drm: debugfs: make drm_debugfs_create_files() never fail
As stated before, there is no need to care if a debugfs function
succeeds or not, and no code logic in the kernel should ever change
based on a debugfs function return value, so make
drm_debugfs_create_files() never fail. If it encounters an
odd/rare/impossible error (i.e. out of memory, or a duplicate debugfs
filename to be created), just keep on moving as if nothing improper had
happened.
Justin Swartz [Wed, 22 May 2019 22:46:29 +0000 (22:46 +0000)]
drm/rockchip: dw_hdmi: add basic rk3228 support
Like the RK3328, RK322x SoCs offer a Synopsis DesignWare HDMI transmitter
and an Innosilicon HDMI PHY.
Add a new dw_hdmi_plat_data struct, rk3228_hdmi_drv_data.
Assign a set of mostly generic rk3228_hdmi_phy_ops functions.
Add dw_hdmi_rk3228_setup_hpd() to enable the HDMI HPD and DDC lines.
Greg Hackmann [Thu, 13 Jun 2019 22:34:07 +0000 (15:34 -0700)]
dma-buf: add DMA_BUF_SET_NAME ioctls
This patch adds complimentary DMA_BUF_SET_NAME ioctls, which lets
userspace processes attach a free-form name to each buffer.
This information can be extremely helpful for tracking and accounting
shared buffers. For example, on Android, we know what each buffer will
be used for at allocation time: GL, multimedia, camera, etc. The
userspace allocator can use DMA_BUF_SET_NAME to associate that
information with the buffer, so we can later give developers a
breakdown of how much memory they're allocating for graphics, camera,
etc.
Greg Hackmann [Thu, 13 Jun 2019 22:34:06 +0000 (15:34 -0700)]
dma-buf: give each buffer a full-fledged inode
By traversing /proc/*/fd and /proc/*/map_files, processes with CAP_ADMIN
can get a lot of fine-grained data about how shmem buffers are shared
among processes. stat(2) on each entry gives the caller a unique
ID (st_ino), the buffer's size (st_size), and even the number of pages
currently charged to the buffer (st_blocks / 512).
In contrast, all dma-bufs share the same anonymous inode. So while we
can count how many dma-buf fds or mappings a process has, we can't get
the size of the backing buffers or tell if two entries point to the same
dma-buf. On systems with debugfs, we can get a per-buffer breakdown of
size and reference count, but can't tell which processes are actually
holding the references to each buffer.
Replace the singleton inode with full-fledged inodes allocated by
alloc_anon_inode(). This involves creating and mounting a
mini-pseudo-filesystem for dma-buf, following the example in fs/aio.c.
Sean Paul [Tue, 11 Jun 2019 16:08:20 +0000 (12:08 -0400)]
drm/rockchip: Use dirtyfb helper
Instead of flushing all vops every time we get a dirtyfb call, use the
damage helper to kick off an atomic commit. Even though we don't use
damage clips, the helper commit will force us through the normal
psr_inhibit_get/put sequence.
Changes in v3:
- Added to the set
Changes in v4:
- None
Changes in v5:
- None
Sean Paul [Wed, 12 Jun 2019 14:50:19 +0000 (10:50 -0400)]
drm: Add helpers to kick off self refresh mode in drivers
This patch adds a new drm helper library to help drivers implement
self refresh. Drivers choosing to use it will register crtcs and
will receive callbacks when it's time to enter or exit self refresh
mode.
In its current form, it has a timer which will trigger after a
driver-specified amount of inactivity. When the timer triggers, the
helpers will submit a new atomic commit to shut the refreshing pipe
off. On the next atomic commit, the drm core will revert the self
refresh state and bring everything back up to be actively driven.
From the driver's perspective, this works like a regular disable/enable
cycle. The driver need only check the 'self_refresh_active' state in
crtc_state. It should initiate self refresh mode on the panel and enter
an off or low-power state.
Changes in v2:
- s/psr/self_refresh/ (Daniel)
- integrated the psr exit into the commit that wakes it up (Jose/Daniel)
- made the psr state per-crtc (Jose/Daniel)
Changes in v3:
- Remove the self_refresh_(active|changed) from connector state (Daniel)
- Simplify loop in drm_self_refresh_helper_alter_state (Daniel)
- Improve self_refresh_aware comment (Daniel)
- s/self_refresh_state/self_refresh_data/ (Daniel)
Changes in v4:
- Move docbook location below panel (Daniel)
- Improve docbook with references and more detailed explanation (Daniel)
- Instead of register/unregister, use init/cleanup (Daniel)
Changes in v5:
- Resolved conflict in drm_atomic_helper.c #include block
- Resolved conflict in rst with HDCP helper docs
Changes in v6:
- Fix include ordering, clean up forward declarations (Sam)
Sean Paul [Tue, 11 Jun 2019 16:08:18 +0000 (12:08 -0400)]
drm: Convert connector_helper_funcs->atomic_check to accept drm_atomic_state
Everyone who implements connector_helper_funcs->atomic_check reaches
into the connector state to get the atomic state. Instead of continuing
this pattern, change the callback signature to just give atomic state
and let the driver determine what it does and does not need from it.
