Update msm_gem and msm_drv to comply with latest 5.15 kernel.
Modify dma_buf_vmap() and dma-buf's vmap callback to use
struct dma_buf_map. Rename dma_resv_get_excl_rcu to _unlocked.
Remove deprecated GEM and PRIME callbacks.
Change-Id: Ifdfc7f872c988d8455f465e7d17dfbcf212bb5f6
Signed-off-by: Bruce Hoo <bingchua@codeaurora.org>
Signed-off-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Reorder registration of various display drivers in the order of
dependency.
Change-Id: Idfa0616d3133f3b03c713e3c15a4fd3956ec2594
Signed-off-by: Shashank Babu Chinta Venkata <sbchin@codeaurora.org>
Drm file is not set to NULL after freeing it from drm
release. This can result in use-after-free issues in
some scenarios. Add a mutex lock and other proper null
checks to prevent such issues.
Change-Id: Ic35b0a76166b0f47a354b1737e6f4c3ac1437ed4
Signed-off-by: Krishna Manikandan <mkrishn@codeaurora.org>
Early wakeup ioctls from perf HAL driver can access display HW
registers and come as a sideband to atomic commits. Since the
atomic validate checks are not part of this code path it's
possible for HW access to occur during a trusted UI session.
Prevent this whenever the HW is not under this VM's ownership.
Also, move the VM ownership check for the register/deregister
event ioctl to cover both the CRTC and connector events since
new connector events are being added which can access HW.
Change-Id: I39660ae60e7e8f8a405e819c43ec42fbac3f492a
Signed-off-by: Steve Cohen <cohens@codeaurora.org>
This change avoids upstream drm issuing drm_atomic_commit in
drm_fb_release which is leading to artifacts on screen or
atomic_check failures due to atomically unstaging each fb
from plane_state and committing remaining planes on hardware.
a) This patch moves the state operations for setting crtc to
connector state to a helper api.
b) This patch clears any dim_layers present in the crtc_state
as part of null commit.
c) This patch removes any framebuffers attached to a drm_file
in msm_preclose whose refcount is not managed by composer kill
inadvertently and issues null flush to hardware in such cases.
d) This patch handles msm_preclose as part of msm_release
operation since legacy feature is not supported
for msm_driver.
Change-Id: Ib2068d74d4b23b73b7c84544858c9f6bb6adfa67
Signed-off-by: Jayaprakash Madisetty <jmadiset@codeaurora.org>
Signed-off-by: Samantha Tran <samtran@codeaurora.org>
This change reverts commit ae2dceb0b6 ("disp: msm:
sde: remove all preclose logic"). This change also
modifies the force disable plane logic since
__drm_atomic_helper_disable_plane is not exposed to
drivers.
Change-Id: I89e19dead9ade724798952b1934b45b5663e1a42
Signed-off-by: Jayaprakash Madisetty <jmadiset@codeaurora.org>
Signed-off-by: Samantha Tran <samtran@codeaurora.org>
In dumping display registers, physical address will be appended after
each block name. This is to support register compare between kernel
and UEFI.
Change-Id: Ic20d3e2bd4c95aa7c71c4b646a149f7e83ad731a
Signed-off-by: Yu Wu <zwy@codeaurora.org>
This change moves mmrm enable flag to the display platform config files.
This allows to selectively enable mmrm in the display driver,
so other platforms not requiring mmrm in their build can compile
with their own config files.
Change-Id: I02415c6a22252dfc483a3da03b623351811ffc01
Signed-off-by: Christina Oliveira <coliveir@codeaurora.org>
Add changes to fix the null dereference in
drm_atomic_get_property caused by connector->state
being NULL. This change allows the drm_mode_config_reset
operation to happen before drm_dev_register to avoid this.
In current scenario, connector->state->crtc is being
accessed due to call to drm_mode_getconnector ioctl with
the drm_mode_config_reset operation pending.
Change-Id: I374d9485819fad85100d1837f4ae22fc2a3ccc40
Signed-off-by: Jayaprakash <jmadiset@codeaurora.org>
Now that we don't need struct_mutex in the free path, we can get rid of
the asynchronous free all together.
