- 20 Jun, 2012 8 commits
-
-
Daniel Vetter authored
This is just the minimal patch to disable all this code so that we can do decent amounts of QA before we rip it all out. The complicating thing is that we need to flush the gpu caches after the batchbuffer is emitted. Which is past the point of no return where execbuffer can't fail any more (otherwise we risk submitting the same batch multiple times). Hence we need to add a flag to track whether any caches associated with that ring are dirty. And emit the flush in add_request if that's the case. Note that this has a quite a few behaviour changes: - Caches get flushed/invalidated unconditionally. - Invalidation now happens after potential inter-ring sync. I've bantered around a bit with Chris on irc whether this fixes anything, and it might or might not. The only thing clear is that with these changes it's much easier to reason about correctness. Also rip out a lone get_next_request_seqno in the execbuffer retire_commands function. I've dug around and I couldn't figure out why that is still there, with the outstanding lazy request stuff it shouldn't be necessary. v2: Chris Wilson complained that I also invalidate the read caches when flushing after a batchbuffer. Now optimized. v3: Added some comments to explain the new flushing behaviour. Cc: Eric Anholt <eric@anholt.net> 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>
-
Daniel Vetter authored
It seems to blow up my ilk in all kinds of strange ways. And now that we're no longer resetting the entire modeset state, it shouldn't be necessary any longer. This essentially reverts commit f817586c Author: Daniel Vetter <daniel.vetter@ffwll.ch> Date: Tue Apr 10 15:50:11 2012 +0200 drm/i915: re-init modeset hw state after gpu reset safe for the introduction of modeset_init_hw, that one is nice to prevent code duplication between driver load and resume. v2: Add a comment to the code to warn future travellers of the dragon dungeon ahead, suggested by Chris Wilson. Acked-by:
-
Daniel Vetter authored
Otherwise userspace has no way to figure this out. Reviewed-by:
Ben Widawsky <ben@bwidawsk.net> Signed-Off-by:
-
Daniel Vetter authored
The idr code already passes us the pointer associated with that id, so no need to look it up again. Also, we'll kill the idr right away, so there's no issue with leaving these dangling pointers behind - the current code does the same. v2: Also drop the file argument, spotted by Ben Widawsky. Reviewed-by:
-
Daniel Vetter authored
It found some unused variables. Reviewed-by:
-
Daniel Vetter authored
This is our customary "no such object" errno, not -EINVAL. Reviewed-by:
-
Daniel Vetter authored
It doesn't hurt and it at least prevents us from OOPSing left and right at quite a few places. This also allows us to simplify the code a bit by folding the only line of context_open into the callsite. We obviuosly also need to run the cleanup code unconditionally, too. Reviewed-by:
-
Daniel Vetter authored
commit 8e96d9c4 Author: Ben Widawsky <ben@bwidawsk.net> Date: Mon Jun 4 14:42:56 2012 -0700 drm/i915: reset the GPU on context fini broke module unload because it reset the gpu before we've stopped touching it. Later on in the unload sequence the ringbuffer code complained that the gpu would idle properly (because intel_gpu_reset only resets the hw and not our sw state). v2: Reorder things so that we reset the gpu _before_ we release the backing storage of the default context. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=51183Reviewed-by:
-
- 18 Jun, 2012 8 commits
-
-
Jesse Barnes authored
Prevents a possible hang: WaDisableL3Bank2xClockGate. v2: only apply to VLV, IVB doesn't need this anymore References: https://bugs.freedesktop.org/show_bug.cgi?id=50245Signed-off-by:
Jesse Barnes <jbarnes@virtuousgeek.org> Reviewed-by:
Eugeni Dodonov <eugeni.dodonov@intel.com> Signed-off-by:
-
Jesse Barnes authored
Another required workaround for a potential hang: WaDisableTDLUnitClockGating. v2: only apply this to VLV, IVB doesn't need it anymore (Eugeni) References: https://bugs.freedesktop.org/show_bug.cgi?id=50245Signed-off-by:
-
Jesse Barnes authored
According to the bspec for MBCTL: Driver must set bit in the following scenarios: - to realod teh h/w boot context every time it gets loaded through OS - after an FLR clears the register (BIOS won't run afterwards) References: https://bugs.freedesktop.org/show_bug.cgi?id=50237Signed-off-by:
-
Jesse Barnes authored
The RCBP workaround still applies on these chips, and we need VDS as well. v2: remove MB boot fetch that snuck in (Daniel) add workaround tags to comments for easier internal tracking (Daniel) v3: only apply RCPB and VDS on SNB and VLV, IVB doesn't need them (Eugeni) References: https://bugs.freedesktop.org/show_bug.cgi?id=50251Signed-off-by: -
Daniel Vetter authored
