drm/amdgpu: add configurable grace period for unmap queues

The HWS schedule allows a grace period for wave completion prior to
preemption for better performance by avoiding CWSR on waves that can
potentially complete quickly. The debugger, on the other hand, will
want to inspect wave status immediately after it actively triggers
preemption (a suspend function to be provided).

To minimize latency between preemption and debugger wave inspection, allow
immediate preemption by setting the grace period to 0.

Note that setting the preepmtion grace period to 0 will result in an
infinite grace period being set due to a CP FW bug so set it to 1 for now.

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>

1 files changed, 43 insertions, 0 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
index 8d7d04704b00..829ee720cc44 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
@@ -739,6 +739,24 @@ uint32_t kgd_gfx_v9_disable_debug_trap(struct amdgpu_device *adev,
 	return 0;
 }
 
+/* kgd_gfx_v9_get_iq_wait_times: Returns the mmCP_IQ_WAIT_TIME1/2 values
+ * The values read are:
+ *     ib_offload_wait_time     -- Wait Count for Indirect Buffer Offloads.
+ *     atomic_offload_wait_time -- Wait Count for L2 and GDS Atomics Offloads.
+ *     wrm_offload_wait_time    -- Wait Count for WAIT_REG_MEM Offloads.
+ *     gws_wait_time            -- Wait Count for Global Wave Syncs.
+ *     que_sleep_wait_time      -- Wait Count for Dequeue Retry.
+ *     sch_wave_wait_time       -- Wait Count for Scheduling Wave Message.
+ *     sem_rearm_wait_time      -- Wait Count for Semaphore re-arm.
+ *     deq_retry_wait_time      -- Wait Count for Global Wave Syncs.
+ */
+void kgd_gfx_v9_get_iq_wait_times(struct amdgpu_device *adev,
+					uint32_t *wait_times)
+
+{
+	*wait_times = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_IQ_WAIT_TIME2));
+}
+
 void kgd_gfx_v9_set_vm_context_page_table_base(struct amdgpu_device *adev,
 			uint32_t vmid, uint64_t page_table_base)
 {
@@ -926,6 +944,29 @@ void kgd_gfx_v9_get_cu_occupancy(struct amdgpu_device *adev, int pasid,
 				adev->gfx.cu_info.max_waves_per_simd;
 }
 
+void kgd_gfx_v9_build_grace_period_packet_info(struct amdgpu_device *adev,
+		uint32_t wait_times,
+		uint32_t grace_period,
+		uint32_t *reg_offset,
+		uint32_t *reg_data)
+{
+	*reg_data = wait_times;
+
+	/*
+	 * The CP cannont handle a 0 grace period input and will result in
+	 * an infinite grace period being set so set to 1 to prevent this.
+	 */
+	if (grace_period == 0)
+		grace_period = 1;
+
+	*reg_data = REG_SET_FIELD(*reg_data,
+			CP_IQ_WAIT_TIME2,
+			SCH_WAVE,
+			grace_period);
+
+	*reg_offset = SOC15_REG_OFFSET(GC, 0, mmCP_IQ_WAIT_TIME2);
+}
+
 void kgd_gfx_v9_program_trap_handler_settings(struct amdgpu_device *adev,
 		uint32_t vmid, uint64_t tba_addr, uint64_t tma_addr, uint32_t inst)
 {
@@ -969,6 +1010,8 @@ const struct kfd2kgd_calls gfx_v9_kfd2kgd = {
 	.set_vm_context_page_table_base = kgd_gfx_v9_set_vm_context_page_table_base,
 	.enable_debug_trap = kgd_gfx_v9_enable_debug_trap,
 	.disable_debug_trap = kgd_gfx_v9_disable_debug_trap,
+	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
+	.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
 	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
 	.program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings,
 };