dfu: add scsi backend

This is extremely similar to the MMC backend, but there are some notable
differences.

Works with a DFU string like

    scsi 4=u-boot-bin part 11

Where "4" is the SCSI dev number (sequential LUN across all SCSI devices)
and "11" is the partition number.

Reviewed-by: Mattijs Korpershoek <mkorpershoek@baylibre.com>
Acked-by: Mattijs Korpershoek <mkorpershoek@baylibre.com>
Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>

doc/usage/dfu.rst

@@ -22,6 +22,7 @@ U-Boot implements this DFU capability (CONFIG_DFU) with the command dfu
 Today the supported DFU backends are:
 
 - MMC (RAW or FAT / EXT2 / EXT3 / EXT4 file system / SKIP / SCRIPT)
+- SCSI (UFS, RAW partition, FAT / EXT2 / EXT3 / EXT4 file system / SKIP / SCRIPT)
 - NAND
 - RAM
 - SF (serial flash)
@@ -51,6 +52,7 @@ The following configuration options are relevant to device firmware upgrade:
 * CONFIG_DFU_MTD
 * CONFIG_DFU_NAND
 * CONFIG_DFU_RAM
+* CONFIG_DFU_SCSI
 * CONFIG_DFU_SF
 * CONFIG_DFU_SF_PART
 * CONFIG_DFU_TIMEOUT
@@ -167,6 +169,36 @@ mmc
     Please note that this means the user will be able to execute any
     arbitrary commands just like in the u-boot's shell.
 
+scsi
+    for UFS storage::
+
+        dfu 0 scsi <dev>
+
+    each element in *dfu_alt_info* being
+
+    * <name> raw <offset> <size>     raw access to SCSI LUN
+    * <name> part <part_id>          raw access to partition
+    * <name> fat <part_id>           file in FAT partition
+    * <name> ext4 <part_id>          file in EXT4 partition
+    * <name> skip 0 0                ignore flashed data
+    * <name> script 0 0              execute commands in shell
+
+    with
+
+    size
+        is the size of the access area (hexadecimal without "0x")
+        or 0 which means whole device
+    partid
+        is the GPT or DOS partition index.
+    dev
+        is the SCSI LU (Logical Unit) index (decimal only)
+
+    A value of environment variable *dfu_alt_info* for UFS could be::
+
+        u-boot part 4;bl2 raw 0x1e 0x1d
+
+    See mmc section above for details on the skip and script types.
+
 nand
     raw slc nand device::
 
@@ -278,6 +310,7 @@ alternate list separated by '&' with the same format for each <alt>::
     mmc <dev>=<alt1>;....;<altN>
     nand <dev>=<alt1>;....;<altN>
     ram <dev>=<alt1>;....;<altN>
+    scsi <dev>=<alt1>;....;<altN>
     sf <dev>=<alt1>;....;<altN>
     mtd <dev>=<alt1>;....;<altN>
     virt <dev>=<alt1>;....;<altN>

drivers/dfu/Kconfig

@@ -89,6 +89,13 @@ config DFU_VIRT
 	  used at board level to manage specific behavior
 	  (OTP update for example).
 
+config DFU_SCSI
+	bool "SCSI flash back end for DFU"
+	help
+	  This option enables using DFU to read and write to SCSI devices
+	  used at board level to manage specific behavior
+	  (OTP update for example).
+
 config SET_DFU_ALT_INFO
 	bool "Dynamic set of DFU alternate information"
 	help

drivers/dfu/Makefile

@@ -11,3 +11,4 @@ obj-$(CONFIG_$(XPL_)DFU_RAM) += dfu_ram.o
 obj-$(CONFIG_$(XPL_)DFU_SF) += dfu_sf.o
 obj-$(CONFIG_$(XPL_)DFU_WRITE_ALT) += dfu_alt.o
 obj-$(CONFIG_$(XPL_)DFU_VIRT) += dfu_virt.o
+obj-$(CONFIG_$(XPL_)DFU_SCSI) += dfu_scsi.o

drivers/dfu/dfu.c

@@ -564,6 +564,9 @@ static int dfu_fill_entity(struct dfu_entity *dfu, char *s, int alt,
 	} else if (strcmp(interface, "virt") == 0) {
 		if (dfu_fill_entity_virt(dfu, devstr, argv, argc))
 			return -1;
+	} else if (strcmp(interface, "scsi") == 0) {
+		if (dfu_fill_entity_scsi(dfu, devstr, argv, argc))
+			return -1;
 	} else {
 		printf("%s: Device %s not (yet) supported!\n",
 		       __func__,  interface);
@@ -660,7 +663,7 @@ int dfu_config_entities(char *env, char *interface, char *devstr)
 const char *dfu_get_dev_type(enum dfu_device_type t)
 {
 	const char *const dev_t[] = {NULL, "eMMC", "OneNAND", "NAND", "RAM",
-				     "SF", "MTD", "VIRT"};
+				     "SF", "MTD", "VIRT", "SCSI"};
 	return dev_t[t];
 }
 

