Evergreen and NI blit copy was broken if the buffer maps to a rectangle
whose one dimension is 16384 (max dimension allowed by these chips).
In the mainline kernel, the problem is exposed only when buffers are
very large (1G), but it's still a problem. The problem could be exposed
for smaller buffers if anyone modifies the algorithm for rectangle
construction in r600_blit_create_rect() (the reason why someone would
modify that algorithm is to tune the performance of buffer moves).
The root cause was in i2f() function which only operated on range between
0 and 16383. Fix this by extending the range of i2f() function to 0 to
32767.
While at it improve the function so that the range can be easily
extended in the future (if it becomes necessary), cleanup lines
over 80 characters, and replace in-line comments with one strategic
comment that explains the crux of the function.
Credits to michel@daenzer.net for pointing out the root cause of
the bug.
v2: Fix I2F_MAX_INPUT constant definition goof and warn only once
if input argument is out of range. Edit the comment a little
bit to avoid some linguistic confusion and make it look better
in general.
Signed-off-by: Ilija Hadzic <ihadzic@research.bell-labs.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Michel Dänzer <michel@daenzer.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This avoid to waste ib pool size and avoid a bunch of wait for
previous ib to finish.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
That naming seems to make more sense, since we not
only want to run PM4 rings with it.
Signed-off-by: Christian König <deathsimple@vodafone.de>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Replace cp, cp1 and cp2 members with just an array
of radeon_cp structs.
Signed-off-by: Christian König <deathsimple@vodafone.de>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Give all asic and radeon_ring_* functions a
radeon_cp parameter, so they know the ring to work with.
Signed-off-by: Christian König <deathsimple@vodafone.de>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Move the TC flush before the texture setup to match mesa and
the ddx. Also, move the TC flush into the texture setup
function.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
blit copy functions deal with GPU pages, not CPU pages,
so rename the variables and parameters accordingly
Signed-off-by: Ilija Hadzic <ihadzic@research.bell-labs.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
reorganize the code such that only the primitives (i.e., the functions
that load the CP ring) are hardware specific; dynamically link the
primitives in a (new) pointer structure inside r600_blit at
blit initialization time so that the functions that control the blit
operations can be made common for r600 and evergreen parts
Signed-off-by: Ilija Hadzic <ihadzic@research.bell-labs.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
some 3d register bits look like magic in r600 blit functions
use predefined constants to make it more intuitive what they are
Signed-off-by: Ilija Hadzic <ihadzic@research.bell-labs.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Covert 4k pages to multiples of 64x64x4 tiles.
This is also more efficient than a scanline based
approach from the MC's perspective.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Ilija Hadzic <ihadzic@research.bell-labs.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
With this patch I'm only about 50k larger with DRM debugging
enables (why is that enabled by default?!?), and slightly
smaller without.
[airlied: moved r100.c additions to radeon_ring.c]
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This commit changed an internal radeon structure, that meant a new driver
in -next had to be fixed up, merge in the commit and fix up the driver.
Also fixes a trivial nouveau merge.
Conflicts:
drivers/gpu/drm/nouveau/nouveau_mem.c
* 'intel/drm-intel-next' of ../drm-next: (755 commits)
drm/i915: Only wait on a pending flip if we intend to write to the buffer
drm/i915/dp: Sanity check eDP existence
drm/i915: Rebind the buffer if its alignment constraints changes with tiling
drm/i915: Disable GPU semaphores by default
drm/i915: Do not overflow the MMADDR write FIFO
Revert "drm/i915: fix corruptions on i8xx due to relaxed fencing"
drm/i915: Don't save/restore hardware status page address register
drm/i915: don't store the reg value for HWS_PGA
drm/i915: fix memory corruption with GM965 and >4GB RAM
Linux 2.6.38-rc7
Revert "TPM: Long default timeout fix"
drm/i915: Re-enable GPU semaphores for SandyBridge mobile
drm/i915: Replace vblank PM QoS with "Interrupt-Based AGPBUSY#"
Revert "drm/i915: Use PM QoS to prevent C-State starvation of gen3 GPU"
drm/i915: Allow relocation deltas outside of target bo
drm/i915: Silence an innocuous compiler warning for an unused variable
fs/block_dev.c: fix new kernel-doc warning
ACPI: Fix build for CONFIG_NET unset
mm: <asm-generic/pgtable.h> must include <linux/mm_types.h>
x86: Use u32 instead of long to set reset vector back to 0
...
