Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

bochs kms driver lacks power management support, thus
the vga display doesn't work any more after S3 resume.

Fix this by adding suspend and resume functions.
Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

cirrus kms driver lacks power management support, thus
the vga display doesn't work any more after S3 resume.

Fix this by adding suspend and resume functions.
Also make the mode_set function unblank the screen.
Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

This is leftover stuff from my previous doc round which I kinda wanted
to do but didn't yet due to rebase hell.

The modeset helpers and the probing helpers a independent and e.g.
i915 uses the probing stuff but has its own modeset infrastructure. It
hence makes to split this up. While at it add a DOC: comment for the
probing libraray.

It would be rather neat to pull some of the DocBook documenting these
two helpers into in-line DOC: comments. But unfortunately kerneldoc
doesn't support markdown or something similar to make nice-looking
documentation, so the current state is better.
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: NDave Airlie <airlied@redhat.com>

After thinking about this topic a bit more I've reached the conclusion
that implementing this doesn't make sense:

- The locking is all wrong: set_config(NULL) will also unlink encoders
  and connectors, but those links are protected with the mode_config
  mutex. In the ->disable_plane callback we only hold all modeset
  locks, but eventually we want to switch to just grabbing the
  per-crtc (and maybe per-plane) locks as needed, maybe based on
  ww_mutexes. Having a callback which absolutely needs all modeset
  locks is bad for this conversion.

  Note that the same isn't true for the provided ->update_plane since
  we've audited the crtc helpers to make sure that not encoder or
  connector links are changed.

- There's no way to re-enable the plane with an ->update_plane: The
  connectors/encoder links are lost and so we can't re-enable the
  CRTC. Even without that issue the driver might have reassigned some
  shared resources (as opposed to e.g. DPMS off, where drivers are not
  allowed to do that to make sure the CRTC can be enabled again).

- The semantics don't make much sense: Userspace asked to scan out
  black (or some other color if the driver supports a background
  color), not that the screen be disabled.

- Implementing proper primary plane support (i.e. actually disabling
  the primary plane without disabling the CRTC) is really simple, at
  least if all the hw needs is flipping a bit. The big task is
  auditing all the interactions with other ioctls when the CRTC is on
  but there's no primary plane (e.g. pageflips). And some of that work
  still needs to be done.

Cc: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NMatt Roper <matthew.d.roper@intel.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

this is a typo vs the ums driver, fix to check correct value.

Found initially by Coverity.
Reported-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Commit 457e77b2 added two checks applied to a
value received from nv_rd32(bios, 0x619f04). But after this new piece of code
is executed, the addr local variable does not hold the same value it used to
hold before the commit. Here is what is was assigned in the original code:
	(u64)(nv_rd32(bios, 0x619f04) & 0xffffff00) << 8
in the committed code it ends up with this value:
	(u64)(nv_rd32(bios, 0x619f04) >> 8) << 8
These expressions are obviously not equivalent.

My Nvidia video card does not show anything on the display when I boot a
kernel containing this commit.

The patch fixes the code so that the new checks are still done, but the
side effect of an incorrect addr value is gone.

Cc: Ben Skeggs <bskeggs@redhat.com>
Cc: Dave Airlie <airlied@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: NSergei Antonov <saproj@gmail.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Avoid a possible segfault.
Noticed-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Avoid a possible segfault.
Noticed-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Otherwise we might be quite off on older chipsets.

v2: keep ref_div minimum
Signed-off-by: NChristian König <christian.koenig@amd.com>

On bo reservation failure, we end up leaking fpriv.

v2 (chk): rebased and added missing free on vm failure as well

Fixes: 5e386b57 ("drm/radeon: fix missing bo reservation")
Cc: stable@vger.kernel.org
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: NQuentin Casasnovas <quentin.casasnovas@oracle.com>
Signed-off-by: NChristian König <christian.koenig@amd.com>

Due to a type mismatch that causes an implicit type conversion, the
upper 32 bits of the GPU address have been zeroed out when adding to the
command buffer.

Picked up by Coverity - CID 1198624.
Signed-off-by: NChristoph Jaeger <christophjaeger@linux.com>

If the new mc ucode is available.
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Fixes mclk stability on certain asics.

v2: print out mc firmware version used and size

bug:
https://bugs.freedesktop.org/show_bug.cgi?id=75992Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

May fix stability issues with some newer cards.

v2: print out mc firmware version used and size
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Letting post and refernce divider get to big is bad for signal stability.

v2: increase the limit to 210
Signed-off-by: NChristian König <christian.koenig@amd.com>

As per internal recommendations.
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Don't try and runtime suspend the APU in PX systems.  We
only want to power down the dGPU.

v2: fix harder
v3: fix stupid typo
v4: consolidate runpm enablement to a single flag

bugs:
https://bugs.freedesktop.org/show_bug.cgi?id=75127
https://bugzilla.kernel.org/show_bug.cgi?id=72701Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Setting higher mclks seems to cause stability issues
on some R7 260X boards.  Disable it for now for stability
until we find a proper fix.

bug:
https://bugs.freedesktop.org/show_bug.cgi?id=75992Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

BIT_WORD() truncates rather than rounds, so the loops in
syncpt_thresh_isr() and _host1x_intr_disable_all_syncpt_intrs() use <=
rather than < in an attempt to process the correct number of registers
when rounding of the conversion of count of bits to count of words is
necessary. However, when rounding isn't necessary because the value is
already a multiple of the divisor (as is the case for all values of
nb_pts the code actually sees), this causes one too many registers to
be processed.

