Hi Hiep,
On Fri, Aug 26, 2016 at 2:58 AM, Hiep Cao Minh <cm-hiep@xxxxxxxxxxx> wrote:
> I'd like to ask you about R-Car Gen2 QSPI driver performance test of single,
> dual and quad.
>
> ------
> commit 4f12b5e529e4ff274eb478ec1c2ef41358ed9577
> Author: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
> Date: Mon Jun 2 15:38:17 2014 +0200
>
> spi: rspi: Add DMA support for QSPI on R-Car Gen2
>
> Enable DMA support for QSPI on R-Car Gen2, for Single, Dual, and Quad
> SPI
> Transfers.
>
> Performance figures for reading from a QSPI FLASH driven at 24.375 MHz
> on r8a7791/koelsch:
> - Single: 1.1 Mbps PIO, 23 Mbps DMA
> - Dual : 12.7 Mbps PIO, 48 Mbps DMA
> - Quad : 13 Mbps PIO, 70 Mbps DMA
> ------
> Could you show me how to test these performances?
As I wanted to measure raw SPI performance, and not overhead (block
device, file system, ...) or other impact (buffer cache!), I used a patch like
the (whitespace-damaged) one at the end of this email, and:
dd if=/dev/mtd2 bs=4M count=1 of=/dev/null
Using a sufficiently large buffer size is recommended for good performance.
All of this was done in r8a7791/koelsch.
> I have tried to reproduce them but I only got the results like this:
>
> Enabled DMA:
> Linux stable v4.7
> Quad-spi： Read 3.2 MB/s Write 601 kB/s
> Dual-spi： Read 3.1 MB/s Write 847 kB/s
> Single-spi： Read 2.5 MB/s Write 850 kB/s
>
> PIO:
> RcarGen2_BSP linux_v3.10.101-ltsi
> Quad-spi： Read 2.2 MB/s Write 1.1 MB/s
> Dual-spi： Read 2.1 MB/s Write 158 kB/s
> Single-spi： Read 173 kB/s Write 159 kB/s
Note that some of these are even faster than mine, probably due to
caching...
diff --git a/drivers/spi/spi-rspi.c b/drivers/spi/spi-rspi.c
index 112f5130ee9579f4..d9d822a3e35cbed2 100644
--- a/drivers/spi/spi-rspi.c
+++ b/drivers/spi/spi-rspi.c
@@ -199,6 +199,8 @@ struct rspi_data {
unsigned dma_callbacked:1;
unsigned byte_access:1;
+
+ u64 msg_start;
};
static void rspi_write8(const struct rspi_data *rspi, u8 data, u16 offset)
@@ -926,6 +972,8 @@ static int rspi_prepare_message(struct spi_master *master,
struct rspi_data *rspi = spi_master_get_devdata(master);
int ret;
+ rspi->msg_start = get_jiffies_64();
+
if (msg->spi->mode &
(SPI_TX_DUAL | SPI_TX_QUAD | SPI_RX_DUAL | SPI_RX_QUAD)) {
/* Setup sequencer for messages with multiple transfer modes */
@@ -943,6 +992,7 @@ static int rspi_unprepare_message(struct spi_master *master,
struct spi_message *msg)
{
struct rspi_data *rspi = spi_master_get_devdata(master);
+ unsigned int elapsed, bps;
/* Disable SPI function */
rspi_write8(rspi, rspi_read8(rspi, RSPI_SPCR) & ~SPCR_SPE, RSPI_SPCR);
@@ -950,6 +1000,13 @@ static int rspi_unprepare_message(struct
spi_master *master,
/* Reset sequencer for Single SPI Transfers */
rspi_write16(rspi, rspi->spcmd, RSPI_SPCMD0);
rspi_write8(rspi, 0, RSPI_SPSCR);
+
+ elapsed = get_jiffies_64() - rspi->msg_start;
+ bps = elapsed ? div_u64(msg->actual_length * 8ULL * HZ + elapsed / 2,
+ elapsed) : 0;
+ printk("RSPI %u bps (%u bytes in %u ticks)\n", bps, msg->actual_length,
+ elapsed);
+
return 0;
}
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
