Add initial ib-following code, polish README for public release, and

add missing PKT3 opcodes.

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>

7 files changed, 201 insertions, 29 deletions

diff --git a/README b/README
index 0a5a38e..2700736 100644
--- a/README
+++ b/README
@@ -1,4 +1,4 @@
-UserMode Register Debugger for AMDGPU Hardware
+User Mode Register Debugger for AMDGPU Hardware
 Copyright (c) 2017 AMD Inc.
 
 Introduction
@@ -18,19 +18,19 @@ to them.
 
 The tool is open source and hosted at
 
-    : TBD
+    : https://cgit.freedesktop.org/amd/umr/
 
 Users may report bugs, feedback, and submit patches to the amd-gfx
 mailing list at:
 
-    https://lists.freedesktop.org/mailman/listinfo/amd-gfx
+    : https://lists.freedesktop.org/mailman/listinfo/amd-gfx
 
 Building
 ---------
 
-To build umr you will need pciaccess and ncurses headers.  Which are
-available in both Fedora and Ubuntu (as well as other distributions).
-To build simply invoke the make command
+To build umr you will need pciaccess and ncurses headers and libraries.
+Which are available in both Fedora and Ubuntu (as well as other
+distributions).  To build simply invoke the make command
 
     $ make
 
@@ -76,10 +76,14 @@ Selecting Hardware
 -------------------
 
 On machines with more than one AMDGPU device umr can be instructed
-which to look at.  If the driver is loaded and display attached
+which to look at.  If the driver is loaded and display attached [*]
 the --instance option can select a device.  The devices can be listed
-with the --enumerate option.  Once one is picked the number can be used
-for instance:
+with the --enumerate option.
+
+[*] Kernels v4.11 and newer only require the amdgpu module to be loaded
+in order to work without '-f'.
+
+Once one is picked the number can be used for instance:
 
     $ umr -i 1 -s uvd6
 
@@ -172,7 +176,7 @@ Wavefront Debugging
 
 The status of valid wavefronts can be displayed with the --waves
 command.  The bits option enables bitfield decoding.  Normally,
-the colums would be pretty printed:
+the columns would be pretty printed:
 
     $ umr -wa | column -t
 
diff --git a/src/app/main.c b/src/app/main.c
index fb71083..5b16946 100644
--- a/src/app/main.c
+++ b/src/app/main.c
@@ -82,6 +82,8 @@ static void parse_options(char *str)
 			options.read_smc = 1;
 		} else if (!strcmp(option, "quiet")) {
 			options.quiet = 1;
+		} else if (!strcmp(option, "follow_ib")) {
+			options.follow_ib = 1;
 		} else {
 			printf("error: Unknown option [%s]\n", option);
 			exit(EXIT_FAILURE);
diff --git a/src/app/ring_read.c b/src/app/ring_read.c
index 52206de..1726573 100644
--- a/src/app/ring_read.c
+++ b/src/app/ring_read.c
@@ -129,7 +129,9 @@ void umr_read_ring(struct umr_asic *asic, char *ringpath)
 
 	pdecoder = decoder.next_ib;
 	while (pdecoder) {
-		umr_dump_ib(asic, pdecoder);
+		if (asic->options.follow_ib) {
+			umr_dump_ib(asic, pdecoder);
+		}
 		ppdecoder = pdecoder->next_ib;
 		free(pdecoder);
 		pdecoder = ppdecoder;
diff --git a/src/lib/dump_ib.c b/src/lib/dump_ib.c
index f88ed3f..4e81dbe 100644
--- a/src/lib/dump_ib.c
+++ b/src/lib/dump_ib.c
@@ -38,6 +38,7 @@ void umr_dump_ib(struct umr_asic *asic, struct umr_ring_decoder *decoder)
 	data = calloc(sizeof(*data), decoder->next_ib_info.size/sizeof(*data));
 	if (data && !umr_read_vram(asic, decoder->next_ib_info.vmid, decoder->next_ib_info.ib_addr, decoder->next_ib_info.size, data)) {
 	// dump IB
+		decoder->pm4.cur_opcode = 0xFFFFFFFF;
 		for (x = 0; x < decoder->next_ib_info.size/4; x++) {
 			printf("IB[%5u] = 0x%08lx ... ", (unsigned)x, (unsigned long)data[x]);
 			umr_print_decode(asic, decoder, data[x]);
diff --git a/src/lib/read_vram.c b/src/lib/read_vram.c
index 705d611..34d25fd 100644
--- a/src/lib/read_vram.c
+++ b/src/lib/read_vram.c
@@ -51,6 +51,144 @@ static void read_via_mmio(struct umr_asic *asic, uint64_t address, uint32_t size
 	}
 }
 
