commit 75569c182e upstream.
Otherwise, the SQ may skip some of the register writes, or shader waves may
be allocated where we don't expect them, so that as a result we don't actually
reset all of the register SRAMs. This can lead to spurious ECC errors later on
if a shader uses an uninitialized register.
Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 5e1df40f40 upstream.
Currently we see sporadic timeouts during CDCLK changing both on BXT and
GLK as reported by the Bugzilla: ticket. It's easy to reproduce this by
changing the frequency in a tight loop after blanking the display. The
upper bound for the completion time is 800us based on my tests, so
increase it from the current 500us to 2ms; with that I couldn't trigger
the problem either on BXT or GLK.
Note that timeouts happened during both the change notification and the
voltage level setting PCODE request. (For the latter one BSpec doesn't
require us to wait for completion before further HW programming.)
This issue is similar to
commit 2c7d0602c8 ("drm/i915/gen9: Fix PCODE polling during CDCLK
change notification")
but there the PCODE request does complete (as shown by the mbox
busy flag), only the reply we get from PCODE indicates a failure.
So there we keep resending the request until a success reply, here we
just have to increase the timeout for the one PCODE request we send.
v2:
- s/snb_pcode_request/sandybridge_pcode_write_timeout/ (Ville)
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: <stable@vger.kernel.org> # v4.9
Acked-by: Chris Wilson <chris@chris-wilson.co.uk> (v1)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103326
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180130142939.17983-1-imre.deak@intel.com
(cherry picked from commit e76019a819)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
(Rebased for v4.9 stable tree due to upstream intel_cdclk.c, cdclk_state and pcu_lock change)
Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 7eb2c4dd54 upstream.
LSPCON adapters in low-power state may ignore the first I2C write during
TMDS output buffer enabling, resulting in a blank screen even with an
otherwise enabled pipe. Fix this by reading back and validating the
written value a few times.
The problem was noticed on GLK machines with an onboard LSPCON adapter
after entering/exiting DC5 power state. Doing an I2C read of the adapter
ID as the first transaction - instead of the I2C write to enable the
TMDS buffers - returns the correct value. Based on this we assume that
the transaction itself is sent properly, it's only the adapter that is
not ready for some reason to accept this first write after waking from
low-power state. In my case the second I2C write attempt always
succeeded.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105854
Cc: Clinton Taylor <clinton.a.taylor@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: stable@vger.kernel.org
Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180416155309.11100-1-imre.deak@intel.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 85e290d92b upstream.
Two years ago I tried an AMD Radeon E8860 embedded GPU with the drm driver.
The dmesg output included driver warnings about an invalid PCIe lane width.
Tracking the problem back led to si_set_pcie_lane_width_in_smc().
The calculation of the lane widths via ATOM_PPLIB_PCIE_LINK_WIDTH_MASK and
ATOM_PPLIB_PCIE_LINK_WIDTH_SHIFT macros did not increment the resulting
value, per the comment in pptable.h ("lanes - 1"), and per usage elsewhere.
Applying the increment silenced the warnings.
The code has not changed since, so either my analysis was incorrect or the
bug has gone unnoticed. Hence submitting this as an RFC.
Acked-by: Christian König <christian.koenig@amd.com>
Acked-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Paul Parsons <lost.distance@yahoo.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 5f9e93fed4 upstream.
Calling request_irq() followed by disable_irq() is usually a bad idea,
specially if the interrupt can be pending, and you're not yet in a
position to handle it.
This is exactly what happens on my kevin system when rebooting in a
second kernel using kexec: Some interrupt is left pending from
the previous kernel, and we take it too early, before disable_irq()
could do anything.
Let's clear the pending interrupts as we initialize the HW, and move
the interrupt request after that point. This ensures that we're in
a sane state when the interrupt is requested.
Cc: stable@vger.kernel.org
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
[adapted to recent rockchip-drm changes]
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20180220130120.5254-2-marc.zyngier@arm.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit a20ee0b1f8 upstream.
If these bos are evicted and are in the validated list
things blow up, so do not put them in there. Notably,
that tries to add the bo to the LRU twice, which results
in a BUG_ON in ttm_bo.c.
While for the bo_list an alternative would be to not allow
always valid bos in there, that does not work for the user
fence.
v2: Fixed whitespace issue pointed out by checkpatch.pl
Signed-off-by: Bas Nieuwenhuizen <basni@chromium.org>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit c02216acf4 upstream.
