Re: [PATCH 18/33] drm/amdkfd: add raise exception event function

[Felix Kuehling <felix.kuehling@xxxxxxx>]

Am 2023-05-25 um 13:27 schrieb Jonathan Kim:
> Exception events can be generated from interrupts or queue activitity.
>
> The raise event function will save exception status of a queue, device
> or process then notify the debugger of the status change by writing to
> a debugger polled file descriptor that the debugger provides during
> debug attach.
>
> For memory violation exceptions, extra exception data will be saved.
>
> The debugger will be able to query the saved exception states by query
> operation that will be provided by follow up patches.
>
> v2: use new kfd_node struct in prototype.
>
> Signed-off-by: Jonathan Kim <jonathan.kim@xxxxxxx>

Reviewed-by: Felix Kuehling <Felix.Kuehling@xxxxxxx>


> ---
>   drivers/gpu/drm/amd/amdkfd/kfd_debug.c   | 104 +++++++++++++++++++++++
>   drivers/gpu/drm/amd/amdkfd/kfd_debug.h   |   7 ++
>   drivers/gpu/drm/amd/amdkfd/kfd_priv.h    |  10 +++
>   drivers/gpu/drm/amd/amdkfd/kfd_process.c |   2 +
>   4 files changed, 123 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
> index 5e2ee2d1acc4..dccb27fc764b 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
> @@ -24,6 +24,107 @@
>   #include "kfd_device_queue_manager.h"
>   #include <linux/file.h>
>   
> +void debug_event_write_work_handler(struct work_struct *work)
> +{
> +	struct kfd_process *process;
> +
> +	static const char write_data = '.';
> +	loff_t pos = 0;
> +
> +	process = container_of(work,
> +			struct kfd_process,
> +			debug_event_workarea);
> +
> +	kernel_write(process->dbg_ev_file, &write_data, 1, &pos);
> +}
> +
> +/* update process/device/queue exception status, write to descriptor
> + * only if exception_status is enabled.
> + */
> +bool kfd_dbg_ev_raise(uint64_t event_mask,
> +			struct kfd_process *process, struct kfd_node *dev,
> +			unsigned int source_id, bool use_worker,
> +			void *exception_data, size_t exception_data_size)
> +{
> +	struct process_queue_manager *pqm;
> +	struct process_queue_node *pqn;
> +	int i;
> +	static const char write_data = '.';
> +	loff_t pos = 0;
> +	bool is_subscribed = true;
> +
> +	if (!(process && process->debug_trap_enabled))
> +		return false;
> +
> +	mutex_lock(&process->event_mutex);
> +
> +	if (event_mask & KFD_EC_MASK_DEVICE) {
> +		for (i = 0; i < process->n_pdds; i++) {
> +			struct kfd_process_device *pdd = process->pdds[i];
> +
> +			if (pdd->dev != dev)
> +				continue;
> +
> +			pdd->exception_status |= event_mask & KFD_EC_MASK_DEVICE;
> +
> +			if (event_mask & KFD_EC_MASK(EC_DEVICE_MEMORY_VIOLATION)) {
> +				if (!pdd->vm_fault_exc_data) {
> +					pdd->vm_fault_exc_data = kmemdup(
> +							exception_data,
> +							exception_data_size,
> +							GFP_KERNEL);
> +					if (!pdd->vm_fault_exc_data)
> +						pr_debug("Failed to allocate exception data memory");
> +				} else {
> +					pr_debug("Debugger exception data not saved\n");
> +					print_hex_dump_bytes("exception data: ",
> +							DUMP_PREFIX_OFFSET,
> +							exception_data,
> +							exception_data_size);
> +				}
> +			}
> +			break;
> +		}
> +	} else if (event_mask & KFD_EC_MASK_PROCESS) {
> +		process->exception_status |= event_mask & KFD_EC_MASK_PROCESS;
> +	} else {
> +		pqm = &process->pqm;
> +		list_for_each_entry(pqn, &pqm->queues,
> +				process_queue_list) {
> +			int target_id;
> +
> +			if (!pqn->q)
> +				continue;
> +
> +			target_id = event_mask & KFD_EC_MASK(EC_QUEUE_NEW) ?
> +					pqn->q->properties.queue_id :
> +							pqn->q->doorbell_id;
> +
