[PATCH 1/3] drm/amdkfd: Set per-process flags only once cik/vi

[PATCH 1/3] drm/amdkfd: Set per-process flags only once cik/vi

[Amber Lin]

From: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>

Set per-process static sh_mem config only once during process
initialization. Move all static changes from update_qpd() which is
called each time a queue is created to set_cache_memory_policy() which
is called once during process initialization.

set_cache_memory_policy() is currently defined only for cik and vi
family. So this commit only focuses on these two. A separate commit will
address other asics.

Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
---
 .../drm/amd/amdkfd/kfd_device_queue_manager.c | 39 +---------
 .../amd/amdkfd/kfd_device_queue_manager_cik.c | 69 ++++++++++++------
 .../amd/amdkfd/kfd_device_queue_manager_vi.c  | 71 ++++++++++++-------
 3 files changed, 94 insertions(+), 85 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index f3f2fd6ee65c..d23c6a358d34 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -2591,14 +2591,6 @@ static int destroy_queue_cpsch(struct device_queue_manager *dqm,
 	return retval;
 }
 
-/*
- * Low bits must be 0000/FFFF as required by HW, high bits must be 0 to
- * stay in user mode.
- */
-#define APE1_FIXED_BITS_MASK 0xFFFF80000000FFFFULL
-/* APE1 limit is inclusive and 64K aligned. */
-#define APE1_LIMIT_ALIGNMENT 0xFFFF
-
 static bool set_cache_memory_policy(struct device_queue_manager *dqm,
 				   struct qcm_process_device *qpd,
 				   enum cache_policy default_policy,
@@ -2613,34 +2605,6 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm,
 
 	dqm_lock(dqm);
 
-	if (alternate_aperture_size == 0) {
-		/* base > limit disables APE1 */
-		qpd->sh_mem_ape1_base = 1;
-		qpd->sh_mem_ape1_limit = 0;
-	} else {
-		/*
-		 * In FSA64, APE1_Base[63:0] = { 16{SH_MEM_APE1_BASE[31]},
-		 *			SH_MEM_APE1_BASE[31:0], 0x0000 }
-		 * APE1_Limit[63:0] = { 16{SH_MEM_APE1_LIMIT[31]},
-		 *			SH_MEM_APE1_LIMIT[31:0], 0xFFFF }
-		 * Verify that the base and size parameters can be
-		 * represented in this format and convert them.
-		 * Additionally restrict APE1 to user-mode addresses.
-		 */
-
-		uint64_t base = (uintptr_t)alternate_aperture_base;
-		uint64_t limit = base + alternate_aperture_size - 1;
-
-		if (limit <= base || (base & APE1_FIXED_BITS_MASK) != 0 ||
-		   (limit & APE1_FIXED_BITS_MASK) != APE1_LIMIT_ALIGNMENT) {
-			retval = false;
-			goto out;
-		}
-
-		qpd->sh_mem_ape1_base = base >> 16;
-		qpd->sh_mem_ape1_limit = limit >> 16;
-	}
-
 	retval = dqm->asic_ops.set_cache_memory_policy(
 			dqm,
 			qpd,
@@ -2649,6 +2613,9 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm,
 			alternate_aperture_base,
 			alternate_aperture_size);
 
+	if (retval)
+		goto out;
+
 	if ((dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) && (qpd->vmid != 0))
 		program_sh_mem_settings(dqm, 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
index d4d95c7f2e5d..32bedef912b3 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c
@@ -27,6 +27,14 @@
 #include "oss/oss_2_4_sh_mask.h"
 #include "gca/gfx_7_2_sh_mask.h"
 
