GPUDirect Storage kernel driver (nvidia-fs) ver-2.23.1 commit

ChangeLog

@@ -1,3 +1,6 @@
+nvidia-fs (2.23.1) RELEASE; urgency=low
+  * Use persistent_p2p APIs >= nvidia driver version 555
+  * Fix bug in copy_to_uer through batch API
 nvidia-fs (2.22.3) RELEASE; urgency=low
   * Switch to nvidia_p2p_get_pages_persistent() call for baremetal on x86
 nvidia-fs (2.20.5) RELEASE; urgency=low

src/GDS_VERSION

@@ -1 +1 @@
-1.11.0.20
+1.12.0.25

src/nvfs-core.c

@@ -85,6 +85,10 @@
 #define MAJ_MIN_P_V(maj, min, patch) maj##.##min##.##patch
 #define MOD_VERS(major, minor, p) MAJ_MIN_P_V(major, minor, p)
 
+#define NVIDIA_DRIVER_PATH "/sys/module/nvidia/version"
+#define NVIDIA_DRIVER_BUF_SIZE 4
+#define NVIDIA_MIN_DRIVER_FOR_VGPU 555
+
 static int major_number;
 static struct class* nvfs_class = NULL;
 static struct device* nvfs_device[MAX_NVFS_DEVICES];
@@ -2287,7 +2291,7 @@ static long nvfs_ioctl(struct file *file, unsigned int ioctl_num,
                         local_param.ioargs.ioctl_return = PTR_ERR(nvfs_batch);
                         if (copy_to_user((void *) ioctl_param,
                                                 (void*) &local_param,
-                                                sizeof(nvfs_ioctl_param_union))) {
+                                                sizeof(nvfs_ioctl_batch_ioargs_t))) {
                                 nvfs_err("%s:%d copy_to_user failed\n", __func__, __LINE__);
                         }
                         nvfs_stat(&nvfs_n_batch_err);
@@ -2306,7 +2310,7 @@ static long nvfs_ioctl(struct file *file, unsigned int ioctl_num,
                         nvfs_stat64(&nvfs_n_batches_ok);
                 }
                 if (copy_to_user((void *) ioctl_param, (void*) &local_param,
-					sizeof(nvfs_ioctl_param_union))) {
+					sizeof(nvfs_ioctl_batch_ioargs_t))) {
 			local_param.ioargs.ioctl_return = -EFAULT;
                         nvfs_stat(&nvfs_n_batch_err);
 			nvfs_err("%s:%d copy_to_user failed\n", __func__, __LINE__);
@@ -2437,20 +2441,61 @@ static char *nvfs_devnode(const struct device *dev, umode_t *mode)
         return NULL;
 }
 
+static int get_nvidia_driver_version(void){
+	struct file *file;
+	char buf[NVIDIA_DRIVER_BUF_SIZE];
+	loff_t pos = 0;
+	ssize_t bytes_read;
+	int value;
+
+	file = filp_open(NVIDIA_DRIVER_PATH, O_RDONLY, 0);
+	if (IS_ERR(file)) {
+		nvfs_err("Error opening nvidia driver version file: %ld\n", PTR_ERR(file));
+		return -1;
+	}
+
+	bytes_read = kernel_read(file, buf, NVIDIA_DRIVER_BUF_SIZE - 1, &pos);
+
+	if (bytes_read < 0) {
+		nvfs_err("Error reading file: %ld\n", bytes_read);
+		filp_close(file, NULL);
+		return -1;
+	} else {
+		buf[bytes_read] = '\0';
+		int ret = kstrtoint(buf,10,&value);
+		if (ret != 0) {
+			nvfs_err("Failed to convert driver version to int: %s, ret: %d\n", buf, ret);
+			filp_close(file, NULL);
+			return -1;
+		}
+
+	}
+	filp_close(file, NULL);
+	return value;
+}
+
 /*
  * Initialize nvfs driver
  */
 static int __init nvfs_init(void)
 {
 	int i;
 
+	int version= get_nvidia_driver_version();
+
+	// if the version is greater than or qequal 555, use persistent_p2p APIs
+	if (version >= NVIDIA_MIN_DRIVER_FOR_VGPU) {
+		nvfs_use_legacy_p2p_allocation = 0;
+	}
+
+	// if the driver version is less than 555, but it's x86 and BM, use the persistent_p2p APIs
 	#if defined(CONFIG_X86_64)
-	if (!cpu_feature_enabled(X86_FEATURE_HYPERVISOR)){
+	if (nvfs_use_legacy_p2p_allocation == 1  && !cpu_feature_enabled(X86_FEATURE_HYPERVISOR)){
 		// X86 and not a VM
 		nvfs_use_legacy_p2p_allocation = 0;
 	}
 	#endif
-	
+
 	pr_info("nvidia_fs: Initializing nvfs driver module\n");
 
 	major_number = register_chrdev(0, DEVICE_NAME, &nvfs_dev_fops);

src/nvfs-mmap.c

@@ -1179,7 +1179,7 @@ nvfs_mgroup_ptr_t nvfs_mgroup_from_page_range(struct page* page, int nblocks, un
 	unsigned block_idx;
 	unsigned cur_page;
 
