b0473699ed
ttm: returning into dummy causes a buffer object leak
...
as nobody ever derefs dummy, however not returning does the deref
correctly.
2007-10-02 15:48:28 +10:00
c4b3a0f602
Merge branch 'master' into pre-superioctl-branch
...
Conflicts:
linux-core/drm_bo.c
linux-core/drm_fence.c
linux-core/drm_objects.h
shared-core/drm.h
2007-09-25 18:03:31 +02:00
bb5516f4f4
drm/ttm: fixup fence class naming and interfaces
...
This is some code for nouveau that Ben Skeggs worked on, and also
fixes the naming (having class in a system header file == C++ keyword == bad plan)
2007-09-25 16:17:17 +10:00
54df1b9ff3
Fix pinned buffer fence class.
2007-09-22 14:30:55 +02:00
0774090d5b
Fix drm_bo.c compiling.
2007-09-22 13:59:56 +02:00
da63f4ba0f
Add fence error member.
...
Modify the TTM backend bind arguments.
Export a number of functions needed for driver-specific super-ioctls.
Add a function to map buffer objects from the kernel, regardless of where they're
currently placed.
A number of error fixes.
2007-09-22 13:57:13 +02:00
0055fd5c35
Merge branch 'master' into bo-set-pin
2007-08-16 09:23:09 -07:00
92084c6e05
Export some useful ttm functions to drivers.
2007-08-06 22:11:18 +10:00
3a0bc518e3
Remove the pinned buffer from the LRU when pinning.
...
Also, be a little safer with setting the pinned flag within the struct lock.
I'm not 100% sure if this is required, but it seems like it might be.
2007-08-02 14:08:04 -07:00
cf4f1a85af
Add a couple of doxygen comments from reading the code.
2007-08-02 13:51:55 -07:00
f9c27aa50b
Copy the important parts of object_validate into object_create().
...
This should let us allocate buffers without holding the hardware lock.
While here, add DRM_DEBUG info for the drm_bo ioctls, so you can see something
more specific than just the cmd value per ioctl.
2007-07-26 11:18:27 -07:00
cf2d569dac
Replace NO_MOVE/NO_EVICT flags to buffer objects with an ioctl to set pinning.
...
This cleans up the create/validate interfaces for this very uncommon path, and
makes pinned object creation much easier to use for the X Server.
2007-07-26 10:15:11 -07:00
4175dd8181
drm/bo: set the req pointer for each buffer to validate
2007-07-26 15:26:36 +10:00
e2d00715cc
drm: fix size argument for copy to/from user
2007-07-26 13:26:04 +10:00
5b38e13416
Replace DRM_IOCTL_ARGS with (dev, data, file_priv) and remove DRM_DEVICE.
...
The data is now in kernel space, copied in/out as appropriate according to the
This results in DRM_COPY_{TO,FROM}_USER going away, and error paths to deal
with those failures. This also means that XFree86 4.2.0 support for i810 DRM
is lost.
2007-07-20 18:16:42 -07:00
c1119b1b09
Replace filp in ioctl arguments with drm_file *file_priv.
...
As a fallout, replace filp storage with file_priv storage for "unique
identifier of a client" all over the DRM. There is a 1:1 mapping, so this
should be a noop. This could be a minor performance improvement, as everything
on Linux dereferenced filp to get file_priv anyway, while only the mmap ioctls
went the other direction.
2007-07-20 13:39:45 -07:00
6ad1df2176
drm: remove drm_u64_t, replace with uint64_t everwhere
...
This might break something, stdint.h inclusion in drm.h maybe required
but I'm not sure yet what platforms have it what ones don't.
2007-07-18 09:42:06 +10:00
191c062933
drm: remove drm_ref_t
2007-07-16 13:45:39 +10:00
be85ad0333
drm: detypedef ttm/bo/fence code
2007-07-16 13:37:02 +10:00
6dce9e0735
drm: remove hashtab/sman and object typedefs
2007-07-16 12:48:44 +10:00
21ee6fbfb8
drm: remove drmP.h internal typedefs
2007-07-16 12:32:51 +10:00
1a07256d60
drm: remove ttm userspace typedefs
2007-07-16 11:30:53 +10:00
2c9e05cf4c
Merge branch 'master' into cleanup
...
