Add mode setting support to the DRM layer.

This is a fairly big chunk of work that allows DRM drivers to provide
full output control and configuration capabilities to userspace.  It was
motivated by several factors:
  - the fb layer's APIs aren't suited for anything but simple
    configurations
  - coordination between the fb layer, DRM layer, and various userspace
    drivers is poor to non-existent (radeonfb excepted)
  - user level mode setting drivers makes displaying panic & oops
    messages more difficult
  - suspend/resume of graphics state is possible in many more
    configurations with kernel level support

This commit just adds the core DRM part of the mode setting APIs.
Driver specific commits using these new structure and APIs will follow.

Co-authors: Jesse Barnes <jbarnes@virtuousgeek.org>, Jakob Bornecrantz <jakob@tungstengraphics.com>
Contributors: Alan Hourihane <alanh@tungstengraphics.com>, Maarten Maathuis <madman2003@gmail.com>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Use the new core GEM object mapping code to allow GTT mapping of GEM
objects on i915.  The fault handler will make sure a fence register is
allocated too, if the object in question is tiled.
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Add core support for mapping of GEM objects.  Drivers should provide a
vm_operations_struct if they want to support page faulting of objects.
The code for handling GEM object offsets was taken from TTM, which was
written by Thomas Hellström.
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Make sure we have the primary node so the device can add maps.
Signed-off-by: NDave Airlie <airlied@redhat.com>

...so drm_getunique() is trying to copy some uninitialized data to
userspace. The ECX register contains the number of words that are
left to copy -- so there are 5 * 4 = 20 bytes left. The offset of the
first uninitialized byte (counting from the start of the string) is
also 20 (i.e. 0xf65d2294&((1 << 5)-1) == 20). So somebody tried to
copy 40 bytes when the string was only 19 long.

In drm_set_busid() we have this code:

        dev->unique_len = 40;
        dev->unique = drm_alloc(dev->unique_len + 1, DRM_MEM_DRIVER);
      ...
        len = snprintf(dev->unique, dev->unique_len, pci:%04x:%02x:%02x.%d",

...so it seems that dev->unique is never updated to reflect the
actual length of the string. The remaining bytes (20 in this case)
are random uninitialized bytes that are copied into userspace.

This patch fixes the problem by setting dev->unique_len after the
snprintf().

airlied- I've had to fix this up to store the alloced size so
we have it for drm_free later.
Reported-by: NSitsofe Wheeler <sitsofe@yahoo.com>
Signed-off-by: NVegard Nossum <vegardno@thuin.ifi.uio.no>
Signed-off-by: NDave Airlie <airlied@redhat.com>

This is step one towards having multiple masters sharing a drm
device in order to get fast-user-switching to work.

It splits out the information associated with the drm master
into a separate kref counted structure, and allocates this when
a master opens the device node. It also allows the current master
to abdicate (say while VT switched), and a new master to take over
the hardware.

It moves the Intel and radeon drivers to using the sarea from
within the new master structures.
Signed-off-by: NDave Airlie <airlied@redhat.com>

The current sub-module unload exit path is a mess, it tries
to abuse the idr. Just keep a list of devices per driver struct
and free them in-order on rmmod.
Signed-off-by: NDave Airlie <airlied@redhat.com>

This check was introduced with the logic the wrong way around.

Fixes regression: http://bugzilla.kernel.org/show_bug.cgi?id=12216Tested-by: NFrançois Valenduc <francois.valenduc@tvcablenet.be>
Signed-off-by: NDave Airlie <airlied@redhat.com>

On PAE systems, GEM allocates pages using shmem, and passes these
pages to be bound into AGP, however the AGP interfaces + the x86
set_memory interfaces all take unsigned long not dma_addr_t.

