> > >  	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
> > >  	if (IS_ERR(topology_state))
> > >  		return PTR_ERR(topology_state);
> > >  
> > > -	/* We cannot rely on port->vcpi.num_slots to update
> > > -	 * topology_state->avail_slots as the port may not exist if
the parent
> > > -	 * branch device was unplugged. This should be fixed by
tracking
> > > -	 * per-port slot allocation in drm_dp_mst_topology_state
instead of
> > > -	 * depending on the caller to tell us how many slots to
release.
> > > -	 */
> > > -	topology_state->avail_slots += slots;
> > > -	DRM_DEBUG_KMS("vcpi slots released=%d, avail=%d\n",
> > > -			slots, topology_state->avail_slots);
> > > +	list_for_each_entry(pos, &topology_state->vcpis, next) {
> > > +		if (pos->port == port) {
> > > +			found = true;
> > > +			break;
> > > +		}
> > > +	}
> > > +	if (WARN_ON(!found)) {
> > > +		DRM_ERROR("no VCPI for [MST PORT:%p] found in mst state
%p\n",
> > > +			  port, &topology_state->base);
> > > +		return -EINVAL;
> > > +	}
> > > +
> > > +	DRM_DEBUG_ATOMIC("[MST PORT:%p] VCPI %d -> 0\n",
port, pos->vcpi);
> > > +	if (pos->vcpi) {
> > > +		/*
> > > +		 * This should be safe-as the previous state will still hold a
> > > +		 * reference to this port until we swap states. If it's
not,
> > > +		 * the driver has a bookkeeping error
> > > +		 */
> > > +		if (WARN_ON(drm_dp_put_port(port))) {
> > 
> > The usual way to do this (instead of adding a return value to the
normal
> > _put function) is something like
> > 
> > 	kref_put(port, function_that_with_just_a_WARNING_DRM_ERROR)
> > 
> > avoids the need for patch 2.
> > 
> > > +			DRM_ERROR("cannot allocate and release VCPI on [MST
> > > PORT:%p] in the same state\n",
> > > +				  port);
> > > +			return -EINVAL;
> > > +		}
> > > +		pos->vcpi = 0;
> > > +	}
> > >  
> > >  	return 0;
> > >  }
> > > @@ -3115,15 +3211,45 @@ static void
drm_dp_destroy_connector_work(struct
> > > work_struct *work)
> > >  static struct drm_private_state *
> > >  drm_dp_mst_duplicate_state(struct drm_private_obj *obj)
> > >  {
> > > -	struct drm_dp_mst_topology_state *state;
> > > +	struct drm_dp_mst_topology_state *state, *old_state > >
> +		to_dp_mst_topology_state(obj->state);
> > > +	struct drm_dp_vcpi_allocation *pos, *vcpi;
> > >  
> > > -	state = kmemdup(obj->state, sizeof(*state), GFP_KERNEL);
> > > +	state = kmemdup(old_state, sizeof(*state), GFP_KERNEL);
> > >  	if (!state)
> > >  		return NULL;
> > >  
> > >  	__drm_atomic_helper_private_obj_duplicate_state(obj,
&state->base);
> > >  
> > > +	INIT_LIST_HEAD(&state->vcpis);
> > > +	state->vcpi_allocated = false;
> > > +
> > > +	list_for_each_entry(pos, &old_state->vcpis, next) {
> > > +		/* Prune leftover freed VCPI allocations */
> > > +		if (!pos->vcpi)
> > > +			continue;
> > > +
> > > +		vcpi = kmemdup(pos, sizeof(*vcpi), GFP_KERNEL);
> > > +		if (!vcpi)
> > > +			goto fail;
> > > +
> > > +		/* Take a plain old kref here and don't validate the port,
as
> > > +		 * it's topology might not even exist anymore
> > > +		 */
> > > +		kref_get(&vcpi->port->kref);
> > > +		list_add(&vcpi->next, &state->vcpis);
> > > +	}
> > > +
> > >  	return &state->base;
> > > +
> > > +fail:
> > > +	list_for_each_entry_safe(pos, vcpi, &state->vcpis, next)
{
> > > +		drm_dp_put_port(pos->port);
> > > +		kfree(pos);
> > > +	}
> > > +	kfree(state);
> > > +
> > > +	return NULL;
> > >  }
> > >  
> > >  static void drm_dp_mst_destroy_state(struct drm_private_obj
*obj,