Conflicts:
drivers/gpu/drm/i915/i915_gem.c
So we used to use lpfn directly to restrict VRAM when we couldn't
access the unmappable area, however this was removed in
93225b0d7b as it also restricted
the gtt placements. However it was only later noticed that this
broke on some hw.
This removes the active_vram_size, and just explicitly sets it
when it changes, TTM/drm_mm will always use the real_vram_size,
and the active vram size will change the TTM size used for lpfn
setting.
We should re-work the fpfn/lpfn to per-placement at some point
I suspect, but that is too late for this kernel.
Hopefully this addresses:
https://bugs.freedesktop.org/show_bug.cgi?id=35254
v2: fix reported useful VRAM size to userspace to be correct.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Unconditionally initialize the drm gem object - it's not
worth the trouble not to for the few kernel objects.
This patch only changes the place of the drm gem object,
access is still done via pointers.
v2: Uncoditionally align the size in radeon_bo_create. At
least the r600/evergreen blit code didn't to this, angering
the paranoid gem code.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Not only is linear aligned supposedly more performant,
linear general is only supported by the CB in single
slice mode. The texture hardware doesn't support
linear general, but I think the hw automatically
upgrades it to linear aligned.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
agd5f: minor cleanups
Signed-off-by: Cédric Cano <ccano@interfaceconcept.com>
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We were previously dropping alignment requests on the floor
when allocating buffers so we always ended up page aligned.
Certain tiling modes on 6xx+ require larger alignment which
wasn't happening before.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: Jerome Glisse <j.glisse@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We should not allocate any object into unmappable vram if we
have no means to access them which on all GPU means having the
CP running and on newer GPU having the blit utility working.
This patch limit the vram allocation to visible vram until
we have acceleration up and running.
Note that it's more than unlikely that we run into any issue
related to that as when acceleration is not woring userspace
should allocate any object in vram beside front buffer which
should fit in visible vram.
V2 use real_vram_size as mc_vram_size could be bigger than
the actual amount of vram
[airlied: fixup r700_cp_stop case]
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This patch implements blit support for bo moves using
the 3D engine. It uses the same method as r6xx/r7xx:
- store the base state in an IB
- emit variable state and vertex buffers to do the blit
This allows the hw to move bos using the 3D engine and allows
full use of vram beyond the pci aperture size.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Move common code to init function.
v2: make sure the bo is pinned after init as well.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This simplify and improve GPU reset for R1XX-R6XX hw, it's
not 100% reliable here are result:
- R1XX/R2XX works bunch of time in a row, sometimes it
seems it can work indifinitly
- R3XX/R3XX the most unreliable one, sometimes you will be
able to reset few times, sometimes not even once
- R5XX more reliable than previous hw, seems to work most
of the times but once in a while it fails for no obvious
reasons (same status than previous reset just no same
happy ending)
- R6XX/R7XX are lot more reliable with this patch, still
it seems that it can fail after a bunch (reset every
2sec for 3hour bring down the GPU & computer)
This have been tested on various hw, for some odd reasons
i wasn't able to lockup RS480/RS690 (while they use to
love locking up).
Note that on R1XX-R5XX the cursor will disapear after
lockup haven't checked why, switch to console and back
to X will restore cursor.
Next step is to record the bogus command that leaded to
the lockup.
V2 Fix r6xx resume path to avoid reinitializing blit
module, use the gpu_lockup boolean to avoid entering
inifinite waiting loop on fence while reiniting the GPU
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
There is 3 different distinct states for an indirect buffer (IB) :
1- free with no fence
2- free with a fence
3- non free (fence doesn't matter)
Previous code mixed case 2 & 3 in a single one leading to possible
catastrophique failure. This patch rework the handling and properly
separate each case. So when you get ib we set the ib as non free and
fence status doesn't matter. Fence become active (ie has a meaning
for the ib code) once the ib is scheduled or free. This patch also
get rid of the alloc bitmap as it was overkill, we know go through
IB pool list like in a ring buffer as the oldest IB is the first
one the will be free.
Fix :
https://bugs.freedesktop.org/show_bug.cgi?id=26438
and likely other bugs.
