This allocates a physical surface for the PCI GART table, this way no
matter what other surface configurations exist the GART table will
always be seen by the hardware properly.
We encode the file pointer of the virtual surface allocate using a
special cookie value, called PCIGART_FILE_PRIV. On the last close, we
release that surface.
Just to be doubly safe, we run the pcigart table setup with the main
surface control register clear.
Based upon ideas from David Airlie and Ben Benjamin Herrenschmidt.
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
The address needs to be a GART relative address, rather than a PCI
DMA address.
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
These are not supposed to be booleans, they are
supposed to be bit masks.
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
The memory behind ring_rptr can either be in ioremapped memory
or a vmalloc() normal kernel memory buffer.
However, the code unconditionally uses DRM_{READ,WRITE}32() (and thus
readl() and writel()) to access it.
Basically, if RADEON_IS_AGP then it's ioremap()'d memory else it's
vmalloc'd memory.
Adjust all of the ring_rptr access code as needed.
While we're here, kill the 'scratch' pointer in drm_radeon_private.
It's only used in the one place where it is initialized.
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
The buffers mapped by the PCI GART can be written to by the device,
not just read.
For example, this happens via the RB_RPTR writeback on Radeon.
So we can't use PCI_DMA_TODEVICE else we'll get protection faults
on IOMMU platforms.
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
The PCI GART table initialization code treats the GART table mapping
unconditionally as a kernel virtual address.
But it could be in the framebuffer, for example, and thus we're
dealing with a PCI MEM space ioremap() cookie. Treating that as a
virtual address is illegal and will crash some system types (such as
sparc64 where the ioremap() return value is actually a physical I/O
address).
So access the area correctly, using gart_info->gart_table_location as
our guide.
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
The kernel shouldn't be in the business of telling user space which
driver to load. The kernel defers mapping PCI IDs to module names
to user space and we should do the same for DRI drivers.
And in fact, that's how it does work today. Nothing uses the
dri_library_name attribute, and the attribute is in fact broken.
For intel devices, it falls back to the default behaviour of returning
the kernel module name as the DRI driver name, which doesn't work for
i965 devices. Nobody has ever hit this problem or filed a bug about this.
Signed-off-by: Kristian Høgsberg <krh@redhat.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>
Under kernel modesetting, we manage the device at all times, regardless
of VT switching and X servers, so the only decent thing to do is to
claim the PCI device. In that case, we call the suspend/resume hooks
directly from the pci driver hooks instead of the current class device detour.
Signed-off-by: Kristian Høgsberg <krh@redhat.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>
This changes drm_local_map to use a resource_size for its "offset"
member instead of an unsigned long, thus allowing 32-bit machines
with a >32-bit physical address space to be able to store there
their register or framebuffer addresses when those are above 4G,
such as when using a PCI video card on a recent AMCC 440 SoC.
This patch isn't as "trivial" as it sounds: A few functions needed
to have some unsigned long/int changed to resource_size_t and a few
printk's had to be adjusted.
But also, because userspace isn't capable of passing such offsets,
I had to modify drm_find_matching_map() to ignore the offset passed
in for maps of type _DRM_FRAMEBUFFER or _DRM_REGISTERS.
If we ever support multiple _DRM_FRAMEBUFFER or _DRM_REGISTERS maps
for a given device, we might have to change that trick, but I don't
think that happens on any current driver.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Dave Airlie <airlied@linux.ie>
Once upon a time, the DRM made the distinction between the drm_map
data structure exchanged with user space and the drm_local_map used
in the kernel.
For some reasons, while the BSD port still has that "feature", the
linux part abused drm_map for kernel internal usage as the local
map only existed as a typedef of the struct drm_map.
This patch fixes it by declaring struct drm_local_map separately
(though its content is currently identical to the userspace variant),
and changing the kernel code to only use that, except when it's a
user<->kernel interface (ie. ioctl).
