Virtual address space are per drm client (opener of /dev/drm).
Client are in charge of virtual address space, they need to
map bo into it by calling DRM_RADEON_GEM_VA ioctl.
First 16M of virtual address space is reserved by the kernel.
Once using 2 level page table we should be able to have a small
vram memory footprint for each pt (there would be one pt for all
gart, one for all vram and then one first level for each virtual
address space).
Plan include using the sub allocator for a common vm page table
area and using memcpy to copy vm page table in & out. Or use
a gart object and copy things in & out using dma.
v2: agd5f fixes:
- Add vram base offset for vram pages. The GPU physical address of a
vram page is FB_OFFSET + page offset. FB_OFFSET is 0 on discrete
cards and the physical bus address of the stolen memory on
integrated chips.
- VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR covers all vmid's >= 1
v3: agd5f:
- integrate with the semaphore/multi-ring stuff
v4:
- rebase on top ttm dma & multi-ring stuff
- userspace is now in charge of the address space
- no more specific cs vm ioctl, instead cs ioctl has a new
chunk
v5:
- properly handle mem == NULL case from move_notify callback
- fix the vm cleanup path
v6:
- fix update of page table to only happen on valid mem placement
v7:
- add tlb flush for each vm context
- add flags to define mapping property (readable, writeable, snooped)
- make ring id implicit from ib->fence->ring, up to each asic callback
to then do ring specific scheduling if vm ib scheduling function
v8:
- add query for ib limit and kernel reserved virtual space
- rename vm->size to max_pfn (maximum number of page)
- update gem_va ioctl to also allow unmap operation
- bump kernel version to allow userspace to query for vm support
v9:
- rebuild page table only when bind and incrementaly depending
on bo referenced by cs and that have been moved
- allow virtual address space to grow
- use sa allocator for vram page table
- return invalid when querying vm limit on non cayman GPU
- dump vm fault register on lockup
v10: agd5f:
- Move the vm schedule_ib callback to a standalone function, remove
the callback and use the existing ib_execute callback for VM IBs.
v11:
- rebase on top of lastest Linus
v12: agd5f:
- remove spurious backslash
- set IB vm_id to 0 in radeon_ib_get()
v13: agd5f:
- fix handling of RADEON_CHUNK_ID_FLAGS
v14:
- fix va destruction
- fix suspend resume
- forbid bo to have several different va in same vm
v15:
- rebase
v16:
- cleanup left over of vm init/fini
v17: agd5f:
- cs checker
v18: agd5f:
- reworks the CS ioctl to better support multiple rings and
VM. Rather than adding a new chunk id for VM, just re-use the
IB chunk id and add a new flags for VM mode. Also define additional
dwords for the flags chunk id to define the what ring we want to use
(gfx, compute, uvd, etc.) and the priority.
v19:
- fix cs fini in weird case of no ib
- semi working flush fix for ni
- rebase on top of sa allocator changes
v20: agd5f:
- further CS ioctl cleanups from Christian's comments
v21: agd5f:
- integrate CS checker improvements
v22: agd5f:
- final cleanups for release, only allow VM CS on cayman
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>
This adds a new optional chunk to the CS ioctl that specifies optional flags
to the CS parser. Why this is useful is explained below. Note that some regs
no longer need the NOP relocation packet if this feature is enabled.
Tested on r300g and r600g with this flag disabled and enabled.
Assume there are two contexts sharing the same mipmapped tiled texture.
One context wants to render into the first mipmap and the other one
wants to render into the last mipmap. As you probably know, the hardware
has a MACRO_SWITCH feature, which turns off macro tiling for small mipmaps,
but that only applies to samplers.
(at least on r300-r500, though later hardware likely behaves the same)
So we want to just re-set the tiling flags before rendering (writing
packets), right? ... No. The contexts run in parallel, so they may
set the tiling flags simultaneously and then fire their command streams
also simultaneously. The last one setting the flags wins, the other one
loses.
Another problem is when one context wants to render into the first and
the last mipmap in one CS. Impossible. It must flush before changing
tiling flags and do the rendering into the smaller mipmaps in another CS.
Yet another problem is that writing copy_blit in userspace would be a mess
involving re-setting tiling flags to please the kernel, and causing races
with other contexts at the same time.
The only way out of this is to send tiling flags with each CS, ideally
with each relocation. But we already do that through the registers.
So let's just use what we have in the registers.
Signed-off-by: Marek Olšák <maraeo@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
With the dropped inlines gccs starts warning about genuinely unused
functions. Remove r600_bpe_from_format, evergreen_cs_track_validate_cb,
evergreen-cs_packet_next_is_pkt3_nop which are all unused.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Fixes
evergreen_cs_parse 4080 23124 +19044
and others compared to a non force inline kernel.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This off by one range check was copy and pasted a couple places.
It's not really harmful, but we should fix it anyway.
Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
- Add some new compute regs
- Add new dispatch packets for evergreen/cayman
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
BC6 and BC7 are described in ARB_texture_compression_bptc.
No idea what FMT_32_AS_32_32_32_32 is good for.
