[PATCH 24/29] drm/amdkfd: add debug set flags operation

Felix Kuehling felix.kuehling at amd.com
Wed Nov 30 00:39:32 UTC 2022


On 2022-10-31 12:23, Jonathan Kim wrote:
> Allow the debugger to set single memory and single ALU operations.
>
> Some exceptions are imprecise (memory violations, address watch) in the
> sense that a trap occurs only when the exception interrupt occurs and
> not at the non-halting faulty instruction.  Trap temporaries 0 & 1 save
> the program counter address, which means that these values will not point
> to the faulty instruction address but to whenever the interrupt was
> raised.
>
> Setting the Single Memory Operations flag will inject an automatic wait
> on every memory operation instruction forcing imprecise memory exceptions
> to become precise at the cost of performance.  This setting is not
> permitted on debug devices that support only a global setting of this
> option.
>
> Likewise, Single ALU Operations will force in-order ALU operations.
> Although this is available on current hardware, it's not required so it
> will be treated as a NOP.

Having a flag in the API that is just ignored is misleading. I think we 
should either remove it from the API for now, or at least make the 
function return an error if a debugger attempts to set the precise-ALU 
flag. This would be consistent with attempting to set a flag that is not 
supported on the HW.

Regards,
   Felix


>
> Return the previous set flags to the debugger as well.
>
> Signed-off-by: Jonathan Kim <jonathan.kim at amd.com>
> ---
>   drivers/gpu/drm/amd/amdkfd/kfd_chardev.c |  2 ++
>   drivers/gpu/drm/amd/amdkfd/kfd_debug.c   | 35 ++++++++++++++++++++++++
>   drivers/gpu/drm/amd/amdkfd/kfd_debug.h   |  1 +
>   3 files changed, 38 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
> index 9b2ea6e9e078..200e11f02382 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
> @@ -2943,6 +2943,8 @@ static int kfd_ioctl_set_debug_trap(struct file *filep, struct kfd_process *p, v
>   				args->clear_node_address_watch.id);
>   		break;
>   	case KFD_IOC_DBG_TRAP_SET_FLAGS:
> +		r = kfd_dbg_trap_set_flags(target, &args->set_flags.flags);
> +		break;
>   	case KFD_IOC_DBG_TRAP_QUERY_DEBUG_EVENT:
>   	case KFD_IOC_DBG_TRAP_QUERY_EXCEPTION_INFO:
>   	case KFD_IOC_DBG_TRAP_GET_QUEUE_SNAPSHOT:
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
> index 68bc1d5bfd05..1f4d3fa0278e 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
> @@ -23,6 +23,7 @@
>   #include "kfd_debug.h"
>   #include "kfd_device_queue_manager.h"
>   #include <linux/file.h>
> +#include <uapi/linux/kfd_ioctl.h>
>   
>   /*
>    * The spinlock protects the per device dev->alloc_watch_ids for multi-process access.
> @@ -355,6 +356,37 @@ static void kfd_dbg_clear_process_address_watch(struct kfd_process *target)
>   			kfd_dbg_trap_clear_dev_address_watch(target->pdds[i], j);
>   }
>   
> +int kfd_dbg_trap_set_flags(struct kfd_process *target, uint32_t *flags)
> +{
> +	uint32_t prev_flags = target->dbg_flags;
> +	int i, r = 0;
> +
> +	for (i = 0; i < target->n_pdds; i++) {
> +		if (!kfd_dbg_is_per_vmid_supported(target->pdds[i]->dev) &&
> +			(*flags & KFD_DBG_TRAP_FLAG_SINGLE_MEM_OP)) {
> +			*flags = prev_flags;
> +			return -EACCES;
> +		}
> +	}
> +
> +	target->dbg_flags = *flags;
> +	*flags = prev_flags;
> +	for (i = 0; i < target->n_pdds; i++) {
> +		struct kfd_process_device *pdd = target->pdds[i];
> +
> +		if (!kfd_dbg_is_per_vmid_supported(pdd->dev))
> +			continue;
> +
> +		r = debug_refresh_runlist(target->pdds[i]->dev->dqm);
> +		if (r) {
> +			target->dbg_flags = prev_flags;
> +			break;
> +		}
> +	}
> +
> +	return r;
> +}
> +
>   
>   /* kfd_dbg_trap_deactivate:
>    *	target: target process
> @@ -369,9 +401,12 @@ void kfd_dbg_trap_deactivate(struct kfd_process *target, bool unwind, int unwind
>   	int i, count = 0;
>   
>   	if (!unwind) {
> +		uint32_t flags = 0;
>   		cancel_work_sync(&target->debug_event_workarea);
>   		kfd_dbg_clear_process_address_watch(target);
>   		kfd_dbg_trap_set_wave_launch_mode(target, 0);
> +
> +		kfd_dbg_trap_set_flags(target, &flags);
>   	}
>   
>   	for (i = 0; i < target->n_pdds; i++) {
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
> index ad677e67e7eb..12b80b6c96d0 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
> @@ -57,6 +57,7 @@ int kfd_dbg_trap_set_dev_address_watch(struct kfd_process_device *pdd,
>   					uint32_t watch_address_mask,
>   					uint32_t *watch_id,
>   					uint32_t watch_mode);
> +int kfd_dbg_trap_set_flags(struct kfd_process *target, uint32_t *flags);
>   int kfd_dbg_send_exception_to_runtime(struct kfd_process *p,
>   					unsigned int dev_id,
>   					unsigned int queue_id,


More information about the amd-gfx mailing list