This allows subsequent changes to the in-kernel format
I've also replaced the use of drm_local_map_t with struct drm_local_map
in a couple of places. Mostly by accident but they are the same (the
former is a typedef of the later) and I have some remote plans and
half finished patch to completely kill the drm_local_map_t typedef
so I left those bits in.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
The DRM uses its own wrappers to obtain resources from PCI devices,
which currently convert the resource_size_t into an unsigned long.
This is broken on 32-bit platforms with >32-bit physical address
space.
This fixes them, along with a few occurences of unsigned long used
to store such a resource in drivers.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Dave Airlie <airlied@linux.ie>
The last 8 fence registers sit at a different offset, so when we went to set
fence number 8 in the lower offset, we instead set PGETBL_CTL, and the GPU
got all sorts of angry at us.
fd.o bug #20567. Easily reproducible by running glxgears and killing it about
6 times.
Signed-off-by: Eric Anholt <eric@anholt.net>
The i915 also uses the fence registers for GPU access to tiled buffers so
we cannot reallocate one whilst it is on the active list. By performing a
LRU scan of the fenced buffers we also avoid waiting the possibility of
waiting on a pinned, or otherwise unusable, buffer.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
We need to check and report if there are no available fences - or else we
spin endlessly waiting for a buffer to magically unpin itself.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
As we may steal the fence register of an unpinned buffer for another,
every time we repin the buffer we need to recheck whether it needs to be
allocated a fence.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
If we wait upon a request and successfully unbind a buffer occupying a
fence register, then that slot will be freed and cause a NULL derefrence
upon rescanning.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
The VGA registers just hit the pipe registers that we already set through
MMIO. This fixes strange colors on resume.
Signed-off-by: Pierre Willenbrock <pierre@pirsoft.de>
Signed-off-by: Eric Anholt <eric@anholt.net>
If userspace passes an object list with the same object appearing more
than once, we end up hitting the BUG_ON() in
i915_gem_object_set_to_gpu_domain() as it gets called a second time
for the same object.
Signed-off-by: Kristian Høgsberg <krh@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
This could be triggered by a client asking to emit an irq when the device
wasn't initialized.
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
This is done by
1) Wake up lock waiters when we close the master file descriptor.
Not when the master structure is removed, since the latter
requires the waiters themselves to release the refcount on the
master structure -> Deadlock.
2) Send a SIGTERM to all clients waiting for the lock.
Normally these clients will get a SIGPIPE when the X server dies,
but clients may also spin trying to grab the DRM lock, without
getting any sort of notification.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>
Currently only one waiter is woken up, leaving other waiters
hanging waiting for the DRM lock.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>
That return code is for in-kernel use only.
Use EINTR instead.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>
Impact: standardize IO on cached ops
On modern CPUs it is almost always a bad idea to use non-temporal stores,
as the regression in this commit has shown it:
30d697f: x86: fix performance regression in write() syscall
The kernel simply has no good information about whether using non-temporal
stores is a good idea or not - and trying to add heuristics only increases
complexity and inserts fragility.
The regression on cached write()s took very long to be found - over two
years. So dont take any chances and let the hardware decide how it makes
use of its caches.
The only exception is drivers/gpu/drm/i915/i915_gem.c: there were we are
absolutely sure that another entity (the GPU) will pick up the dirty
data immediately and that the CPU will not touch that data before the
GPU will.
Also, keep the _nocache() primitives to make it easier for people to
experiment with these details. There may be more clear-cut cases where
non-cached copies can be used, outside of filemap.c.
Cc: Salman Qazi <sqazi@google.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
io_mapping_create_wc can return NULL on error and io_mapping_free() should be
called on one of the error-cleanup path.
Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Dave Airlie <airlied@redhat.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: Eric Anholt <eric@anholt.net>
Cc: Keith Packard <keithp@keithp.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup, enable future change
Add a 'total bytes copied' parameter to __copy_from_user_*nocache(),
and update all the callsites.
The parameter is not used yet - architecture code can use it to
more intelligently decide whether the copy should be cached or
non-temporal.