+/*
+ * Low bits must be 0000/FFFF as required by HW, high bits must be 0 to
+ * stay in user mode.
+ */
+#define APE1_FIXED_BITS_MASK 0xFFFF80000000FFFFULL
+/* APE1 limit is inclusive and 64K aligned. */
+#define APE1_LIMIT_ALIGNMENT 0xFFFF
+
 static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm,
 				   struct qcm_process_device *qpd,
 				   enum cache_policy default_policy,
@@ -84,6 +92,36 @@ static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm,
 {
 	uint32_t default_mtype;
 	uint32_t ape1_mtype;
+	unsigned int temp;
+	bool retval = true;
+
+	if (alternate_aperture_size == 0) {
+		/* base > limit disables APE1 */
+		qpd->sh_mem_ape1_base = 1;
+		qpd->sh_mem_ape1_limit = 0;
+	} else {
+		/*
+		 * In FSA64, APE1_Base[63:0] = { 16{SH_MEM_APE1_BASE[31]},
+		 *			SH_MEM_APE1_BASE[31:0], 0x0000 }
+		 * APE1_Limit[63:0] = { 16{SH_MEM_APE1_LIMIT[31]},
+		 *			SH_MEM_APE1_LIMIT[31:0], 0xFFFF }
+		 * Verify that the base and size parameters can be
+		 * represented in this format and convert them.
+		 * Additionally restrict APE1 to user-mode addresses.
+		 */
+
+		uint64_t base = (uintptr_t)alternate_aperture_base;
+		uint64_t limit = base + alternate_aperture_size - 1;
+
+		if (limit <= base || (base & APE1_FIXED_BITS_MASK) != 0 ||
+		   (limit & APE1_FIXED_BITS_MASK) != APE1_LIMIT_ALIGNMENT) {
+			retval = false;
+			goto out;
+		}
+
+		qpd->sh_mem_ape1_base = base >> 16;
+		qpd->sh_mem_ape1_limit = limit >> 16;
+	}
 
 	default_mtype = (default_policy == cache_policy_coherent) ?
 			MTYPE_NONCACHED :
@@ -97,37 +135,22 @@ static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm,
 			| ALIGNMENT_MODE(SH_MEM_ALIGNMENT_MODE_UNALIGNED)
 			| DEFAULT_MTYPE(default_mtype)
 			| APE1_MTYPE(ape1_mtype);
-
-	return true;
-}
-
-static int update_qpd_cik(struct device_queue_manager *dqm,
-			  struct qcm_process_device *qpd)
-{
-	struct kfd_process_device *pdd;
-	unsigned int temp;
-
-	pdd = qpd_to_pdd(qpd);
-
-	/* check if sh_mem_config register already configured */
-	if (qpd->sh_mem_config == 0) {
-		qpd->sh_mem_config =
-			ALIGNMENT_MODE(SH_MEM_ALIGNMENT_MODE_UNALIGNED) |
-			DEFAULT_MTYPE(MTYPE_NONCACHED) |
-			APE1_MTYPE(MTYPE_NONCACHED);
-		qpd->sh_mem_ape1_limit = 0;
-		qpd->sh_mem_ape1_base = 0;
-	}
-
 	/* On dGPU we're always in GPUVM64 addressing mode with 64-bit
 	 * aperture addresses.
 	 */
-	temp = get_sh_mem_bases_nybble_64(pdd);
+	temp = get_sh_mem_bases_nybble_64(qpd_to_pdd(qpd));
 	qpd->sh_mem_bases = compute_sh_mem_bases_64bit(temp);
 
 	pr_debug("is32bit process: %d sh_mem_bases nybble: 0x%X and register 0x%X\n",
 		qpd->pqm->process->is_32bit_user_mode, temp, qpd->sh_mem_bases);
 
+out:
+	return retval;
+}
+
+static int update_qpd_cik(struct device_queue_manager *dqm,
+			  struct qcm_process_device *qpd)
+{
 	return 0;
 }
 
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
index b291ee0fab94..320518f41890 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c
@@ -27,6 +27,14 @@
 #include "gca/gfx_8_0_sh_mask.h"
 #include "oss/oss_3_0_sh_mask.h"
 
