drm/amdkfd: Add support for more per-process flag

hkasivis · alexdeucher · commit cf6d949a409e · 2025-03-07T15:33:49.000-05:00
Add support for more per-process flags starting with option to configure
MFMA precision for gfx 9.5

v2: Change flag name to KFD_PROC_FLAG_MFMA_HIGH_PRECISION
    Remove unused else condition
v3: Bump the KFD API version
v4: Missed SH_MEM_CONFIG__PRECISION_MODE__SHIFT define. Added it.

Signed-off-by: Harish Kasiviswanathan &lt;Harish.Kasiviswanathan@amd.com&gt;
Reviewed-by: Felix Kuehling &lt;felix.kuehling@amd.com&gt;
Reviewed-by: Amber Lin &lt;Amber.Lin@amd.com&gt;
Signed-off-by: Alex Deucher &lt;alexander.deucher@amd.com&gt;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -606,7 +606,8 @@ static int kfd_ioctl_set_memory_policy(struct file *filep,
 				default_policy,
 				alternate_policy,
 				(void __user *)args->alternate_aperture_base,
-				args->alternate_aperture_size))
+				args->alternate_aperture_size,
+				args->misc_process_flag))
 		err = -EINVAL;
 
 out:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -2596,7 +2596,8 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm,
 				   enum cache_policy default_policy,
 				   enum cache_policy alternate_policy,
 				   void __user *alternate_aperture_base,
-				   uint64_t alternate_aperture_size)
+				   uint64_t alternate_aperture_size,
+				   u32 misc_process_properties)
 {
 	bool retval = true;
 
@@ -2611,7 +2612,8 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm,
 			default_policy,
 			alternate_policy,
 			alternate_aperture_base,
-			alternate_aperture_size);
+			alternate_aperture_size,
+			misc_process_properties);
 
 	if (retval)
 		goto out;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
@@ -174,7 +174,8 @@ struct device_queue_manager_ops {
 					   enum cache_policy default_policy,
 					   enum cache_policy alternate_policy,
 					   void __user *alternate_aperture_base,
-					   uint64_t alternate_aperture_size);
+					   uint64_t alternate_aperture_size,
+					   u32 misc_process_properties);
 
 	int (*process_termination)(struct device_queue_manager *dqm,
 			struct qcm_process_device *qpd);
@@ -210,7 +211,8 @@ struct device_queue_manager_asic_ops {
 					   enum cache_policy default_policy,
 					   enum cache_policy alternate_policy,
 					   void __user *alternate_aperture_base,
-					   uint64_t alternate_aperture_size);
+					   uint64_t alternate_aperture_size,
+					   u32 misc_process_properties);
 	void	(*init_sdma_vm)(struct device_queue_manager *dqm,
 				struct queue *q,
 				struct qcm_process_device *qpd);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c
@@ -40,7 +40,8 @@ static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm,
 				   enum cache_policy default_policy,
 				   enum cache_policy alternate_policy,
 				   void __user *alternate_aperture_base,
-				   uint64_t alternate_aperture_size);
+				   uint64_t alternate_aperture_size,
+				   u32 misc_process_properties);
 static int update_qpd_cik(struct device_queue_manager *dqm,
 			  struct qcm_process_device *qpd);
 static void init_sdma_vm(struct device_queue_manager *dqm,
@@ -88,7 +89,8 @@ static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm,
 				   enum cache_policy default_policy,
 				   enum cache_policy alternate_policy,
 				   void __user *alternate_aperture_base,
-				   uint64_t alternate_aperture_size)
+				   uint64_t alternate_aperture_size,
+				   u32 misc_process_properties)
 {
 	uint32_t default_mtype;
 	uint32_t ape1_mtype;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c
@@ -36,7 +36,8 @@ static bool set_cache_memory_policy_v10(struct device_queue_manager *dqm,
 				   enum cache_policy default_policy,
 				   enum cache_policy alternate_policy,
 				   void __user *alternate_aperture_base,
-				   uint64_t alternate_aperture_size);
+				   uint64_t alternate_aperture_size,
+				   u32 misc_process_properties);
 
