[Mesa-dev] [PATCH 00/15] radv: GPU hangs should now be easier to detect

[Samuel Pitoiset]

Hi,

Currently, it's not easy to debug a GPU hang with radv because only a
trace file is generated (cf. RADV_TRACE_FILE). That file contains all
packets and commands emitted into the IB, that's useful but definitely
not enough. For example, the active shaders at the moment of the hang
are not dumped, same for the descriptors, etc.

This series tries to fix the problem by dumping a bunch information
when a hang is detected (or when a VM fault occured). It generates
a full hang report which contains the active shaders (disassembly only
for now), the annotated shaders using UMR (if found in your path), the
descriptors and more.

Here's a real hang report generated by this series:
https://people.freedesktop.org/~hakzsam/radv_hang_report_example

Please review, Thanks!

Samuel Pitoiset (15):
  radv: add a comment that describes the trace BO layout
  radv: save the bound pipeline pointers into the trace BO
  radv: add radv_cmd_buffer_after_draw() helper function
  radv: add debug flags for syncing shaders after every draw call
  radv: dump the active shaders when a hang occured
  radv: add radv_shader_dump_stats() helper
  radv: dump shader stats when a hang occured
  radv/winsys: add a read_registers() callback
  radv: dump some status MMIO registers when a hang occured
  radeonsi: move si_get_wave_info() to AMD common code
  radv: dump annotated shaders using UMR
  radv: save all descriptor pointers into the trace BO
  radv: dump descriptors when a hang occured
  radv: dump last 60 lines of dmesg when a hang occured
  radv: dump the list of enabled options when a hang occured

 src/amd/common/ac_debug.c                         |  76 +++
 src/amd/common/ac_debug.h                         |  18 +
 src/amd/vulkan/radv_cmd_buffer.c                  | 124 ++++-
 src/amd/vulkan/radv_debug.c                       | 540 +++++++++++++++++++++-
 src/amd/vulkan/radv_debug.h                       |   1 +
 src/amd/vulkan/radv_device.c                      |  15 +
 src/amd/vulkan/radv_pipeline.c                    |  64 +--
 src/amd/vulkan/radv_private.h                     |   7 +
 src/amd/vulkan/radv_radeon_winsys.h               |   3 +
 src/amd/vulkan/radv_shader.c                      |  70 +++
 src/amd/vulkan/radv_shader.h                      |   6 +
 src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.c |  11 +
 src/gallium/drivers/radeonsi/si_debug.c           |  96 +---
 13 files changed, 857 insertions(+), 174 deletions(-)

-- 
2.14.1