DisplayPort specifications are fun. For a while, it's been really
unclear to us what available_pbn actually does. There's a somewhat vague
explanation in the DisplayPort spec (starting from 1.2) that partially
explains it:
The minimum payload bandwidth number supported by the path. Each node
updates this number with its available payload bandwidth number if its
payload bandwidth number is less than that in the Message Transaction
reply.
So, it sounds like available_pbn represents the smallest link rate in
use between the source and the branch device. Cool, so full_pbn is just
the highest possible PBN that the branch device supports right?
Well, we assumed that for quite a while until Sean Paul noticed that on
some MST hubs, available_pbn will actually get set to 0 whenever there's
any active payloads on the respective branch device. This caused quite a
bit of confusion since clearing the payload ID table would end up fixing
the available_pbn value.
So, we just went with that until commit cd82d82cbc ("drm/dp_mst: Add
branch bandwidth validation to MST atomic check") started breaking
people's setups due to us getting erroneous available_pbn values. So, we
did some more digging and got confused until we finally looked at the
definition for full_pbn:
The bandwidth of the link at the trained link rate and lane count
between the DP Source device and the DP Sink device with no time slots
allocated to VC Payloads, represented as a Payload Bandwidth Number. As
with the Available_Payload_Bandwidth_Number, this number is determined
by the link with the lowest lane count and link rate.
That's what we get for not reading specs closely enough, hehe. So, since
full_pbn is definitely what we want for doing bandwidth restriction
checks - let's start using that instead and ignore available_pbn
entirely.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Fixes: cd82d82cbc ("drm/dp_mst: Add branch bandwidth validation to MST atomic check")
Cc: Mikita Lipski <mikita.lipski@amd.com>
Cc: Hans de Goede <hdegoede@redhat.com>
Cc: Sean Paul <sean@poorly.run>
Reviewed-by: Mikita Lipski <mikita.lipski@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200306234623.547525-3-lyude@redhat.com
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Tested-by: Hans de Goede <hdegoede@redhat.com>
[Why]
Noticed this while testing MST with the 4 ports MST hub from
StarTech.com. Sometimes can't light up monitors normally and get the
error message as 'sideband msg build failed'.
Look into aux transactions, found out that source sometimes will send
out another down request before receiving the down reply of the
previous down request. On the other hand, in drm_dp_get_one_sb_msg(),
current code doesn't handle the interleaved replies case. Hence, source
can't build up message completely and can't light up monitors.
[How]
For good compatibility, enforce source to send out one down request at a
time. Add a flag, is_waiting_for_dwn_reply, to determine if the source
can send out a down request immediately or not.
- Check the flag before calling process_single_down_tx_qlock to send out
a msg
- Set the flag when successfully send out a down request
- Clear the flag when successfully build up a down reply
- Clear the flag when find erros during sending out a down request
- Clear the flag when find errors during building up a down reply
- Clear the flag when timeout occurs during waiting for a down reply
- Use drm_dp_mst_kick_tx() to try to send another down request in queue
at the end of drm_dp_mst_wait_tx_reply() (attempt to send out messages
in queue when errors occur)
Cc: Lyude Paul <lyude@redhat.com>
Signed-off-by: Wayne Lin <Wayne.Lin@amd.com>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200113093649.11755-1-Wayne.Lin@amd.com
[why]
Whenever a connector on an MST network is changed or
undergoes a modeset, the DSC configs for each stream on that
topology will be recalculated. This can change their required
bandwidth, requiring a full reprogramming, as though a modeset
was performed, even if that stream did not change timing.
[how]
Adding helper to trigger modesets on MST DSC connectors
by setting mode_changed flag on CRTCs in the same topology
as affected connector
v2: use drm_dp_mst_dsc_aux_for_port function to verify
if the port is DSC capable
v3: - added _must_check attribute
- removed topology manager check
- fix typos and indentations
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Mikita Lipski <mikita.lipski@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[why]
Adding PBN attribute to drm_dp_vcpi_allocation structure to
keep track of how much bandwidth each Port requires.
Adding drm_dp_mst_atomic_check_bw_limit to verify that
state's bandwidth needs doesn't exceed available bandwidth.
The funtion is called in drm_dp_mst_atomic_check after
drm_dp_mst_atomic_check_topology_state to fully verify that
the proposed topology is supported.
v2: Fixing some typos and indenting
v3: Return correct error enums if no bw space available
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Mikita Lipski <mikita.lipski@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Adding a helper function to be called by
drivers outside of DRM to enable DSC on
the MST ports.
Function is called to recalculate VCPI allocation
if DSC is enabled and raise the DSC flag to enable.
In case of disabling DSC the flag is set to false
and recalculation of VCPI slots is expected to be done
in encoder's atomic_check.
v2: squash separate functions into one and call it per
port
v3: Fix comment typos
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Mikita Lipski <mikita.lipski@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[why]
For DSC case we cannot use topology manager's PBN divider
variable. The default divider does not take FEC into account.
Therefore the driver has to calculate its own divider based
on the link rate and lane count its handling, as it is hw specific.
[how]
Pass pbn_div as an argument, which is used if its more than
zero, otherwise default topology manager's pbn_div will be used.
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Mikita Lipski <mikita.lipski@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add drm_dp_mst_dsc_aux_for_port. To enable DSC, the DSC_ENABLED
register might have to be written on the leaf port's DPCD,
its parent's DPCD, or the MST manager's DPCD. This function
finds the correct aux for the job.
As part of this, add drm_dp_mst_is_virtual_dpcd. Virtual DPCD
is a DP feature new in DP v1.4, which exposes certain DPCD
registers on virtual ports.
v2: Remember to unlock mutex on all paths
v3: Refactor to match coding style and increase brevity
v4: - Check DSC capable MST sink connected directly to the device.
- Check branch's port_parent to be set
Cc: Lyude Paul <lyude@redhat.com>
Reviewed-by: Wenjing Liu <Wenjing.Liu@amd.com>
Signed-off-by: David Francis <David.Francis@amd.com>
Signed-off-by: Mikita Lipski <mikita.lipski@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
As of DP1.4, ENUM_PATH_RESOURCES returns a bit indicating
if FEC can be supported up to that point in the MST network.
The bit is the first byte of the ENUM_PATH_RESOURCES ack reply,
bottom-most bit (refer to section 2.11.9.4 of DP standard,
v1.4)
That value is needed for FEC and DSC support
Store it on drm_dp_mst_port
Reviewed-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: David Francis <David.Francis@amd.com>
Signed-off-by: Mikita Lipski <mikita.lipski@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
With DSC, bpp can be fractional in multiples of 1/16.