 void device_queue_manager_init_v10(
 	struct device_queue_manager_asic_ops *asic_ops)
@@ -61,7 +62,8 @@ static bool set_cache_memory_policy_v10(struct device_queue_manager *dqm,
 				   enum cache_policy default_policy,
 				   enum cache_policy alternate_policy,
 				   void __user *alternate_aperture_base,
-				   uint64_t alternate_aperture_size)
+				   uint64_t alternate_aperture_size,
+				   u32 misc_process_properties)
 {
 	qpd->sh_mem_config = (SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
 			      SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) |
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v11.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v11.c
@@ -35,7 +35,8 @@ static bool set_cache_memory_policy_v11(struct device_queue_manager *dqm,
 				   enum cache_policy default_policy,
 				   enum cache_policy alternate_policy,
 				   void __user *alternate_aperture_base,
-				   uint64_t alternate_aperture_size);
+				   uint64_t alternate_aperture_size,
+				   u32 misc_process_properties);
 
 void device_queue_manager_init_v11(
 	struct device_queue_manager_asic_ops *asic_ops)
@@ -60,7 +61,8 @@ static bool set_cache_memory_policy_v11(struct device_queue_manager *dqm,
 				   enum cache_policy default_policy,
 				   enum cache_policy alternate_policy,
 				   void __user *alternate_aperture_base,
-				   uint64_t alternate_aperture_size)
+				   uint64_t alternate_aperture_size,
+				   u32 misc_process_properties)
 {
 	qpd->sh_mem_config = (SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
 			      SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) |
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v12.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v12.c
@@ -35,7 +35,8 @@ static bool set_cache_memory_policy_v12(struct device_queue_manager *dqm,
 				   enum cache_policy default_policy,
 				   enum cache_policy alternate_policy,
 				   void __user *alternate_aperture_base,
-				   uint64_t alternate_aperture_size);
+				   uint64_t alternate_aperture_size,
+				   u32 misc_process_properties);
 
 void device_queue_manager_init_v12(
 	struct device_queue_manager_asic_ops *asic_ops)
@@ -60,7 +61,8 @@ static bool set_cache_memory_policy_v12(struct device_queue_manager *dqm,
 				   enum cache_policy default_policy,
 				   enum cache_policy alternate_policy,
 				   void __user *alternate_aperture_base,
-				   uint64_t alternate_aperture_size)
+				   uint64_t alternate_aperture_size,
+				   u32 misc_process_properties)
 {
 	qpd->sh_mem_config = (SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
 			      SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) |
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c
@@ -35,7 +35,8 @@ static bool set_cache_memory_policy_v9(struct device_queue_manager *dqm,
 				   enum cache_policy default_policy,
 				   enum cache_policy alternate_policy,
 				   void __user *alternate_aperture_base,
-				   uint64_t alternate_aperture_size);
+				   uint64_t alternate_aperture_size,
+				   u32 misc_process_properties);
 
 void device_queue_manager_init_v9(
 	struct device_queue_manager_asic_ops *asic_ops)
@@ -60,7 +61,8 @@ static bool set_cache_memory_policy_v9(struct device_queue_manager *dqm,
 				   enum cache_policy default_policy,
 				   enum cache_policy alternate_policy,
 				   void __user *alternate_aperture_base,
-				   uint64_t alternate_aperture_size)
+				   uint64_t alternate_aperture_size,
+				   u32 misc_process_properties)
 {
 	qpd->sh_mem_config = SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
 				SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
@@ -73,6 +75,11 @@ static bool set_cache_memory_policy_v9(struct device_queue_manager *dqm,
 		KFD_GC_VERSION(dqm->dev->kfd) == IP_VERSION(9, 5, 0))
 		qpd->sh_mem_config |= (1 << SH_MEM_CONFIG__F8_MODE__SHIFT);
 