> > > @@ -3131,14 +3257,116 @@ static void
drm_dp_mst_destroy_state(struct
> > > drm_private_obj *obj,
> > >  {
> > >  	struct drm_dp_mst_topology_state *mst_state > > >  	
to_dp_mst_topology_state(state);
> > > +	struct drm_dp_vcpi_allocation *pos, *tmp;
> > > +
> > > +	list_for_each_entry_safe(pos, tmp, &mst_state->vcpis,
next) {
> > > +		/* We only keep references to ports with non-zero VCPIs */
> > > +		if (pos->vcpi)
> > > +			drm_dp_put_port(pos->port);
> > > +		kfree(pos);
> > > +	}
> > >  
> > >  	kfree(mst_state);
> > >  }
> > >  
> > > -static const struct drm_private_state_funcs mst_state_funcs = {
> > > +static inline int
> > > +drm_dp_mst_atomic_check_topology_state(struct
drm_dp_mst_topology_mgr
> > > *mgr,
> > > +				       struct drm_dp_mst_topology_state
> > > *mst_state)
> > > +{
> > > +	struct drm_dp_vcpi_allocation *vcpi;
> > > +	struct drm_dp_mst_port *port;
> > > +	int avail_slots = 63, ret;
> > > +
> > > +	/* There's no possible scenario where releasing VCPI or
keeping it the
> > > +	 * same would make the state invalid
> > > +	 */
> > > +	if (!mst_state->vcpi_allocated) {
> > 
> > Do we need this? If not I'm always in favour of removing code if
it's not
> > needed :-)
> > 
> > > +		DRM_DEBUG_ATOMIC("[MST MGR:%p] state %p allocates no
VCPI,
> > > check skipped\n",
> > > +				 mgr, &mst_state->base);
> > > +		return 0;
> > > +	}
> > > +
> > > +	list_for_each_entry(vcpi, &mst_state->vcpis, next) {
> > > +		/* Releasing VCPI is always OK-even if the port is gone */
> > > +		if (!vcpi->vcpi) {
> > > +			DRM_DEBUG_ATOMIC("[MST PORT:%p] releases all VCPI
> > > slots\n",
> > > +					 vcpi->port);
> > > +			continue;
> > > +		}
> > > +
> > > +		port = drm_dp_get_validated_port_ref(mgr, vcpi->port);
> > > +		if (!port) {
> > > +			DRM_DEBUG_ATOMIC("[MST PORT:%p] is gone but still has
> > > VCPI, cannot add new VCPI\n",
> > > +					 vcpi->port);
> > > +			return -EINVAL;
> > > +		}
> > 
> > For find_vcpi we've already done this. For unplugged vcpi, we
can't do
> > this here I think - otherwise if you enable 2 mst outputs on the same
> > branch, and unplug both, old non-atomic userspace can never ever again
> > disable them because it can't disable both. The check (to skip
already
> > deallocated ports) above isn't enough to handle that.
> > 
> > > +
> > > +		DRM_DEBUG_ATOMIC("[MST PORT:%p] requires %d vcpi
slots\n",
> > > +				 vcpi->port, vcpi->vcpi);
> > > +
> > > +		avail_slots -= vcpi->vcpi;
> > > +		if (avail_slots < 0) {
> > > +			DRM_DEBUG_ATOMIC("[MST PORT:%p] not enough VCPI slots
> > > in mst state %p (avail=%d)\n",
> > > +					 vcpi->port, mst_state,
> > > +					 avail_slots + vcpi->vcpi);
> > > +			ret = -ENOSPC;
> > > +			goto port_fail;
> > > +		}
> > > +
> > > +		drm_dp_put_port(port);
> > > +	}
> > > +	DRM_DEBUG_ATOMIC("[MST MGR:%p] mst state %p VCPI avail=%d
used=%d\n",
> > > +			 mgr, mst_state, avail_slots,
> > > +			 63 - avail_slots);
> > > +
> > > +	return 0;
> > > +port_fail:
> > > +	drm_dp_put_port(port);
> > > +	return ret;
> > > +}
> > > +
> > > +/**
> > > + * drm_dp_mst_atomic_check - Check that the new state of an MST
> > > topology in an
> > > + * atomic update is valid
> > > + * @state: Pointer to the new &struct
drm_dp_mst_topology_state
> > > + *
> > > + * Checks the given topology state for an atomic update to
ensure that
> > > it's
> > > + * valid. This includes checking whether there's enough
bandwidth to
> > > support
> > > + * the new VCPI allocations in the atomic update.