Change-Id: I82406450e3a5d0d49d3fb753c621f55e8f4af088
Signed-off-by: Rob Clark <robdclark@chromium.org>
Git-commit: c951a9b284b907604759628d273901064c60d09f
Git-repo: git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
Signed-off-by: Samantha Tran <samtran@codeaurora.org>
Rather than relying on the big dev->struct_mutex hammer, introduce a
more specific lock for protecting the bo lists.
Change-Id: I4c876a1c3ae51ff62372703a99a8daff0c4a7950
Signed-off-by: Rob Clark <robdclark@chromium.org>
Git-commit: d984457b31c4c53d2af374d5e78b3eb64debd483
Git-repo: git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
[samtran@codeaurora.org: avoid changes related to debugfs and shrinker]
Signed-off-by: Samantha Tran <samtran@codeaurora.org>
We use a llist and a worker to delay the object cleanup. This avoids
taking mmap_sem and struct mutex in the wrong order when calling
drm_gem_object_put-unlocked() from drm_gem_mmap().
Fixes lockdep problem with copy_from_user() in msm_ioctl_gem_submit().
Change-Id: Idfe54ae8108158b69f3835f26991642d1e21f8ee
Signed-off-by: Kristian H. Kristensen <hoegsberg@chromium.org>
Signed-off-by: Rob Clark <robdclark@chromium.org>
Git-commit: 48e7f18392c66f9b69ebac11c54f1a2e033ced54
Git-repo: git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
[samtran@codeaurora.org: resolve trivial merge conflict]
Signed-off-by: Samantha Tran <samtran@codeaurora.org>
This change adds mmrm api to display driver.
Change-Id: I51f310f8c762edc823930c05c001735056a2cbcd
Signed-off-by: Christina Oliveira <coliveir@codeaurora.org>
Fix dlkm compilation errors that are due to the use of -Werror
flags used by the build system.
Change-Id: I5e1e9bc63c1361d73e4930aab123212717872ecb
Signed-off-by: Tatenda Chipeperekwa <tatendac@codeaurora.org>
Upstream now prefers to utilize vblank enable and disable
through crtc hooks rather than drm driver. Remove
calls from kms vblank enable/disbale.
Change-Id: Ic6911838f14f93a0d277b7bdc2c1968270ec25a2
Signed-off-by: Samantha Tran <samtran@codeaurora.org>
Signed-off-by: Nilaan Gunabalachandran <ngunabal@codeaurora.org>
Trusted-vm uses the cont-splash path to initialize the software
states by reading the hardware registers and populating the
SDE & DRM software states. But drm mode cannot be determined by
register reads, so rely on user-mode to get the mode which comes
as part of the first atomic_commit check phase which handles the
transition. The primary-vm user-mode passes the current mode to
trusted-vm through qrtr which is in-turn passed to driver to
set that particular mode.
Change-Id: I4fb86b40b8a0583c8edb539b0a51dcd33de731ff
Signed-off-by: Veera Sundaram Sankaran <veeras@codeaurora.org>
There is a race between disable commit swap-state on display
thread and vblank work on event thread which may skip applying
the vblank vote if the encoder_mask has already been cleared.
To avoid the race, use a cached encoder_mask that is gauranteed
to be valid between the vblank_on and vblank_off calls.
Also, vblank queue work is initializing a variable that's never
used and msm_disable_outputs has 2 instances of pointers to old
CRTC state. Remove this dead code.
Change-Id: I5e4a482b8f067e272a2aef5afa08cc0e1ab89434
Signed-off-by: Steve Cohen <cohens@codeaurora.org>
Due to power-domain, if rotator driver is registered early,
probe might get deferred several times and get stuck
indefinitely. So, this change adds driver registration
after genpd init to handle such cases.
Change-Id: I8dcb640d0ab0cdf0818cbce1b1fb460c28d8b9e7
Signed-off-by: Yashwanth <yvulapu@codeaurora.org>
Add event to event_list after msm_register_event is successful to avoid
use-after-free vulnerability.
Change-Id: I34fb39c99051978cbab64a852851964691a5ea9e
Signed-off-by: Ping Li <pingli@codeaurora.org>
This reverts commit 91fc51777c67ddc92d0ba1fcddf6cd149f7a0560.
