Commit Graph

158 Commits (500c81d194115fb3c4b97d742519689478eeb4e8)

Author SHA1 Message Date
Keith Packard 6cd0ef06a6 [intel] remove settable use_mi_batchbuffer_start
The driver can know what hardware requires MI_BATCH_BUFFER vs
MI_BATCH_BUFFER_START; there's no reason to let user mode configure this.
2008-06-06 13:26:03 -07:00
Keith Packard 9f46c6935d [intel-gem] Use timers to retire requests periodically.
Without the user IRQ running constantly, there's no wakeup when the ring
empties to go retire requests and free buffers. Use a 1 second timer to make
that happen more often.
2008-06-06 13:00:47 -07:00
Keith Packard a708106c77 [intel] free the hardware status page at driver_unload
This goes with the other hardware status page patch.
2008-06-06 13:00:46 -07:00
Keith Packard 56a96841d0 [intel-gem] Add explicit throttle ioctl
Instead of throttling and execbuffer time, have the application ask to
throttle explicitly. This allows the throttle to happen less often, and
without holding the DRM lock.
2008-06-06 13:00:46 -07:00
Keith Packard 5f5badb26f [intel] Allocate hardware status page at driver load time
I couldn't get the re-allocated HWS to work on my 965GM, so I just gave up
and made it persist across the lifetime of the driver instead.
2008-06-06 13:00:46 -07:00
Keith Packard 84162ccb7d Ignore X server provided mmio address 2008-06-06 13:00:46 -07:00
Eric Anholt 461bfa3da6 Merge commit 'origin/master' into drm-gem
Conflicts:

	linux-core/Makefile.kernel
	shared-core/i915_drv.h
	shared-core/nouveau_state.c
2008-05-30 14:42:08 -07:00
Robert Noland 8cd045079e [FreeBSD] Add vblank-rework support and get drivers building.
The i915 driver now works again.
2008-05-27 14:25:20 -07:00
Keith Packard 7cf3fd29fe [intel-gem] Add DRM_I915_GEM_BUSY ioctl to check for idle buffers.
This new ioctl returns whether re-using the buffer would force a wait.
2008-05-25 20:45:20 -07:00
Keith Packard a51c3a76ff [intel] Add debug code to verify the cached ring tail pointer.
Recording the tail pointer in a local variable improves performance, but if
someone messes up and fails to reload at the right time, the driver will
write commands to the wrong part of the ring and scramble execution badly.

This change (available by setting I915_RING_VALIDATE to 1) checks to make
sure the cached tail pointer matches the hardware tail pointer at each ring
buffer addition, calling BUG_ON when that's not true.
2008-05-22 22:00:21 -07:00
Eric Anholt 5e662f90d1 [gem] Release GEM buffers from work task scheduled from IRQ.
There are now 3 lists.  Active is buffers currently in the ringbuffer.
Flushing is not in the ringbuffer, but needs a flush before unbinding.
Inactive is as before.  This prevents object_free → unbind →
wait_rendering → object_reference and a kernel oops about weird refcounting.

