Let's be optimistic that for future platforms this will remain the same
and reorg a bit.
This reorg in if blocks instead of switch make life easier for future
platform support addition.
v2: Jani pointed out I was missing reg_830 for some gen3 platforms. So let's make
this platforms subcases of Gen checks.
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This patch sets up all of the tracking and copying necessary to
use batch pools with the command parser and dispatches the copied
(shadow) batch to the hardware.
After this patch, the parser is in 'enabling' mode.
Note that performance takes a hit from the copy in some cases
and will likely need some work. At a rough pass, the memcpy
appears to be the bottleneck. Without having done a deeper
analysis, two ideas that come to mind are:
1) Copy sections of the batch at a time, as they are reached
by parsing. Might improve cache locality.
2) Copy only up to the userspace-supplied batch length and
memset the rest of the buffer. Reduces the number of reads.
v2:
- Remove setting the capacity of the pool
- One global pool instead of per-ring pools
- Replace batch_obj with shadow_batch_obj and hook into eb->vmas
- Memset any space in the shadow batch beyond what gets copied
- Rebased on execlist prep refactoring
v3:
- Rebase on chained batch handling
- Squash in setting the secure dispatch flag
- Add a note about the interaction w/secure dispatch pinning
- Check for request->batch_obj == NULL in i915_gem_free_request
v4:
- Fix read domains for shadow_batch_obj
- Remove the set_to_gtt_domain call from i915_parse_cmds
- ggtt_pin/unpin in the parser block to simplify error handling
- Check USES_FULL_PPGTT before setting DISPATCH_SECURE flag
- Remove i915_gem_batch_pool_put calls
v5:
- Move 'pending_read_domains |= I915_GEM_DOMAIN_COMMAND' after
the parser (danvet, from v4 0/7 feedback)
Issue: VIZ-4719
Signed-off-by: Brad Volkin <bradley.d.volkin@intel.com>
Reviewed-By: Jon Bloomfield <jon.bloomfield@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This adds a small module for managing a pool of batch buffers.
The only current use case is for the command parser, as described
in the kerneldoc in the patch. The code is simple, but separating
it out makes it easier to change the underlying algorithms and to
extend to future use cases should they arise.
The interface is simple: init to create an empty pool, fini to
clean it up, get to obtain a new buffer. Note that all buffers are
expected to be inactive before cleaning up the pool.
Locking is currently based on the caller holding the struct_mutex.
We already do that in the places where we will use the batch pool
for the command parser.
v2:
- s/BUG_ON/WARN_ON/ for locking assertions
- Remove the cap on pool size
- Switch from alloc/free to init/fini
v3:
- Idiomatic looping structure in _fini
- Correct handling of purged objects
- Don't return a buffer that's too much larger than needed
v4:
- Rebased to latest -nightly
v5:
- Remove _put() function and clean up comments to match
v6:
- Move purged check inside the loop (danvet, from v4 1/7 feedback)
v7:
- Use single list instead of two. (Chris W)
- s/active_list/cache_list
- Squashed in debug patches (Chris W)
drm/i915: Add a batch pool debugfs file
It provides some useful information about the buffers in
the global command parser batch pool.
v2: rebase on global pool instead of per-ring pools
v3: rebase
drm/i915: Add batch pool details to i915_gem_objects debugfs
To better account for the potentially large memory consumption
of the batch pool.
v8:
- Keep cache in LRU order (danvet, from v6 1/5 feedback)
Issue: VIZ-4719
Signed-off-by: Brad Volkin <bradley.d.volkin@intel.com>
Reviewed-By: Jon Bloomfield <jon.bloomfield@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Things like reliable GGTT mappings and mirrored 2d-on-3d display will need
to map objects into the same address space multiple times.
Added a GGTT view concept and linked it with the VMA to distinguish between
multiple instances per address space.
New objects and GEM functions which do not take this new view as a parameter
assume the default of zero (I915_GGTT_VIEW_NORMAL) which preserves the
previous behaviour.
This now means that objects can have multiple VMA entries so the code which
assumed there will only be one also had to be modified.
Alternative GGTT views are supposed to borrow DMA addresses from obj->pages
which is DMA mapped on first VMA instantiation and unmapped on the last one
going away.
v2:
* Removed per view special casing in i915_gem_ggtt_prepare /
finish_object in favour of creating and destroying DMA mappings
on first VMA instantiation and last VMA destruction. (Daniel Vetter)
* Simplified i915_vma_unbind which does not need to count the GGTT views.