Conflicts:
libdrm/xf86drm.c
linux-core/drm_bo.c
linux-core/drm_fence.c
2007-07-11 11:23:41 +10:00
00f1a66f22
Fence object reference / dereference cleanup.
...
Buffer object dereference cleanup.
Add a struct drm_device member to fence objects:
This can simplify code, particularly in drivers.
2007-06-29 12:58:45 +02:00
3ee31a1f35
Indentation fixes.
2007-06-15 10:31:32 +02:00
d34b2c7b9e
Fix refcounting / lock race.
...
Reported by Steve Wilkins / Michel Dnzer.
2007-06-15 10:26:51 +02:00
e1b8eabeee
Locking fixes and instrumentation.
2007-06-15 10:26:51 +02:00
62082ab3e6
Make sure we read fence->signaled while spinlocked.
2007-06-13 15:38:59 +02:00
5156f1c897
Fix fence object deref race.
2007-06-13 15:19:30 +02:00
f984b1b8d1
Fix some obvious bugs.
2007-06-12 12:30:33 +02:00
b6b5df24b9
Try to make buffer object / fence object ioctl args 64-bit safe.
...
Introduce tile members for future tiled buffer support.
Allow user-space to explicitly define a fence-class.
Remove the implicit fence-class mechanism.
64-bit wide buffer object flag member.
2007-06-12 12:21:38 +02:00
58b2ed7832
Revert "drm/ttm: cleanup mm_ioctl ioctls to be separate ioctls."
...
This reverts commit 3fdef0dc20
2007-05-26 03:48:08 +10:00
3fdef0dc20
drm/ttm: cleanup mm_ioctl ioctls to be separate ioctls.
...
This is the first bunch of ioctls
2007-05-26 03:32:34 +10:00
b2a875ba89
ttm: complete drm buffer object ioctl split
...
retain the op operation for validate/fence operations
2007-05-08 18:25:15 +10:00
25c51f539f
drm/ttm: ioctl cleanup for buffer object - user side only
...
This just cleans up the xf86drm.c to what I want and drm.h,
I need to fix up the kernel internals to suit these changes now.
I've moved to using struct instead of typedefs for the bo and it doesn't look
that bad so I'll do the same thing for mm and fence..
2007-05-08 17:53:58 +10:00
ee8954cb53
drm/ttm: cleanup mm_ioctl ioctls to be separate ioctls.
...
This is the first bunch of ioctls
2007-05-06 11:17:30 +10:00
71d44cda52
drm/ttm: allow drm_buffer_object_create to be used in-kernel
2007-04-26 16:09:24 +10:00
3c384a9ad5
Add new buffer object type for kernel allocations that don't initially have a user mapping.
...
(cherry picked from commit 2e21779992
2007-04-26 16:04:09 +10:00
5a96d59ce9
Don't always free up memory space when we unpin buffers.
2007-04-17 14:15:37 +02:00
5432cc4abf
Fix buffer object reference problems.
...
(Reported by Dave Airlie).
2007-04-17 10:53:19 +02:00
e91ceff6c9
Add a code comment.
2007-04-17 08:46:45 +02:00
e6e4946c82
Require the hardware lock for buffer creation
...
(since that implies a validate).
Fix drm_bo_wait_unfenced error messages and codes.
Fix some return codes from libdrm.
2007-04-16 16:23:05 +02:00
7743af9449
Evicted no-move buffers can get lost if they end up in another
...
memory type than local.
2007-04-03 10:29:14 +02:00
d85e243259
Fix an oops when trying to clean a not yet initialized memory type.
2007-04-03 10:29:14 +02:00
3f70518f0b
drm/bo: avoid oops if the memory manager for this type isn't initialised
2007-03-29 09:25:04 +10:00
2463b03cb4
whitespace cleanup pending a kernel merge
2007-03-19 08:23:43 +11:00
e0f53e59be
Simple fence object sample driver for via, based on idling the GPU.
...
Buffer object driver for via.
Some changes to buffer object driver callbacks.
Improve fence flushing.
2007-02-16 20:22:24 +01:00
7766378d97
Initial support for fence object classes.
...
(Fence objects belonging to different command submission mechanisms).
2007-02-15 12:11:38 +01:00
9efdae317c
More bugfixes.
...
Fixed memory, pinned buffers and unmappable memory now seems
fully functional.
2007-02-13 20:05:32 +01:00
398913dc0e
Lindent.