The initial fix for this was a mess, so we need to do this correctly
for 2.6.29.
Signed-off-by: NDave Airlie <airlied@redhat.com>

These buffers don't have active rendering still occurring to them, they just
need either a flush to be emitted or a retire_requests to occur so that we
notice they're done.  Return unbusy so that one of the two occurs.  The two
expected consumers of this interface (OpenGL and libdrm_intel BO cache) both
want this behavior.
Signed-off-by: NEric Anholt <eric@anholt.net>
Acked-by: NKeith Packard <keithp@keithp.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Since applying the fix suggested by the errata (disabling MSI), we've had
issues with interrupts being stuck on despite IIR being 0 on GM965 hardware.
Most reporters of the issue have confirmed that turning MSI back on fixes
things, and given the difficulties experienced in getting reliable MSI working
on Linux, it's believable that the errata was about software issues and not
actual hardware issues.
Signed-off-by: NDave Airlie <airlied@redhat.com>

Returning -ENOMEM errored all the way out of execbuf, so the rendering never
occurred.
Signed-off-by: NDave Airlie <airlied@redhat.com>

vblank can try and enable the IRQ registers before we've set the interrupt
handler up.
Signed-off-by: NDave Airlie <airlied@redhat.com>

It's only for flushing caches appropriately for GTT access, not for actually
getting it there.  Prevents potential smashing of cpu read/write domains on
unbound objects.
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

If we fail to pin all of the buffers in an execbuffer request, go through
and clear the GTT and try again to see if its just a matter of fragmentation
Signed-off-by: NKeith Packard <keithp@keithp.com>
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

This eliminates the dev_set_domain function and just in-lines it
where its used, with the goal of moving the manipulation and use of
invalidate_domains and flush_domains closer together. This also
avoids calling add_request unless some domain has been flushed.
Signed-off-by: NKeith Packard <keithp@keithp.com>
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Now that the CPU and GTT domain operations are isolated to their own
functions, the previously general-purpose set_domain function is now used
only to set GPU domains. It also has no failure cases, which is important as
this eliminates any possible interruption of the computation of new object
domains and subsequent emmission of the flushing instructions into the ring.
Signed-off-by: NKeith Packard <keithp@keithp.com>
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

This fixes several domain management bugs, including potential lack of cache
invalidation for pread, potential failure to wait for set_domain(CPU, 0),
and more, along with producing more intelligible code.
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

This fixes failure to flush caches in the relocation update path, and
failure to wait in the set_domain ioctl, each of which could lead to incorrect
rendering.
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Otherwise, we would leave the objects in an inconsistent state, such as
write_domain == 0 but on the flushing list.
Signed-off-by: NDave Airlie <airlied@redhat.com>

obj_priv->write_domain is "write domain if the GPU went idle now", not
"write domain at this moment."  By postponing the clear, we confused the
concept, required more storage, and potentially emitted more flushes than
are required.
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

This fixes readpixels and buffer corruption when swapped out and in by
disabling tiling on them.

Now that we know that the bit 17 mode isn't just a mistake of older chipsets,
we'll need to work on a clever fix so that we can get the performance of
tiling on these chipsets, but that will require intrusive changes targeted
at the next kernel release, not this one.
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

drivers/gpu/drm/i915/i915_irq.c: In function 'i915_disable_pipestat':
drivers/gpu/drm/i915/i915_irq.c:101: warning: control may reach end of non-void function 'i915_pipestat' being inlined

Cc: Dave Airlie <airlied@linux.ie>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Introduced in the "Avoid BUG_ONs on VT switch" commit.
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

It fixes suspend/resume failure of xf86-video-intel dri2
branch. As dri2 branch doesn't call I830DRIResume() to restore
hardware status page anymore, we need to preserve
this register across suspend/resume.
Signed-off-by: NPeng Li <peng.li@intel.com>
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

drm vblank initialization keeps track of the changes in driver-supplied
frame counts across vt switch and mode setting, but only if you let it by
not tearing down the drm vblank structure.
Signed-off-by: NKeith Packard <keithp@keithp.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Before we had the notion of pinning objects, we had a kludge around to make
sure all of the objects were still resident in the GTT before we committed
to executing a batch buffer. We don't need this any longer, and it sticks an
error return in the middle of object domain computations that must be
associated with a subsequent flush/invalidate emmission into the ring.
Signed-off-by: NKeith Packard <keithp@keithp.com>
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Because we write pipestat before iir, it's possible that a pipestat
interrupt will occur between the pipestat write and the iir write. This
leaves pipestat with an interrupt status not visible in iir. This may cause
an interrupt flood as we never clear the pipestat event.
Signed-off-by: NKeith Packard <keithp@keithp.com>
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

