Op 10-09-2024 om 16:31 schreef Sathishkumar S:
JPEG_4_0_3 has up to 32 jpeg cores and a single mjpeg video decode
will use all available cores on the hardware. This debugfs entry
helps to disable or enable job submission to a cluster of cores or
one specific core in the ip for debugging. The entry is populated
only if there is at least two or more cores in the jpeg ip.
Signed-off-by: Sathishkumar S <sathishkumar.sundararaju@xxxxxxx>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c | 71 +++++++++++++++++++++
drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h | 1 +
3 files changed, 74 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
index cbef720de779..37d8657f0776 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
@@ -2095,6 +2095,8 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev)
if (amdgpu_umsch_mm & amdgpu_umsch_mm_fwlog)
amdgpu_debugfs_umsch_fwlog_init(adev, &adev->umsch_mm);
+ amdgpu_debugfs_jpeg_sched_mask_init(adev);
+
amdgpu_ras_debugfs_create_all(adev);
amdgpu_rap_debugfs_init(adev);
amdgpu_securedisplay_debugfs_init(adev);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c
index 6df99cb00d9a..edf0a50dd345 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c
@@ -342,3 +342,74 @@ int amdgpu_jpeg_psp_update_sram(struct amdgpu_device *adev, int inst_idx,
return psp_execute_ip_fw_load(&adev->psp, &ucode);
}
+
+/*
+ * debugfs for to enable/disable jpeg job submission to specific core.
+ */
+#if defined(CONFIG_DEBUG_FS)
+static int amdgpu_debugfs_jpeg_sched_mask_set(void *data, u64 val)
+{
+ struct amdgpu_device *adev = (struct amdgpu_device *) data;
+ uint32_t i, j;
+ uint64_t mask = 0;
+ struct amdgpu_ring *ring;
+
+ if (!adev)
+ return -ENODEV;
+
+ mask = (1 << (adev->jpeg.num_jpeg_inst * adev->jpeg.num_jpeg_rings)) - 1;
This expression "1 << ..." is evaluated as 32 bits because of the plain
"1". If
the shift is more than 32 then the result of the shift will be 0.
I suggest to change the expression to something like this (assuming ULL
makes it 64 bits)
mask = (1ULL << (adev->jpeg.num_jpeg_inst * adev->jpeg.num_jpeg_rings)) - 1;
or
mask = ((u64)1 << (adev->jpeg.num_jpeg_inst *
adev->jpeg.num_jpeg_rings)) - 1;
[...]
