upd with pci scan

Makefile

@@ -1,5 +1,6 @@
 all:
-	gcc gddr6.c include/nva/gpu.c include/nva/nva.c include/nva/chipset.c include/nva/regspace.c -o gddr6 -lpciaccess 
+	gcc -O3 -Wall -o gddr6 gddr6.c -lpci
-
 clean:
-	rm -rf gddr6
+	rm -f gddr6
+
+

gddr6.c

@@ -1,92 +1,99 @@
 #include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
 #include <unistd.h>
-#include "include/nva/nva.h"
+#include <string.h>
-#include "include/ampere/ga100/dev_boot.h"
+#include <errno.h>
+#include <fcntl.h>
+#include <sys/mman.h>
+#include <pci/pci.h>
 
-enum {
+struct device 
-    NVIDIA_MMIO_MASK_PMC        = 0x00000FFF,
+{
-    NVIDIA_MMIO_MASK_PBUS       = 0x00001FFF,
+    uint32_t bar0;
-    NVIDIA_MASK_ALL             = 0x0000FFFF,
+    uint32_t offset;
+    uint16_t dev_id;
+    const char *arch;
+    const char *name;
 };
 
-struct pmc_id {
+static struct device devices[] = {
-    uint8_t stepping;
+    { .bar0 = 0xEC000000, .offset = 0x0000E2A8, .dev_id = 0x2684, .arch = "AD102", .name =  "RTX 4090" },
-    uint8_t device_id;
+    { .bar0 = 0xFB000000, .offset = 0x0000E2A8, .dev_id = 0x1337, .arch = "GA102", .name =  "RTX 3090" },
-    uint16_t gpu_id;
+    { .bar0 = 0xFB000000, .offset = 0x0000EE50, .dev_id = 0x1337, .arch = "GA104", .name =  "RTX 3070" },
 };
 
-void PMC_ID (uint32_t value,
+struct device * pci_detect_dev() 
-                 struct pmc_id *id) {
+{
-    id->stepping   = (value << 0) >> 0;
+    struct pci_access *pacc = NULL;
-    id->device_id  = (value << 0) >> 12;
+    struct pci_dev *dev = NULL;
-    id->gpu_id     = (value << 0) >> 20;
+    struct device *device = NULL;
+
+    pacc = pci_alloc();
+    pci_init(pacc);
+    pci_scan_bus(pacc);
+
+    for (dev = pacc->devices; dev; dev = dev->next) {
+        pci_fill_info(dev, PCI_FILL_IDENT | PCI_FILL_BASES | PCI_FILL_CLASS);
+        for (uint32_t i = 0; i < 3; i++) {
+          if (dev->device_id == devices[i].dev_id) {
+            device = &devices[i]; 
+          }
+        }
+    }
+
+    pci_cleanup(pacc);
+    return device;
 }
 
-void read_temps(int cnum, uint32_t offset) {
+int main(int argc, char **argv)
-  uint32_t temp = 0;
+{
-  while(1) {
+    int fd;
-    temp = nva_rd32(cnum, offset);
+    int map_size = 4096UL;
-    temp = (temp & 0x00000fff) / 0x20;
+    void *map_base, *virt_addr;
-    fprintf(stderr, "\rGDDR6 VRAM Temp: %d°c", temp);
+    uint32_t read_result, base_offset, temp;
-    fflush(stderr);
+    uint32_t offset;
-    sleep(1);
+    struct device *device = NULL; 
-  }
-}
 
-int main(int argc, char **argv) {
+    char *MEM_RES = "\x2f\x64\x65\x76\x2f\x6d\x65\x6d";
-    int c;
-    int i;
-    int cnum = 0;
-    int mask = NVIDIA_MASK_ALL;
 
-    if (nva_init()) {
+    device = pci_detect_dev();
-            fprintf (stderr, "PCI init failure!. Are you r00t?\n");
+    offset = device->bar0 + device->offset;
-            return 1;
+
+    if (device == NULL) {
+      printf("No compatible devices device. \n");
+      exit(-1);
     }
 
