From: Akhil P Oommen akhilpo@oss.qualcomm.com
[ Upstream commit f195421318bd00151b3a111af6f315a25c3438a8 ]
CP_ALWAYS_ON counter falls under GX domain which is collapsed during IFPC. So switch to GMU_ALWAYS_ON counter for any CPU reads since it is not impacted by IFPC. Both counters are clocked by same xo clock source.
Signed-off-by: Akhil P Oommen akhilpo@oss.qualcomm.com Patchwork: https://patchwork.freedesktop.org/patch/673373/ Signed-off-by: Rob Clark robin.clark@oss.qualcomm.com Signed-off-by: Sasha Levin sashal@kernel.org ---
LLM Generated explanations, may be completely bogus:
YES
- Problem fixed - The CP_ALWAYS_ON counter lives in the GPU GX power domain and stops/gets reset during IFPC (inter‑frame power collapse). This makes CPU-side reads unreliable or forces the driver to wake the GX domain just to read a timestamp, which is fragile and power- inefficient. - The commit switches CPU reads to the GMU AO (always-on) counter, which is not impacted by IFPC and is clocked from the same XO clock, preserving timing semantics.
- What changed - Added a safe 64-bit read helper for the GMU AO counter with a hi-lo- hi read to avoid torn reads: drivers/gpu/drm/msm/adreno/a6xx_gpu.c:19 - Reads `REG_A6XX_GMU_ALWAYS_ON_COUNTER_{H,L}` via `gmu_read()` and rechecks the high word to ensure atomicity. - Replaced CP counter reads in CPU-side tracepoints: - a6xx_submit now traces with GMU AO: drivers/gpu/drm/msm/adreno/a6xx_gpu.c:392 - a7xx_submit now traces with GMU AO: drivers/gpu/drm/msm/adreno/a6xx_gpu.c:592 - Simplified timestamp retrieval to avoid waking GX or using OOB votes: - a6xx_gmu_get_timestamp now returns the GMU AO counter directly: drivers/gpu/drm/msm/adreno/a6xx_gpu.c:2314 - This removes previous lock/OOB sequences to temporarily block IFPC just to read `REG_A6XX_CP_ALWAYS_ON_COUNTER`. - Importantly, GPU-side emissions that snapshot the CP always-on counter via CP_REG_TO_MEM remain unchanged (they run when the GPU is active and safe): for example the stats reads in submit paths still use `REG_A6XX_CP_ALWAYS_ON_COUNTER` (e.g., drivers/gpu/drm/msm/adreno/a6xx_gpu.c:372-375).
- Why this is a good stable candidate - Real bug impact: CPU reads of CP_ALWAYS_ON during IFPC can be stale, zero, or require disruptive OOB votes that wake the domain; this can cause incorrect timestamps (MSM_PARAM_TIMESTAMP), spurious power- ups, and trace anomalies. Moving to GMU AO fixes this by design. - Small, contained, and low risk: - All changes are local to the MSM Adreno a6xx/a7xx driver and a single source file. - No ABI or feature changes; only the source of the timestamp for CPU reads changes. - The helper uses a standard hi-lo-hi pattern to ensure a correct 64-bit read. - The GMU AO counter is already described in the hardware XML and used elsewhere (e.g., other GMU counters), and the driver already depends on GMU MMIO. - Maintains timing consistency: - Both CP_ALWAYS_ON and GMU_ALWAYS_ON are clocked from XO (19.2 MHz); userspace semantics are preserved. See the driver also treating GMU counters as 19.2 MHz (e.g., drivers/gpu/drm/msm/adreno/a6xx_gpu.c:2361-2369).
- Side effects and regressions - Positive: avoids GMU OOB perfcounter votes and GMU lock/handshake just to read a timestamp, reducing the chance of deadlocks or long- latency paths during IFPC. - No architectural changes; no changes to command submission ordering or power sequencing. - Tracepoints now log the GMU AO value; this improves reliability during IFPC without affecting functionality.
- Dependencies present - `gmu_read()` and the `REG_A6XX_GMU_ALWAYS_ON_COUNTER_{H,L}` macros are already in-tree (drivers/gpu/drm/msm/adreno/a6xx_gmu.h:122 and drivers/gpu/drm/msm/registers/adreno/a6xx_gmu.xml:131-132). - The patch updates only `drivers/gpu/drm/msm/adreno/a6xx_gpu.c`, and aligns with existing GMU usage patterns.
Conclusion: This is a targeted, safe bug fix that improves timestamp reliability and avoids unnecessary power domain manipulations during IFPC. It meets stable backport criteria.
drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 30 ++++++++++++++------------- 1 file changed, 16 insertions(+), 14 deletions(-)
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c index 536da1acf615e..1e363af319488 100644 --- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c +++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c @@ -16,6 +16,19 @@
#define GPU_PAS_ID 13
+static u64 read_gmu_ao_counter(struct a6xx_gpu *a6xx_gpu) +{ + u64 count_hi, count_lo, temp; + + do { + count_hi = gmu_read(&a6xx_gpu->gmu, REG_A6XX_GMU_ALWAYS_ON_COUNTER_H); + count_lo = gmu_read(&a6xx_gpu->gmu, REG_A6XX_GMU_ALWAYS_ON_COUNTER_L); + temp = gmu_read(&a6xx_gpu->gmu, REG_A6XX_GMU_ALWAYS_ON_COUNTER_H); + } while (unlikely(count_hi != temp)); + + return (count_hi << 32) | count_lo; +} + static bool fence_status_check(struct msm_gpu *gpu, u32 offset, u32 value, u32 status, u32 mask) { /* Success if !writedropped0/1 */ @@ -376,8 +389,7 @@ static void a6xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit) OUT_RING(ring, upper_32_bits(rbmemptr(ring, fence))); OUT_RING(ring, submit->seqno);
- trace_msm_gpu_submit_flush(submit, - gpu_read64(gpu, REG_A6XX_CP_ALWAYS_ON_COUNTER)); + trace_msm_gpu_submit_flush(submit, read_gmu_ao_counter(a6xx_gpu));
a6xx_flush(gpu, ring); } @@ -577,8 +589,7 @@ static void a7xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit) }
- trace_msm_gpu_submit_flush(submit, - gpu_read64(gpu, REG_A6XX_CP_ALWAYS_ON_COUNTER)); + trace_msm_gpu_submit_flush(submit, read_gmu_ao_counter(a6xx_gpu));
a6xx_flush(gpu, ring);
@@ -2260,16 +2271,7 @@ static int a6xx_gmu_get_timestamp(struct msm_gpu *gpu, uint64_t *value) struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu); struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
- mutex_lock(&a6xx_gpu->gmu.lock); - - /* Force the GPU power on so we can read this register */ - a6xx_gmu_set_oob(&a6xx_gpu->gmu, GMU_OOB_PERFCOUNTER_SET); - - *value = gpu_read64(gpu, REG_A6XX_CP_ALWAYS_ON_COUNTER); - - a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_PERFCOUNTER_SET); - - mutex_unlock(&a6xx_gpu->gmu.lock); + *value = read_gmu_ao_counter(a6xx_gpu);
return 0; }