Eventually all atomic functions should do this, but that's just too much
busy work for me.
Changes in v3:
- Added to the set
Changes in v4:
- None
Changes in v5:
- intel_digital_connector_atomic_check declaration moved to i915_atomic.h
Sean Paul [Tue, 11 Jun 2019 16:08:17 +0000 (12:08 -0400)]
drm: Add atomic variants for bridge enable/disable
This patch adds atomic variants for all of
pre_enable/enable/disable/post_disable bridge functions. These will be
called from the appropriate atomic helper functions. If the bridge
driver doesn't implement the atomic version of the function, we will
fall back to the vanilla implementation.
Note that some drivers call drm_bridge_disable directly, and these cases
are not covered. It's up to the driver to decide whether to implement
both atomic_disable and disable, or if it's not necessary.
Changes in v3:
- Added to the patchset
Changes in v4:
- Fix up docbook references (Daniel)
Changes in v5:
- None
Add functions to the atomic core to retrieve the old and new connectors
associated with an encoder in a drm_atomic_state. This is useful for
encoders and bridges that need to access the connector, for instance for
the drm_display_info.
The CRTC associated with the encoder can also be retrieved through the
connector state, and from it, the old and new CRTC states.
Changed in v4:
- Added to the set
Changed in v5:
- Fix up docbook (Daniel & Laurent)
Changed in v6:
- Updated commit subject (Sam)
Sean Paul [Tue, 11 Jun 2019 20:49:53 +0000 (16:49 -0400)]
drm: Add atomic variants of enable/disable to encoder helper funcs
This patch adds atomic_enable and atomic_disable callbacks to the
encoder helpers. This will allow encoders to make informed decisions in
their start-up/shutdown based on the committed state.
Aside from the new hooks, this patch also introduces the new signature
for .atomic_* functions going forward. Instead of passing object state
(well, encoders don't have atomic state, but let's ignore that), we pass
the entire atomic state so the driver can inspect more than what's
happening locally.
This is particularly important for the upcoming self refresh helpers.
Changes in v3:
- Added patch to the set
Changes in v4:
- Move atomic_disable above prepare (Daniel)
- Add breadcrumb to .enable() docbook (Daniel)
Changes in v5:
- None
Changes in v6:
- Tweak kerneldoc some more (Sam)
sti: no need to check return value of debugfs_create functions
When calling debugfs functions, there is no need to ever check the
return value. The function can work or not, but the code logic should
never do something different based on this.
drm: no need to check return value of debugfs_create functions
When calling debugfs functions, there is no need to ever check the
return value. The function can work or not, but the code logic should
never do something different based on this.
Because there is no need to check these functions, a number of local
functions can be made to return void to simplify things as nothing can
fail.
Linus Walleij [Sun, 9 Jun 2019 22:32:54 +0000 (00:32 +0200)]
drm/bridge: analogix-anx78xx: Drop of_gpio.h include
This include is only used for some gpio drivers and consumers
that look up GPIO numbers directly from the device tree.
This driver does not use it and only needs <linux/gpio/consumer.h>.
Delete the unused include.
panel: rocktech: no need to check return value of debugfs_create functions
When calling debugfs functions, there is no need to ever check the
return value. The function can work or not, but the code logic should
never do something different based on this.
vga_switcheroo: no need to check return value of debugfs_create functions
When calling debugfs functions, there is no need to ever check the
return value. The function can work or not, but the code logic should
never do something different based on this.
Also, because there is no need to save the file dentry, remove the local
variable and just recursively delete the whole directory when shutting
down.
The cursor handling in mgag200 is complicated to understand. It touches a
number of different BOs, but doesn't really use all of them.
Rewriting the cursor update reduces the amount of cursor state. There are
two BOs for double-buffered HW updates. The source BO updates the one that
is currently not displayed and then switches buffers. Explicit BO locking
has been removed from the code. BOs are simply pinned and unpinned in video
RAM.
drm/mgag200: Pin framebuffer BO during dirty update
Another explicit lock operation of a GEM VRAM BO is located in mgag200's
framebuffer update code. Instead of locking the BO, we pin it to wherever
it is.
drm/ast: Pin and map cursor source BO during update
The ast driver used to lock the cursor source BO during updates. Locking
should be done internally by the BO's implementation, so we pin it instead
to system memory. The mapping information is also stored in the BO. No
need to have an extra argument to the kmap function.