-    while ((c = getopt (argc, argv, "c:")) != -1)
+    printf("Device: %s (%s / 0x%04x) \n", device->name, device->arch, device->dev_id);
-    switch (c) {
+
-            case 'c':
+    while (1) {
-                    sscanf(optarg, "%d", &cnum);
+        if ((fd = open(MEM_RES, O_RDWR | O_SYNC)) == -1) {
-                    break;
+          printf("Can't read memory. Are you r00t?\n");
-            default:
+          exit(-1);
-                    break;
+        }
+
+        base_offset = offset & ~(sysconf(_SC_PAGE_SIZE)-1);
+        map_base = mmap(0, map_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, base_offset);
+
+        if (!map_base) {
+          printf("Can't map memory. Are you r00t?\n");
+          exit(-1);
+        }
+
+        virt_addr = map_base + offset - base_offset;
+        read_result = *((uint32_t *) virt_addr);
+        temp = (read_result & 0x00000fff) / 0x20;
+        printf("\rGDDR6X VRAM Temp: %d°c", temp);
+        fflush(stdout);
+
+        if (munmap(map_base, map_size) == -1) {
+          printf("Can't unmap memory!\n");
+          exit(-1);
+        }
+
+        close(fd);
+        sleep(1);
     }
-
-    if (cnum >= nva_cardsnum) {
-        if (nva_cardsnum)
-            fprintf (stderr, "No valid card.\n");
-        else
-            fprintf (stderr, "No cards found.\n");
-        return 1;
-    }
-
-    if (mask & NVIDIA_MMIO_MASK_PMC) {
-            uint32_t pmc_boot_0 = nva_rd32(cnum, NV_PMC_BOOT_0);
-            struct pmc_id id;
-            PMC_ID(pmc_boot_0, &id);
-
-            switch (id.gpu_id) {
-              uint32_t offset;
-              case 0xb72: { // ga102
-                            // TODO: temps over 100 might fail
-                    offset = 0x0000E2A8;
-                    read_temps(cnum, offset);
-                    }
-                    break;
-
-              case 0xb74: { // ga104
-                    offset = 0x0000EE50;
-                    read_temps(cnum, offset);
-                    }
-                    break;
-
-              default:
-                    fprintf (stderr, "No valid card.\n");
-                    break;
-            }
-    }
-
     return 0;
 }
-

no_deps_version/Makefile

@@ -1,11 +0,0 @@
-all:
-	gcc -c gddr6_asm.s -o gddr6_asm.o
-	gcc gddr6_asm.o -o gddr6_asm
-	gcc -c gddr6.c -o gddr6.o
-	gcc gddr6.o -o gddr6
-
-clean:
-	rm -f *.o
-	rm -f gddr6_asm gddr6
-
-

no_deps_version/gddr6.c

@@ -1,58 +0,0 @@
-#include <stdio.h>
-#include <stdlib.h>
-#include <stdint.h>
-#include <unistd.h>
-#include <string.h>
-#include <errno.h>
-#include <fcntl.h>
-#include <sys/mman.h>
-
-
-int main(int argc, char **argv)
-{
-    int fd;
-    int map_size = 4096UL;
-    void *map_base, *virt_addr;
-    uint32_t read_result, base_offset, temp;
-    //uint32_t bar = 0xFB000000; // GA102 + GA104
-    uint32_t bar = 0xEC000000; // AD102 (RTX 4090)
-    uint32_t offset = bar + 0x0000E2A8; // GA102 + AD102
-    //uint32_t offset = bar + 0x0000EE50; // GA104
-
-    char *MEM_RES = "\x2f\x64\x65\x76\x2f\x6d\x65\x6d";
-
-    while (1)
-    {
-        if ((fd = open(MEM_RES, O_RDWR | O_SYNC)) == -1)
-        {
-          printf("Can't read memory. Are you r00t?\n");
-          exit(-1);
-        }
-
-        base_offset = offset & ~(sysconf(_SC_PAGE_SIZE)-1);
-        map_base = mmap(0, map_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, base_offset);
-
-        if (!map_base)
-        {
-          printf("Can't map memory. Are you r00t?\n");
-          exit(-1);
-        }
-
-        virt_addr = map_base + offset - base_offset;
-        read_result = *((uint32_t *) virt_addr);
-        temp = (read_result & 0x00000fff) / 0x20;
-        printf("\rGDDR6X VRAM Temp: %d°c", temp);
-        fflush(stdout);
-
-        if (munmap(map_base, map_size) == -1)
-        {
-          printf("Can't unmap memory!\n");
-          exit(-1);
-        }
-
-        close(fd);
-        sleep(1);
-    }
-
-    return 0;
-}

