[Question] m25p80 driver versus spi clock rate

On Tuesday 23 June 2009, Steven A. Falco wrote:
> David Brownell wrote:
> > I just looked at that code, and didn't see any obvious issue.
> > It relies on initial setup to be correct, and then restores it
> > after any per-transfer override.

Summary of the key point below:  it does rely on that, but when
used with this ppc4xx driver that setup isn't getting done.

 
> I have an m25p16 spi flash and an Atmel AVR on the spi bus.  These are
> specified in a device tree (DTS file) - the m25p16 has a max frequency
> of 50 MHz, and the AVR has a max frequency of 250 KHz.  When the kernel
> boots, the m25p16 is registered to the m25p80 driver, and the AVR is
> registered to the spidev (userland) driver.
> 
> As the devices are added, I see spi_ppc4xx_setup called for each one.
> The first device discovered happens to be the m25p16, and so the bus
> speed is first set to 50 MHz.  Next, the AVR happens to be discovered
> and the bus is set to 250 KHz.
> 
> The max_speed_hz is set correctly for each device.  But, since the AVR
> is listed second in the device tree, the spi bus winds up at the slower
> clock rate of 250 KHz.  At this point, no actual transfers have been
> done - we are just talking about initialization.

Note the conceptual problem goof there too:  until a transfer to
some device is active, talking about "bus speed" is meaningless.


> If I now access the m25p16 device, we go through the calling hierarchy
> from my previous email.  As I said, speed_hz is 0, because m25p80
> doesn't set it, max_speed_hz is 50 MHz as set by spi_ppc4xx_setup,
> and the bus is still running at 250 KHz as described above.
> 
> bitbang_work never calls spi_ppc4xx_setupxfer, because speed_hz is 0.
> So the transfer that should have happened at 50 MHz happens at 250 KHz.
> 
> > 
> > Maybe the problem is that the OF-to-SPI linkage is still borked.
> > Is it ensuring spi_setup() was called at device setup time?
> > 
> 
> The linkage appears correct - max_speed_hz is set correctly for each
> device.  The problem is that bitbang_work won't call spi_ppc4xx_setupxfer
> unless speed_hz is non-zero, and m25p80 has no way to alter speed_hz.

Or alternatively:  that bitbang_work is missing an initial
call to setup_xfer before the loop *starts* its work...

I think the issue is that few other users have used this
code with multiple devices, which had such mismatches in
device speed that they would have noticed this bug.

See if the below patch resolves this issue.

- Dave


---
 drivers/spi/spi_bitbang.c |   22 +++++++++++++---------
 1 file changed, 13 insertions(+), 9 deletions(-)

Sorry to cross-post this to linuxppc-dev@ozlabs.org in the middle
of the story.  I started this in linux-mtd@lists.infradead.org, but
there are OF issues here, and I'd like the PPC folks to be aware of
the issues.

David Brownell wrote:
> On Tuesday 23 June 2009, Steven A. Falco wrote:
>> David Brownell wrote:
>> The linkage appears correct - max_speed_hz is set correctly for each
>> device.  The problem is that bitbang_work won't call spi_ppc4xx_setupxfer
>> unless speed_hz is non-zero, and m25p80 has no way to alter speed_hz.
> 
> Or alternatively:  that bitbang_work is missing an initial
> call to setup_xfer before the loop *starts* its work...
> 
> I think the issue is that few other users have used this
> code with multiple devices, which had such mismatches in
> device speed that they would have noticed this bug.
> 
> See if the below patch resolves this issue.
> 

Fascinating.  I now get a fatal error:

m25p80 spi0.0: invalid bits-per-word (0)

This message comes from spi_ppc4xx_setupxfer.  I believe your patch
is doing what you intended (i.e. forcing an initial call to
spi_ppc4xx_setupxfer), but it exposes an OF / SPI linkage problem.

Namely, of_register_spi_devices does not support a bits-per-word
property, so bits-per-word is zero.

Since we had never called spi_ppc4xx_setupxfer for the m25p80, we
never saw this until now...