In randconfig testing, we sometimes get this warning:
drivers/gpu/drm/radeon/radeon_object.c: In function 'radeon_bo_create':
drivers/gpu/drm/radeon/radeon_object.c:242:2: error: #warning Please enable CONFIG_MTRR and CONFIG_X86_PAT for better performance thanks to write-combining [-Werror=cpp]
#warning Please enable CONFIG_MTRR and CONFIG_X86_PAT for better performance \
This is rather annoying since almost all other code produces no build-time
output unless we have found a real bug. We already fixed this in the
amdgpu driver in commit 31bb90f1cd ("drm/amdgpu: shut up #warning for
compile testing") by adding a CONFIG_COMPILE_TEST check last year and
agreed to do the same here, but both Michel and I then forgot about it
until I came across the issue again now.
For stable kernels, as this is one of very few remaining randconfig
warnings in 4.14.
Cc: stable@vger.kernel.org
Link: https://patchwork.kernel.org/patch/9550009/
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit d0b1d259a4 ]
If one 'drm_gem_handle_create()' fails, we leak somes handles and some
memory.
In order to fix it:
- move the 'free(bo_state)' at the end of the function so that it is also
called in the eror handling path. This has the side effect to also try
to free it if the first 'kcalloc' fails. This is harmless.
- add a new label, err_delete_handle, in order to delete already
allocated handles in error handling path
- remove the now useless 'err' label
The way the code is now written will also delete the handles if the
'copy_to_user()' call fails.
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Reviewed-by: Eric Anholt <eric@anholt.net>
Link: http://patchwork.freedesktop.org/patch/msgid/20170512123803.1886-1-christophe.jaillet@wanadoo.fr
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit cc8dd7661c ]
omap_gem uses page alignment for buffer stride. The related calculations
are a bit off, though, as byte stride of 4096 gets aligned to 8192,
instead of 4096.
This patch changes the code to use DIV_ROUND_UP(), which fixes those
calculations and makes them more readable.
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 49baeb0747 ]
The generic connectors such as hdmi-connector doesn't have any driver in,
so if they are added to the component list, we will be waiting forever for
a non-existing driver to probe.
Add a list of the connectors we want to ignore when building our component
list.
Reviewed-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 90dd57de4a ]
Amongst its other duties, msm_gem_new_impl adds the newly created
GEM object to the shared inactive list which may also be actively
modifiying the list during submission. All the paths to modify
the list are protected by the mutex except for the one through
msm_gem_import which can end up causing list corruption.
Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
[add extra WARN_ON(!mutex_is_locked(&dev->struct_mutex))]
Signed-off-by: Rob Clark <robdclark@gmail.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 73a88250b7 upstream.
When validating legacy surfaces, the backup bo might be destroyed at
surface validate time. However, the kms resource validation code may have
the bo reserved, so we will destroy a locked mutex. While there shouldn't
be any other users of that mutex when it is destroyed, it causes a lock
leak and thus throws a lockdep error.
Fix this by having the kms resource validation code hold a reference to
the bo while we have it reserved. We do this by introducing a validation
context which might come in handy when the kms code is extended to validate
multiple resources or buffers.
Cc: <stable@vger.kernel.org>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit b7ea6b286c ]
Check the status of the DMM engine after it is reported that the
transaction was completed as in rare cases the engine might not reached a
working state.
The wait_status() will print information in case the DMM is not reached the
expected state and the dmm_txn_commit() will return with an error code to
make sure that we are not continuing with a broken setup.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit c1b9d4c75c ]
The vendor name was "toppoly" but other panels and the vendor list
have defined it as "tpo". So let's fix it in driver and bindings.
We keep the old definition in parallel to stay compatible with
potential older DTB setup.
Signed-off-by: H. Nikolaus Schaller <hns@goldelico.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 4e5ca2d930 ]
Add a check to ensure iowrite64 is only used if it is atomic.
It was decided in [1] that the tilcdc driver should not be using an
atomic operation (so it was left out of this patchset). However, it turns
out that through the drm code, a nonatomic header is actually included:
include/linux/io-64-nonatomic-lo-hi.h
is included from include/drm/drm_os_linux.h:9:0,
from include/drm/drmP.h:74,
from include/drm/drm_modeset_helper.h:26,
from include/drm/drm_atomic_helper.h:33,
from drivers/gpu/drm/tilcdc/tilcdc_crtc.c:19:
And thus, without this change, this patchset would inadvertantly
change the behaviour of the tilcdc driver.