+/*
+ * Low bits must be 0000/FFFF as required by HW, high bits must be 0 to
+ * stay in user mode.
+ */
+#define APE1_FIXED_BITS_MASK 0xFFFF80000000FFFFULL
+/* APE1 limit is inclusive and 64K aligned. */
+#define APE1_LIMIT_ALIGNMENT 0xFFFF
+
 static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
 				       struct qcm_process_device *qpd,
 				       enum cache_policy default_policy,
@@ -85,6 +93,36 @@ static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
 {
 	uint32_t default_mtype;
 	uint32_t ape1_mtype;
+	unsigned int temp;
+	bool retval = true;
+
+	if (alternate_aperture_size == 0) {
+		/* base > limit disables APE1 */
+		qpd->sh_mem_ape1_base = 1;
+		qpd->sh_mem_ape1_limit = 0;
+	} else {
+		/*
+		 * In FSA64, APE1_Base[63:0] = { 16{SH_MEM_APE1_BASE[31]},
+		 *			SH_MEM_APE1_BASE[31:0], 0x0000 }
+		 * APE1_Limit[63:0] = { 16{SH_MEM_APE1_LIMIT[31]},
+		 *			SH_MEM_APE1_LIMIT[31:0], 0xFFFF }
+		 * Verify that the base and size parameters can be
+		 * represented in this format and convert them.
+		 * Additionally restrict APE1 to user-mode addresses.
+		 */
+
+		uint64_t base = (uintptr_t)alternate_aperture_base;
+		uint64_t limit = base + alternate_aperture_size - 1;
+
+		if (limit <= base || (base & APE1_FIXED_BITS_MASK) != 0 ||
+		   (limit & APE1_FIXED_BITS_MASK) != APE1_LIMIT_ALIGNMENT) {
+			retval = false;
+			goto out;
+		}
+
+		qpd->sh_mem_ape1_base = base >> 16;
+		qpd->sh_mem_ape1_limit = limit >> 16;
+	}
 
 	default_mtype = (default_policy == cache_policy_coherent) ?
 			MTYPE_UC :
@@ -100,40 +138,21 @@ static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
 			default_mtype << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT |
 			ape1_mtype << SH_MEM_CONFIG__APE1_MTYPE__SHIFT;
 
-	return true;
-}
-
-static int update_qpd_vi(struct device_queue_manager *dqm,
-			 struct qcm_process_device *qpd)
-{
-	struct kfd_process_device *pdd;
-	unsigned int temp;
-
-	pdd = qpd_to_pdd(qpd);
-
-	/* check if sh_mem_config register already configured */
-	if (qpd->sh_mem_config == 0) {
-		qpd->sh_mem_config =
-				SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
-					SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT |
-				MTYPE_UC <<
-					SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT |
-				MTYPE_UC <<
-					SH_MEM_CONFIG__APE1_MTYPE__SHIFT;
-
-		qpd->sh_mem_ape1_limit = 0;
-		qpd->sh_mem_ape1_base = 0;
-	}
-
 	/* On dGPU we're always in GPUVM64 addressing mode with 64-bit
 	 * aperture addresses.
 	 */
-	temp = get_sh_mem_bases_nybble_64(pdd);
+	temp = get_sh_mem_bases_nybble_64(qpd_to_pdd(qpd));
 	qpd->sh_mem_bases = compute_sh_mem_bases_64bit(temp);
 
 	pr_debug("sh_mem_bases nybble: 0x%X and register 0x%X\n",
 		temp, qpd->sh_mem_bases);
+out:
+	return retval;
+}
 
+static int update_qpd_vi(struct device_queue_manager *dqm,
+			 struct qcm_process_device *qpd)
+{
 	return 0;
 }
 
-- 
2.34.1

[PATCH 2/3] drm/amdkfd: Set per-process flags only once for gfx9/10/11/12

[Amber Lin]

From: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>

Define set_cache_memory_policy() for these asics and move all static
changes from update_qpd() which is called each time a queue is created
to set_cache_memory_policy() which is called once during process
initialization

Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
---
 .../amd/amdkfd/kfd_device_queue_manager_v10.c | 41 +++++++++++--------
 .../amd/amdkfd/kfd_device_queue_manager_v11.c | 41 +++++++++++--------
 .../amd/amdkfd/kfd_device_queue_manager_v12.c | 41 +++++++++++--------
 .../amd/amdkfd/kfd_device_queue_manager_v9.c  | 36 +++++++++++++++-
 4 files changed, 107 insertions(+), 52 deletions(-)

