Commit Graph

24 Commits

Author SHA1 Message Date
Thomas Hellstrom
8bf445cee3 vmwgfx: Break out and comment vmw_execbuf_copy_fence_user
This function will be used also by the upcoming fence event code,
so break it out and add a comment about the functionality.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-10-10 15:46:28 +01:00
Thomas Hellstrom
5bb39e8181 vmwgfx: Handle device surface memory limit
Make surfaces swappable. Make sure we honor the maximum amount of surface
memory the device accepts. This is done by potentially reading back surface
contents not used by the current command submission and storing it
locally in buffer objects.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-10-05 10:17:31 +01:00
Thomas Hellstrom
f18c8840be vmwgfx: Optimize the command submission resource list
Use a list for resources referenced during command submission, instead of
an array.
As long as we don't implement parallell command submission this works fine
and simplifies things a bit.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-10-05 10:17:23 +01:00
Thomas Hellstrom
e2fa3a7683 vmwgfx: Fix up query processing
Previously, query results could be placed in any buffer object, but since
we didn't allow pinned buffer objects, query results could be written when
that buffer was evicted, corrupting data in other buffers.

Now, require that buffers holding query results are no more than two pages
large, and allow one single pinned such buffer. When the command submission
code encounters query result structures in other buffers, the queries in the
pinned buffer will be finished using a query barrier for the last hardware
context using the buffer. Also if the command submission code detects
that a new hardware context is used for queries, all queries of the previous
hardware context is also flushed. Currently we use waiting for a no-op
occlusion query as a query barrier for a specific context.

The query buffer is also flushed and unpinned on context destructions,
master drops and before scanout bo placement.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-10-05 10:17:22 +01:00
Jakob Bornecrantz
0cff60c625 vmwgfx: Disallow user space to send present and readback commands
Signed-off-by: Jakob Bornecrantz <jakob@vmware.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-10-05 10:17:18 +01:00
Jakob Bornecrantz
4084fb89e6 vmwgfx: Expand the command checker to cover screen object commands
Signed-off-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-10-05 10:17:12 +01:00
Thomas Hellstrom
922ade0d16 vmwgfx: Break out execbuf command processing
This will make it easier to execute commands operating on user-space
resources but generated by the kernel.

JB: Added tracking if the sw_context was called from the kernel or userspace.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Jakob Bornecrantz <jakob@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-10-05 10:17:11 +01:00
Jakob Bornecrantz
6070e9fa6d vmwgfx: Some comments and BUG_ON
Signed-off-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-10-05 10:17:10 +01:00
Thomas Hellstrom
2ae7b03c26 vmwgfx: Bump major
This bumps driver major version as a result of previous incompatible
interface changes.

In addition, a leftover command definition is removed from the
vmwgfx_drm.h header.

Also a strict version check is enforced on the exebuf ioctl.

This is intended to be the last major bump before exiting staging.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-09-06 11:51:12 +01:00
Thomas Hellstrom
ae2a104058 vmwgfx: Implement fence objects
Will be needed for queries and drm event-driven throttling.

As a benefit, they help avoid stale user-space fence handles.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-09-06 11:51:11 +01:00
Thomas Hellstrom
6bcd8d3c78 vmwgfx: Fix confusion caused by using "fence" in various places
This is needed before we introduce the fence objects.
Otherwise this will be even more confusing. The plan is to use the following:

seqno: A 32-bit sequence number that may be passed in the fifo.
marker: Objects, carrying a seqno, that track fifo submission time. They
are used for fifo lag based throttling.
fence objects: Kernel space objects, possibly accessible from user-space and
carrying a 32-bit seqno together with signaled status.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-09-06 11:48:40 +01:00
Thomas Hellstrom
be38ab6ea7 vmwgfx: Fix potential execbuf deadlocks
Perform all command stream validation in a bounce buffer separate from the
fifo. This makes the fifo available to all validation-generated commands,
which would otherwise attempt to grab the fifo recursively, causing a
deadlock. This is in preparation for GMR2 and swappable surfaces.

Also maintain references to all surfaces in the command stream until the
command stream has been fired in order to avoid racing with surface
destruction taking place after validation but before submission.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-09-01 09:37:55 +01:00
Marek Olšák
dfadbbdb57 drm/ttm: add a way to bo_wait for either the last read or last write
Sometimes we want to know whether a buffer is busy and wait for it (bo_wait).
However, sometimes it would be more useful to be able to query whether
a buffer is busy and being either read or written, and wait until it's stopped
being either read or written. The point of this is to be able to avoid
unnecessary waiting, e.g. if a GPU has written something to a buffer and is now
reading that buffer, and a CPU wants to map that buffer for read, it needs to
only wait for the last write. If there were no write, there wouldn't be any
waiting needed.