Cc: Salman Qazi <sqazi@google.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
However we still have another issue with ioremap_wc not falling back
properly or somehow doing something else stupid, this probably needs
to be tracked down.
Signed-off-by: Dave Airlie <airlied@redhat.com>
edid->revision == 0 should be valid (at least, so the error message
indicates. :) and wikipedia seems to indicate that EDID 1.0 existed.
We can dump the entire check, since edid->revision is a u8, so
it can't ever be less than 0.
Marko reports in RH bz#476735 that his monitor claims to be
EDID 1.0, and therefore hits the check and is stuck at 800x600 because
of it.
Reported-by: Marko Ristola <marko.ristola@kolumbus.fi>
Signed-off-by: Kyle McMartin <kyle@redhat.com>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Dave Airlie <airlied@linux.ie>
The first time we install a mode, the vblank will be disabled for a pipe
and so drm_vblank_get() in drm_vblank_pre_modeset() will fail. As we
unconditionally call drm_vblank_put() afterwards, the vblank reference
counter becomes unbalanced.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Dave Airlie <airlied@linux.ie>
In some cases we may receive a mode config that has a different
CRTC<->encoder map that the current configuration. In that case, we
need to disable any re-routed encoders before setting the mode,
otherwise they may not pick up the new CRTC (if the output types are
incompatible for example).
Tested-by: Kristian Høgsberg <krh@bitplanet.net>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Dave Airlie <airlied@linux.ie>
We've seen cases in the wild where the VBT sync data is wrong, so add
some code to fix it up in that case, taking care to make sure that the
total is greater than the sync end.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Dave Airlie <airlied@linux.ie>
These are normal; we walk through different values looking for the right
one, so why flood the screen with messages?
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
In the KMS case, enter/leavevt won't fix up the interrupt handler for
us, so we need to do it at suspend/resume time. Make sure we don't fail
the resume if the chip is hung either.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
dev_priv->hw_status_page can be NULL, if i915_gem_retire_requests()
is called from i915_gem_busy_ioctl().
Signed-off-by Karsten Wiese <fzu@wemgehoertderstaat.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
there might be a nicer way to fix this but this is the simplest for now.
Signed-off-by: Pierre Willenbrock <pierre@pirsoft.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The object is dereferenced before the NULL check. Oops.
Fixes http://bugs.freedesktop.org/show_bug.cgi?id=20235
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This ensures that the user gets the latest information from the hardware
on whether the buffer is busy, potentially reducing the working set of objects
that the user chooses.
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
In the KMS case, we need to suspend/resume GEM as well. So on suspend, make
sure we idle GEM and stop any new rendering from coming in, and on resume,
re-init the framebuffer and clear the suspended flag.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The problem was that object_set_to_gpu_domain would set the new write_domains
that are getting set by this batchbuffer, then the accumulated flushes required
for all the objects in preparation for this batchbuffer were posted, and the
brand new write domain would get cleared by the flush being posted. Instead,
hang on to the new (or old if we're not changing it) value and set it after
the flush is queued.
Results from this noticably included conformance test failures from reads
shortly after writes (where the new write domain had been lost and thus not
flushed and waited on), but is a suspected cause of hangs in some apps when
a write domain is lost on a buffer that gets reused for instruction or
commmand state.
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
While not strictly required, it helped while thinking about the following
change. This change should be invariant.
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This fixes a regression reported in bug #12613.
[airlied: not I tweaked the patch slightly and fixed it by etienne did
all the hardwork so gets authorship]
Signed-off-by: etienne <etienne.basset@numericable.fr>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This fixes potential fault at fault time if the object was unreferenced
while the mapping still existed. Now, while the mmap_offset only lives
for the lifetime of the object, the object also stays alive while a vma
exists that needs it.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Need to do this in case the unref ends up doing a free.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Lifted from the DDX modesetting.
Signed-off-by: Kristian Høgsberg <krh@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
They used to be different. Now they're identical.
Signed-off-by: Kristian Høgsberg <krh@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
David Miller