libhsakmt: Implement HSA_FORCE_ASIC_TYPE to overwrite asic type

Force all the GPUs to a certain type, use the below command:
HSA_FORCE_ASIC_TYPE="10.1.0 1 gfx1010 14"
meaning major.minor.step dgpu asic_name asic_id

This will faciliate the cooperation across the teams for bringing up
ASICs which reuse existing device IDs.

Change-Id: I40fe4c9b46d3ccb3e38ea52250e80e82fb50fb0f
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>


[ROCm/ROCR-Runtime commit: d6539ddc24]

projects/rocr-runtime/src/libhsakmt.h

@@ -37,6 +37,10 @@ extern unsigned long kfd_open_count;
 extern pthread_mutex_t hsakmt_mutex;
 extern bool is_dgpu;
 
+extern int force_asic;
+extern char force_asic_name[HSA_PUBLIC_NAME_SIZE];
+extern struct hsa_gfxip_table force_asic_entry;
+
 #undef HSAKMTAPI
 #define HSAKMTAPI __attribute__((visibility ("default")))
 
@@ -90,20 +94,31 @@ extern int hsakmt_debug_level;
 
 enum asic_family_type {
 	CHIP_KAVERI = 0,
-	CHIP_HAWAII,
-	CHIP_CARRIZO,
-	CHIP_TONGA,
-	CHIP_FIJI,
-	CHIP_POLARIS10,
-	CHIP_POLARIS11,
-	CHIP_POLARIS12,
-	CHIP_VEGAM,
-	CHIP_VEGA10,
-	CHIP_VEGA12,
-	CHIP_VEGA20,
-	CHIP_RAVEN,
-	CHIP_ARCTURUS,
-	CHIP_NAVI10,
+	CHIP_HAWAII,	/* 1 */
+	CHIP_CARRIZO,	/* 2 */
+	CHIP_TONGA,	/* 3 */
+	CHIP_FIJI,	/* 4 */
+	CHIP_POLARIS10,	/* 5 */
+	CHIP_POLARIS11,	/* 6 */
+	CHIP_POLARIS12,	/* 7 */
+	CHIP_VEGAM,	/* 8 */
+	CHIP_VEGA10,	/* 9 */
+	CHIP_VEGA12,	/* 10 */
+	CHIP_VEGA20,	/* 11 */
+	CHIP_RAVEN,	/* 12 */
+	CHIP_ARCTURUS,	/* 13 */
+	CHIP_NAVI10,	/* 14 */
+	CHIP_LAST
+};
+
+struct hsa_gfxip_table {
+	uint16_t device_id;		// Device ID
+	unsigned char major;		// GFXIP Major engine version
+	unsigned char minor;		// GFXIP Minor engine version
+	unsigned char stepping;		// GFXIP Stepping info
+	unsigned char is_dgpu;		// Predicate for dGPU devices
+	const char *amd_name;		// CALName of the device
+	enum asic_family_type asic_family;	// Device family id
 };
 
 #define IS_SOC15(chip) ((chip) >= CHIP_VEGA10)

projects/rocr-runtime/src/openclose.c

@@ -82,7 +82,7 @@ static inline void init_page_size(void)
 	PAGE_SHIFT = ffs(PAGE_SIZE) - 1;
 }
 
-static void init_vars_from_env(void)
+static HSAKMT_STATUS init_vars_from_env(void)
 {
 	char *envvar;
 	int debug_level;
@@ -104,6 +104,34 @@ static void init_vars_from_env(void)
 	envvar = getenv("HSA_ZFB");
 	if (envvar)
 		zfb_support = atoi(envvar);
+
+	/* Force all the GPUs to a certain type, use the below command:
+	 * export HSA_FORCE_ASIC_TYPE="10.1.0 1 Navi10 14"
+	 * meaning major.minor.step dgpu asic_name asic_id
+	 */
+	envvar = getenv("HSA_FORCE_ASIC_TYPE");
+	if (envvar) {
+		uint32_t major, minor, step, dgpu, asic_family;
+
+		if ((sscanf(envvar, "%u.%u.%u %u %63s %u", &major, &minor, &step,
+				&dgpu, force_asic_name, &asic_family) != 6)
+			|| (major > 63 || minor > 255 || step > 255)
+			|| dgpu > 1 || asic_family >= CHIP_LAST) {
+			pr_err("HSA_FORCE_ASIC_TYPE %s is invalid\n", envvar);
+			return HSAKMT_STATUS_ERROR;
+		}
+
+		force_asic_entry.major = major;
+		force_asic_entry.minor = minor;
+		force_asic_entry.stepping = step;
+		force_asic_entry.is_dgpu = dgpu;
+
+		force_asic_entry.asic_family = asic_family;
+
+		force_asic = 1;
+	}
+
+	return HSAKMT_STATUS_SUCCESS;
 }
 
 HSAKMT_STATUS HSAKMTAPI hsaKmtOpenKFD(void)
@@ -122,7 +150,9 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtOpenKFD(void)
 		clear_after_fork();
 
 	if (kfd_open_count == 0) {
-		init_vars_from_env();
+		result = init_vars_from_env();
+		if (result != HSAKMT_STATUS_SUCCESS)
+			goto open_failed;
 
 		fd = open(kfd_device_name, O_RDWR | O_CLOEXEC);
 

projects/rocr-runtime/src/topology.c

@@ -84,15 +84,13 @@ static const char *supported_processor_vendor_name[] = {
 static HSAKMT_STATUS topology_take_snapshot(void);
 static HSAKMT_STATUS topology_drop_snapshot(void);
 
-static struct hsa_gfxip_table {
-	uint16_t device_id;		// Device ID
-	unsigned char major;		// GFXIP Major engine version
-	unsigned char minor;		// GFXIP Minor engine version
-	unsigned char stepping;		// GFXIP Stepping info
-	unsigned char is_dgpu;		// Predicate for dGPU devices
-	const char *amd_name;		// CALName of the device
-	enum asic_family_type asic_family;
-} gfxip_lookup_table[] = {
+int force_asic;
+char force_asic_name[HSA_PUBLIC_NAME_SIZE];
+struct hsa_gfxip_table force_asic_entry = {
+	.amd_name = force_asic_name,
+};
+
+static const struct hsa_gfxip_table gfxip_lookup_table[] = {
 	/* Kaveri Family */
 	{ 0x1304, 7, 0, 0, 0, "Spectre", CHIP_KAVERI },
 	{ 0x1305, 7, 0, 0, 0, "Spectre", CHIP_KAVERI },
@@ -722,6 +720,9 @@ static const struct hsa_gfxip_table *find_hsa_gfxip_device(uint16_t device_id)
 {
 	uint32_t i, table_size;
 
+	if (force_asic)
+		return &force_asic_entry;
+
 	table_size = sizeof(gfxip_lookup_table)/sizeof(struct hsa_gfxip_table);
 	for (i = 0; i < table_size; i++) {
 		if (gfxip_lookup_table[i].device_id == device_id)