2007-02-12 20:34:50 +01:00
b0c5339ed6
More bugfixes.
2007-02-12 20:32:03 +01:00
f02f83ee08
Cleanup and fix support for pinned buffers.
2007-02-12 17:47:57 +01:00
85ee2a8d04
Various bugfixes.
2007-02-10 12:06:36 +01:00
57df398072
Reinstate some LRU handling.
2007-02-09 12:43:18 +01:00
d32b21e016
Remove some code that should have gone in
...
commit 6a49d9a8ab
2007-02-09 00:11:53 +01:00
99acdaee48
Fix copyright statements.
2007-02-09 00:07:29 +01:00
6a49d9a8ab
Fix evict_mutex locking range.
...
Implement unmappable buffers. (fault moves them to mappable when needed).
Various bugfixes.
2007-02-09 00:02:02 +01:00
b2bcbf874b
Add an accelerated buffer copy cleanup helper.
...
Export helper functions and make some important buffer-object functions non-static.
Add an i915 accelerated blit buffer move for pci memory buffers.
2007-02-08 21:28:33 +01:00
a0ed808d05
Don't create a ttm just to copy from.
2007-02-08 19:06:39 +01:00
e4b2da4406
A minor function interface change and some memcpy bugfixing.
...
Hooray!! it sort of works with a fixed AGP area as faked VRAM.
2007-02-08 16:21:38 +01:00
1257907fa9
Simplify external ttm page allocation.
...
Implement a memcpy fallback for copying between buffers.
2007-02-08 13:29:08 +01:00
09984ad77b
Update memory compatibility tests.
...
Now only pinned buffers are broken.
2007-02-08 11:55:24 +01:00
c1fbd8a566
Checkpoint commit.
...
Flag handling and memory type selection cleanup.
glxgears won't start.
2007-02-07 17:25:13 +01:00
af24465b2e
Fix a stray unlock_kernel() in drm_vm.c
...
Add a file for memory move helpers, drm_bo_move.c
Implement generic memory move.
Cached, no_move and unmapped memory temporarily broken.
2007-02-07 12:52:23 +01:00
71b9e876f9
Simplify pci map vs no pci map choice.
2007-02-06 16:59:45 +01:00
40ce53dfde
Implement a drm_mem_reg_t substructure in the buffer object type.
2007-02-06 15:56:43 +01:00
609e3b0375
Implement a policy for selecting memory types.
2007-02-06 14:20:33 +01:00
63f2abd721
Make also later kernels work with buffer object vm
...
and clean up some function names.
2007-02-02 19:49:11 +01:00
c269d560e4
Make vm handle buffer objects instead of ttm objects.
...
Remove ttm objects.
Make vm aware of PCI memory type buffer objects.
(Only works for pre 2.6.16 kernels for now).
2007-02-02 14:47:44 +01:00
3024f23c65
memory manager: Make device driver aware of different memory types.
...
Memory types are either fixed (on-card or pre-bound AGP) or not fixed
(dynamically bound) to an aperture. They also carry information about:
1) Whether they can be mapped cached.
2) Whether they are at all mappable.
3) Whether they need an ioremap to be accessible from kernel space.
In this way VRAM memory and, for example, pre-bound AGP appear
identical to the memory manager.
This also makes support for unmappable VRAM simple to implement.
2007-01-31 14:50:57 +01:00
36d50687dd
Fix an error-path oops.
2007-01-31 11:03:53 +01:00
d399fcf46f
Add a buffer object transfer function.
...
Creates a placeholder for the old buffer contents
when it is transfered to / from static memory like VRAM.
2007-01-30 16:20:23 +01:00
9bbdc0fb10
Clean up buffer object destruction somewhat.
2007-01-30 12:35:49 +01:00
9a654e71bd
Use pre-defined list_splice function.
2007-01-29 13:37:02 +01:00
45418bb1b1
s/buf/bo/ for consistency.
2007-01-29 13:37:02 +01:00
1e4c7d69f5
Some cleanup. A buffer object should only have one active memory type.
2007-01-29 13:37:02 +01:00
2dcbf6a599
make build against 2.6.20 hopefully
2007-01-01 11:30:38 +11:00
f6d5fecdd2
Last minute changes to support multi-page size buffer offset alignments.
...
This will come in very handy for tiled buffers on intel hardware.
Also add some padding to interface structures to allow future binary backwards
compatible changes.
