In response to my patch removing the "wait for HPD" logic at the
beginning of the MSM DP transfer() callback [1], we had some debate
about what the "This is an optional function" meant in the
documentation of the wait_hpd_asserted() callback. Let's clarify.
As talked about in the MSM DP patch [1], before wait_hpd_asserted()
was introduced there was no great way for panel drivers to wait for
HPD in the case that the "built-in" HPD signal was used. Panel drivers
could only wait for HPD if a GPIO was used. At the time, we ended up
just saying that if we were using the "built-in" HPD signal that DP
AUX controllers needed to wait for HPD themselves at the beginning of
their transfer() callback. The fact that the wait for HPD at the
beginning of transfer() was awkward/problematic was the whole reason
wait_hpd_asserted() was added.
Let's make it obvious that if a DP AUX controller implements
wait_hpd_asserted() that they don't need a loop waiting for HPD at the
start of their transfer() function. We'll still allow DP controllers
to work the old way but mark it as deprecated.
[1] https://lore.kernel.org/r/20240315143621.v2.3.I535606f6d4f7e3e5588bb75c55996f61980183cd@changeid
Reviewed-by: Abhinav Kumar <quic_abhinavk@quicinc.com>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20240319135836.v2.1.I521dad0693cc24fe4dd14cba0c7048d94f5b6b41@changeid
The array size calculation in pvr_vm_mips_fini() appears to be incorrect based on
taking the size of the pointer rather than the size of the array, which manifests
as a warning about signed integer overflow:
In file included from include/linux/kernel.h:16,
from drivers/gpu/drm/imagination/pvr_rogue_fwif.h:10,
from drivers/gpu/drm/imagination/pvr_ccb.h:7,
from drivers/gpu/drm/imagination/pvr_device.h:7,
from drivers/gpu/drm/imagination/pvr_vm_mips.c:4:
drivers/gpu/drm/imagination/pvr_vm_mips.c: In function 'pvr_vm_mips_fini':
include/linux/array_size.h:11:25: error: overflow in conversion from 'long unsigned int' to 'int' changes value from '18446744073709551615' to '-1' [-Werror=overflow]
11 | #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]) + __must_be_array(arr))
| ^
drivers/gpu/drm/imagination/pvr_vm_mips.c:106:24: note: in expansion of macro 'ARRAY_SIZE'
106 | for (page_nr = ARRAY_SIZE(mips_data->pt_pages) - 1; page_nr >= 0; page_nr--) {
| ^~~~~~~~~~
Just use the number of array elements directly here, and in the corresponding
init function for consistency.
Fixes: 927f3e0253 ("drm/imagination: Implement MIPS firmware processor and MMU support")
Reviewed-by: Donald Robson <donald.robson@imgtec.com>
Link: https://lore.kernel.org/lkml/9df9e4f87727399928c068dbbf614c9895ae15f9.camel@imgtec.com/
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Matt Coster <matt.coster@imgtec.com>
In case the LCDIF is enabled in DT but unused, the clocks used by the
LCDIF are not enabled. Those clocks may even have a use count of 0 in
case there are no other users of those clocks. This can happen e.g. in
case the LCDIF drives HDMI bridge which has no panel plugged into the
HDMI connector.
Do not attempt to disable clocks in the suspend callback and re-enable
clocks in the resume callback unless the LCDIF is enabled and was in
use before the system entered suspend, otherwise the driver might end
up trying to disable clocks which are already disabled with use count
0, and would trigger a warning from clock core about this condition.
Note that the lcdif_rpm_suspend() and lcdif_rpm_resume() functions
internally perform the clocks disable and enable operations and act
as runtime PM hooks too.
Reviewed-by: Liu Ying <victor.liu@nxp.com>
Fixes: 9db35bb349 ("drm: lcdif: Add support for i.MX8MP LCDIF variant")
Signed-off-by: Marek Vasut <marex@denx.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20240226082644.32603-1-marex@denx.de
Commit 95da53d63d ("drm/omapdrm: Use regular fbdev I/O helpers")
stopped console from updating for command mode displays because there is
no damage handling in fb_sys_write() unlike we had earlier in
drm_fb_helper_sys_write().
Let's fix the issue by adding FB_GEN_DEFAULT_DEFERRED_DMAMEM_OPS and
FB_DMAMEM_HELPERS_DEFERRED as suggested by Thomas. We cannot use the
FB_DEFAULT_DEFERRED_OPS as fb_deferred_io_mmap() won't work properly
for write-combine.
