Moving the driver-specific mmap code into a GEM object function allows
for using DRM helpers for various mmap callbacks.
The respective rockchip functions are being removed. The file_operations
structure fops is now being created by the helper macro
DEFINE_DRM_GEM_FOPS().
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
[On rk3288 (pinky), rk3399 (gru-kevin, puma) and rk3328 (rock64)]
Tested-by: Heiko Stuebner <heiko@sntech.de>
[On RK3188/RK3066 (without iommu)]
Tested-by: Alex Bee <knaerzche@gmail.com>
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20210624095502.8945-1-tzimmermann@suse.de
While the pages can't be swapped out, they can be discarded by the shrinker.
Normally such objects are marked with __I915_MADV_PURGED, which can't be
unset, and therefore requires a new object. For kernel internal objects
this is not true, since the madv hint is reset for our special volatile
objects, such that we can re-acquire new pages, if so desired, without
needing a new object. As a result we should probably be paranoid here
and put the object back into the CPU domain when discarding the pages,
and also correctly set cache_dirty, if required.
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211018174508.2137279-8-matthew.auld@intel.com
Non-continuous clock mode doesn't work because driver doesn't support it
properly. The bridge driver programs wrong bitfields that are required by
the non-continuous mode (BTACNTRL1 register bitfields are swapped in the
code), but fixing them doesn't help.
Display panel of ASUS Transformer TF700T tablet supports non-continuous
mode and display doesn't work at all using that mode. There are no
device-trees that are actively using this DSI bridge in upstream yet,
so clearly the broken mode wasn't ever tested properly. It's a bit too
difficult to get LP mode working, hence let's disable the offending mode
for now and fall back to continuous mode.
Tested-by: Andreas Westman Dorcsak <hedmoo@yahoo.com> # Asus TF700T
Tested-by: Maxim Schwalm <maxim.schwalm@gmail.com> #TF700T
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Reviewed-by: Robert Foss <robert.foss@linaro.org>
Signed-off-by: Robert Foss <robert.foss@linaro.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20211002233447.1105-5-digetx@gmail.com
struct gtt_range represents a GEM object and should not be used for GTT
setup. Change psb_gtt_insert() and psb_gtt_remove() to receive all
necessary parameters from their caller. This also eliminates possible
failure from psb_gtt_insert().
There's one exception in psb_gtt_restore(), which requires an upcast
from struct resource to struct gtt_range when restoring the GTT after
hibernation. A possible solution would track the GEM objects that need
restoration separately from the GTT resource.
Rename the functions to psb_gtt_insert_pages() and psb_gtt_remove_pages()
to reflect their similarity to MMU interfaces.
v3:
* restore the comments about locking rules (Patrik)
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Acked-by: Patrik Jakobsson <patrik.r.jakobsson@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211015084053.13708-10-tzimmermann@suse.de
psb_gtt_alloc_range() allocates struct gtt_range, create the GTT resource
and performs some half-baked initialization. Inline the function into its
only caller psb_gem_create(). For creating the GTT resource, introduce a
new helper, psb_gtt_alloc_resource() that hides the details of the GTT.
For psb_gtt_free_range(), inline the function into its only caller
psb_gem_free_object(). While at it, remove the explicit invocation of
drm_gem_free_mmap_offset(). The mmap offset is already released by
drm_gem_object_release().
v3:
* replace offset[static 1] with pointer notation (Patrik)
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Acked-by: Patrik Jakobsson <patrik.r.jakobsson@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211015084053.13708-8-tzimmermann@suse.de
Support private objects for stolen memory in psb_gem_create() and
convert users to psb_gem_create(). For stolen memory, psb_gem_create()
now initializes the GEM object via drm_gem_private_object_init().
In the fbdev setup, replace the open-coded initialization of struct
gtt_range with a call to psb_gem_create(). Use drm_gem_object_put()
for release.
In the cursor setup, use psb_gem_create() and get a real GEM object.
Previously the allocated instance of struct gtt_range was only partially
initialized. Release the cursor GEM object in gma_crtc_destroy(). The
release was missing from the original code.