2006-10-27 11:28:37 +02:00
e8ba62db72
Make sure delayed delete list is empty on lastclose.
...
Fix some refcounting errors.
Fix some error messages.
2006-10-19 16:58:00 +02:00
c34faf224b
Remove max number of locked pages check and call, since
...
that is now handled by the memory accounting.
2006-10-17 20:03:26 +02:00
89b9441798
Lindent.
2006-10-17 19:57:06 +02:00
d515936ea7
Add memory usage accounting to avoid DOS problems.
2006-10-17 19:40:57 +02:00
5443dbe35f
Implement mm_lock and mm_unlock functions.
...
The mm_lock function is used when leaving vt. It evicts _all_ buffers.
Buffers with the DRM_BO_NO_MOVE attribute set will be guaranteed to
get the same offset when / if they are rebound.
2006-10-17 16:00:25 +02:00
db5c671e86
Remove the memory manager parameter from the put_block function, as this
...
makes the client code a lot cleaner. Prepare buffer manager for lock and
unlock calls.
2006-10-17 11:28:48 +02:00
5881ce1b91
Extend generality for more memory types.
...
Fix up init and destruction code.
2006-10-17 11:05:37 +02:00
540c64c378
Bugfixes.
2006-10-12 16:10:47 +02:00
10150df02b
Simplify the AGP backend interface somewhat.
...
Fix buffer bound caching policy changing, Allow
on-the-fly changing of caching policy on bound buffers if the hardware
supports it.
Allow drivers to use driver-specific AGP memory types for TTM AGP pages.
Will make AGP drivers much easier to migrate.
2006-10-12 12:09:16 +02:00
3070389367
Compatibility code for 2.6.15-2.6.18. It is ugly but a little comfort is that
...
it will go away in the mainstream kernel.
Some bugfixes, mainly in error paths.
2006-10-11 22:21:01 +02:00
f2db76e2f2
Big update:
...
Adapt for new functions in the 2.6.19 kernel.
Remove the ability to have multiple regions in one TTM.
This simplifies a lot of code.
Remove the ability to access TTMs from user space.
We don't need it anymore without ttm regions.
Don't change caching policy for evicted buffers. Instead change it only
when the buffer is accessed by the CPU (on the first page fault).
This tremendously speeds up eviction rates.
Current code is safe for kernels <= 2.6.14.
Should also be OK with 2.6.19 and above.
2006-10-11 13:40:35 +02:00
3802f9adbf
Fix buffer manager takedown error.
...
Prepare for the possibility to evict all buffers from vram / agp.
This will be used by the X server when, for example, switching vts.
2006-09-29 11:15:59 +02:00
f2c03ecae6
Fix racy buffer object destruction.
2006-09-27 19:07:55 +02:00
c97149b45b
Fix tt fixed size that slipped through in previous commit.
2006-09-27 09:31:39 +02:00
235f6fc650
Adapt to architecture-specific hooks for gatt pages.
2006-09-27 09:27:31 +02:00
273eb7833d
Add /proc filesystem buffer / fence object accounting.
...
Check for NULL pointer in the i915 flush handler.
Remove i915_sync_flush declaration.
2006-09-25 11:51:08 +02:00
ca1b15d645
Alternative implementation of page table zeroing using zap page_range.
...
(Disabled for now)
Fix bo_wait_idle bug.
Remove stray debug message.
2006-09-18 20:43:31 +02:00
f613022cee
Allow a "native type" to be associated with a fence sequence.
...
In the intel case, we can associate a flush with a sequence.
2006-09-15 16:47:09 +02:00
49fbeb339c
Some bugfixes.
...
Change the fence object interface somewhat to allow some more flexibility.
Make list IOCTLS really restartable.
Try to avoid busy-waits in the kernel using immediate return to user-space with an -EAGAIN.
2006-09-15 11:18:35 +02:00
9adc9584a7
Fix some debug messages.
2006-09-12 17:39:44 +02:00
861b26578c
Use lazy fence wait when possible even for RW fences. Saves some CPU.
...
Lindent.
2006-09-12 16:28:34 +02:00
191e284709
More bugfixes.
...
Disable the i915 IRQ turnoff for now since it seems to be causing problems.
2006-09-12 12:01:00 +02:00
Dave Airlie
Thomas Hellstrom
Dave Airlie
Eric Anholt
Ben Skeggs
Jesse Barnes