spi: wpcm-fiu: Add driver for Nuvoton WPCM450 Flash Interface Unit (FIU)
authorJonathan Neuschäfer <j.neuschaefer@gmx.net>
Thu, 24 Nov 2022 19:13:59 +0000 (20:13 +0100)
committerMark Brown <broonie@kernel.org>
Fri, 25 Nov 2022 13:30:15 +0000 (13:30 +0000)
The Flash Interface Unit (FIU) is the SPI flash controller in the
Nuvoton WPCM450 BMC SoC. It supports four chip selects, and direct
(memory-mapped) access to 16 MiB per chip. Larger flash chips can be
accessed by software-defined SPI transfers.

The FIU in newer NPCM7xx SoCs is not compatible with the WPCM450 FIU.

Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Link: https://lore.kernel.org/r/20221124191400.287918-3-j.neuschaefer@gmx.net
Signed-off-by: Mark Brown <broonie@kernel.org>
drivers/spi/Kconfig
drivers/spi/Makefile
drivers/spi/spi-wpcm-fiu.c [new file with mode: 0644]

index c6fb8cd6d2aebf62cbfd50a3826e51c4d4b723f5..b5412b1af81134c05802a87a39d96641c3131beb 100644 (file)
@@ -635,6 +635,17 @@ config SPI_MTK_SNFI
          is implemented as a SPI-MEM controller with pipelined ECC
          capcability.
 
+config SPI_WPCM_FIU
+       tristate "Nuvoton WPCM450 Flash Interface Unit"
+       depends on ARCH_NPCM || COMPILE_TEST
+       select REGMAP
+       help
+         This enables support got the Flash Interface Unit SPI controller
+         present in the Nuvoton WPCM450 SoC.
+
+         This driver does not support generic SPI. The implementation only
+         supports the spi-mem interface.
+
 config SPI_NPCM_FIU
        tristate "Nuvoton NPCM FLASH Interface Unit"
        depends on ARCH_NPCM || COMPILE_TEST
index 971bd62e44816b13ff09e4b953533bc860b85663..be9ba40ef8d03d2b16dbaccd03946a2ac91f7db6 100644 (file)
@@ -83,6 +83,7 @@ obj-$(CONFIG_SPI_MTK_NOR)             += spi-mtk-nor.o
 obj-$(CONFIG_SPI_MTK_SNFI)             += spi-mtk-snfi.o
 obj-$(CONFIG_SPI_MXIC)                 += spi-mxic.o
 obj-$(CONFIG_SPI_MXS)                  += spi-mxs.o
+obj-$(CONFIG_SPI_WPCM_FIU)             += spi-wpcm-fiu.o
 obj-$(CONFIG_SPI_NPCM_FIU)             += spi-npcm-fiu.o
 obj-$(CONFIG_SPI_NPCM_PSPI)            += spi-npcm-pspi.o
 obj-$(CONFIG_SPI_NXP_FLEXSPI)          += spi-nxp-fspi.o
diff --git a/drivers/spi/spi-wpcm-fiu.c b/drivers/spi/spi-wpcm-fiu.c
new file mode 100644 (file)
index 0000000..e525fe0
--- /dev/null
@@ -0,0 +1,444 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (C) 2022 Jonathan Neuschäfer
+
+#include <linux/clk.h>
+#include <linux/mfd/syscon.h>
+#include <linux/module.h>
+#include <linux/of_address.h>
+#include <linux/of_device.h>
+#include <linux/platform_device.h>
+#include <linux/regmap.h>
+#include <linux/spi/spi-mem.h>
+
+#define FIU_CFG                0x00
+#define FIU_BURST_BFG  0x01
+#define FIU_RESP_CFG   0x02
+#define FIU_CFBB_PROT  0x03
+#define FIU_FWIN1_LOW  0x04
+#define FIU_FWIN1_HIGH 0x06
+#define FIU_FWIN2_LOW  0x08
+#define FIU_FWIN2_HIGH 0x0a
+#define FIU_FWIN3_LOW  0x0c
+#define FIU_FWIN3_HIGH 0x0e
+#define FIU_PROT_LOCK  0x10
+#define FIU_PROT_CLEAR 0x11
+#define FIU_SPI_FL_CFG 0x14
+#define FIU_UMA_CODE   0x16
+#define FIU_UMA_AB0    0x17
+#define FIU_UMA_AB1    0x18
+#define FIU_UMA_AB2    0x19
+#define FIU_UMA_DB0    0x1a
+#define FIU_UMA_DB1    0x1b
+#define FIU_UMA_DB2    0x1c
+#define FIU_UMA_DB3    0x1d
+#define FIU_UMA_CTS    0x1e
+#define FIU_UMA_ECTS   0x1f
+
+#define FIU_BURST_CFG_R16      3
+
+#define FIU_UMA_CTS_D_SIZE(x)  (x)
+#define FIU_UMA_CTS_A_SIZE     BIT(3)
+#define FIU_UMA_CTS_WR         BIT(4)
+#define FIU_UMA_CTS_CS(x)      ((x) << 5)
+#define FIU_UMA_CTS_EXEC_DONE  BIT(7)
+
+#define SHM_FLASH_SIZE 0x02
+#define SHM_FLASH_SIZE_STALL_HOST BIT(6)
+
+/*
+ * I observed a typical wait time of 16 iterations for a UMA transfer to
+ * finish, so this should be a safe limit.