+	if (KFD_GC_VERSION(dqm->dev->kfd) == IP_VERSION(9, 5, 0)) {
+		if (misc_process_properties & KFD_PROC_FLAG_MFMA_HIGH_PRECISION)
+			qpd->sh_mem_config |= 1 << SH_MEM_CONFIG__PRECISION_MODE__SHIFT;
+	}
+
 	qpd->sh_mem_ape1_limit = 0;
 	qpd->sh_mem_ape1_base = 0;
 	qpd->sh_mem_bases = compute_sh_mem_bases_64bit(qpd_to_pdd(qpd));
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c
@@ -40,7 +40,8 @@ static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
 				       enum cache_policy default_policy,
 				       enum cache_policy alternate_policy,
 				       void __user *alternate_aperture_base,
-				       uint64_t alternate_aperture_size);
+				       uint64_t alternate_aperture_size,
+				       u32 misc_process_properties);
 static int update_qpd_vi(struct device_queue_manager *dqm,
 			 struct qcm_process_device *qpd);
 static void init_sdma_vm(struct device_queue_manager *dqm,
@@ -89,7 +90,8 @@ static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
 		enum cache_policy default_policy,
 		enum cache_policy alternate_policy,
 		void __user *alternate_aperture_base,
-		uint64_t alternate_aperture_size)
+		uint64_t alternate_aperture_size,
+		u32 misc_process_properties)
 {
 	uint32_t default_mtype;
 	uint32_t ape1_mtype;
diff --git a/drivers/gpu/drm/amd/include/asic_reg/gc/gc_9_4_3_sh_mask.h b/drivers/gpu/drm/amd/include/asic_reg/gc/gc_9_4_3_sh_mask.h
@@ -2261,11 +2261,13 @@
 #define SH_MEM_CONFIG__ADDRESS_MODE__SHIFT                                                                    0x0
 #define SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT                                                                  0x3
 #define SH_MEM_CONFIG__F8_MODE__SHIFT                                                                         0x8
+#define SH_MEM_CONFIG__PRECISION_MODE__SHIFT                                                                  0x9
 #define SH_MEM_CONFIG__RETRY_DISABLE__SHIFT                                                                   0xc
 #define SH_MEM_CONFIG__PRIVATE_NV__SHIFT                                                                      0xd
 #define SH_MEM_CONFIG__ADDRESS_MODE_MASK                                                                      0x00000001L
 #define SH_MEM_CONFIG__ALIGNMENT_MODE_MASK                                                                    0x00000018L
 #define SH_MEM_CONFIG__F8_MODE_MASK                                                                           0x00000100L
+#define SH_MEM_CONFIG__PRECISION_MODE_MASK                                                                    0x00000200L
 #define SH_MEM_CONFIG__RETRY_DISABLE_MASK                                                                     0x00001000L
 #define SH_MEM_CONFIG__PRIVATE_NV_MASK                                                                        0x00002000L
 //SP_MFMA_PORTD_RD_CONFIG
diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
@@ -43,9 +43,10 @@
  * - 1.15 - Enable managing mappings in compute VMs with GEM_VA ioctl
  * - 1.16 - Add contiguous VRAM allocation flag
  * - 1.17 - Add SDMA queue creation with target SDMA engine ID
+ * - 1.18 - Rename pad in set_memory_policy_args to misc_process_flag
  */
 #define KFD_IOCTL_MAJOR_VERSION 1
-#define KFD_IOCTL_MINOR_VERSION 17
+#define KFD_IOCTL_MINOR_VERSION 18
 
 struct kfd_ioctl_get_version_args {
 	__u32 major_version;	/* from KFD */
@@ -150,14 +151,17 @@ struct kfd_dbg_device_info_entry {
 #define KFD_IOC_CACHE_POLICY_COHERENT 0
 #define KFD_IOC_CACHE_POLICY_NONCOHERENT 1
 
+/* Misc. per process flags */
+#define KFD_PROC_FLAG_MFMA_HIGH_PRECISION (1 << 0)
+
 struct kfd_ioctl_set_memory_policy_args {
 	__u64 alternate_aperture_base;	/* to KFD */
 	__u64 alternate_aperture_size;	/* to KFD */
 
 	__u32 gpu_id;			/* to KFD */
 	__u32 default_policy;		/* to KFD */
 	__u32 alternate_policy;		/* to KFD */
-	__u32 pad;
+	__u32 misc_process_flag;        /* to KFD */
 };
 
 /*