drivers/dfu/dfu_scsi.c

@@ -0,0 +1,435 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * DFU SCSI backend (based on MMC backend).
+ *
+ * Copyright (C) 2012 Samsung Electronics
+ *   author: Lukasz Majewski <l.majewski@samsung.com>
+ * Copyright (C) 2024 Linaro Ltd.
+ */
+
+#include <log.h>
+#include <malloc.h>
+#include <errno.h>
+#include <div64.h>
+#include <dfu.h>
+#include <ext4fs.h>
+#include <fat.h>
+#include <scsi.h>
+#include <part.h>
+#include <command.h>
+#include <linux/printk.h>
+
+static unsigned char *dfu_file_buf;
+static u64 dfu_file_buf_len;
+static u64 dfu_file_buf_offset;
+
+#define scsi_get_blk_desc(dev) ((struct blk_desc *)dev_get_uclass_plat(dev))
+
+#define find_scsi_device(dev_num, scsi) blk_get_device(UCLASS_SCSI, dev_num, scsi)
+
+static int scsi_block_op(enum dfu_op op, struct dfu_entity *dfu, u64 offset, void *buf, long *len)
+{
+	struct udevice *scsi;
+	u32 blk_start, blk_count, n = 0;
+	int ret;
+
+	ret = find_scsi_device(dfu->data.scsi.lun, &scsi);
+	if (ret < 0) {
+		pr_err("Device scsi %d - not found!", dfu->data.scsi.lun);
+		return -ENODEV;
+	}
+
+	/*
+	 * We must ensure that we work in lba_blk_size chunks, so ALIGN
+	 * this value.
+	 */
+	*len = ALIGN(*len, dfu->data.scsi.lba_blk_size);
+
+	blk_start = dfu->data.scsi.lba_start + (u32)lldiv(offset, dfu->data.scsi.lba_blk_size);
+	blk_count = *len / dfu->data.scsi.lba_blk_size;
+	if (blk_start + blk_count > dfu->data.scsi.lba_start + dfu->data.scsi.lba_size) {
+		puts("Request would exceed designated area!\n");
+		return -EINVAL;
+	}
+
+	debug("%s: %s dev: %d start: %d cnt: %d buf: 0x%p\n", __func__,
+	      op == DFU_OP_READ ? "scsi READ" : "scsi WRITE", dfu->data.scsi.lun, blk_start,
+	      blk_count, buf);
+	switch (op) {
+	case DFU_OP_READ:
+		n = blk_dread(scsi_get_blk_desc(scsi), blk_start, blk_count, buf);
+		break;
+	case DFU_OP_WRITE:
+		n = blk_dwrite(scsi_get_blk_desc(scsi), blk_start, blk_count, buf);
+		break;
+	default:
+		pr_err("Operation not supported\n");
+	}
+
+	if (n != blk_count) {
+		pr_err("scsi block operation failed");
+		return -EIO;
+	}
+
+	return 0;
+}
+
+static int scsi_file_op(enum dfu_op op, struct dfu_entity *dfu, u64 offset, void *buf, u64 *len)
+{
+	char dev_part_str[8];
+	int ret;
+	int fstype;
+	loff_t size = 0;
+
+	switch (dfu->layout) {
+	case DFU_FS_FAT:
+		fstype = FS_TYPE_FAT;
+		break;
+	case DFU_FS_EXT4:
+		fstype = FS_TYPE_EXT;
+		break;
+	case DFU_SKIP:
+		return 0;
+	default:
+		printf("%s: Layout (%s) not (yet) supported!\n", __func__,
+		       dfu_get_layout(dfu->layout));
+		return -1;
+	}
+
+	snprintf(dev_part_str, sizeof(dev_part_str), "%d:%d", dfu->data.scsi.dev,
+		 dfu->data.scsi.part);
+
+	ret = fs_set_blk_dev("scsi", dev_part_str, fstype);
+	if (ret) {
+		puts("dfu: fs_set_blk_dev error!\n");
+		return ret;
+	}
+
+	switch (op) {
+	case DFU_OP_READ:
+		ret = fs_read(dfu->name, (size_t)buf, offset, *len, &size);
+		if (ret) {
+			puts("dfu: fs_read error!\n");
+			return ret;
+		}
+		*len = size;
+		break;
+	case DFU_OP_WRITE:
+		ret = fs_write(dfu->name, (size_t)buf, offset, *len, &size);
+		if (ret) {
+			puts("dfu: fs_write error!\n");
+			return ret;
+		}
+		break;
+	case DFU_OP_SIZE:
+		ret = fs_size(dfu->name, &size);
+		if (ret) {
+			puts("dfu: fs_size error!\n");
+			return ret;
+		}
+		*len = size;
+		break;
+	default:
+		return -1;
+	}
+
+	return ret;
+}
+
+static int scsi_file_buf_write(struct dfu_entity *dfu, u64 offset, void *buf, long *len)
+{
+	int ret = 0;
+
+	if (offset == 0) {
+		dfu_file_buf_len = 0;
+		dfu_file_buf_offset = 0;
+	}
+
+	/* Add to the current buffer. */