> > > + *
> > > + * Any atomic drivers supporting DP MST must make sure to call
this
> > > after
> > > + * checking the rest of their state in their
> > > + * &drm_mode_config_funcs.atomic_check() callback.
> > > + *
> > > + * See also:
> > > + * drm_dp_atomic_find_vcpi_slots()
> > > + * drm_dp_atomic_release_vcpi_slots()
> > > + *
> > > + * Returns:
> > > + *
> > > + * 0 if the new state is valid, negative error code otherwise.
> > > + */
> > > +int drm_dp_mst_atomic_check(struct drm_atomic_state *state)
> > > +{
> > > +	struct drm_dp_mst_topology_mgr *mgr;
> > > +	struct drm_dp_mst_topology_state *mst_state;
> > > +	int i, ret = 0;
> > > +
> > > +	for_each_new_mst_mgr_in_state(state, mgr, mst_state, i) {
> > > +		ret = drm_dp_mst_atomic_check_topology_state(mgr, mst_state);
> > > +		if (ret)
> > > +			break;
> > > +	}
> > > +
> > > +	return ret;
> > > +}
> > > +EXPORT_SYMBOL(drm_dp_mst_atomic_check);
> > > +
> > > +const struct drm_private_state_funcs
drm_dp_mst_topology_state_funcs > > > {
> > >  	.atomic_duplicate_state = drm_dp_mst_duplicate_state,
> > >  	.atomic_destroy_state = drm_dp_mst_destroy_state,
> > >  };
> > > +EXPORT_SYMBOL(drm_dp_mst_topology_state_funcs);
> > >  
> > >  /**
> > >   * drm_atomic_get_mst_topology_state: get MST topology state
> > > @@ -3216,13 +3444,11 @@ int drm_dp_mst_topology_mgr_init(struct
> > > drm_dp_mst_topology_mgr *mgr,
> > >  		return -ENOMEM;
> > >  
> > >  	mst_state->mgr = mgr;
> > > -
> > > -	/* max. time slots - one slot for MTP header */
> > > -	mst_state->avail_slots = 63;
> > > +	INIT_LIST_HEAD(&mst_state->vcpis);
> > >  
> > >  	drm_atomic_private_obj_init(&mgr->base,
> > >  				    &mst_state->base,
> > > -				    &mst_state_funcs);
> > > +				    &drm_dp_mst_topology_state_funcs);
> > >  
> > >  	return 0;
> > >  }
> > > diff --git a/drivers/gpu/drm/i915/intel_display.c
> > > b/drivers/gpu/drm/i915/intel_display.c
> > > index 132e978227fb..8e85e04a5948 100644
> > > --- a/drivers/gpu/drm/i915/intel_display.c
> > > +++ b/drivers/gpu/drm/i915/intel_display.c
> > > @@ -12592,6 +12592,10 @@ static int intel_atomic_check(struct
drm_device
> > > *dev,
> > >  				       "[modeset]" : "[fastset]");
> > >  	}
> > >  
> > > +	ret = drm_dp_mst_atomic_check(state);
> > > +	if (ret)
> > > +		return ret;
> > > +
> > >  	if (any_ms) {
> > >  		ret = intel_modeset_checks(state);
> > >  
> > > diff --git a/drivers/gpu/drm/i915/intel_dp_mst.c
> > > b/drivers/gpu/drm/i915/intel_dp_mst.c
> > > index 4de247ddf05f..be2d1ac7d87d 100644
> > > --- a/drivers/gpu/drm/i915/intel_dp_mst.c
> > > +++ b/drivers/gpu/drm/i915/intel_dp_mst.c
> > > @@ -41,8 +41,13 @@ static bool intel_dp_mst_compute_config(struct
> > > intel_encoder *encoder,
> > >  	struct drm_connector *connector = conn_state->connector;
> > >  	void *port = to_intel_connector(connector)->port;
> > >  	struct drm_atomic_state *state = pipe_config->base.state;
> > > +	struct drm_crtc *crtc = pipe_config->base.crtc;
> > > +	struct drm_crtc_state *old_crtc_state > > > +	
drm_atomic_get_old_crtc_state(state, crtc);
> > > +	struct drm_crtc_state *new_crtc_state =
&pipe_config->base;
> > >  	int bpp;
> > > -	int lane_count, slots = 0;
> > > +	int lane_count, slots > > > +	
to_intel_crtc_state(old_crtc_state)->dp_m_n.tu;
> > >  	const struct drm_display_mode *adjusted_mode =
&pipe_config-
> > > >base.adjusted_mode;
> > >  	int mst_pbn;
> > >  	bool constant_n = drm_dp_has_quirk(&intel_dp->desc,
> > > @@ -77,8 +82,12 @@ static bool intel_dp_mst_compute_config(struct
> > > intel_encoder *encoder,
> > >  	mst_pbn = drm_dp_calc_pbn_mode(adjusted_mode->crtc_clock,
bpp);
> > >  	pipe_config->pbn = mst_pbn;
> > >  
> > > -	/* Zombie connectors can't have VCPI slots */
> > > -	if (!drm_connector_is_unregistered(connector)) {
> > > +	/* Only change VCPI allocation on actual mode changes, to
prevent us
> > > +	 * from trying to allocate VCPI to ports that no longer exist
when we
> > > +	 * may just be trying to disable DPMS on them
> > > +	 */
> > > +	if (new_crtc_state->mode_changed ||
> > > +	    new_crtc_state->connectors_changed) {
> > 
> > Why exactly is this needed? We only call compute_config on the
enocoder
> > when enabling an output. Not when disabling one. Similar for the
atomic
> > helpers and the various encoder callbacks (you have the same change in
the
> > nouveau code). Did this actually blow up while testing?