+#define DEBUG printf
+
+static int umr_read_sram(uint64_t address, uint32_t size, void *dst)
+{
+	int fd;
+
+	fd = open("/dev/mem", O_RDWR);
+	if (fd >= 0) {
+		memset(dst, 0, size);
+		lseek(fd, address, SEEK_SET);
+		if (read(fd, dst, size) != size) {
+			close(fd);
+			return -1;
+		}
+		close(fd);
+		return 0;
+	}
+	return -1;
+}
+
+
+static int umr_read_vram_vi(struct umr_asic *asic, uint32_t vmid, uint64_t address, uint32_t size, void *dst)
+{
+	uint64_t start_addr, page_table_start_addr, page_table_base_addr,
+		 page_table_size, pte_idx, pde_idx, pte_entry, pde_entry;
+	uint32_t chunk_size, tmp;
+	int page_table_depth;
+	struct {
+		uint64_t
+			frag_size,
+			pte_base_addr,
+			valid;
+	} pde_fields;
+	struct {
+		uint64_t
+			page_base_addr,
+			fragment,
+			system;
+	} pte_fields;
+	char buf[64];
+	unsigned char *pdst = dst;
+
+	/*
+	 * PTE format on VI:
+	 * 63:40 reserved
+	 * 39:12 4k physical page base address
+	 * 11:7 fragment
+	 * 6 write
+	 * 5 read
+	 * 4 exe
+	 * 3 reserved
+	 * 2 snooped
+	 * 1 system
+	 * 0 valid
+	 *
+	 * PDE format on VI:
+	 * 63:59 block fragment size
+	 * 58:40 reserved
+	 * 39:1 physical base address of PTE
+	 * bits 5:1 must be 0.
+	 * 0 valid
+	 */
+
+	// read vm registers
+	sprintf(buf, "mmVM_CONTEXT%d_PAGE_TABLE_START_ADDR", (int)vmid ? 1 : 0);
+		page_table_start_addr = (uint64_t)umr_read_reg_by_name(asic, buf) << 12;
+
+	sprintf(buf, "mmVM_CONTEXT%d_CNTL", (int)vmid ? 1 : 0);
+		tmp = umr_read_reg_by_name(asic, buf);
+		page_table_depth      = umr_bitslice_reg_by_name(asic, buf, "PAGE_TABLE_DEPTH", tmp);
+		page_table_size       = umr_bitslice_reg_by_name(asic, buf, "PAGE_TABLE_BLOCK_SIZE", tmp);
+
+	sprintf(buf, "mmVM_CONTEXT%d_PAGE_TABLE_BASE_ADDR", (int)vmid);
+		page_table_base_addr  = (uint64_t)umr_read_reg_by_name(asic, buf) << 12;
+
+	DEBUG("PAGE_START_ADDR = %08llx\n", (unsigned long long)page_table_start_addr);
+	DEBUG("BASE_ADDR = 0x%08llx\n", (unsigned long long)page_table_base_addr);
+	DEBUG("BASE_SIZE = %lu\n", page_table_size);
+	DEBUG("PAGE_TABLE_DEPTH = %d\n", page_table_depth);
+
+	address -= page_table_start_addr;
+
+	while (size) {
+		if (page_table_depth == 1) {
+			// decode addr into pte and pde selectors...
+			pde_idx = (address >> (12 + 9 + page_table_size)) & ((1UL << (19 - page_table_size)) - 1);
+			pte_idx = (address >> 12) & ((1UL << (9 + page_table_size)) - 1);
+
+			// read PDE entry
+			umr_read_vram(asic, 0xFFFF, page_table_base_addr + pde_idx * 8, 8, &pde_entry);
+
+			// decode PDE values
+			pde_fields.frag_size     = (pde_entry >> 59) & 0x1F;
+			pde_fields.pte_base_addr = pde_entry & 0xFFFFFFF000;
+			pde_fields.valid         = pde_entry & 1;
+			DEBUG("pde_idx=%llx, frag_size=%u, pte_base_addr=0x%llx, valid=%d\n", (unsigned long long)pde_idx, (unsigned)pde_fields.frag_size, (unsigned long long)pde_fields.pte_base_addr, (int)pde_fields.valid);
+
+			// now read PTE entry for this page
+			umr_read_vram(asic, 0xFFFF, pde_fields.pte_base_addr + pte_idx*8, 8, &pte_entry);
+
+			// decode PTE values
+			pte_fields.page_base_addr = pte_entry & 0xFFFFFFF000;
+			pte_fields.fragment       = (pte_entry >> 7)  & 0x1F;
+			pte_fields.system         = (pte_entry >> 1) & 1;
+			DEBUG("pte_idx=%llx, page_base_addr=0x%llx, fragment=%u, system=%d\n", (unsigned long long)pte_idx, (unsigned long long)pte_fields.page_base_addr, (unsigned)pte_fields.fragment, (int)pte_fields.system);
+
+			// compute starting address
+			start_addr = pte_fields.page_base_addr + (address & 0xFFF);
+		} else {
+			fprintf(stderr, "[BUG] depth 0 page tables are not yet supported\n");
+			return -1;
+		}
+
+		// read upto 4K from it
+		if (((start_addr & 0xFFF) + size) & ~0xFFF) {
+			chunk_size = 0x1000 - (start_addr & 0xFFF);
+		} else {
+			chunk_size = size;
+		}
+		DEBUG("Computed address we will read from: %s:%llx (reading: %lu bytes)\n", pte_fields.system ? "sys" : "vram", (unsigned long long)start_addr, (unsigned long)chunk_size);
+		if (pte_fields.system) {
+			if (umr_read_sram(start_addr, chunk_size, pdst) < 0) {
+				fprintf(stderr, "[ERROR] Cannot read system ram, perhaps CONFIG_STRICT_DEVMEM is set in your kernel config?\n");
+				return -1;
+			}
+		} else {
+			if (umr_read_vram(asic, 0xFFFF, start_addr, chunk_size, pdst) < 0) {
+				fprintf(stderr, "[ERROR] Cannot read from VRAM\n");
+				return -1;
+			}
+		}
+		pdst += chunk_size;
+		size -= chunk_size;
+		address += chunk_size;
+	}
+	return 0;
+}
+
 