The old code was wandering through the active list looking for pinned
buffers; there may be other pinned buffers around. Fortunately, we keep a
count of the total amount of pinned memory and can use that instead.
Signed-off-by: NKeith Packard <keithp@keithp.com>
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Instead, just warn that bad things are happening and do our best to clean up
the mess without the GPU's help.
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

The IMR masking was a technique recommended for avoiding getting stuck with
no interrupts generated again in MSI mode.  It kept new IIR bits from getting
set between the IIR read and the IIR write, which would have otherwise
prevented an MSI from ever getting generated again.  However, this caused a
problem for vblank as the IMR mask would keep the pipe event interrupt from
getting reflected in IIR, even after the IMR mask was brought back down.

Instead, just check the state of IIR after we ack the interrupts we're going
to handle, and restart if we didn't get IIR all the way to zero.
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

The pipestat fields affect reporting of all vblank-related interrupts, so we
have to reset them during the irq_handler, and while enabling vblank
interrupts.  Otherwise, if a pipe status field had been set to non-zero
before enabling reporting, we would never see an interrupt again.

This patch adds i915_enable_pipestat and i915_disable_pipestat to abstract
out the steps needed to change the reported interrupts.
Signed-off-by: NKeith Packard <keithp@keithp.com>
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Addresses in the hardware status page below index 0x20 are reserved for use
by the hardware. The legacy breadcrumb was sitting at index 5. Move it to
index 0x21, and make sure everyone uses the defined value instead of
hard-coded constants.
Signed-off-by: NKeith Packard <keithp@keithp.com>
Signed-off-by: NDave Airlie <airlied@linux.ie>

This fixes hangs on 855-class hardware by avoiding double attachment of the
driver due to the stub second head device having the same pci id as the real
device.

Other DRM drivers probably want this treatment as well, but I'm applying it
just to this one for safety. But we should clean up the drm_pciids.h mess
now so that each driver has its own pci id list header in its own directory.
Lets do that in the next release.
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Now that the radeon driver has suspend/resume functions, it needs to map its
registers at load time or it will likely crash if a suspend operation occurs
before the driver has been initialized.

This patch moves the register mapping code from firstopen to load and makes
the mapping into a _DRM_DRIVER one so that the core won't remove it at
lastclose time.

Fixes (at least partially) kernel bz #11891.
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: NDave Airlie <airlied@linux.ie>

It's not used in any other drivers, and doesn't look like it will be from
drm.git master.
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@linux.ie>

When userland detected that this ioctl was supported (by version number check),
it used it in a racy way -- dispatch delayed swap, wait for vblank, continue
rendering. As there was no mechanism for it to wait for the swap to finish,
sometimes it would render before the swap and garbage would be displayed on
the screen.

By removing the ioctl and returning -EINVAL, userland returns to its previous,
correct rendering path of waiting for a vblank then dispatching a swap.  The
only path that could have used this ioctl correctly was page flipping, which
relied on only one client running and emitting wait-for-vblank-before-rendering
in the command stream.  That path also falls back correctly, at the performance
cost of not being able to queue up rendering before the flip occurs.
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@linux.ie>

This probably just means the chipset doesn't support MSI, which is fine.
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@linux.ie>

This could return early when reading after writing a buffer, if somebody
had already put it on the flushing list (write domains are 0, but still
active), leading to glReadPixels failure.
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDave Airlie <airlied@linux.ie>