Change drm_dp_calc_pbn_mode to reflect this, adding a new
parameter bool dsc. When this parameter is true, treat the
bpp parameter as having units not of bits per pixel, but
1/16 of a bit per pixel
v2: Don't add separate function for this
v3: In the equation divide bpp by 16 as it is expected
not to leave any remainder
v4: Added DSC test parameters for selftest
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: David Francis <David.Francis@amd.com>
Signed-off-by: Mikita Lipski <mikita.lipski@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
UAPI Changes:
- Add support for DMA-BUF HEAPS.
Cross-subsystem Changes:
- mipi dsi definition updates, pulled into drm-intel as well.
- Add lockdep annotations for dma_resv vs mmap_sem and fs_reclaim.
- Remove support for dma-buf kmap/kunmap.
- Constify fb_ops in all fbdev drivers, including drm drivers and drm-core, and media as well.
Core Changes:
- Small cleanups to ttm.
- Fix SCDC definition.
- Assorted cleanups to core.
- Add todo to remove load/unload hooks, and use generic fbdev emulation.
- Assorted documentation updates.
- Use blocking ww lock in ttm fault handler.
- Remove drm_fb_helper_fbdev_setup/teardown.
- Warning fixes with W=1 for atomic.
- Use drm_debug_enabled() instead of drm_debug flag testing in various drivers.
- Fallback to nontiled mode in fbdev emulation when not all tiles are present. (Later on reverted)
- Various kconfig indentation fixes in core and drivers.
- Fix freeing transactions in dp-mst correctly.
- Sean Paul is steping down as core maintainer. :-(
- Add lockdep annotations for atomic locks vs dma-resv.
- Prevent use-after-free for a bad job in drm_scheduler.
- Fill out all block sizes in the P01x and P210 definitions.
- Avoid division by zero in drm/rect, and fix bounds.
- Add drm/rect selftests.
- Add aspect ratio and alternate clocks for HDMI 4k modes.
- Add todo for drm_framebuffer_funcs and fb_create cleanup.
- Drop DRM_AUTH for prime import/export ioctls.
- Clear DP-MST payload id tables downstream when initializating.
- Fix for DSC throughput definition.
- Add extra FEC definitions.
- Fix fake offset in drm_gem_object_funs.mmap.
- Stop using encoder->bridge in core directly
- Handle bridge chaining slightly better.
- Add backlight support to drm/panel, and use it in many panel drivers.
- Increase max number of y420 modes from 128 to 256, as preparation to add the new modes.
Driver Changes:
- Small fixes all over.
- Fix documentation in vkms.
- Fix mmap_sem vs dma_resv in nouveau.
- Small cleanup in komeda.
- Add page flip support in gma500 for psb/cdv.
- Add ddc symlink in the connector sysfs directory for many drivers.
- Add support for analogic an6345, and fix small bugs in it.
- Add atomic modesetting support to ast.
- Fix radeon fault handler VMA race.
- Switch udl to use generic shmem helpers.
- Unconditional vblank handling for mcde.
- Miscellaneous fixes to mcde.
- Tweak debug output from komeda using debugfs.
- Add gamma and color transform support to komeda for DOU-IPS.
- Add support for sony acx424AKP panel.
- Various small cleanups to gma500.
- Use generic fbdev emulation in udl, and replace udl_framebuffer with generic implementation.
- Add support for Logic PD Type 28 panel.
- Use drm_panel_* wrapper functions in exynos/tegra/msm.
- Add devicetree bindings for generic DSI panels.
- Don't include drm_pci.h directly in many drivers.
- Add support for begin/end_cpu_access in udmabuf.
- Stop using drm_get_pci_dev in gma500 and mga200.
- Fixes to UDL damage handling, and use dma_buf_begin/end_cpu_access.
- Add devfreq thermal support to panfrost.
- Fix hotplug with daisy chained monitors by removing VCPI when disabling topology manager.
- meson: Add support for OSD1 plane AFBC commit.
- Stop displaying garbage when toggling ast primary plane on/off.
- More cleanups and fixes to UDL.
- Add D32 suport to komeda.
- Remove globle copy of drm_dev in gma500.
- Add support for Boe Himax8279d MIPI-DSI LCD panel.
- Add support for ingenic JZ4770 panel.
- Small null pointer deference fix in ingenic.
- Remove support for the special tfp420 driver, as there is a generic way to do it.
-----BEGIN PGP SIGNATURE-----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=b15X
-----END PGP SIGNATURE-----
Merge tag 'drm-misc-next-2019-12-16' of git://anongit.freedesktop.org/drm/drm-misc into drm-next
drm-misc-next for v5.6:
UAPI Changes:
- Add support for DMA-BUF HEAPS.
Cross-subsystem Changes:
- mipi dsi definition updates, pulled into drm-intel as well.
- Add lockdep annotations for dma_resv vs mmap_sem and fs_reclaim.
- Remove support for dma-buf kmap/kunmap.
- Constify fb_ops in all fbdev drivers, including drm drivers and drm-core, and media as well.
Core Changes:
- Small cleanups to ttm.
- Fix SCDC definition.
- Assorted cleanups to core.
- Add todo to remove load/unload hooks, and use generic fbdev emulation.
- Assorted documentation updates.
- Use blocking ww lock in ttm fault handler.
- Remove drm_fb_helper_fbdev_setup/teardown.
- Warning fixes with W=1 for atomic.
- Use drm_debug_enabled() instead of drm_debug flag testing in various drivers.
- Fallback to nontiled mode in fbdev emulation when not all tiles are present. (Later on reverted)
- Various kconfig indentation fixes in core and drivers.
- Fix freeing transactions in dp-mst correctly.
- Sean Paul is steping down as core maintainer. :-(
- Add lockdep annotations for atomic locks vs dma-resv.
- Prevent use-after-free for a bad job in drm_scheduler.
- Fill out all block sizes in the P01x and P210 definitions.
- Avoid division by zero in drm/rect, and fix bounds.
- Add drm/rect selftests.
- Add aspect ratio and alternate clocks for HDMI 4k modes.
- Add todo for drm_framebuffer_funcs and fb_create cleanup.
- Drop DRM_AUTH for prime import/export ioctls.
- Clear DP-MST payload id tables downstream when initializating.
- Fix for DSC throughput definition.
- Add extra FEC definitions.
- Fix fake offset in drm_gem_object_funs.mmap.
- Stop using encoder->bridge in core directly
- Handle bridge chaining slightly better.
- Add backlight support to drm/panel, and use it in many panel drivers.