no_deps_version/gddr6_asm.s

@@ -1,127 +0,0 @@
-	.file	"gddr6.s"
-	.text
-	.section	.rodata
-.LC0:
-	.string	"/dev/mem"
-.LC1:
-	.string	"\rGDDR6 VRAM Temp: %d\302\260c"
-	.text
-	.globl	main
-	.type	main, @function
-main:
-.LFB6:
-	.cfi_startproc
-	endbr64
-	pushq	%rbp
-	.cfi_def_cfa_offset 16
-	.cfi_offset 6, -16
-	movq	%rsp, %rbp
-	.cfi_def_cfa_register 6
-	subq	$80, %rsp
-	movl	%edi, -68(%rbp)
-	movq	%rsi, -80(%rbp)
-	movl	$4096, -56(%rbp)
-	movl	$-83886080, -52(%rbp)
-	movl	-52(%rbp), %eax
-	addl	$58024, %eax
-	movl	%eax, -48(%rbp)
-	leaq	.LC0(%rip), %rax
-	movq	%rax, -24(%rbp)
-.L5:
-	movq	-24(%rbp), %rax
-	movl	$1052674, %esi
-	movq	%rax, %rdi
-	movl	$0, %eax
-	call	open@PLT
-	movl	%eax, -44(%rbp)
-	cmpl	$-1, -44(%rbp)
-	jne	.L2
-	movl	$-1, %edi
-	call	exit@PLT
-.L2:
-	movl	$30, %edi
-	call	sysconf@PLT
-	subl	$1, %eax
-	notl	%eax
-	movl	%eax, %edx
-	movl	-48(%rbp), %eax
-	andl	%edx, %eax
-	movl	%eax, -40(%rbp)
-	movl	-40(%rbp), %ecx
-	movl	-56(%rbp), %eax
-	cltq
-	movl	-44(%rbp), %edx
-	movq	%rcx, %r9
-	movl	%edx, %r8d
-	movl	$1, %ecx
-	movl	$3, %edx
-	movq	%rax, %rsi
-	movl	$0, %edi
-	call	mmap@PLT
-	movq	%rax, -16(%rbp)
-	cmpq	$0, -16(%rbp)
-	jne	.L3
-	movl	$-1, %edi
-	call	exit@PLT
-.L3:
-	movl	-48(%rbp), %edx
-	movl	-40(%rbp), %eax
-	subq	%rax, %rdx
-	movq	-16(%rbp), %rax
-	addq	%rdx, %rax
-	movq	%rax, -8(%rbp)
-	movq	-8(%rbp), %rax
-	movl	(%rax), %eax
-	movl	%eax, -36(%rbp)
-	movl	-36(%rbp), %eax
-	andl	$4095, %eax
-	movl	%eax, -32(%rbp)
-	movl	-32(%rbp), %eax
-	shrl	$5, %eax
-	movl	%eax, -28(%rbp)
-	movl	-28(%rbp), %eax
-	movl	%eax, %esi
-	leaq	.LC1(%rip), %rdi
-	movl	$0, %eax
-	call	printf@PLT
-	movq	stdout(%rip), %rax
-	movq	%rax, %rdi
-	call	fflush@PLT
-	movl	-56(%rbp), %eax
-	movslq	%eax, %rdx
-	movq	-16(%rbp), %rax
-	movq	%rdx, %rsi
-	movq	%rax, %rdi
-	call	munmap@PLT
-	cmpl	$-1, %eax
-	jne	.L4
-	movl	$-1, %edi
-	call	exit@PLT
-.L4:
-	movl	-44(%rbp), %eax
-	movl	%eax, %edi
-	call	close@PLT
-	movl	$1, %edi
-	call	sleep@PLT
-	jmp	.L5
-	.cfi_endproc
-.LFE6:
-	.size	main, .-main
-	.ident	"GCC: (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0"
-	.section	.note.GNU-stack,"",@progbits
-	.section	.note.gnu.property,"a"
-	.align 8
-	.long	 1f - 0f
-	.long	 4f - 1f
-	.long	 5
-0:
-	.string	 "GNU"
-1:
-	.align 8
-	.long	 0xc0000002
-	.long	 3f - 2f
-2:
-	.long	 0x3
-3:
-	.align 8
-4:

