From a5713c85bb6dba410e1b3b2ceb74a6263810bf68 Mon Sep 17 00:00:00 2001
From: hkasivis <58193769+hkasivis@users.noreply.github.com>
Date: Sun, 7 Sep 2025 20:04:31 -0400
Subject: [PATCH] Users/hkasivis/sync kfd ioctl header (#848)

* libhsakmt: Update ioctl version to 1.18

Sync with kernel ioctl version.

Also explicitly set the ioctl flag to KFD_PROC_FLAG_MFMA_HIGH_PRECISION

Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>

* libhsakmt: Sync ioctl header by adding kfd_ioctl_profiler

Sync with kernel ioctl version. Add kfd_ioctl_profiler.

Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>

---------

Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
---
 .../include/hsakmt/linux/kfd_ioctl.h          | 35 +++++++++++++++++--
 projects/rocr-runtime/libhsakmt/src/fmm.c     |  2 ++
 2 files changed, 34 insertions(+), 3 deletions(-)

diff --git a/projects/rocr-runtime/libhsakmt/include/hsakmt/linux/kfd_ioctl.h b/projects/rocr-runtime/libhsakmt/include/hsakmt/linux/kfd_ioctl.h
index bc388a7529..66b499d424 100644
--- a/projects/rocr-runtime/libhsakmt/include/hsakmt/linux/kfd_ioctl.h
+++ b/projects/rocr-runtime/libhsakmt/include/hsakmt/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 */
@@ -364,7 +365,7 @@ enum kfd_dbg_trap_exception_code {
 			(!!(KFD_EC_MASK(ecode) & KFD_EC_MASK_PROCESS))
 
 /* Misc. per process flags */
-#define ENABLE_MFMA_HIGH_PRECISION              (1 << 0)
+#define KFD_PROC_FLAG_MFMA_HIGH_PRECISION              (1 << 0)
 
 enum kfd_dbg_runtime_state {
 	DEBUG_RUNTIME_STATE_DISABLED = 0,
@@ -1608,6 +1609,31 @@ struct kfd_ioctl_pc_sample_args {
 	__u32 reserved;
 };
 
+#define KFD_IOC_PROFILER_VERSION_NUM 1
+enum kfd_profiler_ops {
+	KFD_IOC_PROFILER_PMC = 0,
+	KFD_IOC_PROFILER_PC_SAMPLE = 1,
+	KFD_IOC_PROFILER_VERSION = 2,
+};
+
+/**
+ * Enables/Disables GPU Specific profiler settings
+ */
+struct kfd_ioctl_pmc_settings {
+	__u32 gpu_id;             /* This is the user_gpu_id */
+	__u32 lock;               /* Lock GPU for Profiling */
+	__u32 perfcount_enable;   /* Force Perfcount Enable for queues on GPU */
+};
+
+struct kfd_ioctl_profiler_args {
+	__u32 op;						/* kfd_profiler_op */
+	union {
+		struct kfd_ioctl_pc_sample_args pc_sample;
+		struct kfd_ioctl_pmc_settings  pmc;
+		__u32 version;				/* KFD_IOC_PROFILER_VERSION_NUM */
+	};
+};
+
 #define AMDKFD_IOCTL_BASE 'K'
 #define AMDKFD_IO(nr)			_IO(AMDKFD_IOCTL_BASE, nr)
 #define AMDKFD_IOR(nr, type)		_IOR(AMDKFD_IOCTL_BASE, nr, type)
@@ -1747,7 +1773,10 @@ struct kfd_ioctl_pc_sample_args {
 #define AMDKFD_IOC_PC_SAMPLE		\
 		AMDKFD_IOWR(0x85, struct kfd_ioctl_pc_sample_args)
 
+#define AMDKFD_IOC_PROFILER			\
+		AMDKFD_IOWR(0x86, struct kfd_ioctl_profiler_args)
+
 #define AMDKFD_COMMAND_START_2		0x80
-#define AMDKFD_COMMAND_END_2		0x86
+#define AMDKFD_COMMAND_END_2		0x87
 
 #endif
diff --git a/projects/rocr-runtime/libhsakmt/src/fmm.c b/projects/rocr-runtime/libhsakmt/src/fmm.c
index b9b2f8b464..8f3cc65a5d 100644
--- a/projects/rocr-runtime/libhsakmt/src/fmm.c
+++ b/projects/rocr-runtime/libhsakmt/src/fmm.c
@@ -2780,6 +2780,8 @@ HSAKMT_STATUS hsakmt_fmm_init_process_apertures(unsigned int NumNodes)
 	mfmaHighPrecisionModeStr = getenv("HSA_HIGH_PRECISION_MODE");
 	mfma_high_precision_mode = (mfmaHighPrecisionModeStr &&
 				    strcmp(mfmaHighPrecisionModeStr, "0"));
+	mfma_high_precision_mode = mfma_high_precision_mode ?
+					KFD_PROC_FLAG_MFMA_HIGH_PRECISION : 0;
 	/* Sets the max VA alignment order size during mapping. By default the order
 	 * size is set to 18(1G) for GFX950 to reduce TLB hits. If any non-gfx950
 	 * ASIC is found in the system, set back to 9(2MB).