- Increase max number of y420 modes from 128 to 256, as preparation to add the new modes.
Driver Changes:
- Small fixes all over.
- Fix documentation in vkms.
- Fix mmap_sem vs dma_resv in nouveau.
- Small cleanup in komeda.
- Add page flip support in gma500 for psb/cdv.
- Add ddc symlink in the connector sysfs directory for many drivers.
- Add support for analogic an6345, and fix small bugs in it.
- Add atomic modesetting support to ast.
- Fix radeon fault handler VMA race.
- Switch udl to use generic shmem helpers.
- Unconditional vblank handling for mcde.
- Miscellaneous fixes to mcde.
- Tweak debug output from komeda using debugfs.
- Add gamma and color transform support to komeda for DOU-IPS.
- Add support for sony acx424AKP panel.
- Various small cleanups to gma500.
- Use generic fbdev emulation in udl, and replace udl_framebuffer with generic implementation.
- Add support for Logic PD Type 28 panel.
- Use drm_panel_* wrapper functions in exynos/tegra/msm.
- Add devicetree bindings for generic DSI panels.
- Don't include drm_pci.h directly in many drivers.
- Add support for begin/end_cpu_access in udmabuf.
- Stop using drm_get_pci_dev in gma500 and mga200.
- Fixes to UDL damage handling, and use dma_buf_begin/end_cpu_access.
- Add devfreq thermal support to panfrost.
- Fix hotplug with daisy chained monitors by removing VCPI when disabling topology manager.
- meson: Add support for OSD1 plane AFBC commit.
- Stop displaying garbage when toggling ast primary plane on/off.
- More cleanups and fixes to UDL.
- Add D32 suport to komeda.
- Remove globle copy of drm_dev in gma500.
- Add support for Boe Himax8279d MIPI-DSI LCD panel.
- Add support for ingenic JZ4770 panel.
- Small null pointer deference fix in ingenic.
- Remove support for the special tfp420 driver, as there is a generic way to do it.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/ba73535a-9334-5302-2e1f-5208bd7390bd@linux.intel.com
It seems that on certain MST hubs, namely the CableMatters USB-C 2x DP
hub, using the DP_PAYLOAD_ALLOCATE_SET and DP_PAYLOAD_TABLE_UPDATE_STATUS
register ranges to clear any pre-existing payload allocations on the hub isn't
always enough to reset things if the source device has been reset unexpectedly.
Or at least, that's the current running theory. The precise behavior appears to
be that when the source device gets reset unexpectedly, the hub begins reporting
an available_pbn value of 0 for all of its ports. This is a bit inconsistent
with the our theory, since this seems to happen even if previously set PBN
allocations should have resulted in a non-zero available_pbn value. So, it's
possible that something else may be going on here.
Strangely though, sending a CLEAR_PAYLOAD_ID_TABLE broadcast request when
initializing the MST topology seems to bring things into working order and make
available_pbn work again. Since this is a pretty safe solution, let's go ahead
and implement it.
Changes since v1:
* Change indenting on drm_dp_send_clear_payload_id_table() prototype
* Remove some braces in drm_dp_send_clear_payload_id_table()
* Reorganize some variable declarations in drm_dp_send_clear_payload_id_table()
* Don't forget to handle DP_CLEAR_PAYLOAD_ID_TABLE in
drm_dp_sideband_parse_reply()
* Move drm_dp_send_clear_payload_id_table() call into
drm_dp_mst_link_probe_work(), since we can't send sideband messages
while under lock in drm_dp_mst_topology_mgr_set_mst()
* Change commit message
Acked-by: Daniel Vetter <daniel@ffwll.ch>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190829000944.20722-1-lyude@redhat.com
For very subtle mistakes with topology refs, it can be rather difficult
to trace them down with the debugging info that we already have. I had
one such issue recently while trying to implement suspend/resume
reprobing for MST, and ended up coming up with this.
Inspired by Chris Wilson's wakeref tracking for i915, this adds a very
similar feature to the DP MST helpers, which allows for partial tracking
of topology refs for both ports and branch devices. This is a lot less
advanced then wakeref tracking: we merely keep a count of all of the
spots where a topology ref has been grabbed or dropped, then dump out
that history in chronological order when a port or branch device's
topology refcount reaches 0. So far, I've found this incredibly useful
for debugging topology refcount errors.
Since this has the potential to be somewhat slow and loud, we add an
expert kernel config option to enable or disable this feature,
CONFIG_DRM_DEBUG_DP_MST_TOPOLOGY_REFS.
Changes since v1:
* Don't forget to destroy topology_ref_history_lock
Changes since v4:
* Correct order of kref_put()/topology_ref_history_unlock - we can't
unlock the history after kref_put() since the memory might have been
freed by that point
* Don't print message on allocation error failures, the kernel already
does this for us
Changes since v5:
* Get rid of some leftover usages of %px
* Remove a leftover empty return; statement
Cc: Juston Li <juston.li@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Harry Wentland <hwentlan@amd.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191022023641.8026-15-lyude@redhat.com
Finally! For a very long time, our MST helpers have had one very
annoying issue: They don't know how to reprobe the topology state when
coming out of suspend. This means that if a user has a machine connected
to an MST topology and decides to suspend their machine, we lose all
topology changes that happened during that period. That can be a big
problem if the machine was connected to a different topology on the same
port before resuming, as we won't bother reprobing any of the ports and
likely cause the user's monitors not to come back up as expected.
So, we start fixing this by teaching our MST helpers how to reprobe the
link addresses of each connected topology when resuming. As it turns
out, the behavior that we want here is identical to the behavior we want
when initially probing a newly connected MST topology, with a couple of
important differences:
- We need to be more careful about handling the potential races between
events from the MST hub that could change the topology state as we're
performing the link address reprobe
- We need to be more careful about handling unlikely state changes on
ports - such as an input port turning into an output port, something
that would be far more likely to happen in situations like the MST hub
we're connected to being changed while we're suspend
Both of which have been solved by previous commits. That leaves one
requirement:
- We need to prune any MST ports in our in-memory topology state that
were present when suspending, but have not appeared in the post-resume
link address response from their parent branch device
Which we can now handle in this commit by modifying
drm_dp_send_link_address(). We then introduce suspend/resume reprobing
by introducing drm_dp_mst_topology_mgr_invalidate_mstb(), which we call
in drm_dp_mst_topology_mgr_suspend() to traverse the in-memory topology
state to indicate that each mstb needs it's link address resent and PBN
resources reprobed.
On resume, we start back up &mgr->work and have it reprobe the topology
in the same way we would on a hotplug, removing any leftover ports that
no longer appear in the topology state.