I've added a bit of logic such that running the hangman test on chips without any hw reset support at all doesn't wedge the gpu because the reset failed. This relied on checking for non-null stop_rings. Unfortunately I've botched a rebase somewhere and stop_rings is still cleared at the old place before the reset code. Fix this up so that running the i-g-t tests on gen2/3 doesn't result in a wedged gpu. v2: Actually remove the lines instead of adding them twice ... my git license should be revoked immediately. Signed-Off-by: -
Ben Widawsky authored
I'm seeing about a 5% FPS improvement across various benchmarks on my IVB i3. Rumor has it that the higher end parts show even more benefit. This derives from a patch originally given to me by Bernard. The docs are confusing about the definition names (ie. medium really seems like max), but it would seem it gives more cache to the GT at the expense of uncore. This configuration makes the split most in favor of the GT. I've not tried the other IDICOS values. Cc: "Kilarski, Bernard R" <bernard.r.kilarski@intel.com> Acked-by:
Eric Anholt <eric@anholt.net> Signed-off-by:
-
Ben Widawsky authored
This got dropped as a result of the last round of comments. I didn't test it on unsupported HW (which this is likely the case). Note that this prevents hw context from blowing up on any pre-gen6 hw. Signed-off-by:
-
Eugeni Dodonov authored
Somehow this went unnoticed in the past reviews, but the condition would never timeout properly. This was initially introduced in the v2 of original SBI enabling patch. Highly embarrassing. Note that we now actually time out for the read, which resulted in gcc complaining that we can now return unitialized garbage if that happens. There's not much we can do here because there's not much point in thread -EIO all the way down through these functions. Hence simply shut up the compiler. Reported-by:
Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by:
-
- 16 Jun, 2012 1 commit
-
-
Daniel Vetter authored
VGA hotplug detection "works" by measuring the resistance across certain pins. A lot of kvm switches fumble this and wire up cheap resistors with the wrong resistance or don't bother at all. To accomodate these, also try to detect a connected monitor by trying to grab the edid. Contrary to !HAS_HOTPLUG platforms we don't bother with an actual load-detection cycle when the output is life - that would be actual work to implement because things moved around. This is the big difference to Chris Wilson's original approach: commit 9e612a00 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Thu May 31 13:08:53 2012 +0100 drm/i915/crt: Do not rely upon the HPD presence pin This blew up on Linus' machine because it errornously detected a vga screen (without and edid and hence only the default modes), leading to it's prompt removal: commit 8f53369b Author: Linus Torvalds <torvalds@linux-foundation.org> Date: Fri Jun 8 14:53:06 2012 -0700 Revert "drm/i915/crt: Do not rely upon the HPD presence pin" Some digging around in Bspec shows the reason why load detect doesn't work on newer chips - the legacy VGA load detect bit isn't wired up any longer: Public Snb Bspec, Vol3 Part1, 1.1.1 ST00 Input Status 0, bit4: "RGB Comparator / Sense. This bit is here for compatibility and will always return one. Monitor detection must be done be done through the programming of registers in the MMIO space. 0 = Below threshold 1 = Above threshold" v2: Add a comment in the code that load detect on hotplug capable machines is broken and pimp the commit message with a quote of Bspec to show why. Reported-and-tested-by:
Matthieu LAVIE <boiteamadmax@hotmail.com> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=50501Reviewed-by:
-
- 14 Jun, 2012 14 commits
-
-
Ben Widawsky authored
It's the only way we know how to make the GPU actually forget about the default context. Signed-off-by: -
Ben Widawsky authored
Use the rsvd1 field in execbuf2 to specify the context ID associated with the workload. This will allow the driver to do the proper context switch when/if needed. v2: Add checks for context switches on rings not supporting contexts. Before the code would silently ignore such requests. Signed-off-by: -
Ben Widawsky authored
Add the interfaces to allow user space to create and destroy contexts. Contexts are destroyed automatically if the file descriptor for the dri device is closed. Following convention as usual here causes checkpatch warnings. v2: with is_initialized, no longer need to init at create drop the context switch on create (daniel) v3: Use interruptible lock (Chris) return -ENODEV in !GEM case (Chris) Signed-off-by: -
Ben Widawsky authored
To keep things as sane as possible, switch to the default context before idling. This should help free context objects, as well as put things in a more well defined state before suspending. v2: remove seqno from context switch call (daniel) return error on failed context switch instead of WARN+continue (daniel) v3: move idling to i915_gpu idle (from i915_gem_idle) (Chris) Signed-off-by: -
Ben Widawsky authored
Signed-off-by: -
Ben Widawsky authored
With the code to do HW context switches in place have the driver load the default context for the render ring when the driver loads. The default context will be an ever present context that is available to switch to at any time for the given ring. Signed-off-by: -