[1] lkml.kernel.org/r/CAK8P3a2HhO_zCnsTzq7hmWSz5La5Thu19FWZpun16iMnyyNreQ@mail.gmail.com
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Cc: Jyri Sarha <jsarha@ti.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Tomi Valkeinen <tomi.valkeinen@ti.com>
Cc: David Airlie <airlied@linux.ie>
Signed-off-by: Jyri Sarha <jsarha@ti.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 62e3a3e342 ]
get_pages doesn't keep a reference of the pages allocated
when it fails later in the code path. This can lead to
a memory leak. Keep reference of the allocated pages so
that it can be freed when msm_gem_free_object gets called
later during cleanup.
Signed-off-by: Prakash Kamliya <pkamliya@codeaurora.org>
Signed-off-by: Sharat Masetty <smasetty@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 60b95d7095 ]
So far we only allowed for 1 retry and just failed the query
- and thereby high precision vblank timestamping - if we did
not get a reasonable result, as such a failure wasn't considered
all too horrible. There are a few NVidia gpu models out there which
may need a bit more than 1 retry to get a successful query result
under some conditions.
Since Linux 4.4 the update code for vblank counter and timestamp
in drm_update_vblank_count() changed so that the implementation
assumes that high precision vblank timestamping of a kms driver
either consistently succeeds or consistently fails for a given
video mode and encoder/connector combo. Iow. switching from success
to fail or vice versa on a modeset or connector change is ok, but
spurious temporary failure for a given setup can confuse the core
code and potentially cause bad miscounting of vblanks and confusion
or hangs in userspace clients which rely on vblank stuff, e.g.,
desktop compositors.
Therefore change the max retry count to a larger number - more than
any gpu so far is known to need to succeed, but still low enough
so that these queries which do also happen in vblank interrupt are
still fast enough to be not disastrously long if something would
go badly wrong with them.
As such sporadic retries only happen seldom even on affected gpu's,
this could mean a vblank irq could take a few dozen microseconds
longer every few hours of uptime -- better than a desktop compositor
randomly hanging every couple of hours or days of uptime in a hard
to reproduce manner.
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 5108d76840 ]
Kobject created using kobject_create_and_add() can be freed using
kobject_put() when there is no referenece any more. However,
kobject memory allocated with kzalloc() has to set up a release
callback in order to free it when the counter decreases to 0.
Otherwise it causes memory leak.
Signed-off-by: Yong Zhao <yong.zhao@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 6dd47cfd03 ]
The driver currently handles vblank events only when updating planes on
a CRTC. The atomic update API however allows requesting an event when
disabling a CRTC. This currently leads to event objects being leaked in
the kernel and to events not being sent out. Fix it.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit aa74f0687c ]
1. When unsetting a mode, num_connector should be set to zero
2. The pixel_format field needs to be initialized as newer DRM internal
functions checks this field
3. Take the drm_modeset_lock_all() because vmw_fb_kms_detach() can
change current mode
Signed-off-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit c2c139cf43 ]
Fixes a potential race condition in amdgpu that looks as follows:
Task 1: attempt ttm_bo_init, but ttm_bo_validate fails
Task 1: add BO to global list anyway
Task 2: grabs hold of the BO, waits on its reservation lock
Task 1: releases its reference of the BO; never gives up the
reservation lock
The patch "drm/amdgpu: fix a potential deadlock in
amdgpu_bo_create_restricted()" attempts to fix that by releasing
the reservation lock in amdgpu code; unfortunately, it introduces
a use-after-free when this race _doesn't_ happen.
This patch should fix the race properly by never adding the BO
to the global list in the first place.
Cc: zhoucm1 <david1.zhou@amd.com>
Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Tested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 608b205069 ]
On vblank instant-off systems, we can get into a situation where the cost
of enabling and disabling the vblank IRQ around a drmWaitVblank query
dominates. And with the advent of even deeper hardware sleep state,
touching registers becomes ever more expensive. However, we know that if
the user wants the current vblank counter, they are also very likely to
immediately queue a vblank wait and so we can keep the interrupt around
and only turn it off if we have no further vblank requests queued within
the interrupt interval.
After vblank event delivery, this patch adds a shadow of one vblank where
the interrupt is kept alive for the user to query and queue another vblank
event. Similarly, if the user is using blocking drmWaitVblanks, the
interrupt will be disabled on the IRQ following the wait completion.