+ */
+#define UMA_WAIT_ITERATIONS 100
+
+struct wpcm_fiu_spi {
+       struct device *dev;
+       struct clk *clk;
+       void __iomem *regs;
+       struct regmap *shm_regmap;
+};
+
+static void wpcm_fiu_set_opcode(struct wpcm_fiu_spi *fiu, u8 opcode)
+{
+       writeb(opcode, fiu->regs + FIU_UMA_CODE);
+}
+
+static void wpcm_fiu_set_addr(struct wpcm_fiu_spi *fiu, u32 addr)
+{
+       writeb((addr >>  0) & 0xff, fiu->regs + FIU_UMA_AB0);
+       writeb((addr >>  8) & 0xff, fiu->regs + FIU_UMA_AB1);
+       writeb((addr >> 16) & 0xff, fiu->regs + FIU_UMA_AB2);
+}
+
+static void wpcm_fiu_set_data(struct wpcm_fiu_spi *fiu, const u8 *data, unsigned int nbytes)
+{
+       int i;
+
+       for (i = 0; i < nbytes; i++)
+               writeb(data[i], fiu->regs + FIU_UMA_DB0 + i);
+}
+
+static void wpcm_fiu_get_data(struct wpcm_fiu_spi *fiu, u8 *data, unsigned int nbytes)
+{
+       int i;
+
+       for (i = 0; i < nbytes; i++)
+               data[i] = readb(fiu->regs + FIU_UMA_DB0 + i);
+}
+
+/*
+ * Perform a UMA (User Mode Access) operation, i.e. a software-controlled SPI transfer.
+ */
+static int wpcm_fiu_do_uma(struct wpcm_fiu_spi *fiu, unsigned int cs,
+                          bool use_addr, bool write, int data_bytes)
+{
+       int i = 0;
+       u8 cts = FIU_UMA_CTS_EXEC_DONE | FIU_UMA_CTS_CS(cs);
+
+       if (use_addr)
+               cts |= FIU_UMA_CTS_A_SIZE;
+       if (write)
+               cts |= FIU_UMA_CTS_WR;
+       cts |= FIU_UMA_CTS_D_SIZE(data_bytes);
+
+       writeb(cts, fiu->regs + FIU_UMA_CTS);
+
+       for (i = 0; i < UMA_WAIT_ITERATIONS; i++)
+               if (!(readb(fiu->regs + FIU_UMA_CTS) & FIU_UMA_CTS_EXEC_DONE))
+                       return 0;
+
+       dev_info(fiu->dev, "UMA transfer has not finished in %d iterations\n", UMA_WAIT_ITERATIONS);
+       return -EIO;
+}
+
+static void wpcm_fiu_ects_assert(struct wpcm_fiu_spi *fiu, unsigned int cs)
+{
+       u8 ects = readb(fiu->regs + FIU_UMA_ECTS);
+
+       ects &= ~BIT(cs);
+       writeb(ects, fiu->regs + FIU_UMA_ECTS);
+}
+
+static void wpcm_fiu_ects_deassert(struct wpcm_fiu_spi *fiu, unsigned int cs)
+{
+       u8 ects = readb(fiu->regs + FIU_UMA_ECTS);
+
+       ects |= BIT(cs);
+       writeb(ects, fiu->regs + FIU_UMA_ECTS);
+}
+
+struct wpcm_fiu_op_shape {
+       bool (*match)(const struct spi_mem_op *op);
+       int (*exec)(struct spi_mem *mem, const struct spi_mem_op *op);
+};
+
+static bool wpcm_fiu_normal_match(const struct spi_mem_op *op)
+{
+       // Opcode 0x0b (FAST READ) is treated differently in hardware
+       if (op->cmd.opcode == 0x0b)
+               return false;
+
+       return (op->addr.nbytes == 0 || op->addr.nbytes == 3) &&