This also avoids an synchronous extra flush and wait when freeing a buffer
which had a write_domain set (such as a temporary rendered to and then from
using the 2d engine).  It will sit around on the flushing list until the
appropriate flush gets emitted, or we need the GTT space for another
operation.
2008-05-22 22:00:21 -07:00
Eric Anholt af8e087157 [gem] Use a separate sequence number field from classic/ttm
This lets us get some qualities we desire, such as using the full 32-bit
range (except zero), avoiding DRM_WAIT_ON, and a 1:1 mapping of active
sequence numbers to request structs, which will be used soon for throttling
and interrupt-driven list cleanup.
2008-05-20 14:16:26 -07:00
Eric Anholt 3ab152da66 [gem] Rename the GTT LRU lists to active (executing) and inactive (idle). 2008-05-15 11:59:58 -07:00
Keith Packard 7e7ea313c4 [intel] When polling for ring space, sleep for a lot longer (10ms)
If the ring is full, the engine will surely be running for more than 10ms.
2008-05-12 13:03:16 -07:00
Eric Anholt c5c59eab80 GEM: Separate the LRU into execution list and LRU list.
Now, the LRU list has objects that are completely done rendering and ready
to kick out, while the execution list has things with active rendering,
which have associated cookies and reference counts on them.
2008-05-09 17:38:32 -07:00
Eric Anholt d2373b2a34 GEM: Use irq-based fencing rather than syncing and evicting every exec. 2008-05-06 13:28:26 -07:00
Keith Packard ed6657fa8e Monitor ACTHD register while polling for idle ring.
When batch buffers are executing, the ring may be stuck for a long time.
Monitor the ACTHD pointer which will show if the execution engine is
actually hung.
2008-05-05 22:09:34 -07:00
Keith Packard d59a9300ec Remove some debug messages. 2008-05-05 14:32:01 -07:00
Keith Packard 4511e6cd80 Correct execbuffer offset. Add memory barrier and chipset flush. 2008-05-05 11:27:06 -07:00
Keith Packard b6f173c430 Add i915_dispatch_gem_execbuffer (broken).
This function submits a gem-based execbuffer to the ring.
It doesn't work yet.
2008-05-05 10:51:49 -07:00
Eric Anholt 793549116e Add pin/unpin object ioctls for gem. 2008-05-01 15:40:02 -07:00
Eric Anholt 2140e102f9 checkpoint: rename to GEM and a few more i915 bits. 2008-05-01 11:39:20 -07:00
Eric Anholt 1a84067950 Hacking towards hooking up execbuffer. 2008-04-30 16:03:15 -07:00
Dave Airlie ce8c842518 i915: gfx hw and i945gme fixes from upstream
From Jesse and Zhenyu originally.
2008-04-22 16:08:17 +10:00
Dave Airlie b0817a42e7 i915: fix oops on agp=off
Kernel bug 10289.
2008-03-24 18:52:26 +10:00
Thomas Hellstrom 88bd1e4a35 Merge branch 'intel-post-reloc'
Conflicts:

	linux-core/drm_compat.c
	linux-core/drm_compat.h
	linux-core/drm_ttm.c
	shared-core/i915_dma.c

