summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorDavid Hendricks <dhendrix@chromium.org>2013-11-22 18:41:38 -0800
committerIsaac Christensen <isaac.christensen@se-eng.com>2014-09-17 19:56:16 +0200
commit1101a71219794b3c070dc67df2f9530b05c4c0fb (patch)
tree8bc93804dee1d0e9fd1927707ab1c1db8564b57a
parent0108b936bf0f2d185691d8719c30193af39abefe (diff)
downloadcoreboot-staging-1101a71219794b3c070dc67df2f9530b05c4c0fb.zip
coreboot-staging-1101a71219794b3c070dc67df2f9530b05c4c0fb.tar.gz
spi: add Kconfig variable for dual-output read enable
Add a Kconfig variable so that driver code knows whether or not to use dual-output reads. Signed-off-by: David Hendricks <dhendrix@chromium.org> Old-Change-Id: I31d23bfedd91521d719378ec573e33b381ebd2c5 Reviewed-on: https://chromium-review.googlesource.com/177834 Reviewed-by: David Hendricks <dhendrix@chromium.org> Commit-Queue: David Hendricks <dhendrix@chromium.org> Tested-by: David Hendricks <dhendrix@chromium.org> (cherry picked from commit de6869a3350041c6823427787971efc9fcf469b8) tegra124: implement x2 mode for SPI transfers on CBFS media This implements x2 mode when reading CBFS media over SPI. In theory this effectively doubles our throughput, though the initial results were almost negligibly better. Using a logic analyzer we see a pattern of 12 clocks, ~70ns delay, 4 clocks, ~310ns delay. So if we want to see further gains here then we'll probably need to tune AHB arbitration and utilization to eliminate bubbles/stalls when copying from APB DMA. Signed-off-by: David Hendricks <dhendrix@chromium.org> Old-Change-Id: I33d6ae30923fc42b4dc7103d029085985472cf3e Reviewed-on: https://chromium-review.googlesource.com/177835 Reviewed-by: Tom Warren <twarren@nvidia.com> Reviewed-by: David Hendricks <dhendrix@chromium.org> Commit-Queue: David Hendricks <dhendrix@chromium.org> Tested-by: David Hendricks <dhendrix@chromium.org> (cherry picked from commit 29289223362b12e84da5cbb130f285c6b9d314cc) nyan: turn on dual-output reads for SPI flash Nyan's SPI chip is capable of dual-output reads, so let's use it. Signed-off-by: David Hendricks <dhendrix@chromium.org> Old-Change-Id: I51a97c05aa25442d8ddcc4e3e35a2507d91a64df Reviewed-on: https://chromium-review.googlesource.com/177836 Reviewed-by: David Hendricks <dhendrix@chromium.org> Commit-Queue: David Hendricks <dhendrix@chromium.org> Tested-by: David Hendricks <dhendrix@chromium.org> (cherry picked from commit 62de0889a9cfc5686800645d05e21e272e4beb5c) Squashed three commits to enable dual output spi reads for nyan. Also fixed the spi_xfer interface that has been updated to use bytes instead of bits. Change-Id: I750a177576175b297f61e1b10eac6db15e75aa6e Signed-off-by: Isaac Christensen <isaac.christensen@se-eng.com> Reviewed-on: http://review.coreboot.org/6909 Tested-by: build bot (Jenkins) Reviewed-by: David Hendricks <dhendrix@chromium.org>
-rw-r--r--src/drivers/spi/Kconfig9
-rw-r--r--src/mainboard/google/nyan/Kconfig2
-rw-r--r--src/soc/nvidia/tegra124/spi.c53
-rw-r--r--src/soc/nvidia/tegra124/spi.h1
4 files changed, 53 insertions, 12 deletions
diff --git a/src/drivers/spi/Kconfig b/src/drivers/spi/Kconfig
index 6b31053..d832e0c 100644
--- a/src/drivers/spi/Kconfig
+++ b/src/drivers/spi/Kconfig
@@ -103,4 +103,13 @@ config SPI_FLASH_WINBOND
Select this option if your chipset driver needs to store certain
data in the SPI flash and your SPI flash is made by Winbond.
+config SPI_FLASH_FAST_READ_DUAL_OUTPUT_3B
+ bool
+ default n
+ depends on SPI_FLASH
+ help
+ Select this option if your SPI flash supports the fast read dual-
+ output command (opcode 0x3b) where the opcode and address are sent
+ to the chip on MOSI and data is received on both MOSI and MISO.
+
endif # SPI_FLASH
diff --git a/src/mainboard/google/nyan/Kconfig b/src/mainboard/google/nyan/Kconfig
index 1a7b260..320039f 100644
--- a/src/mainboard/google/nyan/Kconfig
+++ b/src/mainboard/google/nyan/Kconfig
@@ -28,6 +28,8 @@ config BOARD_SPECIFIC_OPTIONS # dummy
select MAINBOARD_HAS_BOOTBLOCK_INIT
select MAINBOARD_DO_NATIVE_VGA_INIT
select BOARD_ROMSIZE_KB_1024
+ select SPI_FLASH
+ select SPI_FLASH_FAST_READ_DUAL_OUTPUT_3B
config MAINBOARD_DIR
string
diff --git a/src/soc/nvidia/tegra124/spi.c b/src/soc/nvidia/tegra124/spi.c
index c13f2c3..ae8a9a7 100644
--- a/src/soc/nvidia/tegra124/spi.c
+++ b/src/soc/nvidia/tegra124/spi.c
@@ -710,18 +710,20 @@ static int xfer_finish(struct tegra_spi_channel *spi)
return ret;
}
+unsigned int spi_crop_chunk(unsigned int cmd_len, unsigned int buf_len)
+{
+ return buf_len;
+}
+
int spi_xfer(struct spi_slave *slave, const void *dout,
- unsigned int bitsout, void *din, unsigned int bitsin)
+ unsigned int out_bytes, void *din, unsigned int in_bytes)
{
- unsigned int out_bytes = bitsout / 8, in_bytes = bitsin / 8;
struct tegra_spi_channel *spi = to_tegra_spi(slave->bus);
u8 *out_buf = (u8 *)dout;
u8 *in_buf = (u8 *)din;
unsigned int todo;
int ret = 0, frame_started = 1;
- ASSERT(bitsout % 8 == 0 && bitsin % 8 == 0);
-
/* tegra bus numbers start at 1 */
ASSERT(slave->bus >= 1 && slave->bus <= ARRAY_SIZE(tegra_spi_channels));
@@ -842,19 +844,37 @@ static int tegra_spi_cbfs_close(struct cbfs_media *media)
return 0;
}
-#define JEDEC_READ 0x03
-#define JEDEC_READ_OUTSIZE 0x04
-/* JEDEC_READ_INSIZE : any length */
+#define JEDEC_READ 0x03
+#define JEDEC_READ_OUTSIZE 0x04
+#define JEDEC_FAST_READ_DUAL 0x3b
+#define JEDEC_FAST_READ_DUAL_OUTSIZE 0x05
static size_t tegra_spi_cbfs_read(struct cbfs_media *media, void *dest,
size_t offset, size_t count)
{
struct tegra_spi_media *spi = (struct tegra_spi_media *)media->context;
- u8 spi_read_cmd[JEDEC_READ_OUTSIZE];
+ u8 spi_read_cmd[JEDEC_FAST_READ_DUAL_OUTSIZE];
+ unsigned int read_cmd_bytes;
int ret = count;
+ struct tegra_spi_channel *channel;
+
+ channel = to_tegra_spi(spi->slave->bus);
- /* TODO: Dual mode (BOTH_EN_BIT) and packed mode */
- spi_read_cmd[0] = JEDEC_READ;
+ if (channel->dual_mode) {
+ /*
+ * Command 0x3b will interleave data only, command 0xbb will
+ * interleave the address as well. It's nice to see the address
+ * plainly when debugging, and we're mostly concerned with
+ * large transfers so the optimization of using 0xbb isn't
+ * really worthwhile.
+ */
+ spi_read_cmd[0] = JEDEC_FAST_READ_DUAL;
+ spi_read_cmd[4] = 0x00; /* dummy byte */
+ read_cmd_bytes = JEDEC_FAST_READ_DUAL_OUTSIZE;
+ } else {
+ spi_read_cmd[0] = JEDEC_READ;
+ read_cmd_bytes = JEDEC_READ_OUTSIZE;
+ }
spi_read_cmd[1] = (offset >> 16) & 0xff;
spi_read_cmd[2] = (offset >> 8) & 0xff;
spi_read_cmd[3] = offset & 0xff;
@@ -863,18 +883,23 @@ static size_t tegra_spi_cbfs_read(struct cbfs_media *media, void *dest,
spi_cs_activate(spi->slave);
if (spi_xfer(spi->slave, spi_read_cmd,
- sizeof(spi_read_cmd) * 8, NULL, 0) < 0) {
+ read_cmd_bytes, NULL, 0) < 0) {
ret = -1;
printk(BIOS_ERR, "%s: Failed to transfer %u bytes\n",
__func__, sizeof(spi_read_cmd));
goto tegra_spi_cbfs_read_exit;
}
- if (spi_xfer(spi->slave, NULL, 0, dest, count * 8)) {
+ if (channel->dual_mode) {
+ setbits_le32(&channel->regs->command1, SPI_CMD1_BOTH_EN_BIT);
+ }
+ if (spi_xfer(spi->slave, NULL, 0, dest, count)) {
ret = -1;
printk(BIOS_ERR, "%s: Failed to transfer %u bytes\n",
__func__, count);
}
+ if (channel->dual_mode)
+ clrbits_le32(&channel->regs->command1, SPI_CMD1_BOTH_EN_BIT);
tegra_spi_cbfs_read_exit:
/* de-assert /CS */
@@ -924,6 +949,10 @@ int initialize_tegra_spi_cbfs_media(struct cbfs_media *media,
media->map = tegra_spi_cbfs_map;
media->unmap = tegra_spi_cbfs_unmap;
+#if CONFIG_SPI_FLASH_FAST_READ_DUAL_OUTPUT_3B == 1
+ channel->dual_mode = 1;
+#endif
+
return 0;
}
diff --git a/src/soc/nvidia/tegra124/spi.h b/src/soc/nvidia/tegra124/spi.h
index b722240..fa238f2 100644
--- a/src/soc/nvidia/tegra124/spi.h
+++ b/src/soc/nvidia/tegra124/spi.h
@@ -56,6 +56,7 @@ struct tegra_spi_channel {
/* stuff that is specific to the attached device */
int rx_frame_header_enable;
u8 frame_header;
+ int dual_mode; /* for x2 transfers with bit interleaving */
/* context (used internally) */
u8 *in_buf, *out_buf;
OpenPOWER on IntegriCloud