+              op->dummy.nbytes == 0 && op->data.nbytes <= 4;
+}
+
+static int wpcm_fiu_normal_exec(struct spi_mem *mem, const struct spi_mem_op *op)
+{
+       struct wpcm_fiu_spi *fiu = spi_controller_get_devdata(mem->spi->controller);
+       int ret;
+
+       wpcm_fiu_set_opcode(fiu, op->cmd.opcode);
+       wpcm_fiu_set_addr(fiu, op->addr.val);
+       if (op->data.dir == SPI_MEM_DATA_OUT)
+               wpcm_fiu_set_data(fiu, op->data.buf.out, op->data.nbytes);
+
+       ret = wpcm_fiu_do_uma(fiu, mem->spi->chip_select, op->addr.nbytes == 3,
+                             op->data.dir == SPI_MEM_DATA_OUT, op->data.nbytes);
+
+       if (op->data.dir == SPI_MEM_DATA_IN)
+               wpcm_fiu_get_data(fiu, op->data.buf.in, op->data.nbytes);
+
+       return ret;
+}
+
+static bool wpcm_fiu_fast_read_match(const struct spi_mem_op *op)
+{
+       return op->cmd.opcode == 0x0b && op->addr.nbytes == 3 &&
+              op->dummy.nbytes == 1 &&
+              op->data.nbytes >= 1 && op->data.nbytes <= 4 &&
+              op->data.dir == SPI_MEM_DATA_IN;
+}
+
+static int wpcm_fiu_fast_read_exec(struct spi_mem *mem, const struct spi_mem_op *op)
+{
+       return -EINVAL;
+}
+
+/*
+ * 4-byte addressing.
+ *
+ * Flash view:  [ C  A  A  A   A     D  D  D  D]
+ * bytes:        13 aa bb cc  dd -> 5a a5 f0 0f
+ * FIU's view:  [ C  A  A  A][ C     D  D  D  D]
+ * FIU mode:    [ read/write][      read       ]
+ */
+static bool wpcm_fiu_4ba_match(const struct spi_mem_op *op)
+{
+       return op->addr.nbytes == 4 && op->dummy.nbytes == 0 && op->data.nbytes <= 4;
+}
+
+static int wpcm_fiu_4ba_exec(struct spi_mem *mem, const struct spi_mem_op *op)
+{
+       struct wpcm_fiu_spi *fiu = spi_controller_get_devdata(mem->spi->controller);
+       int cs = mem->spi->chip_select;
+
+       wpcm_fiu_ects_assert(fiu, cs);
+
+       wpcm_fiu_set_opcode(fiu, op->cmd.opcode);
+       wpcm_fiu_set_addr(fiu, op->addr.val >> 8);
+       wpcm_fiu_do_uma(fiu, cs, true, false, 0);
+
+       wpcm_fiu_set_opcode(fiu, op->addr.val & 0xff);
+       wpcm_fiu_set_addr(fiu, 0);
+       if (op->data.dir == SPI_MEM_DATA_OUT)
+               wpcm_fiu_set_data(fiu, op->data.buf.out, op->data.nbytes);
+       wpcm_fiu_do_uma(fiu, cs, false, op->data.dir == SPI_MEM_DATA_OUT, op->data.nbytes);
+
+       wpcm_fiu_ects_deassert(fiu, cs);
+
+       if (op->data.dir == SPI_MEM_DATA_IN)
+               wpcm_fiu_get_data(fiu, op->data.buf.in, op->data.nbytes);
+
+       return 0;
+}
+
+/*
+ * RDID (Read Identification) needs special handling because Linux expects to
+ * be able to read 6 ID bytes and FIU can only read up to 4 at once.
+ *
+ * We're lucky in this case, because executing the RDID instruction twice will
+ * result in the same result.