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
index 245a90dfc2f6..b5f5f141353b 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c
@@ -31,10 +31,17 @@ static int update_qpd_v10(struct device_queue_manager *dqm,
 			 struct qcm_process_device *qpd);
 static void init_sdma_vm_v10(struct device_queue_manager *dqm, struct queue *q,
 			    struct qcm_process_device *qpd);
+static bool set_cache_memory_policy_v10(struct device_queue_manager *dqm,
+				   struct qcm_process_device *qpd,
+				   enum cache_policy default_policy,
+				   enum cache_policy alternate_policy,
+				   void __user *alternate_aperture_base,
+				   uint64_t alternate_aperture_size);
 
 void device_queue_manager_init_v10(
 	struct device_queue_manager_asic_ops *asic_ops)
 {
+	asic_ops->set_cache_memory_policy = set_cache_memory_policy_v10;
 	asic_ops->update_qpd = update_qpd_v10;
 	asic_ops->init_sdma_vm = init_sdma_vm_v10;
 	asic_ops->mqd_manager_init = mqd_manager_init_v10;
@@ -49,27 +56,27 @@ static uint32_t compute_sh_mem_bases_64bit(struct kfd_process_device *pdd)
 		private_base;
 }
 
-static int update_qpd_v10(struct device_queue_manager *dqm,
-			 struct qcm_process_device *qpd)
+static bool set_cache_memory_policy_v10(struct device_queue_manager *dqm,
+				   struct qcm_process_device *qpd,
+				   enum cache_policy default_policy,
+				   enum cache_policy alternate_policy,
+				   void __user *alternate_aperture_base,
+				   uint64_t alternate_aperture_size)
 {
-	struct kfd_process_device *pdd;
-
-	pdd = qpd_to_pdd(qpd);
-
-	/* check if sh_mem_config register already configured */
-	if (qpd->sh_mem_config == 0) {
-		qpd->sh_mem_config =
-			(SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
-				SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) |
-			(3 << SH_MEM_CONFIG__INITIAL_INST_PREFETCH__SHIFT);
-		qpd->sh_mem_ape1_limit = 0;
-		qpd->sh_mem_ape1_base = 0;
-	}
-
-	qpd->sh_mem_bases = compute_sh_mem_bases_64bit(pdd);
+	qpd->sh_mem_config = (SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
+			      SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) |
+			      (3 << SH_MEM_CONFIG__INITIAL_INST_PREFETCH__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));
 
 	pr_debug("sh_mem_bases 0x%X\n", qpd->sh_mem_bases);
+	return true;
+}
 
+static int update_qpd_v10(struct device_queue_manager *dqm,
+			 struct qcm_process_device *qpd)
+{
 	return 0;
 }
 
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
index 2e129da7acb4..f436878d0d62 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v11.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v11.c
@@ -30,10 +30,17 @@ static int update_qpd_v11(struct device_queue_manager *dqm,
 			 struct qcm_process_device *qpd);
 static void init_sdma_vm_v11(struct device_queue_manager *dqm, struct queue *q,
 			    struct qcm_process_device *qpd);
+static bool set_cache_memory_policy_v11(struct device_queue_manager *dqm,
+				   struct qcm_process_device *qpd,
+				   enum cache_policy default_policy,
+				   enum cache_policy alternate_policy,
+				   void __user *alternate_aperture_base,
+				   uint64_t alternate_aperture_size);
 
 void device_queue_manager_init_v11(
 	struct device_queue_manager_asic_ops *asic_ops)
 {
+	asic_ops->set_cache_memory_policy = set_cache_memory_policy_v11;
 	asic_ops->update_qpd = update_qpd_v11;
 	asic_ops->init_sdma_vm = init_sdma_vm_v11;
 	asic_ops->mqd_manager_init = mqd_manager_init_v11;
@@ -48,28 +55,28 @@ static uint32_t compute_sh_mem_bases_64bit(struct kfd_process_device *pdd)
 		private_base;
 }
 
-static int update_qpd_v11(struct device_queue_manager *dqm,
-			 struct qcm_process_device *qpd)
+static bool set_cache_memory_policy_v11(struct device_queue_manager *dqm,
+				   struct qcm_process_device *qpd,
+				   enum cache_policy default_policy,
+				   enum cache_policy alternate_policy,
+				   void __user *alternate_aperture_base,
+				   uint64_t alternate_aperture_size)
 {
-	struct kfd_process_device *pdd;
-
-	pdd = qpd_to_pdd(qpd);
-
-	/* check if sh_mem_config register already configured */
-	if (qpd->sh_mem_config == 0) {
-		qpd->sh_mem_config =
-			(SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
-				SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) |
-			(3 << SH_MEM_CONFIG__INITIAL_INST_PREFETCH__SHIFT);
-
-		qpd->sh_mem_ape1_limit = 0;
-		qpd->sh_mem_ape1_base = 0;
-	}
+	qpd->sh_mem_config = (SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
+			      SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) |
+			      (3 << SH_MEM_CONFIG__INITIAL_INST_PREFETCH__SHIFT);
 