Changes since v4:
* Split indenting changes in drm_dp_mst_topology_mgr_resume() into a
separate patch
* Only fire hotplugs when something has actually changed after a link
address probe
* Don't try to change port->connector at all on ports, just throw out
ports that need their connectors removed to make things easier.
Cc: Juston Li <juston.li@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Harry Wentland <hwentlan@amd.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191022023641.8026-14-lyude@redhat.com
This is a complicated one. Essentially, there's currently a problem in the MST
core that hasn't really caused any issues that we're aware of (emphasis on "that
we're aware of"): locking.
When we go through and probe the link addresses and path resources in a
topology, we hold no locks when updating ports with said information. The
members I'm referring to in particular are:
- ldps
- ddps
- mcs
- pdt
- dpcd_rev
- num_sdp_streams
- num_sdp_stream_sinks
- available_pbn
- input
- connector
Now that we're handling UP requests asynchronously and will be using some of
the struct members mentioned above in atomic modesetting in the future for
features such as PBN validation, this is going to become a lot more important.
As well, the next few commits that prepare us for and introduce suspend/resume
reprobing will also need clear locking in order to prevent from additional
racing hilarities that we never could have hit in the past.
So, let's solve this issue by using &mgr->base.lock, the modesetting
lock which currently only protects &mgr->base.state. This works
perfectly because it allows us to avoid blocking connection_mutex
unnecessarily, and we can grab this in connector detection paths since
it's a ww mutex. We start by having drm_dp_mst_handle_up_req() hold this
when updating ports. For drm_dp_mst_handle_link_address_port() things
are a bit more complicated. As I've learned the hard way, we can grab
&mgr->lock.base for everything except for port->connector. See, our
normal driver probing paths end up generating this rather obvious
lockdep chain:
&drm->mode_config.mutex
-> crtc_ww_class_mutex/crtc_ww_class_acquire
-> &connector->mutex
However, sysfs grabs &drm->mode_config.mutex in order to protect itself
from connector state changing under it. Because this entails grabbing
kn->count, e.g. the lock that the kernel provides for protecting sysfs
contexts, we end up grabbing kn->count followed by
&drm->mode_config.mutex. This ends up creating an extremely rude chain:
&kn->count
-> &drm->mode_config.mutex
-> crtc_ww_class_mutex/crtc_ww_class_acquire
-> &connector->mutex
I mean, look at that thing! It's just evil!!! This gross thing ends up
making any calls to drm_connector_register()/drm_connector_unregister()
impossible when holding any kind of modesetting lock. This is annoying
because ideally, we always want to ensure that
drm_dp_mst_port->connector never changes when doing an atomic commit or
check that would affect the atomic topology state so that it can
reliably and easily be used from future DRM DP MST helpers to assist
with tasks such as scanning through the current VCPI allocations and
adding connectors which need to have their allocations updated in
response to a bandwidth change or the like.
Being able to hold &mgr->base.lock throughout the entire link probe
process would have been _great_, since we could prevent userspace from
ever seeing any states in-between individual port changes and as a
result likely end up with a much faster probe and more consistent
results from said probes. But without some rework of how we handle
connector probing in sysfs it's not at all currently possible. In the
future, maybe we can try using the sysfs locks to protect updates to
connector probing state and fix this mess.
So for now, to protect everything other than port->connector under
&mgr->base.lock and ensure that we still have the guarantee that atomic
check/commit contexts will never see port->connector change we use a
silly trick. See: port->connector only needs to change in order to
ensure that input ports (see the MST spec) never have a ghost connector
associated with them. But, there's nothing stopping us from simply
throwing the entire port out and creating a new one in order to maintain
that requirement while still keeping port->connector consistent across
the lifetime of the port in atomic check/commit contexts. For all
intended purposes this works fine, as we validate ports in any contexts
we care about before using them and as such will end up reporting the
connector as disconnected until it's port's destruction finalizes. So,
we just do that in cases where we detect port->input has transitioned
from true->false. We don't need to worry about the other direction,
since a port without a connector isn't visible to userspace and as such
doesn't need to be protected by &mgr->base.lock until we finish
registering a connector for it.
For updating members of drm_dp_mst_port other than port->connector, we
simply grab &mgr->base.lock in drm_dp_mst_link_probe_work() for already
registered ports, update said members and drop the lock before
potentially registering a connector and probing the link address of it's
children.
Finally, we modify drm_dp_mst_detect_port() to take a modesetting lock
acquisition context in order to acquire &mgr->base.lock under
&connection_mutex and convert all it's users over to using the
.detect_ctx probe hooks.
With that, we finally have well defined locking.
Changes since v4:
* Get rid of port->mutex, stop using connection_mutex and just use our own
modesetting lock - mgr->base.lock. Also, add a probe_lock that comes
before this patch.
* Just throw out ports that get changed from an output to an input, and
replace them with new ports. This lets us ensure that modesetting
contexts never see port->connector go from having a connector to being
NULL.
* Write an extremely detailed explanation of what problems this is
trying to fix, since there's a _lot_ of context here and I honestly
forgot some of it myself a couple times.
* Don't grab mgr->lock when reading port->mstb in
drm_dp_mst_handle_link_address_port(). It's not needed.
Cc: Juston Li <juston.li@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Harry Wentland <hwentlan@amd.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191022023641.8026-7-lyude@redhat.com
Currently, MST lacks locking in a lot of places that really should have
some sort of locking. Hotplugging and link address code paths are some
of the offenders here, as there is actually nothing preventing us from
running a link address probe while at the same time handling a
connection status update request - something that's likely always been
possible but never seen in the wild because hotplugging has been broken
for ages now (with the exception of amdgpu, for reasons I don't think
are worth digging into very far).
Note: I'm going to start using the term "in-memory topology layout" here
to refer to drm_dp_mst_port->mstb and drm_dp_mst_branch->ports.
Locking in these places is a little tougher then it looks though.
Generally we protect anything having to do with the in-memory topology
layout under &mgr->lock. But this becomes nearly impossible to do from
the context of link address probes due to the fact that &mgr->lock is
usually grabbed under random various modesetting locks, meaning that
there's no way we can just invert the &mgr->lock order and keep it
locked throughout the whole process of updating the topology.
Luckily there are only two workers which can modify the in-memory
topology layout: drm_dp_mst_up_req_work() and
drm_dp_mst_link_probe_work(), meaning as long as we prevent these two
workers from traveling the topology layout in parallel with the intent
of updating it we don't need to worry about grabbing &mgr->lock in these
workers for reads. We only need to grab &mgr->lock in these workers for
writes, so that readers outside these two workers are still protected
from the topology layout changing beneath them.