Bump driver minor to 13 due to introduction of new
relocation type.
2008-03-12 11:34:29 +01:00
Thomas Hellstrom 8a18d123f5 Avoid large kmallocs. 2008-03-12 09:49:27 +01:00
Dave Airlie 12574590cd drm: reorganise minor number handling using code from modesetting branch
Rip out the whole head thing and replace it with an idr and drm_minor
structure.
2008-03-06 05:21:50 +10:00
Xiang, Haihao 638353103d i915: Evict if relocatee buffer is CACHED_MAPPED before
writting relocations, otherwise the GPU probably sees some
inconsistent data. Fix fd.o bug#14656
2008-03-05 15:09:17 +08:00
Thomas Hellstrom 2305100c0f More post-ioctl work. 2008-02-29 13:25:55 +01:00
Thomas Hellstrom fd595fa4dc Reinstate buffer idle before applying relocations. 2008-02-27 21:44:40 +01:00
Thomas Hellstrom 72983ff301 Don't wait for buffer idle before applying relocations. 2008-02-27 19:46:28 +01:00
Thomas Hellstrom e87cec1968 [i915] Relocation fixes. 2008-02-26 10:47:05 +01:00
Thomas Hellstrom 56bb29cf37 Make the execbuffer code reasonably safe against errors.
In particular -EAGAINs, which should be common during Xserver operation.
Also handle the fence creation failure case.
2008-02-26 00:05:26 +01:00
Kristian Høgsberg b7086e6ae5 Fix one last occurance of struct _drm_i915_batchbuffer.
Thanks to Todd Merrill for pointing it out.
2008-02-22 11:22:52 -05:00
Kristian Høgsberg 373dbcf8b2 i915: Add a dri2 init path that gets the lock from the dri2 sarea. 2008-02-13 13:34:02 -05:00
Kristian Høgsberg db3f03ae35 i915: Only look up dev_priv->mmio_map if it's not already set up 2008-02-13 13:34:02 -05:00
Kristian Høgsberg ee15459483 i915: Add I915_PARAM_CHIPSET_ID param to get chipset ID. 2008-02-13 13:34:02 -05:00
Kristian Høgsberg 4feb0638f1 i915: Make sarea_priv setup optional. 2008-02-13 13:34:02 -05:00
Thomas Hellstrom 76748efae2 i915: Re-report breadcrumbs on poll to the fence manager,
since a breadcrumb may actually turn up before a corresponding fence object
has been placed on the fence ring.
2008-02-05 10:36:49 +01:00
Thomas Hellstrom 47ee6237fe i915: Avoid calling drm_fence_flush_old excessively. 2008-01-30 22:14:02 +01:00
Thomas Hellstrom f1edb7ad91 Simplify the fencing code and differentiate between flushes and
waiting types.
Add a "command_stream_barrier" method to the bo driver.
2008-01-30 22:06:02 +01:00
Dave Airlie 5b99306452 i915: fix missing header when copying data from userspace 2008-01-24 15:18:09 +10:00
Dave Airlie 34b71eb451 i915 make relocs use copy from user
Switch relocs to using copy from user and remove index and pass buffer
handles in instead.
2008-01-24 14:37:40 +10:00
Thomas Hellstrom 88c511e49d Properly propagate the user-space fence flags.
This avoids a sync flush when user-space has already programmed
and MI_FLUSH in the batchbuffer.
2008-01-15 10:03:41 +01:00
Márton Németh 9ab620d661 drm: cleanup DRM_DEBUG() parameters
As DRM_DEBUG macro already prints out the __FUNCTION__ string (see
drivers/char/drm/drmP.h), it is not worth doing this again. At some
other places the ending "\n" was added.

airlied:- I cleaned up a few that this patch missed also
2008-01-03 16:56:04 +10:00
Keith Packard 5f23519b14 Document drm_bo_handle_validate. Match drm_bo_do_validate parameter order.
Document parameters and usage for drm_bo_handle_validate. Change parameter
order to match drm_bo_do_validate (fence_class has been moved to after
flags, hint and mask values). Existing users of this function have been
changed, but out-of-tree users must be modified separately.
2007-12-15 12:10:42 -08:00
Keith Packard 7dcaf0cdbb Make relocation validate client computed values when debugging 2007-12-11 20:23:00 -08:00
Keith Packard 4ec8f58d04 i915: wait for buffer idle before writing relocations
When writing a relocation entry, make sure the target buffer is idle,
otherwise the GPU may see inconsistent data.
2007-12-11 20:23:00 -08:00
Keith Packard 57b9a54eb6 Allow relocation to be skipped when buffers don't move.
One of the costs of superioctl has been the need to perform relocations
inside the kernel. The cost of mapping the buffers to the CPU and writing
data is fairly high, especially if those buffers have been mapped and read
by the GPU.

If we assume that buffers don't move around very often, we can have the
client compute the relocations itself using the previous GPU address. When
that object doesn't move, the kernel can skip computing and writing the
updated data.

Here's a patch which adds a new field to struct drm_bo_info_req called
'presumed_offset', and a new DRM_BO_HINT_PRESUMED_OFFSET that is set when
this field has been filled in by the client.

There are two separate optimizations performed when the presumed_offset is
correct:

 1. i915_exec_reloc checks to see if all previous buffer offsets were guessed
    correctly. If so, there's no need for it to look at *any* of the
    relocations for a buffer. When this happens, it skips the whole
    relocation process, simply returning success.

 2. i915_apply_reloc checks to see if the target buffer offset was guessed
    correctly. If so, it skips mapping the relocatee, computing the
    relocation and writing the value. If no relocations are needed, the
    relocatee should never be mapped to the CPU, and so the kernel shouldn't
    need to wait for any fences to pass.
2007-12-11 20:23:00 -08:00