Hi Heinrich,
On Sun, Nov 12, 2017 at 08:42:28PM +0100, Heinrich Schuchardt wrote:
On 11/12/2017 04:30 PM, Baruch Siach wrote:
Calling .set_speed with zero speed is definitely a bug. Instead of crashing, set hz to 1 so that we get the lowest possible frequency rate.
Did this actually occur? Why?
Yes. CONFIG_ARCH_MVEBU selects CONFIG_DM_SPI_FLASH. In cmd/sf.c:do_spi_flash_probe(), speed is set to 0 when CONFIG_DM_SPI_FLASH, passed to spi_flash_probe_bus_cs(), and from there to spi_get_bus_and_cs(). Unfortunately, since the ClearFog SPI flash node has no "spi-flash" compatible string, the spi_flash_std driver is not bound, so spi_child_post_bind() returns early without calling spi_slave_ofdata_to_platdata(). The dm_spi_slave_platdata speed field is left uninitialized, and we end up with speed=0 when calling spi_set_speed_mode().
This should be fixed with http://patchwork.ozlabs.org/patch/837360/ for ClearFog, by adding the "spi-flash" compatible. But "spi-flash" is non standard. Most kernel DT files use "jedec,spi-nor" instead. So you can expect this issue to show up again in the future.
A workaround is to provide speed argument in the 'sf probe' command line.
Signed-off-by: Baruch Siach baruch@tkos.co.il
v2: Don't fail; set lowest rate (Jagan)
drivers/spi/kirkwood_spi.c | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/drivers/spi/kirkwood_spi.c b/drivers/spi/kirkwood_spi.c index 0c6bd295cde9..5744849a5ad9 100644 --- a/drivers/spi/kirkwood_spi.c +++ b/drivers/spi/kirkwood_spi.c @@ -257,6 +257,9 @@ static int mvebu_spi_set_speed(struct udevice *bus, uint hz) struct kwspi_registers *reg = plat->spireg; u32 data;
- /* Avoid divide by zero; instead, set the lowest possible speed */
- if (hz == 0)
hz = 1;Shouldn't we write a debug message here like in other drivers?
debug("%s: can't set 0 Hz (actual will be 1 Hz)\n", __func__);
I'm fine with that. The actual speed is platform specific, though. See below.
Does any bus member support 1 Hz? Wouldn't 125 kHz be a more reasonable value?
The code below would write the highest possible prescaler value, KWSPI_CLKPRESCL_MASK, thus setting the lowest possible per-platform frequency value.
Could a division by zero valued max_hz occur in spi_setup_slave too?
I don't know. The prescaler calculation code is duplicated over there, so it might be a good idea to refactor.
/* calculate spi clock prescaller using max_hz */ data = ((CONFIG_SYS_TCLK / 2) / hz) + 0x10; data = data < KWSPI_CLKPRESCL_MIN ? KWSPI_CLKPRESCL_MIN : data;
Thanks for reviewing, baruch