> > 
> > >  		slots = drm_dp_atomic_find_vcpi_slots(state,
> > >  						      &intel_dp->mst_mgr,
> > >  						      port,
> > > @@ -107,36 +116,39 @@ static bool
intel_dp_mst_compute_config(struct
> > > intel_encoder *encoder,
> > >  	return true;
> > >  }
> > >  
> > > -static int intel_dp_mst_atomic_check(struct drm_connector
*connector,
> > > -		struct drm_connector_state *new_conn_state)
> > > +static int
> > > +intel_dp_mst_atomic_check(struct drm_connector *connector,
> > > +			  struct drm_connector_state *new_conn_state)
> > >  {
> > > +	struct intel_connector *intel_connector > > > +	
to_intel_connector(connector);
> > > +	struct drm_dp_mst_topology_mgr *mgr > > > +	
&intel_connector->mst_port->mst_mgr;
> > > +	struct drm_dp_mst_port *port = intel_connector->port;
> > >  	struct drm_atomic_state *state = new_conn_state->state;
> > > -	struct drm_connector_state *old_conn_state;
> > > -	struct drm_crtc *old_crtc;
> > > -	struct drm_crtc_state *crtc_state;
> > > -	int slots, ret = 0;
> > > +	struct drm_connector_state *old_conn_state > > > +	
drm_atomic_get_old_connector_state(state, connector);
> > > +	struct drm_crtc *old_crtc = old_conn_state->crtc,
> > > +			*new_crtc = new_conn_state->crtc;
> > > +	struct drm_crtc_state *old_crtc_state, *new_crtc_state;
> > >  
> > > -	old_conn_state = drm_atomic_get_old_connector_state(state,
connector);
> > > -	old_crtc = old_conn_state->crtc;
> > 
> > Bikeshed: Reworking all the above hides the real changes quite a bit.
> > 
> > >  	if (!old_crtc)
> > > -		return ret;
> > > +		return 0;
> > >  
> > > -	crtc_state = drm_atomic_get_new_crtc_state(state, old_crtc);
> > > -	slots = to_intel_crtc_state(crtc_state)->dp_m_n.tu;
> > > -	if (drm_atomic_crtc_needs_modeset(crtc_state) && slots
> 0) {
> > > -		struct drm_dp_mst_topology_mgr *mgr;
> > > -		struct drm_encoder *old_encoder;
> > > +	old_crtc_state = drm_atomic_get_old_crtc_state(state,
old_crtc);
> > > +	if (!old_crtc_state ||
> > 
> > !old_crtc_state would be a framework bug. Just remove that check.
> > 
> > > +	    !to_intel_crtc_state(old_crtc_state)->dp_m_n.tu)
> > > +		return 0;
> > 
> > Hm, why do we need this? Having a DP port previously enabled with a 0
tu
> > sounds like a pretty big driver bug.
> > >  
> > > -		old_encoder = old_conn_state->best_encoder;
> > > -		mgr = &enc_to_mst(old_encoder)->primary->dp.mst_mgr;
> > > +	/* If we switched CRTCs, clear the previous one's
allocation */
> > > +	new_crtc_state = drm_atomic_get_new_crtc_state(state,
old_crtc);
> > > +	if (new_crtc_state->connectors_changed &&
!new_crtc_state->enable)
> > > +		to_intel_crtc_state(new_crtc_state)->dp_m_n.tu = 0;
> > 
> > No idea why we need to clear this here? Shouldn't we just clear
this
> > unconditionally if we release the allocation.