V2 remove the scheduled list, it's useless now, fix free ib scanning
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Only flush the gpu caches before we submit a fence.
This leads to a small performance boost when we take
the extra gpu cache flushes out of the ddx and mesa.
Once this patch is in and the drm version is bumped,
we can remove the flushes from the ddx and drm.
Also, remove the extra cache flushes from the blit
routine.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
If an error happen in r600_blit_prepare_copy report it rather
than WARNING and keeping execution. For instance if ib allocation
failed we did just warn about but then latter tried to access
NULL ib ptr causing oops. This patch also protect r600_copy_blit
with a mutex as otherwise one process might overwrite blit temporary
data with new one possibly leading to GPU lockup.
Should partialy or totaly fix:
https://bugzilla.redhat.com/show_bug.cgi?id=553279
V2 failing blit initialization is not fatal, fallback to memcpy when
this happen
V3 init blit before startup as we pin in startup, remove duplicate
code (this one was actualy tested unlike V2)
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
r600 blit cleanup path need to check if a bo was allocated before
trying to free or unpin it. This patch add this check and avoid
oops when the initialization on r6xx or r7xx hw fails.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
R300 family will hard lockup if host path read cache flush is
done through MMIO to HOST_PATH_CNTL. But scheduling same flush
through ring seems harmless. This patch remove the hdp_flush
callback and add a flush after each fence emission which means
a flush after each IB schedule. Thus we should have same behavior
without the hard lockup.
Tested on R100,R200,R300,R400,R500,R600,R700 family.
V2: Adjust fence counts in r600_blit_prepare_copy()
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The locking & protection of radeon object was somewhat messy.
This patch completely rework it to now use ttm reserve as a
protection for the radeon object structure member. It also
shrink down the various radeon object structure by removing
field which were redondant with the ttm information. Last it
converts few simple functions to inline which should with
performances.
airlied: rebase on top of r600 and other changes.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This enables the use of interrupts on r6xx/r7xx hardware.
Interrupts are implemented via a ring buffer. The GPU adds
interrupts vectors to the ring and the host reads them off
in the interrupt handler. The interrupt controller requires
firmware like the CP. This firmware must be installed and
accessble to the firmware loader for interrupts to function.
MSIs don't seem to work on my RS780. They work fine on all
my discrete cards. I'm not sure about other RS780s or
RS880s. I've disabled MSIs on RS780 and RS880, but it would
probably be worth checking on some other systems.
v2 - fix some checkpatch.pl problems;
re-read the disp int status reg if we restart the ih;
v3 - remove the irq handler if r600_irq_init() fails;
remove spinlock in r600_ih_ring_fini();
move ih rb overflow check to r600_get_ih_wptr();
move irq ack to separate function;
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Lots of cases were wrong or missing.
v2: rebased against drm-next
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
There is no need to assign vb before you know that space is available.
[agd5f: adapted for kernel tree.]
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
a) don't zero gart table on gart enable
b) move pinning shader object into resume path
c) unpin shader object on suspend
d) set cp ready to false after cp shutdown on suspend.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Some people were seeing
*ERROR* radeon: writting more dword to ring than expected
after certain blits, the loops calculation didn't take
into account that we do a separate blit for the remainder
after doing the aligned blits.
Acked-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We sometimes lock IB then the ring and sometimes the ring then
the IB. This is mostly due to the IB locking not being well defined
about what data in the structs it actually locks. Define what I
believe is the correct behaviour and gets rid of the lock dep ordering
warning.
Signed-off-by: Dave Airlie <airlied@redhat.com>
rv6xx emits two extra dwords in the render target setup.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>
select the correct max number of bytes per blit based
on whether the size is multiple of 4 bytes. This
determines whether we can use 8 or 32 bit pixels for
the blit.
airlied: also merged the IB padding patch +
correcting the VS offset for context
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This adds the r600 KMS + CS support to the Linux kernel.
The r600 TTM support is quite basic and still needs more
work esp around using interrupts, but the polled fencing
should work okay for now.
Also currently TTM is using memcpy to do VRAM moves,
the code is here to use a 3D blit to do this, but
isn't fully debugged yet.
Authors:
Alex Deucher <alexdeucher@gmail.com>
Dave Airlie <airlied@redhat.com>
Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Ilija Hadzic