+	if (dfu_file_buf_len + *len > CONFIG_SYS_DFU_MAX_FILE_SIZE)
+		*len = CONFIG_SYS_DFU_MAX_FILE_SIZE - dfu_file_buf_len;
+	memcpy(dfu_file_buf + dfu_file_buf_len, buf, *len);
+	dfu_file_buf_len += *len;
+
+	if (dfu_file_buf_len == CONFIG_SYS_DFU_MAX_FILE_SIZE) {
+		ret = scsi_file_op(DFU_OP_WRITE, dfu, dfu_file_buf_offset, dfu_file_buf,
+				   &dfu_file_buf_len);
+		dfu_file_buf_offset += dfu_file_buf_len;
+		dfu_file_buf_len = 0;
+	}
+
+	return ret;
+}
+
+static int scsi_file_buf_write_finish(struct dfu_entity *dfu)
+{
+	int ret = scsi_file_op(DFU_OP_WRITE, dfu, dfu_file_buf_offset, dfu_file_buf,
+			       &dfu_file_buf_len);
+
+	/* Now that we're done */
+	dfu_file_buf_len = 0;
+	dfu_file_buf_offset = 0;
+
+	return ret;
+}
+
+int dfu_write_medium_scsi(struct dfu_entity *dfu, u64 offset, void *buf, long *len)
+{
+	int ret = -1;
+
+	switch (dfu->layout) {
+	case DFU_RAW_ADDR:
+		ret = scsi_block_op(DFU_OP_WRITE, dfu, offset, buf, len);
+		break;
+	case DFU_FS_FAT:
+	case DFU_FS_EXT4:
+		ret = scsi_file_buf_write(dfu, offset, buf, len);
+		break;
+	case DFU_SCRIPT:
+		ret = run_command_list(buf, *len, 0);
+		break;
+	case DFU_SKIP:
+		ret = 0;
+		break;
+	default:
+		printf("%s: Layout (%s) not (yet) supported!\n", __func__,
+		       dfu_get_layout(dfu->layout));
+	}
+
+	return ret;
+}
+
+int dfu_flush_medium_scsi(struct dfu_entity *dfu)
+{
+	int ret = 0;
+
+	switch (dfu->layout) {
+	case DFU_FS_FAT:
+	case DFU_FS_EXT4:
+		ret = scsi_file_buf_write_finish(dfu);
+		break;
+	case DFU_SCRIPT:
+		/* script may have changed the dfu_alt_info */
+		dfu_reinit_needed = true;
+		break;
+	case DFU_RAW_ADDR:
+	case DFU_SKIP:
+		break;
+	default:
+		printf("%s: Layout (%s) not (yet) supported!\n", __func__,
+		       dfu_get_layout(dfu->layout));
+	}
+
+	return ret;
+}
+
+int dfu_get_medium_size_scsi(struct dfu_entity *dfu, u64 *size)
+{
+	int ret;
+
+	switch (dfu->layout) {
+	case DFU_RAW_ADDR:
+		*size = dfu->data.scsi.lba_size * dfu->data.scsi.lba_blk_size;
+		return 0;
+	case DFU_FS_FAT:
+	case DFU_FS_EXT4:
+		ret = scsi_file_op(DFU_OP_SIZE, dfu, 0, NULL, size);
+		if (ret < 0)
+			return ret;
+		return 0;
+	case DFU_SCRIPT:
+	case DFU_SKIP:
+		return 0;
+	default:
+		printf("%s: Layout (%s) not (yet) supported!\n", __func__,
+		       dfu_get_layout(dfu->layout));
+		return -1;
+	}
+}
+
+static int scsi_file_buf_read(struct dfu_entity *dfu, u64 offset, void *buf, long *len)
+{
+	int ret;
+
+	if (offset == 0 || offset >= dfu_file_buf_offset + dfu_file_buf_len ||
+	    offset + *len < dfu_file_buf_offset) {
+		u64 file_len = CONFIG_SYS_DFU_MAX_FILE_SIZE;
+
+		ret = scsi_file_op(DFU_OP_READ, dfu, offset, dfu_file_buf, &file_len);
+		if (ret < 0)
+			return ret;
+		dfu_file_buf_len = file_len;
+		dfu_file_buf_offset = offset;
+	}
+	if (offset + *len > dfu_file_buf_offset + dfu_file_buf_len)
+		return -EINVAL;
+
+	/* Add to the current buffer. */
+	memcpy(buf, dfu_file_buf + offset - dfu_file_buf_offset, *len);
+
+	return 0;
+}
+
+int dfu_read_medium_scsi(struct dfu_entity *dfu, u64 offset, void *buf, long *len)
+{
+	int ret = -1;
+
+	switch (dfu->layout) {
+	case DFU_RAW_ADDR:
+		ret = scsi_block_op(DFU_OP_READ, dfu, offset, buf, len);
+		break;
+	case DFU_FS_FAT:
+	case DFU_FS_EXT4:
+		ret = scsi_file_buf_read(dfu, offset, buf, len);
+		break;
+	default:
+		printf("%s: Layout (%s) not (yet) supported!\n", __func__,
+		       dfu_get_layout(dfu->layout));
+	}
+
+	return ret;
+}
+
+void dfu_free_entity_scsi(struct dfu_entity *dfu)
+{
+	if (dfu_file_buf) {
+		free(dfu_file_buf);
+		dfu_file_buf = NULL;
+	}
+}
+
+/*