The ast driver's data structures store unused or uncecessary cursor
state. Most of the cursor state is already stored elsewhere and can
be retrieved when necessary. Remove the obsolete fields and adapt
users accordingly.
drm/gem-vram: Support pinning buffers to current location
Pinning a buffer prevents it from being moved to a different memory
location. For some operations, such as buffer updates, it is not
important where the buffer is located. Setting the pin function's
pl_flag argument to 0 will pin the buffer to whereever it is stored.
I copied the kerneldoc for encoder_funcs.atomic_enable from encoder_funcs.enable
in a recent patch [1]. Sam rightly pointed out in the review that "for symmetry
with" text is awkward [2]. So here's a patch to fix up the source of the awkward
language.
Yannick Fertré [Mon, 27 May 2019 10:21:39 +0000 (12:21 +0200)]
drm/stm: dsi: add power on/off phy ops
These new physical operations are helpful to power_on/off the dsi
wrapper. If the dsi wrapper is powered in video mode, the display
controller (ltdc) register access will hang when DSI fifos are full.
On Rockchip rk3288-based Chromebooks when you do a suspend/resume
cycle:
1. You lose the ability to detect an HDMI device being plugged in.
2. If you're using the i2c bus built in to dw_hdmi then it stops
working.
Let's call the core dw-hdmi's suspend/resume functions to restore
things.
NOTE: in downstream Chrome OS (based on kernel 3.14) we used the
"late/early" versions of suspend/resume because we found that the VOP
was sometimes resuming before dw_hdmi and then calling into us before
we were fully resumed. For now I have gone back to the normal
suspend/resume because I can't reproduce the problems.
On Rockchip rk3288-based Chromebooks when you do a suspend/resume
cycle:
1. You lose the ability to detect an HDMI device being plugged in.
2. If you're using the i2c bus built in to dw_hdmi then it stops
working.
Let's add a hook to the core dw-hdmi driver so that we can call it in
dw_hdmi-rockchip in the next commit.
NOTE: the exact set of steps I've done here in resume come from
looking at the normal dw_hdmi init sequence in upstream Linux plus the
sequence that we did in downstream Chrome OS 3.14. Testing show that
it seems to work, but if an extra step is needed or something here is
not needed we could improve it.
As part of this change we'll refactor the hardware init bits of
dw-hdmi to happen all in one function and all at the same time. Since
we need to init the interrupt mutes before we request the IRQ, this
means moving the hardware init earlier in the function, but there
should be no problems with that. Also as part of this we now
unconditionally init the "i2c" parts of dw-hdmi, but again that ought
to be fine.
Noralf Trønnes [Sat, 8 Jun 2019 15:26:53 +0000 (17:26 +0200)]
drm/fb-helper: Remove drm_fb_helper_connector
All drivers add all their connectors so there's no need to keep around an
array of available connectors. Instead we just put the useable (not
writeback) connectors in a temporary array using
drm_client_for_each_connector_iter() everytime we probe the outputs.
Other places where it's necessary to look at the connectors, we just
iterate over them using the same iterator function.
Rename functions which signature is changed since they will be moved to
drm_client in a later patch.
David Riley [Mon, 10 Jun 2019 21:18:10 +0000 (14:18 -0700)]
drm/virtio: Add memory barriers for capset cache.
After data is copied to the cache entry, atomic_set is used indicate
that the data is the entry is valid without appropriate memory barriers.
Similarly the read side was missing the corresponding memory barriers.
Sam Ravnborg [Sat, 8 Jun 2019 08:02:39 +0000 (10:02 +0200)]
drm/radeon: prepare header files for drmP.h removal
While removing drmP.h from all .c files the list of
header files are also sorted alphabetically.
To allow this adjust the header files to pull in
the dependencies they needed to allow this.
Sam Ravnborg [Sat, 8 Jun 2019 08:02:37 +0000 (10:02 +0200)]
drm/radeon: drop dependency on drm_os_linux.h
The drm_os_linux.h header file is deprecated.
Drop all uses of symbols from this file.
All macros are just opencoded.
A few adjustments to include files was required as the obsolete
drm_os_linux.h included these headers.
Sam Ravnborg [Sat, 8 Jun 2019 08:02:36 +0000 (10:02 +0200)]
drm: drm_debugfs.h self-contained
While removing drmP.h from drm/radeon a few files ended
up including drm_debugfs.h as the first file.
