drm/amdgpu: prepare map process for multi-process debug devices

Unlike single process debug devices, multi-process debug devices allow
debug mode setting per-VMID (non-device-global).

Because the HWS manages PASID-VMID mapping, the new MAP_PROCESS API allows
the KFD to forward the required SPI debug register write requests.

To request a new debug mode setting change, the KFD must be able to
preempt all queues then remap all queues with these new setting
requests for MAP_PROCESS to take effect.

Note that by default, trap enablement in non-debug mode must be disabled
for performance reasons for multi-process debug devices due to setup
overhead in FW.
Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdkfd/kfd_debug.h

@@ -29,4 +29,9 @@ 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,
 			uint32_t *runtime_info_size);
+static inline bool kfd_dbg_is_per_vmid_supported(struct kfd_node *dev)
+{
+	return KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 2);
+}
+
 #endif

drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c

@@ -36,6 +36,7 @@
 #include "kfd_kernel_queue.h"
 #include "amdgpu_amdkfd.h"
 #include "mes_api_def.h"
+#include "kfd_debug.h"
 
 /* Size of the per-pipe EOP queue */
 #define CIK_HPD_EOP_BYTES_LOG2 11
@@ -2594,6 +2595,56 @@ int release_debug_trap_vmid(struct device_queue_manager *dqm,
 	return r;
 }
 
+int debug_lock_and_unmap(struct device_queue_manager *dqm)
+{
+	int r;
+
+	if (dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) {
+		pr_err("Unsupported on sched_policy: %i\n", dqm->sched_policy);
+		return -EINVAL;
+	}
+
+	if (!kfd_dbg_is_per_vmid_supported(dqm->dev))
+		return 0;
+
+	dqm_lock(dqm);
+
+	r = unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, 0, 0, false);
+	if (r)
+		dqm_unlock(dqm);
+
+	return r;
+}
+
+int debug_map_and_unlock(struct device_queue_manager *dqm)
+{
+	int r;
+
+	if (dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) {
+		pr_err("Unsupported on sched_policy: %i\n", dqm->sched_policy);
+		return -EINVAL;
+	}
+
+	if (!kfd_dbg_is_per_vmid_supported(dqm->dev))
+		return 0;
+
+	r = map_queues_cpsch(dqm);
+
+	dqm_unlock(dqm);
+
+	return r;
+}
+
+int debug_refresh_runlist(struct device_queue_manager *dqm)
+{
+	int r = debug_lock_and_unmap(dqm);
+
+	if (r)
+		return r;
+
+	return debug_map_and_unlock(dqm);
+}
+
 #if defined(CONFIG_DEBUG_FS)
 
 static void seq_reg_dump(struct seq_file *m,

drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h

@@ -290,6 +290,9 @@ int reserve_debug_trap_vmid(struct device_queue_manager *dqm,
 			struct qcm_process_device *qpd);
 int release_debug_trap_vmid(struct device_queue_manager *dqm,
 			struct qcm_process_device *qpd);
+int debug_lock_and_unmap(struct device_queue_manager *dqm);
+int debug_map_and_unlock(struct device_queue_manager *dqm);
+int debug_refresh_runlist(struct device_queue_manager *dqm);
 
 static inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd)
 {

drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c

@@ -88,6 +88,10 @@ static int pm_map_process_aldebaran(struct packet_manager *pm,
 {
 	struct pm4_mes_map_process_aldebaran *packet;
 	uint64_t vm_page_table_base_addr = qpd->page_table_base;
+	struct kfd_dev *kfd = pm->dqm->dev->kfd;
+	struct kfd_process_device *pdd =
+			container_of(qpd, struct kfd_process_device, qpd);
+	int i;
 
 	packet = (struct pm4_mes_map_process_aldebaran *)buffer;
 	memset(buffer, 0, sizeof(struct pm4_mes_map_process_aldebaran));
@@ -102,6 +106,16 @@ static int pm_map_process_aldebaran(struct packet_manager *pm,
 	packet->bitfields14.num_oac = qpd->num_oac;
 	packet->bitfields14.sdma_enable = 1;
 	packet->bitfields14.num_queues = (qpd->is_debug) ? 0 : qpd->queue_count;
+	packet->spi_gdbg_per_vmid_cntl = pdd->spi_dbg_override |
+						pdd->spi_dbg_launch_mode;
+
+	if (pdd->process->debug_trap_enabled) {
+		for (i = 0; i < kfd->device_info.num_of_watch_points; i++)
+			packet->tcp_watch_cntl[i] = pdd->watch_points[i];
+
+		packet->bitfields2.single_memops =
+				!!(pdd->process->dbg_flags & KFD_DBG_TRAP_FLAG_SINGLE_MEM_OP);
+	}
 
 	packet->sh_mem_config = qpd->sh_mem_config;
 	packet->sh_mem_bases = qpd->sh_mem_bases;

drivers/gpu/drm/amd/amdkfd/kfd_priv.h

@@ -816,6 +816,12 @@ struct kfd_process_device {
 	uint64_t faults;
 	uint64_t page_in;
 	uint64_t page_out;
+
+	/* Tracks debug per-vmid request settings */
+	uint32_t spi_dbg_override;
+	uint32_t spi_dbg_launch_mode;
+	uint32_t watch_points[4];
+
 	/*
 	 * If this process has been checkpointed before, then the user
 	 * application will use the original gpu_id on the
@@ -952,6 +958,9 @@ struct kfd_process {
 
 	bool xnack_enabled;
 
+	/* Tracks debug per-vmid request for debug flags */
+	bool dbg_flags;
+
 	atomic_t poison;
 	/* Queues are in paused stated because we are in the process of doing a CRIU checkpoint */
 	bool queues_paused;

drivers/gpu/drm/amd/amdkfd/kfd_process.c

@@ -1612,6 +1612,11 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_node *dev,
 	}
 
 	p->pdds[p->n_pdds++] = pdd;
+	if (kfd_dbg_is_per_vmid_supported(pdd->dev))
+		pdd->spi_dbg_override = pdd->dev->kfd2kgd->disable_debug_trap(
+							pdd->dev->adev,
+							false,
+							0);
 
 	/* Init idr used for memory handle translation */
 	idr_init(&pdd->alloc_idr);