Ben Widawsky authored
From http://intellinuxgraphics.org/documentation/SNB/IHD_OS_Vol1_Part3.pdf [DevSNB] If Flush TLB invalidation Mode is enabled it's the driver's responsibility to invalidate the TLBs at least once after the previous context switch after any GTT mappings changed (including new GTT entries). This can be done by a pipelined PIPE_CONTROL with TLB inv bit set immediately before MI_SET_CONTEXT. On GEN7 the invalidation mode is explicitly set, but this appears to be lacking for GEN6. Since I don't know the history on this, I've decided to dynamically read the value at ring init time, and use that value throughout. v2: better comment (daniel) Signed-off-by:
-
Ben Widawsky authored
This has showed up in several other patches. It's required for the next context workaround. I tested this one on its own and saw no differences in basic tests (performance or otherwise). This patch is relatively likely to cause regressions, hence why it's split out. Signed-off-by: -
Ben Widawsky authored
The workaround itself applies to gen7 only (according to the docs) and as Eric Anholt points out shouldn't be required since we don't use HW scheduling features, and therefore arbitration. Though since it is a small, and simple addition, and we don't really understand the issue, just do it. FWIW, I eventually want to play with some of the arbitration stuff, and I'd hate to forget about this. Signed-off-by: -
Daniel Vetter authored
This way round we don't introduce and ugly layering violations and use the interface as I planned to use it. Signed-Off-by: -
Ben Widawsky authored
Implement the context switch code as well as the interfaces to do the context switch. This patch also doesn't match 1:1 with the RFC patches. The main difference is that from Daniel's responses the last context object is now stored instead of the last context. This aids in allows us to free the context data structure, and context object independently. There is room for optimization: this code will pin the context object until the next context is active. The optimal way to do it is to actually pin the object, move it to the active list, do the context switch, and then unpin it. This allows the eviction code to actually evict the context object if needed. The context switch code is missing workarounds, they will be implemented in future patches. v2: actually do obj->dirty=1 in switch (daniel) Modified comment around above Remove flags to context switch (daniel) Move mi_set_context code to i915_gem_context.c (daniel) Remove seqno , use lazy request instead (daniel) v3: use i915_gem_request_next_seqno instead of outstanding_lazy_request (Daniel) remove id's from trace events (Daniel) Put the context BO in the instruction domain (Daniel) Don't unref the BO is context switch fails (Chris) Signed-off-by: -
Ben Widawsky authored
Invent an abstraction for a hw context which is passed around through the core functions. The main bit a hw context holds is the buffer object which backs the context. The rest of the members are just helper functions. Specifically the ring member, which could likely go away if we decide to never implement whatever other hw context support exists. Of note here is the introduction of the 64k alignment constraint for the BO. If contexts become heavily used, we should consider tweaking this down to 4k. Until the contexts are merged and tested a bit though, I think 64k is a nice start (based on docs). Since we don't yet switch contexts, there is really not much complexity here. Creation/destruction works pretty much as one would expect. An idr is used to generate the context id numbers which are unique per file descriptor. v2: add DRM_DEBUG_DRIVERS to distinguish ENOMEM failures (ben) convert a BUG_ON to WARN_ON, default destruction is still fatal (ben) Signed-off-by: -
Ben Widawsky authored
Very basic code for context setup/destruction in the driver. Adds the file i915_gem_context.c This file implements HW context support. On gen5+ a HW context consists of an opaque GPU object which is referenced at times of context saves and restores. With RC6 enabled, the context is also referenced as the GPU enters and exists from RC6 (GPU has it's own internal power context, except on gen5). Though something like a context does exist for the media ring, the code only supports contexts for the render ring. In software, there is a distinction between contexts created by the user, and the default HW context. The default HW context is used by GPU clients that do not request setup of their own hardware context. The default context's state is never restored to help prevent programming errors. This would happen if a client ran and piggy-backed off another clients GPU state. The default context only exists to give the GPU some offset to load as the current to invoke a save of the context we actually care about. In fact, the code could likely be constructed, albeit in a more complicated fashion, to never use the default