-        nvfs_dbg("setting for %d nblocks from page: %p and start offset :%u\n", nblocks, page, start_offset);
+        nvfs_dbg("setting metadata for %d nblocks from page: %p and start offset :%u\n", nblocks, page, start_offset);
 	nvfs_mgroup = __nvfs_mgroup_from_page(page, false);
 	if (!nvfs_mgroup)
 	       return NULL;

src/nvfs-mmap.h

@@ -131,7 +131,7 @@ struct nvfs_gpu_args {
         u64 gpuvaddr;                               // GPU Buffer address
         u64 gpu_buf_len;                            // length of gpu buffer
         struct page *end_fence_page;                // end fence addr pinned page
-        u32 offset_in_page;			    // end_fence_addr byte offset in end_fence_page
+	u32 offset_in_page;			    // end_fence_addr byte offset in end_fence_page
         atomic_t io_state;                    	    // IO state transitions
         atomic_t dma_mapping_in_progress;	    // Mapping in progress for a specific PCI device
 	atomic_t callback_invoked;

src/nvfs-rdma.c

@@ -40,7 +40,7 @@ int nvfs_set_rdma_reg_info_to_mgroup(
 	uint32_t nkeys;
 	int ret = -EINVAL;
 
-	nvfs_dbg("SG: %s CPU vaddr: %llx \n", __func__, rdma_reg_info_args->cpuvaddr);	
+	nvfs_dbg("%s CPU vaddr: %llx \n", __func__, rdma_reg_info_args->cpuvaddr);
 
 	nvfs_mgroup = nvfs_get_mgroup_from_vaddr(rdma_reg_info_args->cpuvaddr);
 	if(nvfs_mgroup == NULL || unlikely(IS_ERR(nvfs_mgroup))) {
@@ -58,7 +58,7 @@ int nvfs_set_rdma_reg_info_to_mgroup(
 	shadow_buf_size = (nvfs_mgroup->nvfs_blocks_count) * NVFS_BLOCK_SIZE;
 
 
-	nvfs_dbg("SG: %s nvfs_mgroup = %p\n GPU vaddr: %llx", __func__,
+	nvfs_dbg("%s nvfs_mgroup = %p\n GPU vaddr: %llx", __func__,
 		 nvfs_mgroup, nvfs_mgroup->gpu_info.gpuvaddr);
 
 
@@ -130,7 +130,7 @@ int nvfs_get_rdma_reg_info_from_mgroup(
 
 	nvfs_mgroup = nvfs_get_mgroup_from_vaddr(rdma_reg_info_args->cpuvaddr);
 	if(nvfs_mgroup == NULL || unlikely(IS_ERR(nvfs_mgroup))) {
-		printk("SG Error: nvfs_mgroup NULL\n");
+		nvfs_err("Error: nvfs_mgroup NULL\n");
 		return -EINVAL;
 	}
 	shadow_buf_size = (nvfs_mgroup->nvfs_blocks_count) * NVFS_BLOCK_SIZE;
@@ -214,7 +214,6 @@ int nvfs_clear_rdma_reg_info_in_mgroup(
 	nvfs_mgroup = nvfs_get_mgroup_from_vaddr(rdma_clear_info_args->cpuvaddr);
 	if(nvfs_mgroup == NULL || unlikely(IS_ERR(nvfs_mgroup))) {
 		nvfs_err("%s Error:  nvfs_mgroup NULL\n", __func__);
-		printk("SG Error: nvfs_mgroup NULL\n");
 		return -1;
 	}
 

src/nvfs-vers.h

@@ -26,10 +26,10 @@
 
 #define NVFS_DRIVER_MAJOR_VERSION   2 //2-bytes
 
-#define NVFS_DRIVER_MINOR_VERSION   22 //2-bytes
+#define NVFS_DRIVER_MINOR_VERSION   23 //2-bytes
 
 // template for build version
-#define NVFS_DRIVER_PATCH_VERSION  3
+#define NVFS_DRIVER_PATCH_VERSION  1
 
 static inline unsigned int nvfs_driver_version(void) {
     return (NVFS_DRIVER_MAJOR_VERSION << 16) | NVFS_DRIVER_MINOR_VERSION;