Signed-off-by: Marek Olšák <maraeo@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Simplify the way the return value is set a number of times (mostly on
error).
Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
* '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
have to read values from the IB in order as we could cross
a page boundary at any time and won't be able to go backwards.
Signed-off-by: Dave Airlie <airlied@redhat.com>
There are a bunch of off by one errors in the sanity checks here.
Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
the texture checking code didn't work for block formats like s3tc,
this overhauls it to work for all types.
v2: add texture array support.
v3: add subsampled formats
Signed-off-by: Dave Airlie <airlied@redhat.com>
Makes debugging CS rejections much easier.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
There are too many strange corner cases triggered in old userspace
drivers out there to that it's nearly impossible to not break some
obscure app.
Cc: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This reverts commit d33ef52d9d.
This change seems to expose a bug in the 3D driver tiggered by
certain apps, so revert it to keep userspace working.
Reported-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Looks like a typo in:
drm/radeon/r600: fix tiling issues in CS checker.
(f30df2fad0)
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Tiled buffers have the same alignment requirements regardless of
whether the surface is for db, cb, or textures. Previously, the
calculations where inconsistent for each buffer type.
- Unify the alignment calculations in a common function
- Standardize the alignment units (pixels for pitch/height/depth,
bytes for base)
- properly check the buffer base alignments
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Calculate height based on the slice bitfield rather than the size.
Same as Dave's CB fix.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
broken by:
drm/radeon/r600: fix tiling issues in CS checker.
v2: only apply it to 1D tiling case.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
macro tile heights are aligned to num channels, not num banks.
Noticed by Dave Airlie.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
The CS checker had some incorrect alignment requirements for 2D surfaces,
this made rendering to mipmap levels that were 2D broken.
Also the CB height was being worked out from the BO size, this doesn't work
at all when rendering mipmap levels, instead we work out what height userspace
wanted from slice max and use that to check it fits inside the BO, however
the DDX send the wrong slice max for an unaligned buffer so we have to workaround
for that even though its a userspace bug.
Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The texture base address registers are in units of 256 bytes.
The original CS checker treated these offsets as bytes, so the
original check was wrong. I fixed the units in a patch during
the 2.6.36 cycle, but this ended up breaking some existing
userspace (probably due to a bug in either userspace texture allocation
or the drm texture mipmap checker). So for now, until we come
up with a better fix, just warn if the mipmap size it too large.
This will keep existing userspace working and it should be just
as safe as before when we were checking the wrong units. These
are GPU MC addresses, so if they fall outside of the VRAM or
GART apertures, they end up at the GPU default page, so this should
be safe from a security perspective.
v2: Just disable the warning. It just spams the log and there's
nothing the user can do about it.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: Jerome Glisse <glisse@freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This never really got fixed in mesa, and the kernel deals with the problem
just fine, so don't got reporting things that confuse people.
Signed-off-by: Dave Airlie <airlied@redhat.com>
add default case for buffer formats
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: Andre Maasikas <amaasikas@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
- buffer offsets in the base regs are 256b aligned so
shift properly when comparing, fixed by Andre Maasikas
- mipmap size was calculated wrong when nlevel=0
- texture bo offsets were used after the bo base address was added
- vertex resource size register is size - 1, not size
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: Andre Maasikas <amaasikas@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Check for relocs for DB_DEPTH_INFO, CB_COLOR*_INFO, and texture
resources.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We only add/remove crtcs at driver load, you cannot remove when
the GPU is running a CS packet since the fd is open, when
GPU hotplugging on radeons actually is needed all this locking
needs a review and I've started re-working kms core locking to deal
with this better. But for now avoid long delays in CS processing when
hotplug detect is happening in a different thread.
this fixes a regression introduced with hotplug detection.
Signed-off-by: Dave Airlie <airlied@redhat.com>
- Drop some more safe regs taht userspace shouldn't hit
- Constant base regs need relocs. This allows us to use
constant buffers rather than the constant register file.
Also we don't want userspace to be able to set arbitrary
mc base values for the const caches.
- Track SQ_CONFIG so we know whether userspace is using
the cfile or constant buffers.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
drivers/gpu/drm/radeon/r600_cs.c: In function ‘r600_cs_track_check’:
drivers/gpu/drm/radeon/r600_cs.c:166: warning: ‘bpe’ may be used uninitialized in this function
drivers/gpu/drm/radeon/r600_cs.c:166: note: ‘bpe’ was declared here
drivers/gpu/drm/radeon/r600_cs.c: In function ‘r600_cs_parse’:
drivers/gpu/drm/radeon/r600_cs.c:938: warning: ‘bpe’ may be used uninitialized in this function
drivers/gpu/drm/radeon/r600_cs.c:938: note: ‘bpe’ was declared here
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
radeon_cs kfree the tracker structure but for r6xx/r7xx we want
to kfree it inside the parse function because we share it with
the UMS code path. Set tracker to NULL after freeing it will
avoid double free.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This patch add cs checker to r600/r700 hw. Command stream checking
will rewrite some of the cs value in order to restrict GPU access
to BO size. This doesn't break old userspace but just enforce safe
value. It should break any things that was using the r600/r700 cs
ioctl to do forbidden things (malicious software), though we are
not aware of such things.