> +			if (pqn->q->device != dev || target_id != source_id)
> +				continue;
> +
> +			pqn->q->properties.exception_status |= event_mask;
> +			break;
> +		}
> +	}
> +
> +	if (process->exception_enable_mask & event_mask) {
> +		if (use_worker)
> +			schedule_work(&process->debug_event_workarea);
> +		else
> +			kernel_write(process->dbg_ev_file,
> +					&write_data,
> +					1,
> +					&pos);
> +	} else {
> +		is_subscribed = false;
> +	}
> +
> +	mutex_unlock(&process->event_mutex);
> +
> +	return is_subscribed;
> +}
> +
>   static int kfd_dbg_set_queue_workaround(struct queue *q, bool enable)
>   {
>   	struct mqd_update_info minfo = {0};
> @@ -99,6 +200,9 @@ static void kfd_dbg_trap_deactivate(struct kfd_process *target, bool unwind, int
>   {
>   	int i;
>   
> +	if (!unwind)
> +		cancel_work_sync(&target->debug_event_workarea);
> +
>   	for (i = 0; i < target->n_pdds; i++) {
>   		struct kfd_process_device *pdd = target->pdds[i];
>   
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
> index 3e56225f6ef6..66ee7b95d08a 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
> @@ -25,6 +25,11 @@
>   
>   #include "kfd_priv.h"
>   
> +bool kfd_dbg_ev_raise(uint64_t event_mask,
> +			struct kfd_process *process, struct kfd_node *dev,
> +			unsigned int source_id, bool use_worker,
> +			void *exception_data,
> +			size_t exception_data_size);
>   int kfd_dbg_trap_disable(struct kfd_process *target);
>   int kfd_dbg_trap_enable(struct kfd_process *target, uint32_t fd,
>   			void __user *runtime_info,
> @@ -35,6 +40,8 @@ static inline bool kfd_dbg_is_per_vmid_supported(struct kfd_node *dev)
>   	       KFD_GC_VERSION(dev) >= IP_VERSION(11, 0, 0);
>   }
>   
> +void debug_event_write_work_handler(struct work_struct *work);
> +
>   /*
>    * If GFX off is enabled, chips that do not support RLC restore for the debug
>    * registers will disable GFX off temporarily for the entire debug session.
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
> index f0a45d184c8f..b18cd4bf76bf 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
> @@ -529,6 +529,7 @@ struct queue_properties {
>   	uint32_t ctl_stack_size;
>   	uint64_t tba_addr;
>   	uint64_t tma_addr;
> +	uint64_t exception_status;
>   };
>   
>   #define QUEUE_IS_ACTIVE(q) ((q).queue_size > 0 &&	\
> @@ -820,6 +821,11 @@ struct kfd_process_device {
>   	uint64_t page_in;
>   	uint64_t page_out;
>   
> +	/* Exception code status*/
> +	uint64_t exception_status;
> +	void *vm_fault_exc_data;
> +	size_t vm_fault_exc_data_size;
> +
>   	/* Tracks debug per-vmid request settings */
>   	uint32_t spi_dbg_override;
>   	uint32_t spi_dbg_launch_mode;
> @@ -955,12 +961,16 @@ struct kfd_process {
>   
>   	/* Exception code enable mask and status */
>   	uint64_t exception_enable_mask;
> +	uint64_t exception_status;
>   
>   	/* shared virtual memory registered by this process */
>   	struct svm_range_list svms;
>   
>   	bool xnack_enabled;
>   
> +	/* Work area for debugger event writer worker. */
> +	struct work_struct debug_event_workarea;
> +
>   	/* Tracks debug per-vmid request for debug flags */
>   	bool dbg_flags;
>   
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> index a3846bb87510..fa19c1218748 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> @@ -1509,6 +1509,8 @@ static struct kfd_process *create_process(const struct task_struct *thread)
>   	kfd_unref_process(process);
>   	get_task_struct(process->lead_thread);
>   
> +	INIT_WORK(&process->debug_event_workarea, debug_event_write_work_handler);
> +
>   	return process;
>   
>   err_register_notifier: