spi_flash: Add Dual SPI support

This patch adds support to read SPI flash in Dual SPI mode, where both
MISO and MOSI lines are used for output mode (specifically Fast Read
Dual Output (0x3b) where the command is still sent normally, not Fast
Read Dual I/O (0xbb) whose additional benefit should be extremely
marginal for our use cases but which would be more complicated to
implement). This feature needs to be supported by both the flash chip
and the controller, so we add a new dual_spi flag (and a new flags field
to hold it) to the spi_flash structure and a new optional xfer_dual()
function pointer to the spi_ctrlr structure. When both are provided,
Dual SPI mode is used automatically, otherwise things work as before.

This patch only adds the dual_spi flag exemplary to all Winbond and
Gigadevice chips, other vendors need to be added as needed.

Change-Id: Ic6808224c99af32b6c5c43054135c8f4c03c1feb
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/33283
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-by: Furquan Shaikh <furquan@google.com>

6 files changed, 87 insertions, 4 deletions

diff --git a/src/drivers/spi/gigadevice.c b/src/drivers/spi/gigadevice.c
index cc4cf1dddb..1ff594a24a 100644
--- a/src/drivers/spi/gigadevice.c
+++ b/src/drivers/spi/gigadevice.c
@@ -41,7 +41,9 @@
 
 struct gigadevice_spi_flash_params {
 	uint16_t	id;
-	uint8_t		l2_page_size_shift;
+	uint8_t		dual_spi : 1;
+	uint8_t		_reserved_for_flags : 3;
+	uint8_t		l2_page_size_shift : 4;
 	uint8_t		pages_per_sector_shift : 4;
 	uint8_t		sectors_per_block_shift : 4;
 	uint8_t		nr_blocks_shift;
@@ -63,6 +65,7 @@ static const struct gigadevice_spi_flash_params gigadevice_spi_flash_table[] = {
 		.pages_per_sector_shift		= 4,
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 4,
+		.dual_spi			= 1,
 		.name				= "GD25Q80",
 	},					/* also GD25Q80B */
 	{
@@ -71,6 +74,7 @@ static const struct gigadevice_spi_flash_params gigadevice_spi_flash_table[] = {
 		.pages_per_sector_shift		= 4,
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 5,
+		.dual_spi			= 1,
 		.name				= "GD25Q16",
 	},					/* also GD25Q16B */
 	{
@@ -79,6 +83,7 @@ static const struct gigadevice_spi_flash_params gigadevice_spi_flash_table[] = {
 		.pages_per_sector_shift		= 4,
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 6,
+		.dual_spi			= 1,
 		.name				= "GD25Q32B",
 	},					/* also GD25Q32B */
 	{
@@ -87,6 +92,7 @@ static const struct gigadevice_spi_flash_params gigadevice_spi_flash_table[] = {
 		.pages_per_sector_shift		= 4,
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 7,
+		.dual_spi			= 1,
 		.name				= "GD25Q64",
 	},					/* also GD25Q64B, GD25B64C */
 	{
@@ -95,6 +101,7 @@ static const struct gigadevice_spi_flash_params gigadevice_spi_flash_table[] = {
 		.pages_per_sector_shift		= 4,
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 8,
+		.dual_spi			= 1,
 		.name				= "GD25Q128",
 	},					/* also GD25Q128B */
 	{
@@ -103,6 +110,7 @@ static const struct gigadevice_spi_flash_params gigadevice_spi_flash_table[] = {
 		.pages_per_sector_shift		= 4,
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 4,
+		.dual_spi			= 1,
 		.name				= "GD25VQ80C",
 	},
 	{
@@ -111,6 +119,7 @@ static const struct gigadevice_spi_flash_params gigadevice_spi_flash_table[] = {
 		.pages_per_sector_shift		= 4,
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 5,
+		.dual_spi			= 1,
 		.name				= "GD25VQ16C",
 	},
 	{
@@ -119,6 +128,7 @@ static const struct gigadevice_spi_flash_params gigadevice_spi_flash_table[] = {
 		.pages_per_sector_shift		= 4,
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 4,
+		.dual_spi			= 1,
 		.name				= "GD25LQ80",
 	},
 	{
@@ -127,6 +137,7 @@ static const struct gigadevice_spi_flash_params gigadevice_spi_flash_table[] = {
 		.pages_per_sector_shift		= 4,
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 5,
+		.dual_spi			= 1,
 		.name				= "GD25LQ16",
 	},
 	{
@@ -135,6 +146,7 @@ static const struct gigadevice_spi_flash_params gigadevice_spi_flash_table[] = {
 		.pages_per_sector_shift		= 4,
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 6,
+		.dual_spi			= 1,
 		.name				= "GD25LQ32",
 	},
 	{
@@ -143,6 +155,7 @@ static const struct gigadevice_spi_flash_params gigadevice_spi_flash_table[] = {
 		.pages_per_sector_shift		= 4,
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 7,
+		.dual_spi			= 1,
 		.name				= "GD25LQ64C",
 	},					/* also GD25LB64C */
 	{
@@ -151,6 +164,7 @@ static const struct gigadevice_spi_flash_params gigadevice_spi_flash_table[] = {
 		.pages_per_sector_shift		= 4,
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 8,
+		.dual_spi			= 1,
 		.name				= "GD25LQ128",
 	},
 };
diff --git a/src/drivers/spi/spi_flash.c b/src/drivers/spi/spi_flash.c
index 5e42a375f6..a81306e386 100644
--- a/src/drivers/spi/spi_flash.c
+++ b/src/drivers/spi/spi_flash.c
@@ -59,6 +59,34 @@ static int do_spi_flash_cmd(const struct spi_slave *spi, const void *dout,
 	return ret;
 }
 
+static int do_dual_read_cmd(const struct spi_slave *spi, const void *dout,
+			    size_t bytes_out, void *din, size_t bytes_in)
+{
+	int ret;
+
+	/*
+	 * spi_xfer_vector() will automatically fall back to .xfer() if
+	 * .xfer_vector() is unimplemented. So using vector API here is more
+	 * flexible, even though a controller that implements .xfer_vector()
+	 * and (the non-vector based) .xfer_dual() but not .xfer() would be
+	 * pretty odd.
+	 */
+	struct spi_op vector = { .dout = dout, .bytesout = bytes_out,
+				 .din = NULL, .bytesin = 0 };
+
+	ret = spi_claim_bus(spi);
+	if (ret)
+		return ret;
+
+	ret = spi_xfer_vector(spi, &vector, 1);
+
+	if (!ret)
+		ret = spi->ctrlr->xfer_dual(spi, NULL, 0, din, bytes_in);
+
+	spi_release_bus(spi);
+	return ret;
+}
+
 int spi_flash_cmd(const struct spi_slave *spi, u8 cmd, void *response, size_t len)
 {
 	int ret = do_spi_flash_cmd(spi, &cmd, sizeof(cmd), response, len);
@@ -105,6 +133,11 @@ static int spi_flash_read_chunked(const struct spi_flash *flash, u32 offset,
 		cmd_len = 4;
 		cmd[0] = CMD_READ_ARRAY_SLOW;
 		do_cmd = do_spi_flash_cmd;
+	} else if (flash->flags.dual_spi && flash->spi.ctrlr->xfer_dual) {
+		cmd_len = 5;
+		cmd[0] = CMD_READ_FAST_DUAL_OUTPUT;
+		cmd[4] = 0;
+		do_cmd = do_dual_read_cmd;
 	} else {
 		cmd_len = 5;
 		cmd[0] = CMD_READ_ARRAY_FAST;
@@ -347,8 +380,12 @@ int spi_flash_probe(unsigned int bus, unsigned int cs, struct spi_flash *flash)
 		return -1;
 	}
 
-	printk(BIOS_INFO, "SF: Detected %s with sector size 0x%x, total 0x%x\n",
-			flash->name, flash->sector_size, flash->size);
+	const char *mode_string = "";
+	if (flash->flags.dual_spi && spi.ctrlr->xfer_dual)
+		mode_string = " (Dual SPI mode)";
+	printk(BIOS_INFO,
+	       "SF: Detected %s with sector size 0x%x, total 0x%x%s\n",
+		flash->name, flash->sector_size, flash->size, mode_string);
 	if (bus == CONFIG_BOOT_DEVICE_SPI_FLASH_BUS
 			&& flash->size != CONFIG_ROM_SIZE) {
 		printk(BIOS_ERR, "SF size 0x%x does not correspond to"
diff --git a/src/drivers/spi/spi_flash_internal.h b/src/drivers/spi/spi_flash_internal.h
index f15c737461..4a9e289029 100644
--- a/src/drivers/spi/spi_flash_internal.h
+++ b/src/drivers/spi/spi_flash_internal.h
@@ -23,6 +23,8 @@
 #define CMD_READ_ARRAY_FAST		0x0b
 #define CMD_READ_ARRAY_LEGACY		0xe8
 
+#define CMD_READ_FAST_DUAL_OUTPUT	0x3b
+
 #define CMD_READ_STATUS			0x05
 #define CMD_WRITE_ENABLE		0x06
 
diff --git a/src/drivers/spi/winbond.c b/src/drivers/spi/winbond.c
index d0ef3cd6e0..9e9bb00464 100644
--- a/src/drivers/spi/winbond.c
+++ b/src/drivers/spi/winbond.c
@@ -26,7 +26,9 @@
 
 struct winbond_spi_flash_params {
 	uint16_t id;
-	uint8_t l2_page_size_shift;
+	uint8_t dual_spi : 1;
+	uint8_t _reserved_for_flags : 3;
+	uint8_t l2_page_size_shift : 4;
 	uint8_t pages_per_sector_shift : 4;
 	uint8_t sectors_per_block_shift : 4;
 	uint8_t nr_blocks_shift;
@@ -123,6 +125,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 4,
 		.name				= "W25X80",
+		.dual_spi			= 1,
 	},
 	{
 		.id				= 0x3015,
@@ -131,6 +134,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 5,
 		.name				= "W25X16",
+		.dual_spi			= 1,
 	},
 	{
 		.id				= 0x3016,
@@ -139,6 +143,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 6,
 		.name				= "W25X32",
+		.dual_spi			= 1,
 	},
 	{
 		.id				= 0x3017,
@@ -147,6 +152,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 7,
 		.name				= "W25X64",
+		.dual_spi			= 1,
 	},
 	{
 		.id				= 0x4014,
@@ -155,6 +161,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 4,
 		.name				= "W25Q80_V",
+		.dual_spi			= 1,
 	},
 	{
 		.id				= 0x4015,
@@ -163,6 +170,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 5,
 		.name				= "W25Q16_V",
+		.dual_spi			= 1,
 		.protection_granularity_shift	= 16,
 		.bp_bits			= 3,
 	},
@@ -173,6 +181,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 5,
 		.name				= "W25Q16DW",
+		.dual_spi			= 1,
 		.protection_granularity_shift	= 16,
 		.bp_bits			= 3,
 	},
@@ -183,6 +192,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 6,
 		.name				= "W25Q32_V",
+		.dual_spi			= 1,
 		.protection_granularity_shift	= 16,
 		.bp_bits			= 3,
 	},
@@ -193,6 +203,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 6,
 		.name				= "W25Q32DW",
+		.dual_spi			= 1,
 		.protection_granularity_shift	= 16,
 		.bp_bits			= 3,
 	},
@@ -203,6 +214,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 7,
 		.name				= "W25Q64_V",
+		.dual_spi			= 1,
 		.protection_granularity_shift	= 17,
 		.bp_bits			= 3,
 	},
@@ -213,6 +225,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 7,
 		.name				= "W25Q64DW",
+		.dual_spi			= 1,
 		.protection_granularity_shift	= 17,
 		.bp_bits			= 3,
 	},
@@ -223,6 +236,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 8,
 		.name				= "W25Q128_V",
+		.dual_spi			= 1,
 		.protection_granularity_shift	= 18,
 		.bp_bits			= 3,
 	},
@@ -233,6 +247,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 8,
 		.name				= "W25Q128FW",
+		.dual_spi			= 1,
 		.protection_granularity_shift	= 18,
 		.bp_bits			= 3,
 	},
@@ -243,6 +258,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 8,
 		.name				= "W25Q128J",
+		.dual_spi			= 1,
 		.protection_granularity_shift	= 18,
 		.bp_bits			= 3,
 	},
@@ -253,6 +269,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 9,
 		.name				= "W25Q256_V",
+		.dual_spi			= 1,
 		.protection_granularity_shift	= 16,
 		.bp_bits			= 4,
 	},
@@ -263,6 +280,7 @@ static const struct winbond_spi_flash_params winbond_spi_flash_table[] = {
 		.sectors_per_block_shift	= 4,
 		.nr_blocks_shift		= 9,
 		.name				= "W25Q256J",
+		.dual_spi			= 1,
 		.protection_granularity_shift	= 16,
 		.bp_bits			= 4,
 	},
@@ -681,6 +699,8 @@ int spi_flash_probe_winbond(const struct spi_slave *spi, u8 *idcode,
 	flash->erase_cmd = CMD_W25_SE;
 	flash->status_cmd = CMD_W25_RDSR;
 
+	flash->flags.dual_spi = params->dual_spi;
+
 	flash->ops = &spi_flash_ops;
 	flash->driver_private = params;
 
diff --git a/src/include/spi-generic.h b/src/include/spi-generic.h
index c24aadd926..d0f957f1f9 100644
--- a/src/include/spi-generic.h
+++ b/src/include/spi-generic.h
@@ -125,6 +125,7 @@ enum {
  * setup:		Setup given SPI device bus.
  * xfer:		Perform one SPI transfer operation.
  * xfer_vector:	Vector of SPI transfer operations.
+ * xfer_dual:		(optional) Perform one SPI transfer in Dual SPI mode.
  * max_xfer_size:	Maximum transfer size supported by the controller
  *			(0 = invalid,
  *			 SPI_CTRLR_DEFAULT_MAX_XFER_SIZE = unlimited)
@@ -145,6 +146,8 @@ struct spi_ctrlr {
 		    size_t bytesout, void *din, size_t bytesin);
 	int (*xfer_vector)(const struct spi_slave *slave,
 			struct spi_op vectors[], size_t count);
+	int (*xfer_dual)(const struct spi_slave *slave, const void *dout,
+			 size_t bytesout, void *din, size_t bytesin);
 	uint32_t max_xfer_size;
 	uint32_t flags;
 	int (*flash_probe)(const struct spi_slave *slave,
diff --git a/src/include/spi_flash.h b/src/include/spi_flash.h
index 936b0abe85..3a0c383676 100644
--- a/src/include/spi_flash.h
+++ b/src/include/spi_flash.h
@@ -90,6 +90,13 @@ struct spi_flash_ops {
 struct spi_flash {
 	struct spi_slave spi;
 	u8 vendor;
+	union {
+		u8 raw;
+		struct {
+			u8 dual_spi	: 1;
+			u8 _reserved	: 7;
+		};
+	} flags;
 	u16 model;
 	const char *name;
 	u32 size;