Implement xrt-smi query to report preemption counters #368
Conversation
| pci_dev_impl.ioctl(DRM_IOCTL_AMDXDNA_GET_INFO, &query_telemetry); | ||
|
|
||
| if (telemetry.major != NPU_TELEMETRY_VERSION_MAJOR || telemetry.minor != NPU_TELEMETRY_VERSION_MINOR) { | ||
| memset(&telemetry, 0, sizeof(telemetry)); |
There was a problem hiding this comment.
why do you need to clear the buffer?
There was a problem hiding this comment.
This was done to align with the Windows driver implementation for the Phoenix devices. Since we don't have preemption support in Phoenix, the expectation is that querying the preemption counter should return zeros instead of an exception.
There was a problem hiding this comment.
You want to clear the buffer, send it down to driver and return to xrt-smi with whatever is returned from driver. On PHX, since there is no preemption support, driver should have already put all zeros in the buffer, no?
There was a problem hiding this comment.
XDNA driver do put all zeros in the buffer.
- buff = dma_alloc_noncoherent()
- memset(buff, 0, aligned_buffer_size)
- query telemetry data from firmware
- copy_to_user(args->buffer, buff, args->buffer_size)
There was a problem hiding this comment.
As @mamin506 mentioned, the driver does initialize the buffer to all 0s. However, the firmware memsets it to all 0xFF values by default.
Implement xrt-smi query to report preemption counters. Signed-off-by: Nishad Saraf <[email protected]>
Implement xrt-smi query to report preemption counters.
PS: Error "Subcommand not found" is due to the recent changes in xrt, unrelated to this change.