 int umr_read_vram(struct umr_asic *asic, uint32_t vmid, uint64_t address, uint32_t size, void *dst)
 {
@@ -74,5 +212,9 @@ int umr_read_vram(struct umr_asic *asic, uint32_t vmid, uint64_t address, uint32
 		}
 		return 0;
 	}
+
+	if (asic->family == FAMILY_VI)
+		return umr_read_vram_vi(asic, vmid, address, size, dst);
+
 	return 0;
 }
diff --git a/src/lib/ring_decode.c b/src/lib/ring_decode.c
index f2c27d0..401d6ca 100644
--- a/src/lib/ring_decode.c
+++ b/src/lib/ring_decode.c
@@ -55,11 +55,11 @@ static const char *pm4_pkt3_opcode_names[] = {
 	"UNK", // 1a
 	"UNK", // 1b
 	"UNK", // 1c
-	"UNK", // 1d
-	"UNK", // 1e
+	"PKT3_ATOMIC_GDS", // 1d
+	"PKT3_ATOMIC_MEM", // 1e
 	"PKT3_OCCLUSION_QUERY", // 1f
 	"PKT3_SET_PREDICATION", // 20
-	"UNK", // 21
+	"PKT3_REG_RMW", // 21
 	"PKT3_COND_EXEC", // 22
 	"PKT3_PRED_EXEC", // 23
 	"PKT3_DRAW_INDIRECT", // 24
@@ -99,8 +99,8 @@ static const char *pm4_pkt3_opcode_names[] = {
 	"PKT3_EVENT_WRITE", // 46
 	"PKT3_EVENT_WRITE_EOP", // 47
 	"PKT3_EVENT_WRITE_EOS", // 48
-	"UNK", // 49
-	"UNK", // 4a
+	"PKT3_RELEASE_MEM", // 49
+	"PKT3_PREAMBLE_CNTL", // 4a
 	"UNK", // 4b
 	"UNK", // 4c
 	"UNK", // 4d
@@ -115,15 +115,15 @@ static const char *pm4_pkt3_opcode_names[] = {
 	"UNK", // 56
 	"PKT3_ONE_REG_WRITE", // 57
 	"PKT3_ACQUIRE_MEM", // 58
-	"UNK", // 59
+	"PKT3_REWIND", // 59
 	"UNK", // 5a
 	"UNK", // 5b
 	"UNK", // 5c
 	"UNK", // 5d
-	"UNK", // 5e
-	"UNK", // 5f
-	"UNK", // 60
-	"UNK", // 61
+	"PKT3_LOAD_UCONFIG_REG", // 5e
+	"PKT3_LOAD_SH_REG", // 5f
+	"PKT3_LOAD_CONFIG_REG", // 60
+	"PKT3_LOAD_CONTEXT_REG", // 61
 	"UNK", // 62
 	"UNK", // 63
 	"UNK", // 64
@@ -141,12 +141,12 @@ static const char *pm4_pkt3_opcode_names[] = {
 	"UNK", // 70
 	"UNK", // 71
 	"UNK", // 72
-	"UNK", // 73
+	"PKT3_SET_CONTEXT_REG_INDIRECT", // 73
 	"UNK", // 74
 	"UNK", // 75
 	"PKT3_SET_SH_REG", // 76
 	"PKT3_SET_SH_REG_OFFSET", // 77
-	"UNK", // 78
+	"PKT3_SET_QUEUE_REG", // 78
 	"PKT3_SET_UCONFIG_REG", // 79
 	"UNK", // 7a
 	"UNK", // 7b
@@ -162,10 +162,10 @@ static const char *pm4_pkt3_opcode_names[] = {
 	"PKT3_INCREMENT_DE_COUNTER", // 85
 	"PKT3_WAIT_ON_CE_COUNTER", // 86
 	"UNK", // 87
-	"UNK", // 88
+	"PKT3_WAIT_ON_DE_COUNTER_DIFF", // 88
 	"UNK", // 89
 	"UNK", // 8a
-	"UNK", // 8b
+	"PKT3_SWITCH_BUFFER", // 8b
 	"UNK", // 8c
 	"UNK", // 8d
 	"UNK", // 8e
@@ -186,9 +186,9 @@ static const char *pm4_pkt3_opcode_names[] = {
 	"UNK", // 9d
 	"UNK", // 9e
 	"UNK", // 9f
-	"UNK", // a0
+	"PKT3_SET_RESOURCES", // a0
 	"UNK", // a1
-	"UNK", // a2
+	"PKT3_MAP_QUEUES", // a2
 	"UNK", // a3
 	"UNK", // a4
 	"UNK", // a5
@@ -312,7 +312,7 @@ char *umr_reg_name(struct umr_asic *asic, uint64_t addr)
 