(Daniel Vetter)
* Also moved obj->map_and_fenceable reset under the same check.
* Checkpatch cleanups.
v3:
* Only retire objects once the last VMA is unbound.
v4:
* Keep scatter-gather table for alternative views persistent for the
lifetime of the VMA.
* Propagate binding errors to callers and handle appropriately.
v5:
* Explicitly look for normal GGTT view in i915_gem_obj_bound to align
usage in i915_gem_object_ggtt_unpin. (Michel Thierry)
* Change to single if statement in i915_gem_obj_to_ggtt. (Michel Thierry)
* Removed stray semi-colon in i915_gem_object_set_cache_level.
For: VIZ-4544
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
[danvet: Drop hunk from i915_gem_shrink since it's just prettification
but upsets a __must_check warning.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Release struct_mutex if init_rings() fails.
This is a regression introduced in
commit 35a57ffbb1
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date: Thu Nov 20 00:33:07 2014 +0100
drm/i915: Only init engines once
Reported-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
So apparently jiffies<->nsec<->ktime isn't accurate or something. At
elast if we timeout there's occasionally still a few hundred us left
(in a 2 second timeout).
Stuff I've tried and thrown out again:
- Sampling the before timestamp before jiffies. Doesn't improve test
path rate at all.
- Using jiffies. Way to inaccurate, which means way too much drift
with signals plus automatic ioctl restarting in userspace. In
hindsight we should have used an absolute timeout, but hey we need
something for v3 of the i915 gem wait interfaces ;-)
- Trying to figure out where accuracy gets lost. gl testcase really
don't care all that much about this (as long as isn't not massively
off), it's just that the testcase gets a bit upset if it receives an
EITME with timeout > 0.
So as long as we're in the ballbark it's good enough. So patch
everything up if we're at most one jiffies off. I get's me a solid
test again.
This regression is probably introduced in
commit 5ed0bdf21a
Author: Thomas Gleixner <tglx@linutronix.de>
Date: Wed Jul 16 21:05:06 2014 +0000
drm: i915: Use nsec based interfaces
Use ktime_get_raw_ns() and get rid of the back and forth timespec
conversions.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: John Stultz <john.stultz@linaro.org>
Probably because I'm too lazy to confirm myself and still waiting for
QA ;-)
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Stultz <john.stultz@linaro.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=82749
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
We've lost the +1 required for correct timeouts in
commit 5ed0bdf21a
Author: Thomas Gleixner <tglx@linutronix.de>
Date: Wed Jul 16 21:05:06 2014 +0000
drm: i915: Use nsec based interfaces
Use ktime_get_raw_ns() and get rid of the back and forth timespec
conversions.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: John Stultz <john.stultz@linaro.org>
So fix this up by reinstating our handrolled _timeout function. While
at it bother with handling MAX_JIFFIES.
v2: Convert to usecs (we don't care about the accuracy anyway) first
to avoid overflow issues Dave Gordon spotted.
v3: Drop the explicit MAX_JIFFY_OFFSET check, usecs_to_jiffies should
take care of that already. It might be a bit too enthusiastic about it
though.
v4: Chris has a much nicer color, so use his implementation.
This requires to export nsec_to_jiffies from time.c.
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Dave Gordon <david.s.gordon@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=82749
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Stultz <john.stultz@linaro.org>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Acked-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Multiple GGTT VMAs per object will be introduced in the near future which will
make it impossible to guarantee normal GGTT view is at the head of the list.
Purpose of this patch is to break this assumption straight away so any
potential hidden assumptions in the code base can be bisected to this
simple patch.
For: VIZ-4544
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Suggested-by: Daniel Vetter <daniel@ffwll.ch>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We need to do that every time we resume the rings, not just at load.
I've overlooked this in my untangling of the ring init code.
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Reviewed-by: Dave Gordon <david.s.gordon@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We can do this.
And now there's finally the clean split between software setup and
hardware setup I kinda wanted since multi-ring support was merged
aeons ago. It only took almost 5 years.
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Reviewed-by: Dave Gordon <david.s.gordon@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Updated the trace_irq code to use requests instead of seqnos. This includes
reference counting the request object to ensure it sticks around when required.
