spi: stm32-ospi: Optimize FIFO accesses using u16 or u32

FIFO accesses uses u8 only for read/write.
In order to optimize throughput, add u16 or u32 read/write
accesses when possible.

Running mtd_speedtest on a 4MB sNOR partition using a
stm32mp257f-ev1 board gives the following results:

           before        after   gain
Read :  5693 KiB/s  21139 KiB/s   371%
Write:   765 KiB/s    910 KiB/s    19%

Signed-off-by: Patrice Chotard <patrice.chotard@foss.st.com>
Link: https://patch.msgid.link/20251208-upstream_qspi_ospi_updates-v2-2-62526c9467dc@foss.st.com
Signed-off-by: Mark Brown <broonie@kernel.org>

diff --git a/drivers/spi/spi-stm32-ospi.c b/drivers/spi/spi-stm32-ospi.c
index d733e37f0435f867b60c015cafc1e3b508f0171a..a6f53f06200e7dd39c6da4c549e5ddc39f36143c 100644 (file)
--- a/drivers/spi/spi-stm32-ospi.c
+++ b/drivers/spi/spi-stm32-ospi.c
@@ -142,14 +142,32 @@ struct stm32_ospi {
        struct mutex lock;
 };
 
-static void stm32_ospi_read_fifo(u8 *val, void __iomem *addr)
+static void stm32_ospi_read_fifo(void *val, void __iomem *addr, u8 len)
 {
-       *val = readb_relaxed(addr);
+       switch (len) {
+       case sizeof(u32):
+               *((u32 *)val) = readl_relaxed(addr);
+               break;
+       case sizeof(u16):
+               *((u16 *)val) = readw_relaxed(addr);
+               break;
+       case sizeof(u8):
+               *((u8 *)val) = readb_relaxed(addr);
+       };
 }
 
-static void stm32_ospi_write_fifo(u8 *val, void __iomem *addr)
+static void stm32_ospi_write_fifo(void *val, void __iomem *addr, u8 len)
 {
-       writeb_relaxed(*val, addr);
+       switch (len) {
+       case sizeof(u32):
+               writel_relaxed(*((u32 *)val), addr);
+               break;
+       case sizeof(u16):
+               writew_relaxed(*((u16 *)val), addr);
+               break;
+       case sizeof(u8):
+               writeb_relaxed(*((u8 *)val), addr);
+       };
 }
 
 static int stm32_ospi_abort(struct stm32_ospi *ospi)
@@ -172,19 +190,20 @@ static int stm32_ospi_abort(struct stm32_ospi *ospi)
        return timeout;
 }
 
-static int stm32_ospi_poll(struct stm32_ospi *ospi, u8 *buf, u32 len, bool read)
+static int stm32_ospi_poll(struct stm32_ospi *ospi, void *buf, u32 len, bool read)
 {
        void __iomem *regs_base = ospi->regs_base;
-       void (*fifo)(u8 *val, void __iomem *addr);
+       void (*fifo)(void *val, void __iomem *addr, u8 len);
        u32 sr;
        int ret;
+       u8 step;
 
        if (read)
                fifo = stm32_ospi_read_fifo;
        else
                fifo = stm32_ospi_write_fifo;
 
-       while (len--) {
+       while (len) {
                ret = readl_relaxed_poll_timeout_atomic(regs_base + OSPI_SR,
                                                        sr, sr & SR_FTF, 1,
                                                        STM32_FIFO_TIMEOUT_US);
@@ -193,7 +212,17 @@ static int stm32_ospi_poll(struct stm32_ospi *ospi, u8 *buf, u32 len, bool read)
                                len, sr);
                        return ret;
                }
-               fifo(buf++, regs_base + OSPI_DR);
+
+               if (len >= sizeof(u32))
+                       step = sizeof(u32);
+               else if (len >= sizeof(u16))
+                       step = sizeof(u16);
+               else
+                       step = sizeof(u8);
+
+               fifo(buf, regs_base + OSPI_DR, step);
+               len -= step;
+               buf += step;
        }
 
        return 0;
@@ -408,7 +437,7 @@ static int stm32_ospi_xfer(struct stm32_ospi *ospi, const struct spi_mem_op *op)
        if (op->data.dir == SPI_MEM_DATA_IN)
                buf = op->data.buf.in;
        else
-               buf = (u8 *)op->data.buf.out;
+               buf = (void *)op->data.buf.out;
 
        return stm32_ospi_poll(ospi, buf, op->data.nbytes,
                               op->data.dir == SPI_MEM_DATA_IN);