This failed build due to missing dependencies in drm_debugfs.h.
Sam Ravnborg [Sat, 8 Jun 2019 08:02:35 +0000 (10:02 +0200)]
drm: drm_crtc.h self-contained
While removing drmP.h from drm/radeon a few files ended
up including drm_crtc.h as the first file.
This failed build due to a missing dependency in drm_crtc.h.
Douglas Anderson [Mon, 10 Jun 2019 17:52:34 +0000 (10:52 -0700)]
drm/bridge/synopsys: dw-hdmi: Fix unwedge crash when no pinctrl entries
In commit 50f9495efe30 ("drm/bridge/synopsys: dw-hdmi: Add "unwedge"
for ddc bus") I stupidly used IS_ERR() to check for whether we have an
"unwedge" pinctrl state even though on most flows through the driver
the unwedge state will just be NULL.
Fix it so that we consistently use NULL for no unwedge state.
Chris Wilson [Sun, 9 Jun 2019 11:00:02 +0000 (12:00 +0100)]
dma-fence: Signal all callbacks from dma_fence_release()
This is an illegal scenario, to free the fence whilst there are pending
callbacks. Currently, we emit a WARN and then cast aside the callbacks
leaving them dangling. Alternatively, we could set an error on the fence
and then signal fence so that any dependency chains from the fence can
be tidied up, and if they care they can check for the error.
Prepare for moving drm_fb_helper modesetting code to drm_client.
drm_client will be linked to drm.ko, so move
__drm_atomic_helper_disable_plane() and __drm_atomic_helper_set_config()
out of drm_kms_helper.ko.
While at it, fix two checkpatch complaints:
- WARNING: Block comments use a trailing */ on a separate line
- CHECK: Alignment should match open parenthesis
Colin Ian King [Mon, 3 Jun 2019 14:21:02 +0000 (15:21 +0100)]
drm/bridge: sii902x: fix comparision of u32 with less than zero
The less than check for the variable num_lanes is always going to be
false because the variable is a u32. Fix this by making num_lanes an
int and also make loop index i an int too.
Yannick Fertré [Mon, 3 Jun 2019 08:32:02 +0000 (10:32 +0200)]
drm/stm: support runtime power management
This patch enables runtime power management (runtime PM) support for
the display controller. pm_runtime_enable() and pm_runtime_disable()
are added during ltdc load and unload respectively.
pm_runtime_get_sync() and pm_runtime_put_sync() are added for ltdc
register access.
we fixed the vblank counter to give accurate results outside of
drm_crtc_handle_vblank, which fixed bugs around vblank timestamps
being off-by-one and causing the vblank counter to jump when it
shouldn't.
The trouble is that this completely broke crc generation. Shayenne and
Rodrigo tracked this down to the vblank timestamp going backwards in
time somehow. Which then resulted in an underflow in drm_vblank.c
code, which resulted in all kinds of things breaking really badly.
The reason for this is that once we've called drm_crtc_handle_vblank
and the hrtimer isn't forwarded yet, we're returning a vblank
timestamp in the past. This race is really hard to hit since it's
small, except when you enable crc generation: In that case there's a
call to drm_crtc_accurate_vblank right in-betwen, so we're guaranteed
to hit the bug.
The fix is to roll the hrtimer forward _before_ we do the vblank
processing (which has a side-effect of incrementing the vblank
counter), and we always subtract one frame from the hrtimer - since
now it's always one frame in the future.
To make sure we don't hit this again also add a WARN_ON checking for
whether our timestamp is somehow moving into the past, which is never
should.
This also aligns more with how real hw works:
1. first all registers are updated with the new timestamp/vblank
counter values.
2. then an interrupt is generated
3. kernel interrupt handler eventually fires.
So doing this aligns vkms closer with what drm_vblank.c expects.
Document this also in a comment.
Rodrigo Siqueira [Fri, 15 Mar 2019 16:11:42 +0000 (13:11 -0300)]
drm/vkms: Remove useless call to drm_connector_register/unregister()
The function vkms_output_init() is invoked during the module
initialization, and it handles the creation/configuration of the vkms
essential elements (e.g., connectors, encoder, etc). Among the
initializations, this function tries to initialize a connector and
register it by calling drm_connector_register(). However, inside the
drm_connector_register(), at the beginning of this function there is the
following validation:
if (!connector->dev->registered)
return 0;
In this sense, invoke drm_connector_register() after initializing the
connector has no effect because the register field is false. The
connector register happens when drm_dev_register() is invoked; the same
issue exists with drm_connector_unregister(). Therefore, this commit
removes the unnecessary call to drm_connector_register() and
drm_connector_unregister().