The original fix missed to add the client to list in cases when count is
larger than zero and client doesn't need to be deleted from list if
msm_register_event fails as it was not added to list yet.
Revert this change to upload the correct fix.
Change-Id: I00923d65387753b05b6b03efec109c019826f5d5
Signed-off-by: Ping Li <pingli@codeaurora.org>
Add event to event_list after msm_register_event is successful to avoid
use-after-free vulnerability.
Change-Id: I144ae82c657c1e2cf16608c0e8768b12a7d27974
Signed-off-by: Ping Li <pingli@codeaurora.org>
Display clocks and IRQs are disabled during idle state
on command mode for power saving, and will be enabled
when a new frame commits to display driver. But enable
display clocks and IRQs will cause some latency.
So add a new SDE custom IOCTL for user-space to early wake
up display before first frame commits to kernel.
Change-Id: I6ca0188d321c4964f29c46e588b64d06b9634c59
Signed-off-by: Lei Chen <chenlei@codeaurora.org>
Besides SDE, other subdrivers may be interested in participating in
the VM switch. This change provides framework for display dependent
drivers like DSI, DP and RSCC to register for various VM switch
event hooks.
The following hooks to provided through msm_vm_ops:
post_hw_acquire: invoked before the first frame push after gaining
HW access.
pre_hw_release: invoked after the last frame commit before releasing
the HW.
check: check with vm clients for their readiness for HW
releasing.
Change-Id: I616db04e979f78f76f6f97ee3b068dd348339ab6
Signed-off-by: Jeykumar Sankaran <jsanka@codeaurora.org>
If a SIG_TERM/SIG_KILL or other signal is sent to the final drm
client, the driver will be force-closed. A -ERESTARTSYS error is
occasionally seen from the interruptible wait in
msm_atomic_commit when this occurs, and causes the lastclose
cleanup to fail if any crtc is busy at that point. To prevent
this, wait for any pending crtcs to complete before calling
the lastclose cleanup commits.
Change-Id: Ib6a5e55a4b737213756cb9ed8364d5c34ab47c16
Signed-off-by: Steve Cohen <cohens@codeaurora.org>
Vblank refcount can reach out of sync with below case
1. event_thread triggers the vblank_enable
2. commit_thread triggers the modeset
2.a modeset resets the vblank refcount with mode_set
3. event_thread triggers the vblank_disable
Event 2.a resets the vblank refcount and vblank disable
request after 2.a is going to fail. This can be fixed
by avoiding concurrency between mode_set call and vblank
request.
Change-Id: Ibb810ec90e81d63feee443f1c37dd736d5cfac0d
Signed-off-by: Dhaval Patel <pdhaval@codeaurora.org>
Allocate DSI/LUTDMA buffers from VRAM when IOMMU is not
available. Add checks in msm_gem to avoid few operations
when aspace is not available due to no IOMMU. Parse the
VRAM size from device tree, when available.
Change-Id: Iedf5749b71c2e772ac5434048520a34705c54b45
Signed-off-by: Veera Sundaram Sankaran <veeras@codeaurora.org>
Currently DP driver determines if a mode is DSC capable
based on a DTSI entry and the required number of DSC
to support it. This approach does not scale when there
is an overlap in DSC requirement between DSI displays
and external DP display, thus causing one of the display to
report modes that cannot be supported.
This change compares the resources reserved for DP driver
calculated at initialization time and the currently available
ones to determine the correct number of resources that DP driver
can use. It also adds DSC and topology filtering logic and moves
DSC hardware specific from DP driver to SDE driver.
Change-Id: I8e601de33422b7c6d786826f7bfe152c4af8a6b5
Signed-off-by: Amine Najahi <anajahi@codeaurora.org>
This defensive check is now required since the allocation
of driver private data has moved from bind to probe phase.
There is a chance that device shutdown can be called before
bind phase and this can cause invalid memory access.
Change-Id: I2d471a4aff93ae03291e19f6f902f1e5c28dc50f
Signed-off-by: Abhijit Kulkarni <kabhijit@codeaurora.org>
If vblank work is queued to event thread, during modeset,
there might be a case where vblank is enabled in between
resulting in vblank ref count mismatch. This change queues
vblank request to disp thread during modeset.