 	for (i = 0; i < asic->no_blocks; i++)
 	for (j = 0; j < asic->blocks[i]->no_regs; j++)
-		if (asic->blocks[i]->regs[j].addr == addr)
+		if (asic->blocks[i]->regs[j].type == REG_MMIO && asic->blocks[i]->regs[j].addr == addr)
 			return asic->blocks[i]->regs[j].regname;
 	return "<unknown>";
 }
@@ -409,6 +409,26 @@ static void print_decode_pm4_pkt3(struct umr_asic *asic, struct umr_ring_decoder
 				default: printf("Invalid word for opcode 0x%02lx", (unsigned long)decoder->pm4.cur_opcode);
 			}
 			break;
+		case 0x49: // RELEASE_MEM
+			switch(decoder->pm4.cur_word) {
+				case 0: printf("EOP_TCL1_ACTION: %lu, EOP_TC_ACTION: %lu, EOP_TC_WB_ACTION: %lu, EVENT_TYPE: %lu, EVENT_INDEX: %lu",
+				BITS(ib, 16, 17), BITS(ib, 17, 18), BITS(ib, 15, 16), BITS(ib, 0, 7), BITS(ib, 8, 15));
+					break;
+				case 1:
+					printf("DATA_SEL+INT_SEL: 0x%08lx", (unsigned long)ib);
+//DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0)
+					break;
+				case 2: printf("ADDR_LO: 0x%08lx", (unsigned long)ib);
+					break;
+				case 3: printf("ADDR_HI: 0x%08lx", (unsigned long)ib);
+					break;
+				case 4: printf("SEQ_LO: 0x%08lx", (unsigned long)ib);
+					break;
+				case 5: printf("SEQ_HI: 0x%08lx", (unsigned long)ib);
+					break;
+				default: printf("Invalid word for opcode 0x%02lx", (unsigned long)decoder->pm4.cur_opcode);
+			}
+			break;
 		default:
 			printf("PKT3 DATA");
 			break;
diff --git a/src/umr.h b/src/umr.h
index bf33a27..f2bce13 100644
--- a/src/umr.h
+++ b/src/umr.h
@@ -152,7 +152,8 @@ struct umr_options {
 	    use_pci,
 	    use_colour,
 	    read_smc,
-	    quiet;
+	    quiet,
+	    follow_ib;
 	unsigned
 	    instance_bank,
 	    se_bank,