Solve this by using and explicit DIV_ROUND_UP() call, rather than
BIT_WORD(), and comparing with < rather than <=.

Fixes: 7ede0b0b ("gpu: host1x: Add syncpoint wait and interrupts")
Cc: <stable@vger.kernel.org> # 3.10
Signed-off-by: NStephen Warren <swarren@nvidia.com>
Acked-By: NTerje Bergstrom <tbergstrom@nvidia.com>
Signed-off-by: NThierry Reding <treding@nvidia.com>

With the recent primary-plane changes for drm, the primary plane's
framebuffer needs to be ref counted the same way as for
non-primary-planes. This was not done by the omapdrm driver, which
caused the ref count to drop to 0 too early, causing problems.

This patch moves the fb unref and ref from omap_plane_update to
omap_plane_mode_set. This way the fb refs are updated for both primary
and non-primary cases, as omap_plane_update calls omap_plane_mode_set.
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>

The vblank_cb callback and the page_flip ioctl can occur together in different
CPU contexts. vblank_cb uses takes tje drm device's event_lock spinlock when
sending the vblank event and updating omap_crtc->event and omap_crtc->od_fb.

Use the same spinlock in page_flip, to make sure the above omap_crtc parameters
are configured sequentially.
Signed-off-by: NArchit Taneja <archit@ti.com>
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>

omap_crtc->old_fb is used to check whether the previous page flip has completed
or not. However, it's never initialized to anything, so it's always NULL. This
results in the check to always succeed, and the page_flip to proceed.

Initialize old_fb to the fb that we intend to flip to through page_flip, and
therefore prevent a future page flip to proceed if the last one didn't
complete.
Signed-off-by: NArchit Taneja <archit@ti.com>
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>

The channel_names list didn't have a string populated for LCD3 manager, this
results in a crash when the display's output is connected to LCD3. Add an entry
for LCD3.
Reported-by: NSomnath Mukherjee <somnath@ti.com>
Signed-off-by: NArchit Taneja <archit@ti.com>
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>

A waiter of the type OMAP_GEM_READ should wait for a buffer to be completely
written, and only then proceed with reading it. A similar logic applies for
waiters with OMAP_GEM_WRITE flag.

Currently the function is_waiting() waits on the read_complete/read_target
counts in the sync object.

This should be the other way round, as a reader should wait for users who are
'writing' to this buffer, and vice versa.

Make readers of the buffer(OMAP_GEM_READ) wait on the write counters, and
writers to the buffer(OMAP_GEM_WRITE) wait on the read counters in is_waiting()
Signed-off-by: NArchit Taneja <archit@ti.com>
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>

In omap_gem_op_async(), if a waiter is not added to the wait list, it needs to
be free'd in the function itself.

Make sure we free the waiter for this case.
Signed-off-by: NSubhajit Paul <subhajit_paul@ti.com>
Signed-off-by: NArchit Taneja <archit@ti.com>
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>

Patch dfe96ddc (omapdrm: simplify locking in
the fb debugfs file) removed taking locks when using omapdrm's debugfs
to dump fb objects.

However, in omap_gem_describe we give a WARN is the lock has not been
taken, so that WARN is now seen every time omapdrm debugfs is used.

So, presuming the removal of locks is ok, we can also remove the WARN.
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>

All the planes, including primary planes, are now destroyed by the drm
framework. Thus we no longer need the explicit call to plane->destroy
from the crtc's destroy function.

This patch removes the call, thus fixing the crash caused by double
freeing the plane.

remove omap_crtc->plane->funcs->destroy(omap_crtc->plane)
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>

Print a warning when the user tries to rotate a non-TILER framebuffer.

Also set the rotation to 0, to avoid constant flood of the warnings in
case of page flipping.
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>

omap_fbdev_create() takes a reference to the fb's gem object with
omap_gem_get_paddr(). However, it never releases it with
omap_gem_put_paddr().

This patch adds the missing omap_gem_put_paddr() to omap_fbdev_free().
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>

Plane rotation with omapdrm is currently broken.
It seems omap_plane_mode_set() expects width and height in screen
coordinates, so pass it like that.
Signed-off-by: NGrazvydas Ignotas <notasas@gmail.com>
Reviewed-by: NRob Clark <robdclark@gmail.com>
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>

At the moment the omap_crtc_pre_apply() handles the enabling, disabling
and configuring of encoders and panels separately from the CRTC (i.e.
the overlay manager).