> > 
> > Also naming convention, I'd call this old_crtc_new_state, and
maybe
> > old_crtc_old_state. To make it clear these are the old/new states for
> > old_crtc. Instead of old/new state for the new crtc.
> > 
> > >  
> > > -		ret = drm_dp_atomic_release_vcpi_slots(state, mgr, slots);
> > > -		if (ret)
> > > -			DRM_DEBUG_KMS("failed releasing %d vcpi
slots:%d\n",
> > > slots, ret);
> > > -		else
> > > -			to_intel_crtc_state(crtc_state)->dp_m_n.tu = 0;
> > > -	}
> > > -	return ret;
> > > +	if (new_crtc)
> > > +		return 0;
> > 
> > Not sure why we shouldn't release this? If we handle this in
helpers, it
> > would remove piles of special casing code here I think. Plus remove
> > special code in the hlpers.
> > 
> > 
> > > +
> > > +	return drm_dp_atomic_release_vcpi_slots(state, mgr, port);
> > >  }
> > >  
> > >  static void intel_mst_disable_dp(struct intel_encoder *encoder,
> > > diff --git a/include/drm/drm_dp_mst_helper.h
> > > b/include/drm/drm_dp_mst_helper.h
> > > index 3faceb66f5cb..2392e12b6bce 100644
> > > --- a/include/drm/drm_dp_mst_helper.h
> > > +++ b/include/drm/drm_dp_mst_helper.h
> > > @@ -406,9 +406,16 @@ struct drm_dp_payload {
> > >  
> > >  #define to_dp_mst_topology_state(x) container_of(x, struct
> > > drm_dp_mst_topology_state, base)
> > >  
> > > +struct drm_dp_vcpi_allocation {
> > 
> > I think a comment here that this pointer holds a full reference (even
past
> > when the connector is unplugged) would be good. Otherwise the loop in
> > release_vcpi would be buggy.
> > 
> > > +	struct drm_dp_mst_port *port;
> > > +	int vcpi;
> > > +	struct list_head next;
> > > +};
> > > +
> > >  struct drm_dp_mst_topology_state {
> > >  	struct drm_private_state base;
> > > -	int avail_slots;
> > > +	struct list_head vcpis;
> > > +	u8 vcpi_allocated;
> > >  	struct drm_dp_mst_topology_mgr *mgr;
> > >  };
> > >  
> > > @@ -619,14 +626,17 @@ void drm_dp_mst_topology_mgr_suspend(struct
> > > drm_dp_mst_topology_mgr *mgr);
> > >  int drm_dp_mst_topology_mgr_resume(struct
drm_dp_mst_topology_mgr
> > > *mgr);
> > >  struct drm_dp_mst_topology_state
> > > *drm_atomic_get_mst_topology_state(struct drm_atomic_state
*state,
> > >  								    struct
> > > drm_dp_mst_topology_mgr *mgr);
> > > -int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state
*state,
> > > -				  struct drm_dp_mst_topology_mgr *mgr,
> > > -				  struct drm_dp_mst_port *port, int pbn);
> > > -int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state
*state,
> > > -				     struct drm_dp_mst_topology_mgr *mgr,
> > > -				     int slots);
> > > +int __must_check
> > > +drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
> > > +			      struct drm_dp_mst_topology_mgr *mgr,
> > > +			      struct drm_dp_mst_port *port, int pbn);
> > > +int __must_check
> > > +drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
> > > +				 struct drm_dp_mst_topology_mgr *mgr,
> > > +				 struct drm_dp_mst_port *port);
> > >  int drm_dp_send_power_updown_phy(struct drm_dp_mst_topology_mgr
*mgr,
> > >  				 struct drm_dp_mst_port *port, bool power_up);
> > > +int __must_check drm_dp_mst_atomic_check(struct drm_atomic_state
> > > *state);
> > >  
> > >  extern const struct drm_private_state_funcs
> > > drm_dp_mst_topology_state_funcs;
> > >  
> > > -- 
> > > 2.19.1
> > > 
> > 
> > -- 
> > Daniel Vetter
> > Software Engineer, Intel Corporation
> > http://blog.ffwll.ch
-- 
Cheers,
	Lyude Paul