Change-Id: I98b87b4209e1f7f535980eef22a50df17c346c23
Signed-off-by: Yashwanth <yvulapu@codeaurora.org>
msm_drv should be probe deferred until rsc driver
probe is done. OF_POPULATED flag only checks if platform
driver is registered and hence additional condition to
check if driver is attached to this device, is needed.
Change-Id: I070965cc8d3ba3984032ad3704ec86066680eef0
Signed-off-by: Abhijit Kulkarni <kabhijit@codeaurora.org>
Make msm_drm into single module and all child driver
registers and unregisters are handled from parent's
register and unregister respectively.
Change-Id: I017513d1de3b6b25dd5543d7fa7741c0bac1740d
Signed-off-by: Shashank Babu Chinta Venkata <sbchin@codeaurora.org>
Signed-off-by: Narendra Muppalla <NarendraM@codeaurora.org>
Signed-off-by: Chandan Uddaraju <chandanu@codeaurora.org>
sync_state driver disables any resources that
don't have any votes after driver probe is completed.
Move MDSS resource votes to probe so that any resources
that are needed for continuous splash are intact until
the bind of all the components is complete.
Change-Id: I0056bf1ec56bcd6a1b620a81143d4b49d7ea2921
Signed-off-by: Chandan Uddaraju <chandanu@codeaurora.org>
Msm_drv device node has dependency on sde_rsc device
node for power resource enable if sde_rsc device
device is enabled. This change moves the msm_drv device
to probe defer state during component addition if
sde_rsc device is not probed yet. It also removes
the master_drm device node access from sde_rsc device.
Change-Id: Ibb7457279b3376f45e82e5c9373aabd84a37ed36
Signed-off-by: Dhaval Patel <pdhaval@codeaurora.org>
Add support to allow creation of debugfs node only if
CONFIG_DEBUG_FS is enabled.
Change-Id: Iaeaf51b3654c9458cf8131a9756e6b905007c4ae
Signed-off-by: Krishna Manikandan <mkrishn@codeaurora.org>
Signed-off-by: Samantha Tran <samtran@codeaurora.org>
This change addresses out of range and null checks in
sde driver.
Change-Id: I905d795edf6715aa990dd7bbaf061734e95ddea6
Signed-off-by: Narendra Muppalla <NarendraM@codeaurora.org>
This change queues vblank work to event thread rather than
the display thread. This allows vblank events to be processed
without potentially being delayed by crtc commit being processed.
Change-Id: I9cbf9b3de646716e4698af9f338a7140346b65fd
Signed-off-by: Samantha Tran <samtran@codeaurora.org>
There is potential byte alignment problem on 32bits OS which will
cause incorrect histogram blob id getting from the user space.
This change fixes byte alignment problem on 32bits OS by keeping
the struct members in one complete struct as the user space using.
In previous code start address of 'info' would align to multiple of 8,
so there would be 4 bytes padding between member 'event' and 'info'
on 32 bits OS. It is affected by member 'base'. But user space just
uses struct drm_msm_event_resp only, not affected by member 'base'.
Driver code intends to send drm_msm_event_resp, but actually send
unpacked members.
Change-Id: I5410f5b1626387707a57cbcef9c651aa11914d9d
Signed-off-by: Renchao Liu <rencliu@codeaurora.org>
This change moves the msm_driver power resource initialization from
bind time to probe time. This keeps the resource vote on until all the
devices are bound. This is required since the regulator and clock
sync_state driver will remove the proxy votes as soon as msm_driver
has probed.
Change-Id: Icb0e59e4ff0290ef0c1bd3914d6fdbf99bf5d9fa
Signed-off-by: Abhijit Kulkarni <kabhijit@codeaurora.org>
With the introduction of sync state drivers, msm_drm probe can be
moved back to module level from late init. It is now guaranteed
that the dependencies are probed before the actual driver is.
Change-Id: Ib7c3e0e3c3d32b37b7672b6119bcf5371783659d
Signed-off-by: Satya Rama Aditya Pinapala <psraditya30@codeaurora.org>