So, add the new &mgr->probe_lock and use it in both
drm_dp_mst_link_probe_work() and drm_dp_mst_up_req_work(). Additionally,
add some more detailed explanations for how this locking is intended to
work to drm_dp_mst_port->mstb and drm_dp_mst_branch->ports.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Sean Paul <sean@poorly.run>
Cc: Juston Li <juston.li@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Harry Wentland <hwentlan@amd.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20191022023641.8026-6-lyude@redhat.com
Once upon a time, hotplugging devices on MST branches actually worked in
DRM. Now, it only works in amdgpu (likely because of how it's hotplug
handlers are implemented). On both i915 and nouveau, hotplug
notifications from MST branches are noticed - but trying to respond to
them causes messaging timeouts and causes the whole topology state to go
out of sync with reality, usually resulting in the user needing to
replug the entire topology in hopes that it actually fixes things.
The reason for this is because the way we currently handle UP requests
in MST is completely bogus. drm_dp_mst_handle_up_req() is called from
drm_dp_mst_hpd_irq(), which is usually called from the driver's hotplug
handler. Because we handle sending the hotplug event from this function,
we actually cause the driver's hotplug handler (and in turn, all
sideband transactions) to block on
drm_device->mode_config.connection_mutex. This makes it impossible to
send any sideband messages from the driver's connector probing
functions, resulting in the aforementioned sideband message timeout.
There's even more problems with this beyond breaking hotplugging on MST
branch devices. It also makes it almost impossible to protect
drm_dp_mst_port struct members under a lock because we then have to
worry about dealing with all of the lock dependency issues that ensue.
So, let's finally actually fix this issue by handling the processing of
up requests asyncronously. This way we can send sideband messages from
most contexts without having to deal with getting blocked if we hold
connection_mutex. This also fixes MST branch device hotplugging on i915,
finally!
Cc: Juston Li <juston.li@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Harry Wentland <hwentlan@amd.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191022023641.8026-5-lyude@redhat.com
Since we're going to be implementing suspend/resume reprobing very soon,
we need to make sure we are extra careful to ensure that our locking
actually protects the topology state where we expect it to. Turns out
this isn't the case with drm_dp_port_setup_pdt() and
drm_dp_port_teardown_pdt(), both of which change port->mstb without
grabbing &mgr->lock.
Additionally, since most callers of these functions are just using it to
teardown the port's previous PDT and setup a new one we can simplify
things a bit and combine drm_dp_port_setup_pdt() and
drm_dp_port_teardown_pdt() into a single function:
drm_dp_port_set_pdt(). This function also handles actually ensuring that
we grab the correct locks when we need to modify port->mstb.
Cc: Juston Li <juston.li@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Harry Wentland <hwentlan@amd.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191022023641.8026-4-lyude@redhat.com
When reprobing an MST topology during resume, we have to account for the
fact that while we were suspended it's possible that mstbs may have been
removed from any ports in the topology. Since iterating downwards in the
topology requires that we hold &mgr->lock, destroying MSTBs from this
context would result in attempting to lock &mgr->lock a second time and
deadlocking.
So, fix this by first moving destruction of MSTBs into
destroy_connector_work, then rename destroy_connector_work and friends
to reflect that they now destroy both ports and mstbs.
Note that even though this means that MSTBs will still be accessible for
a short period of time between their removal from the topology and
delayed destruction, we are still protected against referencing a MSTB
with a refcount of 0 since we use kref_get_unless_zero() in most places.
Changes since v1:
* s/destroy_connector_list/destroy_port_list/
s/connector_destroy_lock/delayed_destroy_lock/
s/connector_destroy_work/delayed_destroy_work/
s/drm_dp_finish_destroy_branch_device/drm_dp_delayed_destroy_mstb/
s/drm_dp_finish_destroy_port/drm_dp_delayed_destroy_port/
- danvet
* Use two loops in drm_dp_delayed_destroy_work() - danvet
* Better explain why we need to do this - danvet
* Use cancel_work_sync() instead of flush_work() - flush_work() doesn't
account for work requeing
Cc: Juston Li <juston.li@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Harry Wentland <hwentlan@amd.com>
Cc: Daniel Vetter <daniel@ffwll.ch>
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191022023641.8026-2-lyude@redhat.com
These are most certainly accessed from far more than the mgr work. In
fact, up_req_recv is -only- ever accessed from outside the mgr work.
Cc: Juston Li <juston.li@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Harry Wentland <hwentlan@amd.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Sean Paul <sean@poorly.run>
Link: https://patchwork.freedesktop.org/patch/msgid/20190903204645.25487-19-lyude@redhat.com
Unfortunately the DP MST helpers do not have much in the way of
debugging utilities. So, let's add some!
This adds basic debugging output for down sideband requests that we send
from the driver, so that we can actually discern what's happening when
sideband requests timeout.
Since there wasn't really a good way of testing that any of this worked,
I ended up writing simple selftests that lightly test sideband message
encoding and decoding as well. Enjoy!
Changes since v1:
* Clean up DO_TEST() and sideband_msg_req_encode_decode() - danvet
* Get rid of pr_fmt(), just define a prefix string instead and use
drm_printf()
* Check highest bit of VCPI in drm_dp_decode_sideband_req() - danvet
* Make the switch case order between drm_dp_decode_sideband_req() and
drm_dp_encode_sideband_req() the same - danvet
* Only check DRM_UT_DP - danvet
* Clean up sideband_msg_req_equal() from selftests a bit, and add
comments explaining why we can't just use memcmp - danvet
Cc: Juston Li <juston.li@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Harry Wentland <hwentlan@amd.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190903204645.25487-8-lyude@redhat.com
All available downstream ports - physical and logical - are exposed for
each MST device. They are listed in /dev/, following the same naming
scheme as SST devices by appending an incremental ID.
Although all downstream ports are exposed, only some will work as
expected. Consider the following topology:
+---------+
| ASIC |
+---------+
Conn-0|
|
+----v----+
+----| MST HUB |----+
| +---------+ |
| |
|Port-1 Port-2|
+-----v-----+ +-----v-----+
| MST | | SST |
| Display | | Display |
+-----------+ +-----------+
|Port-1
x
MST Path | MST Device
----------+----------------------------------
sst:0 | MST Hub
mst:0-1 | MST Display
mst:0-1-1 | MST Display's disconnected DP out
mst:0-1-8 | MST Display's internal sink
mst:0-2 | SST Display
On certain MST displays, the upstream physical port will ACK DPCD reads.