Note that getting access to the reference counting functions means moving the
inline i915_trace_irq_get() function from intel_ringbuffer.h to i915_drv.h.
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Daniel <Thomas.Daniel@intel.com>
[danvet: Resolve conflict due to shuffled merge order.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The ring member of the object structure was always updated with the
last_read_seqno member. Thus with the conversion to last_read_req, obj->ring is
now a direct copy of obj->last_read_req->ring. This makes it somewhat redundant
and potentially misleading (especially as there was no comment to explain its
purpose).
This checkin removes the redundant field. Many uses were simply testing for
non-null to see if the object is active on the GPU. Some of these have been
converted to check 'obj->active' instead. Others (where the last_read_req is
about to be used anyway) have been changed to check obj->last_read_req. The rest
simply pull the ring out from the request structure and proceed as before.
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Daniel <Thomas.Daniel@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Almost everywhere that caled i915_seqno_passed() was really asking 'has the
given seqno popped out of the hardware yet?'. Thus it had to query the current
hardware seqno and then do a signed delta comparison (which copes with wrapping
around zero but not with seqno values more than 2GB apart, although the latter
is unlikely!).
Now that the majority of seqno instances have been replaced with request
structures, it is possible to convert this test to be request based as well.
There is now a 'i915_gem_request_completed()' function which takes a request and
returns true or false as appropriate. Note that this currently just wraps up the
original _passed() test but a later patch in the series will reduce this to
simply returning a cached internal value, i.e.:
_completed(req) { return req->completed; }'
This checkin converts almost all _seqno_passed() calls. The only one left is in
the semaphore code which still requires seqnos not request structures.
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Daniel <Thomas.Daniel@intel.com>
[danvet: Drop hunk touching the trace_irq code since I've dropped the
patch which converts that, and resolve resulting conflict.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
It makes a lot more sense (and makes future seqno -> request conversion patches
simpler) to fill in the 'ring' field of the request structure at the point of
creation rather than submission. Given that the request structure is assigned by
ring specific code and thus is locked to a ring from the start, there really is
no reason to defer this assignment.
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Daniel <Thomas.Daniel@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
All the code above is now using requests not seqnos so it is possible to convert
the trace functions across. Note that rather than get into problematic reference
counting issues, the trace code only saves the seqno and ring values from the
request structure not the structure pointer itself.
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Daniel <Thomas.Daniel@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
There is no longer any need to retrieve a seqno value from an i915_add_request()
call. The calling code already knows which request structure is being processed
(it can only be ring->OLR). And as the request itself is now used in preference
to the basic seqno value, the latter is now redundant in this situation.
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Daniel <Thomas.Daniel@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Now that all code above is using request structures instead of seqno values, it
is possible to convert __wait_seqno() itself. Internally, it is still calling
i915_seqno_passed(), this will be updated later in the series. This step is just
changing the parameter list and function name.
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Daniel <Thomas.Daniel@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Updated i915_wait_seqno() to take a request structure instead of a seqno value
and renamed it accordingly. Internally, it just pulls the seqno out of the
request and calls on to __wait_seqno() as before. However, all the code further
up the stack is now simplified as it can just pass the request object straight
through without having to peek inside.
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Daniel <Thomas.Daniel@intel.com>
[danvet: Squash in hunk from an earlier patch which was rebased
wrongly.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Updated the _check_olr() function to actually take a request object and compare
it to the OLR rather than extracting seqnos and comparing those.
Note that there is one use case where the request object being processed is no
longer available at that point in the call stack. Hence a temporary copy of the
original function is still present (but called _check_ols() instead). This will
be removed in a subsequent patch.
Also, downgraded a BUG_ON to a WARN_ON as apparently the former is frowned upon
for shipping code.
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Daniel <Thomas.Daniel@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The OLS value is now obsolete. Exactly the same value is guarateed to be always
available as PLR->seqno. Thus it is safe to remove the OLS completely. And also
to rename the PLR to OLR to keep the 'outstanding lazy ...' naming convention
valid.
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Daniel <Thomas.Daniel@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Added reference counting of the request structure around __wait_seqno() calls.