-	qpd->sh_mem_bases = compute_sh_mem_bases_64bit(pdd);
+	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));
 
 	pr_debug("sh_mem_bases 0x%X\n", qpd->sh_mem_bases);
+	return true;
+}
 
+static int update_qpd_v11(struct device_queue_manager *dqm,
+			 struct qcm_process_device *qpd)
+{
 	return 0;
 }
 
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
index 4f3295b29dfb..62ca1c8fcbaf 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v12.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v12.c
@@ -30,10 +30,17 @@ static int update_qpd_v12(struct device_queue_manager *dqm,
 			 struct qcm_process_device *qpd);
 static void init_sdma_vm_v12(struct device_queue_manager *dqm, struct queue *q,
 			    struct qcm_process_device *qpd);
+static bool set_cache_memory_policy_v12(struct device_queue_manager *dqm,
+				   struct qcm_process_device *qpd,
+				   enum cache_policy default_policy,
+				   enum cache_policy alternate_policy,
+				   void __user *alternate_aperture_base,
+				   uint64_t alternate_aperture_size);
 
 void device_queue_manager_init_v12(
 	struct device_queue_manager_asic_ops *asic_ops)
 {
+	asic_ops->set_cache_memory_policy = set_cache_memory_policy_v12;
 	asic_ops->update_qpd = update_qpd_v12;
 	asic_ops->init_sdma_vm = init_sdma_vm_v12;
 	asic_ops->mqd_manager_init = mqd_manager_init_v12;
@@ -48,28 +55,28 @@ static uint32_t compute_sh_mem_bases_64bit(struct kfd_process_device *pdd)
 		private_base;
 }
 
-static int update_qpd_v12(struct device_queue_manager *dqm,
-			 struct qcm_process_device *qpd)
+static bool set_cache_memory_policy_v12(struct device_queue_manager *dqm,
+				   struct qcm_process_device *qpd,
+				   enum cache_policy default_policy,
+				   enum cache_policy alternate_policy,
+				   void __user *alternate_aperture_base,
+				   uint64_t alternate_aperture_size)
 {
-	struct kfd_process_device *pdd;
-
-	pdd = qpd_to_pdd(qpd);
-
-	/* check if sh_mem_config register already configured */
-	if (qpd->sh_mem_config == 0) {
-		qpd->sh_mem_config =
-			(SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
-				SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) |
-			(3 << SH_MEM_CONFIG__INITIAL_INST_PREFETCH__SHIFT);
-
-		qpd->sh_mem_ape1_limit = 0;
-		qpd->sh_mem_ape1_base = 0;
-	}
+	qpd->sh_mem_config = (SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
+			      SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) |
+			      (3 << SH_MEM_CONFIG__INITIAL_INST_PREFETCH__SHIFT);
 
-	qpd->sh_mem_bases = compute_sh_mem_bases_64bit(pdd);
+	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));
 
 	pr_debug("sh_mem_bases 0x%X\n", qpd->sh_mem_bases);
+	return true;
+}
 
+static int update_qpd_v12(struct device_queue_manager *dqm,
+			 struct qcm_process_device *qpd)
+{
 	return 0;
 }
 