However, this doesn't work correctly. The encoder driver has to be in
control of its video input (i.e. the crtc) for correct operation.

This problem causes bugs with (at least) HDMI: the HDMI encoder supplies
pixel clock for DISPC, and DISPC supplies video stream for HDMI. The
current code first enables the HDMI encoder, and CRTC after that.
However, the encoder expects the video stream to start during the
encoder's enable, and if it doesn't, there will be sync lost errors.

The encoder enables its video source by calling src->enable(), and this
call goes to omapdrm (omap_crtc_enable), but omapdrm doesn't do anything
in that function. Similarly for disable, which goes to
omap_crtc_disable().

This patch moves the code to setup and enable/disable the crtc to
omap_crtc_enable. and omap_crtc_disable().
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>
Reviewed-by: NRob Clark <robdclark@gmail.com>

When an encoder is no longer connected to a crtc, the driver will leave
the encoder enabled.

This patch adds code to track the encoder used for a crtc, and when the
encoder changes, the old one is disabled.
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>
Reviewed-by: NRob Clark <robdclark@gmail.com>

At module unload, omap_fbdev_free() gets called which releases the
framebuffers. However, the framebuffers are still used by crtcs, and
will be released only later at vsync. The driver doesn't wait for this,
and goes on to release the rest of the resources, which often
causes a crash.

This patchs adds a omap_crtc_flush() function which waits until the crtc
has finished with its apply queue and page flips.

The function utilizes a simple polling while-loop, as the performance is
not an issue here.
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>
Reviewed-by: NRob Clark <robdclark@gmail.com>

At the moment the DMM driver is never unregistered, even if it's
registered in the omapdrm module's init function. This means we'll get
errors when reloading the omapdrm module.

Fix this by unregistering the DMM driver properly, and also change the
module init to fail if DMM driver cannot be registered, simplifying the
unregister path as we don't need to keep the state whether we registered
the DMM driver or not.
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>
Reviewed-by: NRob Clark <robdclark@gmail.com>

When unloading omapdrm driver, the omapdrm platform device is
uninitialized last, after the displays have been disconnected omap_crtc
callbacks have been removed. As the omapdrm pdev uninitialization needs
the features uninitialized in earlier steps, a crash is guaranteed.

This patch fixes the uninitialize order so that the omapdrm pdev is
removed first.
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>
Reviewed-by: NRob Clark <robdclark@gmail.com>

At the moment it's quite easy to get the following errors when the HDMI
output is enabled or disabled:

[drm:omap_crtc_error_irq] *ERROR* tv: errors: 00008000

The reason for the errors is that the omapdrm driver doesn't properly
handle the sync-lost irqs that happen when enabling the DIGIT crtc,
which is used for HDMI and analog TV. The driver does disable the
sync-lost irq properly, but it fails to wait until the output has been
fully enabled (i.e. the first vsync), so the sync-lost errors are still
seen occasionally.

This patch makes the omapdrm act the same way as the omapfb does:

- When enabling a display, we'll wait for the first vsync.
- When disabling a display, we'll wait for framedone if available, or
  odd and even vsyncs.

These changes make sure the output is fully enabled or disabled at the
end of the function.
Signed-off-by: NTomi Valkeinen <tomi.valkeinen@ti.com>
Reported-by: NSanjay Singh Rawat <sanjay.rawat@linaro.org>
Reviewed-by: NRob Clark <robdclark@gmail.com>

This patch fixes I/O errors with the sym53c8xx_2 driver when the disk
returns QUEUE FULL status.

When the controller encounters an error (including QUEUE FULL or BUSY
status), it aborts all not yet submitted requests in the function
sym_dequeue_from_squeue.

This function aborts them with DID_SOFT_ERROR.

If the disk has full tag queue, the request that caused the overflow is
aborted with QUEUE FULL status (and the scsi midlayer properly retries
it until it is accepted by the disk), but the sym53c8xx_2 driver aborts
the following requests with DID_SOFT_ERROR --- for them, the midlayer
does just a few retries and then signals the error up to sd.

The result is that disk returning QUEUE FULL causes request failures.

The error was reproduced on 53c895 with COMPAQ BD03685A24 disk
(rebranded ST336607LC) with command queue 48 or 64 tags.  The disk has
64 tags, but under some access patterns it return QUEUE FULL when there
are less than 64 pending tags.  The SCSI specification allows returning
QUEUE FULL anytime and it is up to the host to retry.
Signed-off-by: NMikulas Patocka <mpatocka@redhat.com>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: James Bottomley <JBottomley@Parallels.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This patch fixes the initialization of an array used in the TX
datapath that was mistakenly initialized together with the
RX datapath arrays. An out of range array access could happen
when RX and TX rings had different sizes.
Signed-off-by: NVincenzo Maffione <v.maffione@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When the device is unplugged, the driver would try to disable the
device. Add checking the flag of RTL8152_UNPLUG to skip setting
the device when it is unplugged. This could shorten the time of
unloading the driver.
Signed-off-by: NHayes Wang <hayeswang@realtek.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>