However, reads on the local logical port to the internal sink will
*NAK*. i.e. reading mst:0-1 ACKs, but mst:0-1-8 NAKs.
There may also be duplicates. Some displays will return the same GUID
when reading DPCD from both mst:0-1 and mst:0-1-8.
There are some device-dependent behavior as well. The MST hub used
during testing will actually *ACK* read requests on a disconnected
physical port, whereas the MST displays will NAK.
In light of these discrepancies, it's simpler to expose all downstream
ports - both physical and logical - and let the user decide what to use.
v3 changes:
* Change WARN_ON_ONCE -> DRM_ERROR on dpcd read errors
* Docstring and cosmetic fixes
v2 changes:
Moved remote aux device (un)registration to new mst connector late
register and early unregister helpers. Drivers should call these from
their own mst connector function hooks.
This is to solve an issue during driver unload, where mst connector
devices are unregistered before the remote aux devices are. In a setup
where aux devices are created as children of connector devices, the aux
device would be removed too early, and uncleanly. Doing so in
early_unregister solves this issue, as that is called before connector
unregistration.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Leo Li <sunpeng.li@amd.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Harry Wentland <harry.wentland@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190723232808.28128-3-sunpeng.li@amd.com
-----BEGIN PGP SIGNATURE-----
iQFSBAABCAA8FiEEq68RxlopcLEwq+PEeb4+QwBBGIYFAlxqHJYeHHRvcnZhbGRz
QGxpbnV4LWZvdW5kYXRpb24ub3JnAAoJEHm+PkMAQRiGWl8H/jPI4EipzD2GbnjZ
GaFpMBBjcXBaVmoA+Y69so+7BHx1Ql+5GQtqbK0RHJRb9qEPLw3FBhHNjM/N8Sgf
nSrK+GnBZp9s+k/NR/Yf2RacUR3jhz+Q9JEoQd3u9bFUeQyvE8Rf3vgtoBBwFOfz
+t7N1memYVF3asLGWB4e4sP1YVMGfseTQpSPojvM30YWM86Bv+QtSx1AGgHczQIM
kMKealR8ZPelN6JAXgLhQ5opDojBrE4YKB98pwsMDI6abz0Tz2JLFEUTTxsv5XNN
o/Iz+XDoylskEyxN2unNWfHx7Swkvoklog8J/hDg5XlTvipL/WkT66PHBgcGMNvj
BW9GgU8=
=ZizU
-----END PGP SIGNATURE-----
Merge v5.0-rc7 into drm-next
Backmerging for nouveau and imx that needed some fixes for next pulls.
Signed-off-by: Dave Airlie <airlied@redhat.com>
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 v9:
* Add some missing changes that were requested by danvet that I forgot
about after I redid all of the kref stuff:
* Remove unnecessary state changes in intel_dp_mst_atomic_check
* Cleanup atomic check logic for VCPI allocations - all we need to check in
compute_config is whether or not this state disables a CRTC, then free
VCPI based off that
Changes since v8:
* Fix compile errors, whoops!
Changes since v7:
- Don't check for mixed stale/valid VCPI allocations, just rely on
connector registration to stop such erroneous modesets
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.
- 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@redhat.com>
Reviewed-by: Daniel Vetter <daniel@ffwll.ch>
Cc: David Airlie <airlied@redhat.com>
Cc: Jerry Zuo <Jerry.Zuo@amd.com>
Cc: Harry Wentland <harry.wentland@amd.com>
Cc: Juston Li <juston.li@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190111005343.17443-19-lyude@redhat.com
Changes since v6:
- Move EXPORT_SYMBOL() for drm_dp_mst_topology_state_funcs to this
commit
- Document __drm_dp_mst_state_iter_get() and note that it shouldn't be
called directly
Signed-off-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: David Airlie <airlied@redhat.com>
Cc: Jerry Zuo <Jerry.Zuo@amd.com>
Cc: Harry Wentland <harry.wentland@amd.com>
Cc: Juston Li <juston.li@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190111005343.17443-18-lyude@redhat.com
The current way of handling refcounting in the DP MST helpers is really
confusing and probably just plain wrong because it's been hacked up many
times over the years without anyone actually going over the code and
seeing if things could be simplified.
To the best of my understanding, the current scheme works like this:
drm_dp_mst_port and drm_dp_mst_branch both have a single refcount. When
this refcount hits 0 for either of the two, they're removed from the
topology state, but not immediately freed. Both ports and branch devices
will reinitialize their kref once it's hit 0 before actually destroying
themselves. The intended purpose behind this is so that we can avoid
problems like not being able to free a remote payload that might still
be active, due to us having removed all of the port/branch device
structures in memory, as per:
commit 91a25e4631 ("drm/dp/mst: deallocate payload on port destruction")
Which may have worked, but then it caused use-after-free errors. Being
new to MST at the time, I tried fixing it;
commit 263efde31f ("drm/dp/mst: Get validated port ref in drm_dp_update_payload_part1()")
But, that was broken: both drm_dp_mst_port and drm_dp_mst_branch structs
are validated in almost every DP MST helper function. Simply put, this
means we go through the topology and try to see if the given
drm_dp_mst_branch or drm_dp_mst_port is still attached to something
before trying to use it in order to avoid dereferencing freed memory
(something that has happened a LOT in the past with this library).
Because of this it doesn't actually matter whether or not we keep keep
the ports and branches around in memory as that's not enough, because
any function that validates the branches and ports passed to it will
still reject them anyway since they're no longer in the topology
structure. So, use-after-free errors were fixed but payload deallocation
was completely broken.
Two years later, AMD informed me about this issue and I attempted to
come up with a temporary fix, pending a long-overdue cleanup of this
library:
commit c54c7374ff ("drm/dp_mst: Skip validating ports during destruction, just ref")
But then that introduced use-after-free errors, so I quickly reverted
it:
commit 9765635b30 ("Revert "drm/dp_mst: Skip validating ports during destruction, just ref"")
And in the process, learned that there is just no simple fix for this:
the design is just broken. Unfortunately, the usage of these helpers are
quite broken as well. Some drivers like i915 have been smart enough to
avoid accessing any kind of information from MST port structures, but
others like nouveau have assumed, understandably so, that
drm_dp_mst_port structures are normal and can just be accessed at any
time without worrying about use-after-free errors.
After a lot of discussion, me and Daniel Vetter came up with a better
idea to replace all of this.
