Rejoice! We have found a trick to make 830gm/845g stable at long last. Ever since the switch to GEM and dynamic video memory, those early second generation chipsets have been plagued by instability. The lack of flushing cachelines from the CPU to GMCH was eventually solved by using an undocmented bit, but 830/845 were still hanging under memory pressure. These deaths were all due to garbage finding its way into the command streamer, and they go away if we take a leaf out of the original driver and never reuse those pages for anything else. So for the first time ever, I have been able to complete running the test suite on an 845g, even whilst thrashing the page and buffer caches! -Chris * Run the SF stage as single-threaded on gen4 to workaround a few issues https://bugs.freedesktop.org/show_bug.cgi?id=57410 * Keep the scanout SURFACE_STATE separate to avoid overriding its memory access control on gen6/7 (i.e. writes to the scanout need to be kept out of the render cache) * Tune batch flushing after an operation to an exported surface under a compositor. * Make sure the source is on the CPU for inplace composition of trapezoids using the CPU https://bugs.freedesktop.org/show_bug.cgi?id=56825 * Immediately flush in the block hander after a split batch to reduce latency between the two halves of an operation. https://bugs.freedesktop.org/show_bug.cgi?id=51718 * Install a fallback config if we fail to install the desired config at VT switch (i.e. booting, after resume with 3 incompatible pipes on Ivybridge) * Pin batches to avoid CS incoherence on 830/845 https://bugs.freedesktop.org/show_bug.cgi?id=26345 Chris Wilson (53): sna: Tidy addition of fake GTF modes for panels sna/gen4: Avoid emitting URB_FENCE across a cache-line sna/gen4: Remove unused CC viewport sna/gen4: Special case solids through the general vertex emitter sna/gen4: Workaround render corruption with multiple SF threads sna/gen6+: Cache the scanout targets separately to avoid override PTE caching sna: Assume that future hardware only gets more flexible sna: Don't disable CPU bo if supported on unknown hw Refactor the common probe methods for scrn construction sna/gen4+: Add common glyph-to-dst emitters Fix compilation of UMS probe following 13f47008ec Remove the default log message sna: Mark proxies as dirty on first relocation sna: Only flush at the low apeture watermark if idle sna: Only flush before adding fresh surfaces to the batch sna: Only inspect the target ring for busyness sna: Convert the ring from BLT/3D to the internal index for kgem_ring_is_idle() sna: Flush upon change of target if GPU is idle sna: Replace remaining kgem_is_idle() with kgem_ring_is_idle() sna/gen4+: Refine test for preferring GPU spans sna: Move source to CPU prior to referencing for inplace trapezoids sna/sprite: Add a DBG to report whether the kernel supports sprites sna: Immediately flush a split batch sna: Compromise and only flush a split batch if writing to scanout sna: Avoid reusing the same 'busy' bit for two different meanings. sna: Try installing a fallback config on VT enter in case full desiredMode fails sna/dri: Only special case 'divisor && msc-passed' for immediate flipping sna/dri: Disable name exchanges for SwapBuffers sna/dri: Query current msc before use sna/dri: Fix handling of current_msc > target_msc sna/gen4: Use the single-threaded SF w/a for spans as well sna/gen3+: Use nearest for unscaled videos sna/gen2: STIPPLE requires an argument sna: Pin some batches to avoid CS incoherence on 830/845 sna: Fix the error path in kgem_init_pinned_batches() to use the right iter sna: Improve the initialisation failure path for pinned batches sna: Only flush the batch after an actual relocation sna: Fix typo for 830/845 gen sna: Fix up BLT overwrite detection to use target_handle sna/gen2: Align surface sizes to an even tile sna/gen2: Program solid mask using the DIFFUSE component sna/gen3: Remove incorrect optimisation of an opaque source for CA sna/gen2: Assertions sna/gen2: Initialise channel->is_affine for solid sna/gen2: Reorder reuse_source() to avoid NULL dereference for solids sna/gen3: Remove stray setting of vertex_start sna/gen3: Don't combine primitives if beginning a ca 2-pass sna/gen2+: Experiment with not forcing migration to GPU after CPU rasterisation sna: Reduce fence watermarks sna/dri: Store and check size of front/back bo attached to a DRI2 drawable sna/dri: Honour TripleBuffer Option sna/dri: Fallback to a blit after a failed flip 2.20.16 release Jesse Barnes (1): dri2: don't schedule a flip prematurely at ScheduleSwap time git tag: 2.20.16 http://xorg.freedesktop.org/archive/individual/driver/xf86-video-intel-2.20.16.tar.bz2 MD5: bff3599b51e44c43c4213cb8d04fa13d xf86-video-intel-2.20.16.tar.bz2 SHA1: 70f8aba9b0e603276380df6c80a796fa3d7faf8d xf86-video-intel-2.20.16.tar.bz2 SHA256: 308646d1d4b07e5c581cd7cca133b0e3410a1a3911a58d7d49b8ea106b64d0ba xf86-video-intel-2.20.16.tar.bz2 http://xorg.freedesktop.org/archive/individual/driver/xf86-video-intel-2.20.16.tar.gz MD5: 87190671c5bdc2606cb9ca99406c660a xf86-video-intel-2.20.16.tar.gz SHA1: 1ae6c0ed2a4b586159019b2c321dc1e18ca13d40 xf86-video-intel-2.20.16.tar.gz SHA256: 5b278c5bbdc27d2aef3e0e87625da83163551052aa7d257525f4919e0c0d91ef xf86-video-intel-2.20.16.tar.gz -- Chris Wilson, Intel Open Source Technology Centre -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 198 bytes Desc: Digital signature URL: <http://lists.freedesktop.org/archives/intel-gfx/attachments/20121215/7d927034/attachment.pgp>