This, or course, requires user space drivers to send read/write flags
with each relocation (like we have read/write domains in radeon, so we can
actually use those for something useful now).

Now how this patch works:

The read/write flags should passed to ttm_validate_buffer. TTM maintains
separate sync objects of the last read and write for each buffer, in addition
to the sync object of the last use of a buffer. ttm_bo_wait then operates
with one the sync objects.

Signed-off-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-08-31 19:25:35 +01:00
Thomas Hellstrom
6570596202 drm/ttm/vmwgfx: Have TTM manage the validation sequence.
Rather than having the driver supply the validation sequence, leave that
responsibility to TTM. This saves some confusion and a function argument.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2010-11-22 13:25:21 +10:00
Kulikov Vasiliy
dccb2a952b drm: vmwgfx: fix information leak to userland
Structure drm_vmw_fence_rep is copied to userland with field "pad64"
uninitialized.  It leads to leaking of contents of kernel stack memory.

Signed-off-by: Vasiliy Kulikov <segooon@gmail.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2010-11-09 13:42:15 +10:00
Thomas Hellstrom
135cba0dc3 vmwgfx: Implement a proper GMR eviction mechanism
Use Ben's new range manager hooks to implement a manager for
GMRs that manages ids rather than ranges.
This means we can use the standard TTM code for binding, unbinding and
eviction.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2010-10-27 11:07:46 +10:00
Dan Carpenter
9b8eb4d147 drm/vmwgfx: return -EFAULT for copy_to_user errors
copy_to/from_user() returns the number of bytes remaining to be copied
but we want to return a negative error code here.  This gets returned to
userspace.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2010-06-08 09:33:41 +10:00
Thomas Hellstrom
1925d45658 drm/vmwgfx: Add kernel throttling support. Bump minor.
The throttle_us member in the execbuf argument is now honored.
If the member is 0, no waiting for lag will occur, which
guarantees backwards compatibility with well-behaved clients.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2010-06-01 09:37:15 +10:00
Jerome Glisse
9d87fa2138 drm/ttm: split no_wait argument in 2 GPU or reserve wait
There is case where we want to be able to wait only for the
GPU while not waiting for other buffer to be unreserved. This
patch split the no_wait argument all the way down in the whole
ttm path so that upper level can decide on what to wait on or
not.

[airlied: squashed these 4 for bisectability reasons.]
drm/radeon/kms: update to TTM no_wait splitted argument
drm/nouveau: update to TTM no_wait splitted argument
drm/vmwgfx: update to TTM no_wait splitted argument
[vmwgfx patch: Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>]

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Acked-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2010-04-08 10:21:19 +10:00
Thomas Hellstrom
4e4ddd4777 drm/vmwgfx: Fix queries if no dma buffer thrashing is occuring.
Intercept query commands and apply relocations to their guest pointers.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2010-02-23 15:42:36 +10:00
Thomas Hellstrom
8ba5152a3a drm/vmwgfx: Optimize memory footprint for DMA buffers.
Use VRAM whenever there is free space for DMA buffers,
but use system GMR memory if using VRAM would cause an eviction.

This significantly reduces the guest system memory usage for
VMs with a large amount of VRAM allocated.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2010-01-25 16:04:39 +10:00
Thomas Hellstrom
7a73ba7469 drm/vmwgfx: Use TTM handles instead of SIDs as user-space surface handles.
Improve the command verifier to catch all occurences of surface handles,
and translate to SIDs.

This way DMA buffers and 3D surfaces share a common handle space,
which makes it possible for the kms code to differentiate.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-12-23 10:06:24 +10:00
Thomas Hellstrom
3d3a5b3290 drm/vmwgfx: Return -ERESTARTSYS when interrupted by a signal.
Fixes for TTM API change.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-12-23 10:05:47 +10:00
Jakob Bornecrantz
fb1d9738ca drm/vmwgfx: Add DRM driver for VMware Virtual GPU
This commit adds the vmwgfx driver for the VWware Virtual GPU aka SVGA.
The driver is under staging the same as Nouveau and Radeon KMS. Hopefully
the 2D ioctls are bug free and don't need changing, so that part of the
API should be stable. But there there is a pretty big chance that the 3D API
will change in the future.

Signed-off-by: Thomas Hellström <thellstrom@vmware.com>
Signed-off-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2009-12-15 08:38:43 +10:00