However, if the user is simply querying the current vblank counter and
timestamp, the interrupt will be disabled after every IRQ and the user
will enabled it again on the first query following the IRQ.
v2: Mario Kleiner -
After testing this, one more thing that would make sense is to move
the disable block at the end of drm_handle_vblank() instead of at the
top.
Turns out that if high precision timestaming is disabled or doesn't
work for some reason (as can be simulated by echo 0 >
/sys/module/drm/parameters/timestamp_precision_usec), then with your
delayed disable code at its current place, the vblank counter won't
increment anymore at all for instant queries, ie. with your other
"instant query" patches. Clients which repeatedly query the counter
and wait for it to progress will simply hang, spinning in an endless
query loop. There's that comment in vblank_disable_and_save:
"* Skip this step if there isn't any high precision timestamp
* available. In that case we can't account for this and just
* hope for the best.
*/
With the disable happening after leading edge of vblank (== hw counter
increment already happened) but before the vblank counter/timestamp
handling in drm_handle_vblank, that step is needed to keep the counter
progressing, so skipping it is bad.
Now without high precision timestamping support, a kms driver must not
set dev->vblank_disable_immediate = true, as this would cause problems
for clients, so this shouldn't matter, but it would be good to still
make this robust against a future kms driver which might have
unreliable high precision timestamping, e.g., high precision
timestamping that intermittently doesn't work.
v3: Patch before coffee needs extra coffee.
Testcase: igt/kms_vblank
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Michel Dänzer <michel@daenzer.net>
Cc: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Cc: Dave Airlie <airlied@redhat.com>,
Cc: Mario Kleiner <mario.kleiner.de@gmail.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/20170315204027.20160-1-chris@chris-wilson.co.uk
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 4c7f16d14a ]
The TCON driver calls sun4i_tcon_init_regmap and sun4i_tcon_init_clocks
in its bind function. The former creates a regmap and writes to several
register to clear its configuration to a known default. The latter
initializes various clocks. This includes enabling the bus clock for
register access and creating the dotclock.
In order for the first step's writes to work, the bus clock must be
enabled which is done in the second step. but the dotclock's ops use
the regmap created in the first step.
Rearrange the function calls such that the clocks are initialized before
the regmap, and split out the dot clock creation to after the regmap is
initialized.
Fixes: 9026e0d122 ("drm: Add Allwinner A10 Display Engine support")
Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 7544860733 ]
The way drm_of_find_possible_crtcs works is it tries to match the
remote-endpoint of the given node's various endpoints to all the
crtc's .port field. Thus we need to set drm_crtc.port to the output
port node of the underlying TCON.
Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 9d56defb44 ]
The master bind function calls numerous drm functions which initialize
underlying structures. It also tries to bind the various components
of the display pipeline, some of which may add additional drm objects.
This patch adds proper cleanup functions in the error path of the
master bind function.
This requires the patch "drm/sun4i: Move drm_mode_config_cleanup call
to main driver", which splits out drm_mode_config_cleanup from
sun4i_framebuffer_free so we can call it separately.
Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 15734feff2 upstream.
radeon's ->runtime_suspend hook calls drm_kms_helper_poll_disable(),
which waits for the output poll worker to finish if it's running.
The output poll worker meanwhile calls pm_runtime_get_sync() in
radeon's ->detect hooks, which waits for the ongoing suspend to finish,
causing a deadlock.
Fix by not acquiring a runtime PM ref if the ->detect hooks are called
in the output poll worker's context. This is safe because the poll
worker is only enabled while runtime active and we know that
->runtime_suspend waits for it to finish.
Stack trace for posterity:
INFO: task kworker/0:3:31847 blocked for more than 120 seconds
Workqueue: events output_poll_execute [drm_kms_helper]
Call Trace:
schedule+0x3c/0x90
rpm_resume+0x1e2/0x690
__pm_runtime_resume+0x3f/0x60
radeon_lvds_detect+0x39/0xf0 [radeon]
output_poll_execute+0xda/0x1e0 [drm_kms_helper]
process_one_work+0x14b/0x440
worker_thread+0x48/0x4a0
INFO: task kworker/2:0:10493 blocked for more than 120 seconds.