+ *
+ * What we do is as follows (C: write command/opcode byte, D: read data byte,
+ * A: write address byte):
+ *
+ *  1. C D D D
+ *  2. C A A A D D D
+ */
+static bool wpcm_fiu_rdid_match(const struct spi_mem_op *op)
+{
+       return op->cmd.opcode == 0x9f && op->addr.nbytes == 0 &&
+              op->dummy.nbytes == 0 && op->data.nbytes == 6 &&
+              op->data.dir == SPI_MEM_DATA_IN;
+}
+
+static int wpcm_fiu_rdid_exec(struct spi_mem *mem, const struct spi_mem_op *op)
+{
+       struct wpcm_fiu_spi *fiu = spi_controller_get_devdata(mem->spi->controller);
+       int cs = mem->spi->chip_select;
+
+       /* First transfer */
+       wpcm_fiu_set_opcode(fiu, op->cmd.opcode);
+       wpcm_fiu_set_addr(fiu, 0);
+       wpcm_fiu_do_uma(fiu, cs, false, false, 3);
+       wpcm_fiu_get_data(fiu, op->data.buf.in, 3);
+
+       /* Second transfer */
+       wpcm_fiu_set_opcode(fiu, op->cmd.opcode);
+       wpcm_fiu_set_addr(fiu, 0);
+       wpcm_fiu_do_uma(fiu, cs, true, false, 3);
+       wpcm_fiu_get_data(fiu, op->data.buf.in + 3, 3);
+
+       return 0;
+}
+
+/*
+ * With some dummy bytes.
+ *
+ *  C A A A  X*  X D D D D
+ * [C A A A  D*][C D D D D]
+ */
+static bool wpcm_fiu_dummy_match(const struct spi_mem_op *op)
+{
+       // Opcode 0x0b (FAST READ) is treated differently in hardware
+       if (op->cmd.opcode == 0x0b)
+               return false;
+
+       return (op->addr.nbytes == 0 || op->addr.nbytes == 3) &&
+              op->dummy.nbytes >= 1 && op->dummy.nbytes <= 5 &&
+              op->data.nbytes <= 4;
+}
+
+static int wpcm_fiu_dummy_exec(struct spi_mem *mem, const struct spi_mem_op *op)
+{
+       struct wpcm_fiu_spi *fiu = spi_controller_get_devdata(mem->spi->controller);
+       int cs = mem->spi->chip_select;
+
+       wpcm_fiu_ects_assert(fiu, cs);
+
+       /* First transfer */
+       wpcm_fiu_set_opcode(fiu, op->cmd.opcode);
+       wpcm_fiu_set_addr(fiu, op->addr.val);
+       wpcm_fiu_do_uma(fiu, cs, op->addr.nbytes != 0, true, op->dummy.nbytes - 1);
+
+       /* Second transfer */
+       wpcm_fiu_set_opcode(fiu, 0);
+       wpcm_fiu_set_addr(fiu, 0);
+       wpcm_fiu_do_uma(fiu, cs, false, false, op->data.nbytes);
+       wpcm_fiu_get_data(fiu, op->data.buf.in, op->data.nbytes);
+
+       wpcm_fiu_ects_deassert(fiu, cs);
+
+       return 0;
+}
+
+static const struct wpcm_fiu_op_shape wpcm_fiu_op_shapes[] = {
+       { .match = wpcm_fiu_normal_match, .exec = wpcm_fiu_normal_exec },
+       { .match = wpcm_fiu_fast_read_match, .exec = wpcm_fiu_fast_read_exec },
+       { .match = wpcm_fiu_4ba_match, .exec = wpcm_fiu_4ba_exec },
+       { .match = wpcm_fiu_rdid_match, .exec = wpcm_fiu_rdid_exec },
+       { .match = wpcm_fiu_dummy_match, .exec = wpcm_fiu_dummy_exec },
+};
+
+static const struct wpcm_fiu_op_shape *wpcm_fiu_find_op_shape(const struct spi_mem_op *op)
+{
+       size_t i;
+
+       for (i = 0; i < ARRAY_SIZE(wpcm_fiu_op_shapes); i++) {
+               const struct wpcm_fiu_op_shape *shape = &wpcm_fiu_op_shapes[i];
+
+               if (shape->match(op))
+                       return shape;
+       }
+
+       return NULL;
+}
+
+static bool wpcm_fiu_supports_op(struct spi_mem *mem, const struct spi_mem_op *op)
+{
+       if (!spi_mem_default_supports_op(mem, op))
+               return false;
+
+       if (op->cmd.dtr || op->addr.dtr || op->dummy.dtr || op->data.dtr)
+               return false;
+
+       if (op->cmd.buswidth > 1 || op->addr.buswidth > 1 ||
+           op->dummy.buswidth > 1 || op->data.buswidth > 1)
+               return false;
+
+       return wpcm_fiu_find_op_shape(op) != NULL;
+}
+
+/*
+ * In order to ensure the integrity of SPI transfers performed via UMA,
+ * temporarily disable (stall) memory accesses coming from the host CPU.