This is a precursor to updating the wait code itself to take the request rather
than a seqno. At that point, it would be a Bad Idea for a request object to be
retired and freed while the wait code is still using it.
v3:
Note that even though the mutex lock is held during a call to i915_wait_seqno(),
it is still necessary to explicitly bump the reference count. It appears that
the shrinker can asynchronously retire items even though the mutex is locked.
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Daniel <Thomas.Daniel@intel.com>
[danvet: Remove wrongly squashed hunk which breaks the build.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Convert the throttle code to use the request structure rather than extracting a
ring/seqno pair from it and using those. This is in preparation for
__wait_seqno() becoming __wait_request().
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Daniel <Thomas.Daniel@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The object structure contains the last read, write and fenced seqno values for
use in syncrhonisation operations. These have now been replaced with their
request structure counterparts.
Note that to ensure that objects do not end up with dangling pointers, the
assignments of last_*_req include reference count updates. Thus a request cannot
be freed if an object is still hanging on to it for any reason.
v2: Corrected 'last_rendering_' to 'last_read_' in a number of comments that did
not get updated when 'last_rendering_seqno' became 'last_read|write_seqno'
several millenia ago.
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Daniel <Thomas.Daniel@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The plan is to use request structures everywhere that seqno values were
previously used. This means saving pointers to structures in places that used to
be simple integers. In turn, that means that the target structure now needs much
more stringent lifetime tracking. That is, it must not be freed while some other
random object still holds a pointer to it.
To achieve this tracking, a reference count needs to be added. Whenever a
pointer to the structure is saved away, the count must be incremented and the
free must only occur when all references have been released.
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Daniel <Thomas.Daniel@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Before suspending, we wait upon the outstanding GPU requests and flush
our pending idle handlers. This should downclock the GPU to its lowest
power state. Add a WARN to check that the delayed tasks were run and did
their job properly.
Suggested-by: Akash Goel <akash.goel@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Dynamic context pinning for LRCs introduced a leak in legacy mode.
Reinstate context unreference in i915_gem_free_request for legacy contexts.
Leak reported by i-g-t/drv_module_reload fixed by this patch.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=86507
Signed-off-by: Thomas Daniel <thomas.daniel@intel.com>
Reviewed-by: John Harrison<John.C.Harrison@Intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The problem here is that SNA pins batchbuffers to etch out a bit more
performance. Iirc it started out as a w/a for i830M (which we've
implemented in the kernel since a long time already). The problem is
that the pin ioctl wasn't added in
commit d23db88c3a
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Fri May 23 08:48:08 2014 +0200
drm/i915: Prevent negative relocation deltas from wrapping
Fix this by simply disallowing pinning from userspace so that the
kernel is in full control of batch placement again. Especially since
distros are moving towards running X as non-root, so most users won't
even be able to see any benefits.
UMS support is dead now, but we need this minimal patch for
backporting. Follow-up patch will remove the pin ioctl code
completely.
Note to backporters: You must have both
commit b45305fce5
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date: Mon Dec 17 16:21:27 2012 +0100
drm/i915: Implement workaround for broken CS tlb on i830/845
which laned in 3.8 and
commit c4d69da167
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Mon Sep 8 14:25:41 2014 +0100
drm/i915: Evict CS TLBs between batches
which is also marked cc: stable. Otherwise this could introduce a
regression by disabling the userspace w/a without the kernel w/a being
fully functional on i830/45.
References: https://bugs.freedesktop.org/show_bug.cgi?id=76554#c116
Cc: stable@vger.kernel.org # requires c4d69da167 and v3.8
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
drm-intel-next-2014-11-21:
- infoframe tracking (for fastboot) from Jesse
- start of the dri1/ums support removal
- vlv forcewake timeout fixes (Imre)
- bunch of patches to polish the rps code (Imre) and improve it on bdw (Tom
O'Rourke)
- on-demand pinning for execlist contexts
- vlv/chv backlight improvements (Ville)
- gen8+ render ctx w/a work from various people
- skl edp programming (Satheeshakrishna et al.)