context, though that limits the driver's ability to swap out, and/or destroy other contexts. All other contexts are created as a request by the GPU client. These contexts store GPU state, and thus allow GPU clients to not re-emit state (and potentially query certain state) at any time. The kernel driver makes certain that the appropriate commands are inserted. There are 4 entry points into the contexts, init, fini, open, close. The names are self-explanatory except that init can be called during reset, and also during pm thaw/resume. As we expect our context to be preserved across these events, we do not reinitialize in this case. As Adam Jackson pointed out, The cutoff of 1MB where a HW context is considered too big is arbitrary. The reason for this is even though context sizes are increasing with every generation, they have yet to eclipse even 32k. If we somehow read back way more than that, it probably means BIOS has done something strange, or we're running on a platform that wasn't designed for this. v2: rename load/unload to init/fini (daniel) remove ILK support for get_size() (indirectly daniel) add HAS_HW_CONTEXTS macro to clarify supported platforms (daniel) added comments (Ben) Signed-off-by: -
Ben Widawsky authored
The GPUs can have different default context layouts, and the sizes could vary based on platform or BIOS. In order to back the context object with a properly sized BO, we must read this register in order to find out a sufficient size. Thankfully (sarcarm!), the register moves and changes meanings throughout generations. CTX and CXT differences are intentional as that is how it is in the documentation (prior to GEN6 it was CXT). Signed-off-by:
-
- 13 Jun, 2012 3 commits
-
-
Seth Forshee authored
The Lenovo Thinkpad T410 has the LVDS_PIPEB_SELECT bit set in the LVDS register when booted with the lid closed, even though the LVDS hasn't really been initialized. Ignore this bit so that the VBT value will be used instead. Signed-off-by:
Seth Forshee <seth.forshee@canonical.com> Signed-off-by:
-
Chris Wilson authored
As we switch on/off the primary plane if it is completely obscured by an overlapping video sprite, we also nee to make sure that we update the FBC configuration at the same time. v2: Not all crtcs are intel_crtcs, as spotted by Daniel. v3: Boot testing rules. References: https://bugs.freedesktop.org/show_bug.cgi?id=50238Signed-off-by:
Jani Nikula <jani.nikula@linux.intel.com> Signed-off-by:
-
Daniel Vetter authored
Especially vesafb likes to map everything as uc- (yikes), and if that mapping hangs around still while we try to map the gtt as wc the kernel will downgrade our request to uc-, resulting in abyssal performance. Unfortunately we can't do this as early as readon does (i.e. as the first thing we do when initializing the hw) because our fb/mmio space region moves around on a per-gen basis. So I've had to move it below the gtt initialization, but that seems to work, too. The important thing is that we do this before we set up the gtt wc mapping. Now an altogether different question is why people compile their kernels with vesafb enabled, but I guess making things just work isn't bad per se ... v2: - s/radeondrmfb/inteldrmfb/ - fix up error handling v3: Kill #ifdef X86, this is Intel after all. Noticed by Ben Widawsky. v4: Jani Nikula complained about the pointless bool primary initialization. v5: Don't oops if we can't allocate, noticed by Chris Wilson. v6: Resolve conflicts with agp rework and fixup whitespace. Reported-and-tested-by:
"Kilarski, Bernard R" <bernard.r.kilarski@intel.com> Reviewed-by:
-
- 12 Jun, 2012 6 commits
-
-
Daniel Vetter authored
When drm/i915 is in control of the gtt, we need to call the enable function at all the relevant places ourselves. Reviewed-by:
-
Daniel Vetter authored
We need this thing much earlier, and it doesn't make sense in the hw enabling function intel_enable_gtt - this does not change over a suspend/resume cycle ... Reviewed-by:
-
Daniel Vetter authored
To be able to directly set up the intel-gtt code from drm/i915 and avoid setting up the fake-agp driver we need to prepare a few things: - pass both the bridge and gpu pci_dev to the probe function and add code to handle the gpu pdev both being present (for drm/i915) and not present (fake agp). - add refcounting to the remove function so that unloading drm/i915 doesn't kill the fake agp driver v2: Fix up the cleanup and refcount, noticed by Jani Nikula. Reviewed-by:
-
Daniel Vetter authored
We only need it to fake the agp interface and don't actually use it in the driver anywhere. Hence conditionalize that. This is just a prep patch to eventually disable the fake agp driver on gen6+. Reviewed-by:
-
Daniel Vetter authored
For that to work we need to export the base address of the gtt mmio window from intel-gtt. Also replace all other uses of dev->agp by values we already have at hand. Reviewed-by:
-
Daniel Vetter authored
This is a leftover from the conversion of the i81x fake agp driver over to the new intel-gtt code layoute. Reviewed-by:
-