+ */
+static void wpcm_fiu_stall_host(struct wpcm_fiu_spi *fiu, bool stall)
+{
+       if (fiu->shm_regmap) {
+               int res = regmap_update_bits(fiu->shm_regmap, SHM_FLASH_SIZE,
+                                            SHM_FLASH_SIZE_STALL_HOST,
+                                            stall ? SHM_FLASH_SIZE_STALL_HOST : 0);
+               if (res)
+                       dev_warn(fiu->dev, "Failed to (un)stall host memory accesses: %d\n", res);
+       }
+}
+
+static int wpcm_fiu_exec_op(struct spi_mem *mem, const struct spi_mem_op *op)
+{
+       struct wpcm_fiu_spi *fiu = spi_controller_get_devdata(mem->spi->controller);
+       const struct wpcm_fiu_op_shape *shape = wpcm_fiu_find_op_shape(op);
+
+       wpcm_fiu_stall_host(fiu, true);
+
+       if (shape)
+               return shape->exec(mem, op);
+
+       wpcm_fiu_stall_host(fiu, false);
+
+       return -ENOTSUPP;
+}
+
+static int wpcm_fiu_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
+{
+       if (op->data.nbytes > 4)
+               op->data.nbytes = 4;
+
+       return 0;
+}
+
+static const struct spi_controller_mem_ops wpcm_fiu_mem_ops = {
+       .adjust_op_size = wpcm_fiu_adjust_op_size,
+       .supports_op = wpcm_fiu_supports_op,
+       .exec_op = wpcm_fiu_exec_op,
+};
+
+static void wpcm_fiu_hw_init(struct wpcm_fiu_spi *fiu)
+{
+       /* Deassert all manually asserted chip selects */
+       writeb(0x0f, fiu->regs + FIU_UMA_ECTS);
+}
+
+static int wpcm_fiu_probe(struct platform_device *pdev)
+{
+       struct device *dev = &pdev->dev;
+       struct spi_controller *ctrl;
+       struct wpcm_fiu_spi *fiu;
+       struct resource *res;
+
+       ctrl = devm_spi_alloc_master(dev, sizeof(*fiu));
+       if (!ctrl)
+               return -ENOMEM;
+
+       fiu = spi_controller_get_devdata(ctrl);
+       fiu->dev = dev;
+
+       res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "control");
+       fiu->regs = devm_ioremap_resource(dev, res);
+       if (IS_ERR(fiu->regs)) {
+               dev_err(dev, "Failed to map registers\n");
+               return PTR_ERR(fiu->regs);
+       }
+
+       fiu->clk = devm_clk_get_enabled(dev, NULL);
+       if (IS_ERR(fiu->clk))
+               return PTR_ERR(fiu->clk);
+
+       fiu->shm_regmap = syscon_regmap_lookup_by_phandle_optional(dev->of_node, "nuvoton,shm");
+
+       wpcm_fiu_hw_init(fiu);
+
+       ctrl->bus_num = -1;
+       ctrl->mem_ops = &wpcm_fiu_mem_ops;
+       ctrl->num_chipselect = 4;
+       ctrl->dev.of_node = dev->of_node;
+
+       /*
+        * The FIU doesn't include a clock divider, the clock is entirely
+        * determined by the AHB3 bus clock.
+        */
+       ctrl->min_speed_hz = clk_get_rate(fiu->clk);
+       ctrl->max_speed_hz = clk_get_rate(fiu->clk);
+
+       return devm_spi_register_controller(dev, ctrl);
+}
+
+static const struct of_device_id wpcm_fiu_dt_ids[] = {
+       { .compatible = "nuvoton,wpcm450-fiu", },
+       { }
+};
+MODULE_DEVICE_TABLE(of, wpcm_fiu_dt_ids);
+
+static struct platform_driver wpcm_fiu_driver = {
+       .driver = {
+               .name   = "wpcm450-fiu",
+               .bus    = &platform_bus_type,
+               .of_match_table = wpcm_fiu_dt_ids,
+       },
+       .probe      = wpcm_fiu_probe,
+};
+module_platform_driver(wpcm_fiu_driver);
+
+MODULE_DESCRIPTION("Nuvoton WPCM450 FIU SPI controller driver");
+MODULE_AUTHOR("Jonathan Neuschäfer <j.neuschaefer@gmx.net>");
+MODULE_LICENSE("GPL");