Changes since v2:
* Remove unnecessary call to drm_connector_unregister()
* Remove unused label
Ville Syrjälä [Wed, 29 May 2019 11:02:04 +0000 (14:02 +0300)]
drm/edid: Ignore "DFP 1.x" bit for EDID 1.2 and earlier
From VESA EDID implementation guide v1.0:
"For EDID version 1 revision 2 or earlier data structures when offset 14h
bit 7 is set to one, the value of bits 6-0 are undefined, and therefore
cannot be interpreted to mean anything."
And since EDID 1.4 redefines that bit let's consult it only for
EDID 1.3.
Hans de Goede [Fri, 24 May 2019 12:57:59 +0000 (14:57 +0200)]
drm: panel-orientation-quirks: Add quirk for GPD MicroPC
GPD has done it again, make a nice device (good), use way too generic
DMI strings (bad) and use a portrait screen rotated 90 degrees (ugly).
Because of the too generic DMI strings this entry is also doing bios-date
matching, so the gpd_micropc data struct may very well need to be updated
with some extra bios-dates in the future.
Hans de Goede [Fri, 24 May 2019 12:57:58 +0000 (14:57 +0200)]
drm: panel-orientation-quirks: Add quirk for GPD pocket2
GPD has done it again, make a nice device (good), use way too generic
DMI strings (bad) and use a portrait screen rotated 90 degrees (ugly).
Because of the too generic DMI strings this entry is also doing bios-date
matching, so the gpd_pocket2 data struct may very well need to be updated
with some extra bios-dates in the future.
Changes in v2:
-Add one more known BIOS date to the list of BIOS dates
drm: Ignore drm_gem_vram_mm_funcs in generated documentation
The documentation tools interpret drm_gem_vram_mm_funcs as function and
there appears to be no way of inline-documenting constants. This results
in a warning by 'make htmldocs'. For now, removing drm_gem_vram_mm_funcs
from generated documentation works around the issue.
David Riley [Wed, 5 Jun 2019 23:44:22 +0000 (16:44 -0700)]
drm/virtio: Fix cache entry creation race.
virtio_gpu_cmd_get_capset would check for the existence of an entry
under lock. If it was not found, it would unlock and call
virtio_gpu_cmd_get_capset to create a new entry. The new entry would
be added it to the list without checking if it was added by another
task during the period where the lock was not held resulting in
duplicate entries.
Compounding this issue, virtio_gpu_cmd_capset_cb would stop iterating
after find the first matching entry. Multiple callbacks would modify
the first entry, but any subsequent entries and their associated waiters
would eventually timeout since they don't become valid, also wasting
memory along the way.
David Riley [Wed, 5 Jun 2019 23:44:20 +0000 (16:44 -0700)]
drm/virtio: Ensure cached capset entries are valid before copying.
virtio_gpu_get_caps_ioctl could return success with invalid data if a
second caller to the function occurred after the entry was created in
virtio_gpu_cmd_get_capset but prior to the virtio_gpu_cmd_capset_cb
callback being called. This could leak contents of memory as well
since the caps_cache allocation is done without zeroing.
drm/bridge/synopsys: dw-hdmi: Add "unwedge" for ddc bus
See the PhD thesis in the comments in this patch for details, but to
summarize this adds a hacky "unwedge" feature to the dw_hdmi i2c bus to
workaround what appears to be a hardware errata. This relies on a
pinctrl entry to help change around muxing to perform the unwedge.
NOTE that the specific TV this was tested on was the "Samsung
UN40HU6950FXZA" and the specific port was the "STB" port.
dt-bindings: drm/bridge/synopsys: dw-hdmi: Add "unwedge" for ddc bus
In certain situations it was seen that we could wedge up the DDC bus
on the HDMI adapter on rk3288. The only way to unwedge was to mux one
of the pins over to GPIO output-driven-low temporarily and then
quickly mux back. Full details can be found in the patch
("drm/bridge/synopsys: dw-hdmi: Add "unwedge" for ddc bus").
Since unwedge requires remuxing the pins, we first need to add to the
bindings so that we can specify what state the pins should be in for
unwedging.
Chris Wilson [Tue, 4 Jun 2019 12:53:23 +0000 (13:53 +0100)]
dma-buf: Discard old fence_excl on retrying get_fences_rcu for realloc
If we have to drop the seqcount & rcu lock to perform a krealloc, we
have to restart the loop. In doing so, be careful not to lose track of
the already acquired exclusive fence.
projects / linux.git / log