To summarize, since this is documented far more indepth in the
documentation this patch introduces, we make it so that drm_dp_mst_port
and drm_dp_mst_branch structures have two different classes of
refcounts: topology_kref, and malloc_kref. topology_kref corresponds to
the lifetime of the given drm_dp_mst_port or drm_dp_mst_branch in it's
given topology. Once it hits zero, any associated connectors are removed
and the branch or port can no longer be validated. malloc_kref
corresponds to the lifetime of the memory allocation for the actual
structure, and will always be non-zero so long as the topology_kref is
non-zero. This gives us a way to allow callers to hold onto port and
branch device structures past their topology lifetime, and dramatically
simplifies the lifetimes of both structures. This also finally fixes the
port deallocation problem, properly.
Additionally: since this now means that we can keep ports and branch
devices allocated in memory for however long we need, we no longer need
a significant amount of the port validation that we currently do.
Additionally, there is one last scenario that this fixes, which couldn't
have been fixed properly beforehand:
- CPU1 unrefs port from topology (refcount 1->0)
- CPU2 refs port in topology(refcount 0->1)
Since we now can guarantee memory safety for ports and branches
as-needed, we also can make our main reference counting functions fix
this problem by using kref_get_unless_zero() internally so that topology
refcounts can only ever reach 0 once.
Changes since v4:
* Change the kernel-figure summary for dp-mst/topology-figure-1.dot a
bit - danvet
* Remove figure numbers - danvet
Changes since v3:
* Remove rebase detritus - danvet
* Split out purely style changes into separate patches - hwentlan
Changes since v2:
* Fix commit message - checkpatch
* s/)-1/) - 1/g - checkpatch
Changes since v1:
* Remove forward declarations - danvet
* Move "Branch device and port refcounting" section from documentation
into kernel-doc comments - danvet
* Export internal topology lifetime functions into their own section in
the kernel-docs - danvet
* s/@/&/g for struct references in kernel-docs - danvet
* Drop the "when they are no longer being used" bits from the kernel
docs - danvet
* Modify diagrams to show how the DRM driver interacts with the topology
and payloads - danvet
* Make suggested documentation changes for
drm_dp_mst_topology_get_mstb() and drm_dp_mst_topology_get_port() -
danvet
* Better explain the relationship between malloc refs and topology krefs
in the documentation for drm_dp_mst_topology_get_port() and
drm_dp_mst_topology_get_mstb() - danvet
* Fix "See also" in drm_dp_mst_topology_get_mstb() - danvet
* Rename drm_dp_mst_topology_get_(port|mstb)() ->
drm_dp_mst_topology_try_get_(port|mstb)() and
drm_dp_mst_topology_ref_(port|mstb)() ->
drm_dp_mst_topology_get_(port|mstb)() - danvet
* s/should/must in docs - danvet
* WARN_ON(refcount == 0) in topology_get_(mstb|port) - danvet
* Move kdocs for mstb/port structs inline - danvet
* Split drm_dp_get_last_connected_port_and_mstb() changes into their own
commit - danvet
Signed-off-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Daniel Vetter <daniel@ffwll.ch>
Cc: David Airlie <airlied@redhat.com>
Cc: Jerry Zuo <Jerry.Zuo@amd.com>
Cc: Juston Li <juston.li@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190111005343.17443-7-lyude@redhat.com
Since I've had to fix two cases of drivers not checking the return code
from this function, let's make the compiler complain so this doesn't
come up again in the future.
Changes since v1:
* Remove unneeded __must_check in function declaration - danvet
Signed-off-by: Lyude Paul <lyude@redhat.com>
Cc: Jerry Zuo <Jerry.Zuo@amd.com>
Reviewed-by: Daniel Vetter <daniel@ffwll.ch>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190108211133.32564-4-lyude@redhat.com
When everyone implements it exactly the same way, among all 4
implementations, there's not really a need to overwrite this at all.
Aside: drm_kms_helper_hotplug_event is pretty much core functionality
at this point. Probably should move it there.
Reviewed-by: Lyude Paul <lyude@redhat.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181128221234.15054-1-daniel.vetter@ffwll.ch
There's no reason to track the atomic state three times. Unfortunately,
this is currently what we're doing, and even worse is that there is only
one actually correct state pointer: the one in mst_state->base.state.
mgr->state never seems to be used, along with the one in
mst_state->state.
This confused me for over 4 hours until I realized there was no magic
behind these pointers. So, let's save everyone else from the trouble.
Signed-off-by: Lyude Paul <lyude@redhat.com>.
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20181023231251.16883-3-lyude@redhat.com
The POWER_DOWN_PHY and POWER_UP_PHY sideband message transactions allow
the source to reqest any node in a mst path or a whole path to be
powered down or up. This allows drivers to target a specific sink in the
MST topology, an improvement over just power managing the imediate
downstream device. Secondly, since the request-reply protocol waits for an
ACK, we can be sure that a downstream sink has enough time to respond to a
power up/down request.
v2: Fix memory leak (Lyude)
Cc: Lyude <lyude@redhat.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20170907001458.9399-1-dhinakaran.pandiyan@intel.com
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Make the atomic private object stuff less special by introducing proper
base classes for the object and its state. Drivers can embed these in
their own appropriate objects, after which these things will work
exactly like the plane/crtc/connector states during atomic operations.
v2: Reorder to not depend on drm_dynarray (Daniel)
Cc: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch> #v1
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170712155102.26276-3-ville.syrjala@linux.intel.com
drm_dp_atomic_find_vcpi_slots() should be called from ->atomic_check() to
check there are sufficient vcpi slots for a mode and to add that to the
state. This should be followed by a call to drm_dp_mst_allocate_vcpi()
in ->atomic_commit() to initialize a struct vcpi for the port.
drm_dp_atomic_release_vcpi_slots() should be called from
->atomic_check() to release a port's vcpi slot allocation from the
state.
Drivers that do not make use of this atomic helper are expected to call
drm_dp_find_vcpi_slots() instead before calling
drm_dp_mst_allocate_vcpi().
v3: drm_dp_atomic_release_vcpi_slots() now needs to know how many slots
to release as we may not have a valid reference to port.
v2:
Added checks for verifying the port reference is valid
Moved get_mst_topology_state() into the helpers (Daniel)
Changed find_vcpi_slots() to not depend on current allocation
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Archit Taneja <architt@codeaurora.org>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Harry Wentland <Harry.wentland@amd.com>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1492753893-3748-4-git-send-email-dhinakaran.pandiyan@intel.com
Link bandwidth is shared between multiple display streams in DP MST
configurations. The DP MST topology manager structure maintains the
shared link bandwidth for a primary link directly connected to the GPU. For
atomic modesetting drivers, checking if there is sufficient link bandwidth
for a mode needs to be done during the atomic_check phase to avoid failed
modesets. Let's encapsulate the available link bw information in a
private state structure so that bw can be allocated and released atomically
for each of the ports sharing the primary link.
v3: WARN_ON() if connection_mutex is not held (Archit)
v2: Included kernel doc, moved state initialization and switched to
kmemdup() for allocation (Daniel)
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Archit Taneja <architt@codeaurora.org>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Harry Wentland <Harry.wentland@amd.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1492753893-3748-3-git-send-email-dhinakaran.pandiyan@intel.com
drm_dp_mst_allocate_vcpi() apart from setting up the vcpi structure,
also finds if there are enough slots available. This check is a duplicate
of that implemented in drm_dp_mst_find_vcpi_slots(). Let's move this check
out and reuse the existing drm_dp_mst_find_vcpi_slots() function to check
if there are enough vcpi slots before allocating them.