- psr docbook (Rodrigo)
- piles of little fixes and improvements all over, as usual
* tag 'drm-intel-next-2014-11-21-fixed' of git://anongit.freedesktop.org/drm-intel: (117 commits)
drm/i915: Don't pin LRC in GGTT when dumping in debugfs
drm/i915: Update DRIVER_DATE to 20141121
drm/i915/g4x: fix g4x infoframe readout
drm/i915: Only call mod_timer() if not already pending
drm/i915: Don't rely upon encoder->type for infoframe hw state readout
drm/i915: remove the IRQs enabled WARN from intel_disable_gt_powersave
drm/i915: Use ggtt error obj capture helper for gen8 semaphores
drm/i915: vlv: increase timeout when setting idle GPU freq
drm/i915: vlv: fix cdclk setting during modeset while suspended
drm/i915: Dump hdmi pipe_config state
drm/i915: Gen9 shadowed registers
drm/i915/skl: Gen9 multi-engine forcewake
drm/i915: Read power well status before other registers for drpc info
drm/i915: Pin tiled objects for L-shaped configs
drm/i915: Update ring freq for full gpu freq range
drm/i915: change initial rps frequency for gen8
drm/i915: Keep min freq above floor on HSW/BDW
drm/i915: Use efficient frequency for HSW/BDW
drm/i915: Can i915_gem_init_ioctl
drm/i915: Sanitize ->lastclose
...
It happens on occasion that developers of generic user-space applications
abuse the dumb buffer API to get hold of drm buffers that they can both
mmap() and use for GPU acceleration, using the assumptions that dumb buffers
and buffers available for GPU are
a) The same type and can be aribtrarily type-casted.
b) fully coherent.
This patch makes the most widely used drivers warn nicely when that happens,
the next step will be to fail.
v2: Move drmP.h changes to drm_gem.h. Fix Radeon dumb mmap breakage.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Let's just throw in the towel on this one and take the cheap way out.
Based on a patch from Chris Wilson, but checking for a different bit.
Chris' patch checked for even bank layout, this one here for a magic
bit. Given the evidence we've gathered (not much) both work I think,
but checking for the magic bit might be more accurate.
Anyway, works on my gm45 here.
For paranoi restrict to gen4 (and mobile), since we've only ever seen
this on gm45 and i965gm.
Also add some debugfs output so that we can skip the tiled swapping
tests properly in these cases.
v2: Clean up the quirk'ed pin count in free_object to avoid upsetting
the WARN_ON. Spotted by Chris.
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=28813
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=45092
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Found one more!
With this we can clear up the ggtt init code a bit, yay!
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
With this all the ums nonsense around gem setup/teardown has
disappeared, yay!
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Again just complicates gem init functions and makes a general mess out
of everything.
Good riddance!
v2: In my enthusiasm to start removing dri1/ums crud I went overboard a
bit and killed parts of hangcheck. Resurrect it.
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
We've killed ums support by now, it's time to reap the benefits. This
one here is getting in the way of doing some ring init cleanup.
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
With the deprecation of UMS, and by association DRI1, we have a tough
choice when updating the ring access routines. We either rewrite the
DRI1 routines blindly without testing (so likely to be broken) or take
the liberty of declaring them no longer supported and remove them
entirely. This takes the latter approach.
v2: Also remove the DRI1 sarea updates
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Fix rebase conflicts.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Up until now, we have pinned every logical ring context backing object
during creation, and left it pinned until destruction. This made my life
easier, but it's a harmful thing to do, because we cause fragmentation
of the GGTT (and, eventually, we would run out of space).
This patch makes the pinning on-demand: the backing objects of the two
contexts that are written to the ELSP are pinned right before submission
and unpinned once the hardware is done with them. The only context that
is still pinned regardless is the global default one, so that the HWS can
still be accessed in the same way (ring->status_page).
v2: In the early version of this patch, we were pinning the context as
we put it into the ELSP: on the one hand, this is very efficient because
only a maximum two contexts are pinned at any given time, but on the other
hand, we cannot really pin in interrupt time :(
v3: Use a mutex rather than atomic_t to protect pin count to avoid races.
Do not unpin default context in free_request.
v4: Break out pin and unpin into functions. Fix style problems reported
by checkpatch
v5: Remove unpin_lock as all pinning and unpinning is done with the struct
mutex already locked. Add WARN_ONs to make sure this is the case in future.
Issue: VIZ-4277
Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Signed-off-by: Thomas Daniel <thomas.daniel@intel.com>
Reviewed-by: Akash Goel <akash.goels@gmail.com>
Reviewed-by: Deepak S<deepak.s@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
No longer create a work item to clean each execlist queue item.