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
index 67137e674f1d..c734eb9b505f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c
@@ -30,10 +30,17 @@ static int update_qpd_v9(struct device_queue_manager *dqm,
 			 struct qcm_process_device *qpd);
 static void init_sdma_vm_v9(struct device_queue_manager *dqm, struct queue *q,
 			    struct qcm_process_device *qpd);
+static bool set_cache_memory_policy_v9(struct device_queue_manager *dqm,
+				   struct qcm_process_device *qpd,
+				   enum cache_policy default_policy,
+				   enum cache_policy alternate_policy,
+				   void __user *alternate_aperture_base,
+				   uint64_t alternate_aperture_size);
 
 void device_queue_manager_init_v9(
 	struct device_queue_manager_asic_ops *asic_ops)
 {
+	asic_ops->set_cache_memory_policy = set_cache_memory_policy_v9;
 	asic_ops->update_qpd = update_qpd_v9;
 	asic_ops->init_sdma_vm = init_sdma_vm_v9;
 	asic_ops->mqd_manager_init = mqd_manager_init_v9;
@@ -48,10 +55,37 @@ static uint32_t compute_sh_mem_bases_64bit(struct kfd_process_device *pdd)
 		private_base;
 }
 
+static bool set_cache_memory_policy_v9(struct device_queue_manager *dqm,
+				   struct qcm_process_device *qpd,
+				   enum cache_policy default_policy,
+				   enum cache_policy alternate_policy,
+				   void __user *alternate_aperture_base,
+				   uint64_t alternate_aperture_size)
+{
+	qpd->sh_mem_config = SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
+				SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
+
+	if (dqm->dev->kfd->noretry)
+		qpd->sh_mem_config |= 1 << SH_MEM_CONFIG__RETRY_DISABLE__SHIFT;
+
+	if (KFD_GC_VERSION(dqm->dev->kfd) == IP_VERSION(9, 4, 3) ||
+		KFD_GC_VERSION(dqm->dev->kfd) == IP_VERSION(9, 4, 4) ||
+		KFD_GC_VERSION(dqm->dev->kfd) == IP_VERSION(9, 5, 0))
+		qpd->sh_mem_config |= (1 << SH_MEM_CONFIG__F8_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));
+
+	pr_debug("sh_mem_bases 0x%X sh_mem_config 0x%X\n", qpd->sh_mem_bases,
+		 qpd->sh_mem_config);
+	return true;
+}
+
 static int update_qpd_v9(struct device_queue_manager *dqm,
 			 struct qcm_process_device *qpd)
 {
-	struct kfd_process_device *pdd;
+	struct kfd_process_device *pdd = qpd_to_pdd(qpd);
 
 	pdd = qpd_to_pdd(qpd);
 
-- 
2.34.1

[PATCH 3/3] drm/amdkfd: Add support for more per-process flag

[Amber Lin]

From: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>

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

Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_chardev.c              |  3 ++-
 drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c |  6 ++++--
 drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h |  6 ++++--
 .../gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c |  6 ++++--
 .../gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c |  6 ++++--
 .../gpu/drm/amd/amdkfd/kfd_device_queue_manager_v11.c |  6 ++++--
 .../gpu/drm/amd/amdkfd/kfd_device_queue_manager_v12.c |  6 ++++--
 .../gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c  | 11 +++++++++--
 .../gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c  |  6 ++++--
 include/uapi/linux/kfd_ioctl.h                        |  5 ++++-
 10 files changed, 43 insertions(+), 18 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 8c2e92378b49..1e9dd00620bf 100644
--- 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
index d23c6a358d34..2afcc1b4856a 100644
--- 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
index 7146e227e2c1..122eb745e9c4 100644
--- 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
index 32bedef912b3..0508ef5a41d7 100644
--- 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
index b5f5f141353b..ba6e3d747ccd 100644
--- 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
index f436878d0d62..8b447d04558f 100644
--- 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
index 62ca1c8fcbaf..3550da3a46f9 100644
--- 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
index c734eb9b505f..4635077aa905 100644
--- 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
index 320518f41890..dad83356e976 100644
--- 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/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
index b0160b09987c..e7b4b1551529 100644
--- a/include/uapi/linux/kfd_ioctl.h
+++ b/include/uapi/linux/kfd_ioctl.h
@@ -150,6 +150,9 @@ 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 */
@@ -157,7 +160,7 @@ struct kfd_ioctl_set_memory_policy_args {
 	__u32 gpu_id;			/* to KFD */
 	__u32 default_policy;		/* to KFD */
 	__u32 alternate_policy;		/* to KFD */
-	__u32 pad;
+	__u32 misc_process_flag;        /* to KFD */
 };
 
 /*
-- 
2.34.1