Opened 5 years ago
Closed 5 years ago
#13072 closed enhancement (fixed)
mesa-19.3.3
Reported by: | Douglas R. Reno | Owned by: | Bruce Dubbs |
---|---|---|---|
Priority: | normal | Milestone: | 9.1 |
Component: | BOOK | Version: | SVN |
Severity: | normal | Keywords: | |
Cc: |
Description
New point version
Hi list, I'd like to announce mesa 19.3.3. This release was delayed due to bugs caught in CI that needed to be resolved before the release could be made. Due to the slightly longer cycle there's slightly more patches than would normally be present in the release. I've also started using a new script to find the patches in master to pick, so please ignore any .pick_status.json: commits, they're generated by the new script. There's plenty of changes here, but intel, docs, radeonsi, and aco are the biggest sets of changes. Dylan Shortlog ======== Adam Jackson (1): drisw: Cache the depth of the X drawable Andrii Simiklit (1): mesa/st: fix a memory leak in get_version Bas Nieuwenhuizen (2): radv: Disable VK_EXT_sample_locations on GFX10. radv: Remove syncobj_handle variable in header. Caio Marcelo de Oliveira Filho (1): intel/fs: Only use SLM fence in compute shaders Daniel Schürmann (2): aco: fix unconditional demote_to_helper aco: rework lower_to_cssa() Dylan Baker (5): docs: add SHA256 sums for 19.3.2 cherry-ignore: Update for 19.3.3 .pick_status.json: Update to c787b8d2a16d5e2950f209b1fcbec6e6c0388845 docs: Add relnotes for 19.3.3 release VERSION: bump version to 19.3.3 Eric Anholt (1): mesa: Fix detection of invalidating both depth and stencil. Eric Engestrom (1): meson: use github URL for wraps instead of completely unreliable wrapdb Erik Faye-Lund (8): docs: fix typo in html tag name docs: fix paragraphs docs: open paragraph before closing it docs: use code-tag instead of pre-tag docs: use code-tags instead of pre-tags docs: use code-tags instead of pre-tags docs: move paragraph closing tag docs: remove double-closed definition-list Francisco Jerez (3): glsl: Fix software 64-bit integer to 32-bit float conversions. intel/fs/gen11+: Handle ROR/ROL in lower_simd_width(). intel/fs/gen8+: Fix r127 dst/src overlap RA workaround for EOT message payload. Hyunjun Ko (1): turnip: fix invalid VK_ERROR_OUT_OF_POOL_MEMORY Jan Vesely (1): clover: Initialize Asm Parsers Jason Ekstrand (8): anv: Flag descriptors dirty when gl_NumWorkgroups is used intel/vec4: Support scoped_memory_barrier intel/blorp: Fill out all the dwords of MI_ATOMIC anv: Don't over-advertise descriptor indexing features anv: Memset array properties anv/blorp: Rename buffer image stride parameters anv: Canonicalize buffer formats for image/buffer copies anv: Stop allocating WSI event fences off the instance Jonathan Marek (1): st/mesa: don't lower YUV when driver supports it natively Kenneth Graunke (2): intel/compiler: Fix illegal mutation in get_nir_image_intrinsic_image intel: Fix aux map alignments on 32-bit builds. Lasse Lopperi (1): freedreno/drm: Fix memory leak in softpin implementation Lionel Landwerlin (4): anv: fix intel perf queries availability writes anv: only use VkSamplerCreateInfo::compareOp if enabled intel/perf: expose timestamp begin for mdapi intel/perf: report query split for mdapi Marek Olšák (4): ac/gpu_info: always use distributed tessellation on gfx10 radeonsi: work around an LLVM crash when using llvm.amdgcn.icmp.i64.i1 radeonsi: clean up how internal compute dispatches are handled radeonsi: don't invoke decompression inside internal launch_grid Nataraj Deshpande (1): egl/android: Restrict minimum triple buffering for android color_buffers Pierre-Eric Pelloux-Prayer (8): radeonsi: release saved resources in si_retile_dcc radeonsi: release saved resources in si_compute_expand_fmask radeonsi: release saved resources in si_compute_clear_render_target radeonsi: release saved resources in si_compute_copy_image radeonsi: release saved resources in si_compute_do_clear_or_copy radeonsi: fix fmask expand compute shader radeonsi: make sure fmask expand is done if needed util: call bind_sampler_states before setting sampler_views Rhys Perry (8): aco: set vm for pos0 exports on GFX10 aco: fix imageSize()/textureSize() with large buffers on GFX8 aco: fix uninitialized data in the binary aco: set exec_potentially_empty for demotes aco: disable add combining for ds_swizzle_b32 aco: don't DCE atomics with return values aco: check if multiplication/clamp is live when applying output modifier aco: fix off-by-one error when initializing sgpr_live_in Samuel Pitoiset (2): radv: only use VkSamplerCreateInfo::compareOp if enabled radv: fix double free corruption in radv_alloc_memory() Samuel Thibault (1): meson: Do not require libdrm for DRI2 on hurd Tapani Pälli (1): egl/android: fix buffer_count for applications setting max count Thong Thai (1): mesa: Prevent _MaxLevel from being less than zero Timur Kristóf (1): aco/gfx10: Fix VcmpxExecWARHazard mitigation. git tag: mesa-19.3.3
Attachments (1)
Change History (4)
comment:1 by , 5 years ago
Owner: | changed from | to
---|---|
Status: | new → assigned |
comment:2 by , 5 years ago
by , 5 years ago
Attachment: | mesa-19.3.3-fix_svga_vmwgfx_segfaults-1.patch added |
---|
Fix SVGA and VMWGFX segmentation faults that prevent X from starting
Note:
See TracTickets
for help on using tickets.
The SVGA driver needs a patch to work, and has been broken since 19.2.0 with the introduction of OpenGL 4.5 and Vulkan 1.1, but it has gotten worse with the introduction of OpenGL 4.6. Without this, trying to start X on anything that needs the SVGA driver results in a segmentation fault like so:
I traced this down originally to one of three things last night: The Kernel (ruled out after upgrading to 5.5 and reverting to the previous version of 5.2.14 that was on this system), Xorg-Server (downgraded to 1.20.5 and had no difference), and Mesa (I tried versions all the way back to 19.1.6 which was the last known one to work with this system). After this, I did some digging this morning to find out that OpenGL 4.5 and Vulkan support was facilitated in the kernel via additional memory management routines in Linux-5.3+, and did some more digging through PRs in the SVGA driver for Mesa (SVGA == vmwgfx in the tree).
After reaching out to them on IRC early this morning, some fixes were made that resulted in the patch that I've attached to allow it to function. It's also confirmed that it's been broken since 19.2.0.