With the conversion of all callers to psb_gem_create(), the extern
declarations of psb_gtt_alloc_range, psb_gtt_free_range and
psb_gem_object_func are not required any longer. Declare them as
static.
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Acked-by: Patrik Jakobsson <patrik.r.jakobsson@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211015084053.13708-5-tzimmermann@suse.de
The link training delays are different and/or available in different
DPCD offsets depending on:
- Clock recovery vs. channel equalization
- DPRX vs. LTTPR
- 128b/132b vs. 8b/10b
- DPCD 1.4+ vs. earlier
Add helpers to get the correct delays in us, reading DPCD if
necessary. This is more straightforward than trying to retrofit the
existing helpers to take 128b/132b into account.
Having to pass in the DPCD receiver cap field seems unavoidable, because
reading it involves checking the revision and reading extended receiver
cap. So unfortunately the interface is mixed cached and read as needed.
v2: Remove delay_us < 0 check and the whole local var (Ville)
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211014150059.28957-1-jani.nikula@intel.com
Since commit 43c2de1002 ("drm/rockchip: dsi: move all lane config except
LCDC mux to bind()"), we perform most HW configuration in the bind()
function. This configuration may be lost on suspend/resume, so we
need to call it again. That may lead to errors like this after system
suspend/resume:
dw-mipi-dsi-rockchip ff968000.mipi: failed to write command FIFO
panel-kingdisplay-kd097d04 ff960000.mipi.0: failed write init cmds: -110
Tested on Acer Chromebook Tab 10 (RK3399 Gru-Scarlet).
Note that early mailing list versions of this driver borrowed Rockchip's
downstream/BSP solution, to do HW configuration in mode_set() (which
*is* called at the appropriate pre-enable() times), but that was
discarded along the way. I've avoided that still, because mode_set()
documentation doesn't suggest this kind of purpose as far as I can tell.
Fixes: 43c2de1002 ("drm/rockchip: dsi: move all lane config except LCDC mux to bind()")
Cc: <stable@vger.kernel.org>
Signed-off-by: Brian Norris <briannorris@chromium.org>
Reviewed-by: Chen-Yu Tsai <wenst@chromium.org>
Tested-by: Nícolas F. R. A. Prado <nfraprado@collabora.com>
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20210928143413.v3.2.I4e9d93aadb00b1ffc7d506e3186a25492bf0b732@changeid
In commit 43c2de1002 ("drm/rockchip: dsi: move all lane config except
LCDC mux to bind()"), we moved most HW configuration to bind(), but we
didn't move the runtime PM management. Therefore, depending on initial
boot state, runtime-PM workqueue delays, and other timing factors, we
may disable our power domain in between the hardware configuration
(bind()) and when we enable the display. This can cause us to lose
hardware state and fail to configure our display. For example:
dw-mipi-dsi-rockchip ff968000.mipi: failed to write command FIFO
panel-innolux-p079zca ff960000.mipi.0: failed to write command 0
or:
dw-mipi-dsi-rockchip ff968000.mipi: failed to write command FIFO
panel-kingdisplay-kd097d04 ff960000.mipi.0: failed write init cmds: -110
We should match the runtime PM to the lifetime of the bind()/unbind()
cycle.
Tested on Acer Chrometab 10 (RK3399 Gru-Scarlet), with panel drivers
built either as modules or built-in.
Side notes: it seems one is more likely to see this problem when the
panel driver is built into the kernel. I've also seen this problem
bisect down to commits that simply changed Kconfig dependencies, because
it changed the order in which driver init functions were compiled into
the kernel, and therefore the ordering and timing of built-in device
probe.
Fixes: 43c2de1002 ("drm/rockchip: dsi: move all lane config except LCDC mux to bind()")
Link: https://lore.kernel.org/linux-rockchip/9aedfb528600ecf871885f7293ca4207c84d16c1.camel@gmail.com/
Reported-by: <aleksandr.o.makarov@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Brian Norris <briannorris@chromium.org>
Tested-by: Nícolas F. R. A. Prado <nfraprado@collabora.com>
Reviewed-by: Chen-Yu Tsai <wenst@chromium.org>
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20210928143413.v3.1.Ic2904d37f30013a7f3d8476203ad3733c186827e@changeid