nva/Makefile

@@ -0,0 +1,5 @@
+all:
+	gcc gddr6.c include/nva/gpu.c include/nva/nva.c include/nva/chipset.c include/nva/regspace.c -o gddr6 -lpciaccess 
+
+clean:
+	rm -rf gddr6

nva/gddr6.c

@@ -0,0 +1,92 @@
+#include <stdio.h>
+#include <unistd.h>
+#include "include/nva/nva.h"
+#include "include/ampere/ga100/dev_boot.h"
+
+enum {
+    NVIDIA_MMIO_MASK_PMC        = 0x00000FFF,
+    NVIDIA_MMIO_MASK_PBUS       = 0x00001FFF,
+    NVIDIA_MASK_ALL             = 0x0000FFFF,
+};
+
+struct pmc_id {
+    uint8_t stepping;
+    uint8_t device_id;
+    uint16_t gpu_id;
+};
+
+void PMC_ID (uint32_t value,
+                 struct pmc_id *id) {
+    id->stepping   = (value << 0) >> 0;
+    id->device_id  = (value << 0) >> 12;
+    id->gpu_id     = (value << 0) >> 20;
+}
+
+void read_temps(int cnum, uint32_t offset) {
+  uint32_t temp = 0;
+  while(1) {
+    temp = nva_rd32(cnum, offset);
+    temp = (temp & 0x00000fff) / 0x20;
+    fprintf(stderr, "\rGDDR6 VRAM Temp: %d°c", temp);
+    fflush(stderr);
+    sleep(1);
+  }
+}
+
+int main(int argc, char **argv) {
+    int c;
+    int i;
+    int cnum = 0;
+    int mask = NVIDIA_MASK_ALL;
+
+    if (nva_init()) {
+            fprintf (stderr, "PCI init failure!. Are you r00t?\n");
+            return 1;
+    }
+
+    while ((c = getopt (argc, argv, "c:")) != -1)
+    switch (c) {
+            case 'c':
+                    sscanf(optarg, "%d", &cnum);
+                    break;
+            default:
+                    break;
+    }
+
+    if (cnum >= nva_cardsnum) {
+        if (nva_cardsnum)
+            fprintf (stderr, "No valid card.\n");
+        else
+            fprintf (stderr, "No cards found.\n");
+        return 1;
+    }
+
+    if (mask & NVIDIA_MMIO_MASK_PMC) {
+            uint32_t pmc_boot_0 = nva_rd32(cnum, NV_PMC_BOOT_0);
+            struct pmc_id id;
+            PMC_ID(pmc_boot_0, &id);
+
+            switch (id.gpu_id) {
+              uint32_t offset;
+              case 0xb72: { // ga102
+                            // TODO: temps over 100 might fail
+                    offset = 0x0000E2A8;
+                    read_temps(cnum, offset);
+                    }
+                    break;
+
+              case 0xb74: { // ga104
+                    offset = 0x0000EE50;
+                    read_temps(cnum, offset);
+                    }
+                    break;
+
+              default:
+                    fprintf (stderr, "No valid card.\n");
+                    break;
+            }
+    }
+
+    return 0;
+}
+