Here is the list of thing we check :
- enforcing resource size
- enforcing color buffer slice tile max, will restrict cb access
- enforcing db buffer slice tile max, will restrict db access
We don't check for shader bigger than the BO in which they are
supposed to be, such use would lead to GPU lockup and is harmless
from security POV, as far as we can tell (note that even checking
for this wouldn't prevent someone to write bogus shader that lead
to lockup).
This patch has received as much testing as humanly possible with
old userspace to check that it didn't break such configuration.
However not all the applications out there were tested, thus it
might broke some odd, rare applications.
[airlied: fix rules for cs checker for parallel builds]
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This patch workaround a possible security issue which can allow
user to abuse drm on r6xx/r7xx hw to access any system ram memory.
This patch doesn't break userspace, it detect "valid" old use of
CB_COLOR[0-7]_FRAG & CB_COLOR[0-7]_TILE registers and overwritte
the address these registers are pointing to with the one of the
last color buffer. This workaround will work for old mesa &
xf86-video-ati and any old user which did use similar register
programming pattern as those (we expect that there is no others
user of those ioctl except possibly a malicious one). This patch
add a warning if it detects such usage, warning encourage people
to update their mesa & xf86-video-ati. New userspace will submit
proper relocation.
Fix for xf86-video-ati / mesa (this kernel patch is enough to
prevent abuse, fix for userspace are to set proper cs stream and
avoid kernel warning) :
http://cgit.freedesktop.org/xorg/driver/xf86-video-ati/commit/?id=95d63e408cc88b6934bec84a0b1ef94dfe8bee7bhttp://cgit.freedesktop.org/mesa/mesa/commit/?id=46dc6fd3ed5ef96cda53641a97bc68c3bc104a9f
Abusing this register to perform system ram memory is not easy,
here is outline on how it could be achieve. First attacker must
have access to the drm device and be able to submit command stream
throught cs ioctl. Then attacker must build a proper command stream
for r6xx/r7xx hw which will abuse the FRAG or TILE buffer to
overwrite the GPU GART which is in VRAM. To achieve so attacker
as to setup CB_COLOR[0-7]_FRAG or CB_COLOR[0-7]_TILE to point
to the GPU GART, then it has to find a way to write predictable
value into those buffer (with little cleverness i believe this
can be done but this is an hard task). Once attacker have such
program it can overwritte GPU GART to program GPU gart to point
anywhere in system memory. It then can reusse same method as he
used to reprogram GART to overwritte the system ram through the
GART mapping. In the process the attacker has to be carefull to
not overwritte any sensitive area of the GART table, like ring
or IB gart entry as it will more then likely lead to GPU lockup.
Bottom line is that i think it's very hard to use this flaw
to get system ram access but in theory one can achieve so.
Side note: I am not aware of anyone ever using the GPU as an
attack vector, nevertheless we take great care in the opensource
driver to try to detect and forbid malicious use of GPU. I don't
think the closed source driver are as cautious as we are.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>
Use kzalloc rather than kcalloc(1,...)
The use of the allocated memory that looks like an array is &p->relocs[0],
but this should be the same as p->relocs.
The semantic patch that makes this change is as follows:
(http://coccinelle.lip6.fr/)
// <smpl>
@@
@@
- kcalloc(1,
+ kzalloc(
...)
// </smpl>
Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Convert the rv515 asic support to new init path also add an explanation
in radeon.h about the new init path. There is also few cleanups
associated with this change (others asic calling rv515 helper
functions).
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>
The legacy r600 path shares code, but doesn't share quite enough
to get the freeing correct. Free the pages here also.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Need add reloc offset to the offset in the actual
packet. Fixes use of the DRAW_INDEX packet by the 3D
driver.
[airlied: modified first one where idx_value == ib[idx+0]
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>
Provides support for anti-tearing functionality
in the ddx.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>
this avoids reading back the IB on AGP, also it avoids
the race where since we haven't fetched the page from the main IB
and written it to the gpu one, reading back fetches 0.
Signed-off-by: Dave Airlie <airlied@redhat.com>
This avoids needing to do a kmalloc > PAGE_SIZE for the main
indirect buffer chunk, it adds an accessor for all reads from
the chunk and caches a single page at a time for subsequent
reads.
changes since v1:
Use a two page pool which should be the most common case
a single packet spanning > PAGE_SIZE will be hit, but I'm
having trouble seeing anywhere we currently generate anything like that.
hopefully proper short page copying at end
added parser_error flag to set deep errors instead of having to test
every ib value fetch.
fixed bug in patch that went to list.
Signed-off-by: Dave Airlie <airlied@redhat.com>
radeon_share.h was begining to give problem with include order in
respect of radeon.h. It's easier and also i think cleaner to move
what was in radeon_share.h into radeon.h. At the same time use the
extern keyword for function shared accross the module.
Signed-off-by: Jerome Glisse <jglisse@redhat.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>
Jerome Glisse