Here is part of spi_ppc4xx_setupxfer:

	/*
	 * Allow platform reduce the interrupt load on the CPU during SPI
	 * transfers. We do not target maximum performance, but rather allow
	 * platform to limit SPI bus frequency and interrupt rate.
	 */
	bpw = t ? t->bits_per_word : spi->bits_per_word;
	cs->speed_hz = t ? min(t->speed_hz, spi->max_speed_hz) :
		spi->max_speed_hz;

	if (bpw != 8) {
		dev_err(&spi->dev, "invalid bits-per-word (%d)\n", bpw);
		return -EINVAL;
	}

	if (cs->speed_hz == 0) {
		dev_err(&spi->dev, "invalid speed_hz (must be non-zero)\n");
		return -EINVAL;
	}

Actually, the problem is not entirely with of_register_spi_devices.
bitbang_work will call spi_ppc4xx_setupxfer with a non-null 
spi_transfer.  So, the above code will always set bpw based on
t->bits_per_word.  If t->bits_per_word is zero, it wouldn't even matter
if of_register_spi_devices set spi->bits_per_word, because the
transfer would override it.

How about:

	bpw = t && t->bits_per_word ? t->bits_per_word : spi->bits_per_word;

Now, t->bits_per_word would have to be non-zero in order to override
spi->bits_per_word.

We would still need a patch to of_register_spi_devices to allow (require)
a bits-per-word property, along with device tree patches to add the
property.  But that should take care of it.

I'm adding the ppc list as a CC, since this is turning into an OF
discussion.

	Steve

> - Dave
> 
> 
> ---
>  drivers/spi/spi_bitbang.c |   22 +++++++++++++---------
>  1 file changed, 13 insertions(+), 9 deletions(-)
> 
> --- a/drivers/spi/spi_bitbang.c
> +++ b/drivers/spi/spi_bitbang.c
> @@ -258,6 +258,11 @@ static void bitbang_work(struct work_str
>  	struct spi_bitbang	*bitbang =
>  		container_of(work, struct spi_bitbang, work);
>  	unsigned long		flags;
> +	int			do_setup = -1;
> +	int			(*setup_transfer)(struct spi_device *,
> +					struct spi_transfer *);
> +
> +	setup_transfer = bitbang->setup_transfer;
>  
>  	spin_lock_irqsave(&bitbang->lock, flags);
>  	bitbang->busy = 1;
> @@ -269,8 +274,6 @@ static void bitbang_work(struct work_str
>  		unsigned		tmp;
>  		unsigned		cs_change;
>  		int			status;
> -		int			(*setup_transfer)(struct spi_device *,
> -						struct spi_transfer *);
>  
>  		m = container_of(bitbang->queue.next, struct spi_message,
>  				queue);
> @@ -286,19 +289,19 @@ static void bitbang_work(struct work_str
>  		tmp = 0;
>  		cs_change = (spi != bitbang->exclusive);
>  		status = 0;
> -		setup_transfer = NULL;
>  
>  		list_for_each_entry (t, &m->transfers, transfer_list) {
>  
> -			/* override or restore speed and wordsize */
> -			if (t->speed_hz || t->bits_per_word) {
> -				setup_transfer = bitbang->setup_transfer;
> +			/* override speed or wordsize? */
> +			if (t->speed_hz || t->bits_per_word)
> +				do_setup = 1;
> +
> +			/* init or override transfer params */
> +			if (do_setup != 0) {
>  				if (!setup_transfer) {
>  					status = -ENOPROTOOPT;
>  					break;
>  				}
> -			}
> -			if (setup_transfer) {
>  				status = setup_transfer(spi, t);
>  				if (status < 0)
>  					break;
> @@ -362,8 +365,9 @@ static void bitbang_work(struct work_str
>  		m->status = status;
>  
>  		/* restore speed and wordsize */
> -		if (setup_transfer)
> +		if (do_setup == 1)
>  			setup_transfer(spi, NULL);
> +		do_setup = 0;
>  
>  		/* normally deactivate chipselect ... unless no error and
>  		 * cs_change has hinted that the next message will probably