This brings the check to one place. Additionally drivers that will use MST
state tracking for atomic modesets can use the atomic version of
find_vcpi_slots() and reuse drm_dp_mst_allocate_vcpi()
Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1489648231-30700-4-git-send-email-dhinakaran.pandiyan@intel.com
The avail_slots member in the MST topology manager is never updated to
reflect the available vcpi slots. The check is effectively against
total slots, 63. So, let's make that check obvious and remove
avail_slots. While at it, make debug messages more descriptive.
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Archit Taneja <architt@codeaurora.org>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Harry Wentland <Harry.wentland@amd.com>
Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1489648231-30700-3-git-send-email-dhinakaran.pandiyan@intel.com
The total vcpi time slots is always 63 and does not depend on the link BW,
remove total_slots from MST topology manager struct. The next change is to
remove total_pbn which is hardcoded to 2560. The total PBN that the
topology manager allocates from depends on the link rate and is not a
constant. So, fix this by removing the total_pbn member itself.
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Archit Taneja <architt@codeaurora.org>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Harry Wentland <Harry.wentland@amd.com>
Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1489648231-30700-2-git-send-email-dhinakaran.pandiyan@intel.com
I just learned that &struct_name.member_name works and looks pretty
even. It doesn't (yet) link to the member directly though, which would
be really good for big structures or vfunc tables (where the
per-member kerneldoc tends to be long).
Also some minor drive-by polish where it makes sense, I read a lot
of docs ...
v2: Comments from Gustavo.
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Rewiewed-by: Gustavo Padovan <gustavo.padovan@collabora.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170125062657.19270-2-daniel.vetter@ffwll.ch
struct drm_dp_mst_topology_mgr currently stores a pointer to struct dev.
Changing this to instead hold a pointer to drm_device is more useful as it
gives access to DRM structures. This also makes it consistent with other
DRM structures like drm_crtc, drm_connector etc.
Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1485301777-3465-2-git-send-email-dhinakaran.pandiyan@intel.com
Previous implementation does not handle case below: boot up one MST branch
to DP connector of ASIC. After boot up, hot plug 2nd MST branch to DP output
of 1st MST, GUID is not created for 2nd MST branch. When downstream port of
2nd MST branch send upstream request, it fails because 2nd MST branch GUID
is not available.
New Implementation: only create GUID for MST branch and save it within Branch.
Signed-off-by: Hersen Wu <hersenxs.wu@amd.com>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Cc: stable@vger.kernel.org
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Since your main drm-next pull isn't out of the door yet I figured I might
as well flush out drm-misc instead of delaying for 4.6. It's really just
random stuff all over, biggest thing probably connector_mask tracking from
Maarten.
* tag 'topic/drm-misc-2016-01-17' of git://anongit.freedesktop.org/drm-intel: (24 commits)
drm/fb_cma_helper: Remove implicit call to disable_unused_functions
drm/sysfs: use kobj_to_dev()
drm/i915: Init power domains early in driver load
drm: Do not set connector->encoder in drivers
apple-gmux: Add initial documentation
drm: move MODULE_PARM_DESC to other file
drm/edid: index CEA/HDMI mode tables using the VIC
drm/atomic: Remove drm_atomic_connectors_for_crtc.
drm/i915: Update connector_mask during readout, v2.
drm: Remove opencoded drm_gem_object_release_handle()
drm: Do not set outparam on error during GEM handle allocation
drm/docs: more leftovers from the big vtable documentation pile
drm/atomic-helper: Reject legacy flips on a disabled pipe
drm/atomic: add connector mask to drm_crtc_state.
drm/tegra: Use __drm_atomic_helper_reset_connector for subclassing connector state, v2.
drm/atomic: Add __drm_atomic_helper_connector_reset, v2.
drm/i915: Set connector_state->connector using the helper.
drm: Use a normal idr allocation for the obj->name
drm: Only bump object-reference count when adding first handle
drm: Balance error path for GEM handle allocation
...
We should always send reply for UP request in order
to make downstream device clean-up resources appropriately.
Issue was that reply for UP request was sent only once.
Acked-by: Dave Airlie <airlied@gmail.com>
Signed-off-by: Mykola Lysenko <Mykola.Lysenko@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
The drm_dp_mst_topology_cbs structures are never modified, so declare them
as const.
Done with the help of Coccinelle.
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This adds code to initialise the SDP streams
for a sink in the simplest ordering.
I've no idea how you'd want to control the
ordering at this level, so don't bother
until someone comes up with a use case.
Reviewed-by: Ander Conselvan de Oliveira <conselvan2@gmail.com>
Signed-off-by: Libin Yang <libin.yang@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1449036584-105393-1-git-send-email-libin.yang@linux.intel.com
This zeroes the msg so no random stack data ends up getting
sent, it also limits the function to not accepting > 4
i2c msgs.
Cc: stable@vger.kernel.org
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
In order to cache the EDID properly for tiled displays, we
need to retrieve it before we register the connector with
userspace, otherwise userspace can call get resources
and try and get the edid before we've even cached it.
This fixes some problems when hotplugging mst monitors,
with X/mutter running. As mutter seems to get 0 modes
for one of the monitors in the tile.
v2: fix warning in radeon
handle tile setting in cached path rather than
get edid path.
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: stable@vger.kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
I've only seen this once, and I failed to capture the
lockdep backtrace, but I did some investigations.
If we are calling into the MST layer from EDID probing,
we have the mode_config mutex held, if during that EDID
probing, the MST hub goes away, then we can get a deadlock
where the connector destruction function in the driver
tries to retake the mode config mutex.
This offloads connector destruction to a workqueue,
and avoid the subsequenct lock ordering issue.
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: stable@vger.kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
radeon requires this to get the slots for later filling
out a table on every transition.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