+ * @param s Parameter string containing space-separated arguments:
+ *	1st:
+ *		raw	(raw read/write)
+ *		fat	(files)
+ *		ext4	(^)
+ *		part	(partition image)
+ *	2nd and 3rd:
+ *		lba_start and lba_size, for raw write
+ *		scsi_dev and scsi_part, for filesystems and part
+ */
+int dfu_fill_entity_scsi(struct dfu_entity *dfu, char *devstr, char **argv, int argc)
+{
+	const char *entity_type;
+	ssize_t second_arg;
+	ssize_t third_arg = -1;
+	struct udevice *scsi;
+	struct blk_desc *blk_dev;
+	int ret;
+	char *s;
+
+	if (argc < 2) {
+		pr_err("Need at least one argument\n");
+		return -EINVAL;
+	}
+
+	dfu->data.scsi.lun = dectoul(devstr, &s);
+	if (*s)
+		return -EINVAL;
+
+	entity_type = argv[0];
+	/*
+	 * Base 0 means we'll accept (prefixed with 0x or 0) base 16, 8,
+	 * with default 10.
+	 */
+	second_arg = simple_strtol(argv[1], &s, 0);
+	if (*s)
+		return -EINVAL;
+	if (argc >= 3) {
+		third_arg = simple_strtoul(argv[2], &s, 0);
+		if (*s)
+			return -EINVAL;
+	}
+
+	if (scsi_scan(false)) {
+		pr_err("Couldn't init scsi device.\n");
+		return -ENODEV;
+	}
+
+	ret = find_scsi_device(dfu->data.scsi.lun, &scsi);
+	if (ret < 0) {
+		pr_err("Couldn't find scsi device no. %d.\n", dfu->data.scsi.lun);
+		return -ENODEV;
+	}
+
+	blk_dev = scsi_get_blk_desc(scsi);
+	if (!blk_dev) {
+		pr_err("Couldn't get block device for scsi device no. %d.\n", dfu->data.scsi.lun);
+		return -ENODEV;
+	}
+
+	/* if it's NOT a raw write */
+	if (strcmp(entity_type, "raw")) {
+		dfu->data.scsi.dev = (second_arg != -1) ? second_arg : dfu->data.scsi.lun;
+		dfu->data.scsi.part = third_arg;
+	}
+
+	if (!strcmp(entity_type, "raw")) {
+		dfu->layout = DFU_RAW_ADDR;
+		dfu->data.scsi.lba_start = second_arg;
+		if (third_arg < 0) {
+			pr_err("raw requires two arguments\n");
+			return -EINVAL;
+		}
+		dfu->data.scsi.lba_size = third_arg;
+		dfu->data.scsi.lba_blk_size = blk_dev->blksz;
+
+		/*
+		 * In case the size is zero (i.e. scsi raw 0x10 0),
+		 * assume the user intends to use whole device.
+		 */
+		if (third_arg == 0)
+			dfu->data.scsi.lba_size = blk_dev->lba;
+
+	} else if (!strcmp(entity_type, "part")) {
+		struct disk_partition partinfo;
+		int scsipart = second_arg;
+
+		if (third_arg >= 0) {
+			pr_err("part only accepts one argument\n");
+			return -EINVAL;
+		}
+
+		if (part_get_info(blk_dev, scsipart, &partinfo) != 0) {
+			pr_err("Couldn't find part #%d on scsi device #%d\n", scsipart,
+			       dfu->data.scsi.lun);
+			return -ENODEV;
+		}
+
+		dfu->layout = DFU_RAW_ADDR;
+		dfu->data.scsi.lba_start = partinfo.start;
+		dfu->data.scsi.lba_size = partinfo.size;
+		dfu->data.scsi.lba_blk_size = partinfo.blksz;
+	} else if (!strcmp(entity_type, "fat")) {
+		dfu->layout = DFU_FS_FAT;
+	} else if (!strcmp(entity_type, "ext4")) {
+		dfu->layout = DFU_FS_EXT4;
+	} else if (!strcmp(entity_type, "skip")) {
+		dfu->layout = DFU_SKIP;
+	} else if (!strcmp(entity_type, "script")) {
+		dfu->layout = DFU_SCRIPT;
+	} else {
+		pr_err("Memory layout (%s) not supported!\n", entity_type);
+		return -ENODEV;
+	}
+
+	dfu->dev_type = DFU_DEV_SCSI;
+	dfu->get_medium_size = dfu_get_medium_size_scsi;
+	dfu->read_medium = dfu_read_medium_scsi;
+	dfu->write_medium = dfu_write_medium_scsi;
+	dfu->flush_medium = dfu_flush_medium_scsi;
+	dfu->inited = 0;
+	dfu->free_entity = dfu_free_entity_scsi;
+
+	/* Check if file buffer is ready */
+	if (!dfu_file_buf) {
+		dfu_file_buf = memalign(CONFIG_SYS_CACHELINE_SIZE, CONFIG_SYS_DFU_MAX_FILE_SIZE);
+		if (!dfu_file_buf) {
+			pr_err("Could not memalign 0x%x bytes\n", CONFIG_SYS_DFU_MAX_FILE_SIZE);
+			return -ENOMEM;
+		}
+	}
+
+	return 0;
+}