Instead, move retired execlist requests to a queue and clean up the
items during retire_requests.
v2: Fix legacy ring path broken during overzealous cleanup
v3: Update idle detection to take execlists queue into account
v4: Grab execlist lock when checking queue state
v5: Fix leaking requests by freeing in execlists_retire_requests.
Issue: VIZ-4274
Signed-off-by: Thomas Daniel <thomas.daniel@intel.com>
Reviewed-by: Deepak S <deepak.s@linux.intel.com>
Reviewed-by: Akash Goel <akash.goels@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Currently objects for which the hardware needs a contiguous physical
address are allocated a shadow backing storage to satisfy the contraint.
This shadow buffer is not wired into the normal obj->pages and so the
physical object is incoherent with accesses via the GPU, GTT and CPU. By
setting up the appropriate scatter-gather table, we can allow userspace
to access the physical object via either a GTT mmaping of or by rendering
into the GEM bo. However, keeping the CPU mmap of the shmemfs backing
storage coherent with the contiguous shadow is not yet possible.
Fortuituously, CPU mmaps of objects requiring physical addresses are not
expected to be coherent anyway.
This allows the physical constraint of the GEM object to be transparent
to userspace and allow it to efficiently render into or update them via
the GTT and GPU.
v2: Fix leak of pci handle spotted by Ville
v3: Remove the now duplicate call to detach_phys_object during free.
v4: Wait for rendering before pwrite. As this patch makes it possible to
render into the phys object, we should make it correct as well!
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
So that it can be used by the flip code to wait for rendering without
holding any locks.
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Always require PIN_GLOBAL when we want a mappable offset (PIN_MAPPABLE).
This causes the pin to fixup the global binding in cases were the vma
was already bound (and due to the proceeding bug, we considered it to be
already mappable).
References: https://bugs.freedesktop.org/show_bug.cgi?id=85671
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Add WARN_ON to check that PIN_MAP implies PIN_GLOBAL as
discussed on irc.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We use the obj->map_and_fenceable hint for when we already have a
valid mapping of this object in the aperture. This hint can only apply
to the GGTT and not to the aliasing-ppGTT. One user of the hint is
execbuffer relocation, which began to fail when it tried to follow the
hint and perform the relocate through the non-existent GGTT mapping.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=85671
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
An earlier commit (c8725f3dc0: Do not call
retire_requests from wait_for_rendering) removed the use of the ring parameter
within wait_rendering__tail() but did not remove the parameter itself. As the
plan is to remove obj->ring which is where this parameter comes from, it is
simpler to just remove the parameter completely than to update it with a new
source.
For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
CC: Chris Wilson <chris@chris-wilson.co.uk>
CC: Brad Volkin <bradley.d.volkin@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
If these flags are on the object level it will be more difficult to allow
for multiple VMAs per object.
v2: Simplification and cleanup after code review comments (Chris Wilson).
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Not having checks for this isn't good.
I've checked igt and libdrm and they all already clear flags properly.
So we're lucky and should be able to sneak this ABI clarification in.
Testcase: igt/gem_wait/invalid-flags
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=85280
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Too many new drm driver writers seem to look at i915 for inspiration.
But we have two ways to do mmap, so discourage readers from the old,
ugly version. In a new driver we'd just expose two mmap offsets per
object, one for the gtt map and the other for the cpu map.
v2: Make it clear that i915 does cpu mmaps this way for past
cluelessness^W^W historical reasons. Asked for by Jani.
Cc: "Cheng, Yao" <yao.cheng@intel.com>
Cc: David Herrmann <dh.herrmann@gmail.com>
Reviewed-by: David Herrmann <dh.herrmann@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
If we are not able to free anything (the shrinker leaves nothing on the
global object lists), do not log anything. This is useful when other
subsystems are being stress-tested for their oom behaviour and i915.ko
is shouting into the logs about doing nothing.
Reported-by: Dave Jones <davej@redhat.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The shrinker reports the number of pages freed, but we try to log the
number of bytes - which leads to some nonsense values being reportedly
freed during oom.
Reported-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We can use the same logic to walk the different bound/unbound lists
during shrinker (as the unbound list is a degenerate case of the bound
list), slightly compacting the code.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
SKL stage 1 patches still need polish so will likely miss the 3.18
merge window. We've decided to postpone to 3.19 so let's pull this in
to make patch merging and conflict handling easier.
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