Fixes: 95da53d63d ("drm/omapdrm: Use regular fbdev I/O helpers")
Suggested-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240228063540.4444-3-tony@atomide.com
I've recently been seeing some unexplained GSP errors on my RTX 6000 from
failed aux transactions:
[ 132.915867] nouveau 0000:1f:00.0: gsp: cli:0xc1d00002 obj:0x00730000
ctrl cmd:0x00731341 failed: 0x0000ffff
While the cause of these is not yet clear, these messages made me notice
that the aux transactions causing these transactions were succeeding - not
failing. As it turns out, this is because we're currently not returning the
correct variable when r535_dp_aux_xfer() hits an error - causing us to
never propagate GSP errors for failed aux transactions to userspace.
So, let's fix that.
Fixes: 4ae3a20102 ("nouveau/gsp: don't free ctrl messages on errors")
Signed-off-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240315212104.776936-1-lyude@redhat.com
It's found that some panels have variants that they share the same panel id
although their EDID and names are different. Besides panel id, now we need
more information from the EDID base block to distinguish these panel
variants.
Add drm_edid_read_base_block() to return the EDID base block, which is
wrapped in struct drm_edid.
Caller can further use it to get panel id or check if the block contains
certain strings, such as panel name.
Merge drm_edid_get_panel_id() and edid_extract_panel_id() into one
function.
Signed-off-by: Hsin-Yi Wang <hsinyi@chromium.org>
Reviewed-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20240307230653.1807557-2-hsinyi@chromium.org
Pin and vmap are distinct operations. Do not perform a pin as part
of the vmap call. This used to be necessary to keep the fbdev buffer
in place while it is being updated. Fbdev emulation has meanwhile
been fixed to lock the buffer correctly. Same for vunmap.
For refactoring the code, remove the pin calls from the helper's
vmap implementation in drm_gem_vram_vmap() and inline the call to
drm_gem_vram_kmap_locked(). This gives a vmap helper that only
maps the buffer object's memory pages without pinning or locking.
Do a similar refactoring for vunmap.
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Tested-by: Dmitry Osipenko <dmitry.osipenko@collabora.com> # virtio-gpu
Acked-by: Zack Rusin <zack.rusin@broadcom.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240227113853.8464-13-tzimmermann@suse.de
Temporarily lock the fbdev buffer object during updates to prevent
memory managers from evicting/moving the buffer. Moving a buffer
object while update its content results in undefined behaviour.
Fbdev-generic updates its buffer object from a shadow buffer. Gem-shmem
and gem-dma helpers do not move buffer objects, so they are safe to be
used with fbdev-generic. Gem-vram and qxl are based on TTM, but pin
buffer objects are part of the vmap operation. So both are also safe
to be used with fbdev-generic.
Amdgpu and nouveau do not pin or lock the buffer object during an
update. Their TTM-based memory management could move the buffer object
while the update is ongoing.
The new vmap_local and vunmap_local helpers hold the buffer object's
reservation lock during the buffer update. This prevents moving the
buffer object on all memory managers.
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Tested-by: Dmitry Osipenko <dmitry.osipenko@collabora.com> # virtio-gpu
Acked-by: Zack Rusin <zack.rusin@broadcom.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240227113853.8464-11-tzimmermann@suse.de
Acquire the buffer object's reservation lock in drm_gem_pin() and
remove locking the drivers' GEM callbacks where necessary. Same for
unpin().
DRM drivers and memory managers modified by this patch will now have
correct dma-buf locking semantics: the caller is responsible for
holding the reservation lock when calling the pin or unpin callback.
DRM drivers and memory managers that are not modified will now be
protected against concurent invocation of their pin and unpin callbacks.
PRIME does not implement struct dma_buf_ops.pin, which requires
the caller to hold the reservation lock. It does implement struct
dma_buf_ops.attach, which requires to callee to acquire the
reservation lock. The PRIME code uses drm_gem_pin(), so locks
are now taken as specified. Same for unpin and detach.
The patch harmonizes GEM pin and unpin to have non-interruptible
reservation locking across all drivers, as is already the case for
vmap and vunmap. This affects gem-shmem, gem-vram, loongson, qxl and
radeon.
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Zack Rusin <zack.rusin@broadcom.com>
Reviewed-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Tested-by: Dmitry Osipenko <dmitry.osipenko@collabora.com> # virtio-gpu
Link: https://patchwork.freedesktop.org/patch/msgid/20240227113853.8464-10-tzimmermann@suse.de