include/dfu.h

@@ -24,6 +24,7 @@ enum dfu_device_type {
 	DFU_DEV_SF,
 	DFU_DEV_MTD,
 	DFU_DEV_VIRT,
+	DFU_DEV_SCSI,
 };
 
 enum dfu_layout {
@@ -99,6 +100,19 @@ struct virt_internal_data {
 	int dev_num;
 };
 
+struct scsi_internal_data {
+	int lun;
+
+	/* RAW programming */
+	unsigned int lba_start;
+	unsigned int lba_size;
+	unsigned int lba_blk_size;
+
+	/* FAT/EXT */
+	unsigned int dev; // Always 0???
+	unsigned int part;
+};
+
 #if defined(CONFIG_DFU_NAME_MAX_SIZE)
 #define DFU_NAME_SIZE			CONFIG_DFU_NAME_MAX_SIZE
 #else
@@ -126,6 +140,7 @@ struct dfu_entity {
 		struct ram_internal_data ram;
 		struct sf_internal_data sf;
 		struct virt_internal_data virt;
+		struct scsi_internal_data scsi;
 	} data;
 
 	int (*get_medium_size)(struct dfu_entity *dfu, u64 *size);
@@ -516,6 +531,18 @@ static inline int dfu_fill_entity_virt(struct dfu_entity *dfu, char *devstr,
 }
 #endif
 
+#if CONFIG_IS_ENABLED(DFU_SCSI)
+int dfu_fill_entity_scsi(struct dfu_entity *dfu, char *devstr,
+			 char **argv, int argc);
+#else
+static inline int dfu_fill_entity_scsi(struct dfu_entity *dfu, char *devstr,
+				       char **argv, int argc)
+{
+	puts("SCSI support not available!\n");
+	return -1;
+}
+#endif
+
 extern bool dfu_reinit_needed;
 extern bool dfu_alt_info_changed;