Workqueue: pm pm_runtime_work
Call Trace:
schedule+0x3c/0x90
schedule_timeout+0x1b3/0x240
wait_for_common+0xc2/0x180
wait_for_completion+0x1d/0x20
flush_work+0xfc/0x1a0
__cancel_work_timer+0xa5/0x1d0
cancel_delayed_work_sync+0x13/0x20
drm_kms_helper_poll_disable+0x1f/0x30 [drm_kms_helper]
radeon_pmops_runtime_suspend+0x3d/0xa0 [radeon]
pci_pm_runtime_suspend+0x61/0x1a0
vga_switcheroo_runtime_suspend+0x21/0x70
__rpm_callback+0x32/0x70
rpm_callback+0x24/0x80
rpm_suspend+0x12b/0x640
pm_runtime_work+0x6f/0xb0
process_one_work+0x14b/0x440
worker_thread+0x48/0x4a0
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94147
Fixes: 10ebc0bc09 ("drm/radeon: add runtime PM support (v2)")
Cc: stable@vger.kernel.org # v3.13+: 27d4ee0307: workqueue: Allow retrieval of current task's work struct
Cc: stable@vger.kernel.org # v3.13+: 25c058ccaf: drm: Allow determining if current task is output poll worker
Cc: Ismo Toijala <ismo.toijala@gmail.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: Dave Airlie <airlied@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Link: https://patchwork.freedesktop.org/patch/msgid/64ea02c44f91dda19bc563902b97bbc699040392.1518338789.git.lukas@wunner.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit d61a5c1063 upstream.
nouveau's ->runtime_suspend hook calls drm_kms_helper_poll_disable(),
which waits for the output poll worker to finish if it's running.
The output poll worker meanwhile calls pm_runtime_get_sync() in
nouveau_connector_detect() which waits for the ongoing suspend to finish,
causing a deadlock.
Fix by not acquiring a runtime PM ref if nouveau_connector_detect() is
called in the output poll worker's context. This is safe because
the poll worker is only enabled while runtime active and we know that
->runtime_suspend waits for it to finish.
Other contexts calling nouveau_connector_detect() do require a runtime
PM ref, these comprise:
status_store() drm sysfs interface
->fill_modes drm callback
drm_fb_helper_probe_connector_modes()
drm_mode_getconnector()
nouveau_connector_hotplug()
nouveau_display_hpd_work()
nv17_tv_set_property()
Stack trace for posterity:
INFO: task kworker/0:1:58 blocked for more than 120 seconds.
Workqueue: events output_poll_execute [drm_kms_helper]
Call Trace:
schedule+0x28/0x80
rpm_resume+0x107/0x6e0
__pm_runtime_resume+0x47/0x70
nouveau_connector_detect+0x7e/0x4a0 [nouveau]
nouveau_connector_detect_lvds+0x132/0x180 [nouveau]
drm_helper_probe_detect_ctx+0x85/0xd0 [drm_kms_helper]
output_poll_execute+0x11e/0x1c0 [drm_kms_helper]
process_one_work+0x184/0x380
worker_thread+0x2e/0x390
INFO: task kworker/0:2:252 blocked for more than 120 seconds.
Workqueue: pm pm_runtime_work
Call Trace:
schedule+0x28/0x80
schedule_timeout+0x1e3/0x370
wait_for_completion+0x123/0x190
flush_work+0x142/0x1c0
nouveau_pmops_runtime_suspend+0x7e/0xd0 [nouveau]
pci_pm_runtime_suspend+0x5c/0x180
vga_switcheroo_runtime_suspend+0x1e/0xa0
__rpm_callback+0xc1/0x200
rpm_callback+0x1f/0x70
rpm_suspend+0x13c/0x640
pm_runtime_work+0x6e/0x90
process_one_work+0x184/0x380
worker_thread+0x2e/0x390
Bugzilla: https://bugs.archlinux.org/task/53497
Bugzilla: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=870523
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70388#c33
Fixes: 5addcf0a5f ("nouveau: add runtime PM support (v0.9)")
Cc: stable@vger.kernel.org # v3.12+: 27d4ee0307: workqueue: Allow retrieval of current task's work struct
Cc: stable@vger.kernel.org # v3.12+: 25c058ccaf: drm: Allow determining if current task is output poll worker
Cc: Ben Skeggs <bskeggs@redhat.com>
Cc: Dave Airlie <airlied@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Link: https://patchwork.freedesktop.org/patch/msgid/b7d2cbb609a80f59ccabfdf479b9d5907c603ea1.1518338789.git.lukas@wunner.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Sean Paul