Re: [flashrom] [PATCH] linux_spi.c: set SPI mode, word justification and bits per word on init.

[Carl-Daniel Hailfinger]

Am 02.03.2012 23:51 schrieb Stefan Tauner:
> On Fri, 02 Mar 2012 22:43:44 +0100
> Carl-Daniel Hailfinger  wrote:
> > Am 02.03.2012 00:43 schrieb Stefan Tauner:
> > > Previously we relied on a correctly set up state.
> > >
> > > untested.
> > > Signed-off-by: Stefan Tauner
> > >
> > > linux_spi.c |   23 +++++++++++++++++++++++
> > >  1 files changed, 23 insertions(+), 0 deletions(-)
> > >
> > > diff --git a/linux_spi.c b/linux_spi.c
> > > index d994389..d29c59a 100644
> > > --- a/linux_spi.c
> > > +++ b/linux_spi.c
> > > @@ -57,6 +57,8 @@ int linux_spi_init(void)
> > >  {
> > >    char *p, *endp, *dev;
> > >    uint32_t speed = 0;
> > > /* FIXME: make the following configurable by CLI options. */
> > > uint8_t mode = SPI_MODE_0, lsb = 0, bits = 0; /* mode 0, msb first, 8 bits */
> >
> > Can you move that comment above the variable definitions?
>
> done, reads now:
> /* SPI mode 0, msb first, 8 bits (i.e. value=0) */
> > Where should we note that SPI_MODE_0 also implies CS# active low?
>
> it does not. the test program seems to be outdated, the actual code
> masks the CPOL/CPHA bits.

Let me rephrase that: Setting the SPI mode handles CPOL/CPHA/CS/LSB/...
all in one. The bits which are not set have their default value (0).
http://tomoyo.sourceforge.jp/cgi-bin/lxr/source/drivers/spi/spidev.c#L72
#define SPI_MODE_MASK (SPI_CPHA | SPI_CPOL | SPI_CS_HIGH | SPI_LSB_FIRST
| SPI_3WIRE | SPI_LOOP | SPI_NO_CS | SPI_READY)
> > > dev = extract_programmer_param("dev");
> > >    if (!dev || !strlen(dev)) {
> > > @@ -92,6 +94,27 @@ int linux_spi_init(void)
> > >    	msg_pdbg("Using %d kHz clock\n", speed);
> > >    }
> > > if (ioctl(fd, SPI_IOC_WR_MODE, &mode) == -1) {
> > > msg_perr("%s: failed to set SPI mode to %u: %s\n",
> > > 	 __func__, mode, strerror(errno));
> > > close(fd);
> > > return 1;
> > > }
> > > if (ioctl(fd, SPI_IOC_WR_LSB_FIRST, &lsb) == -1) {
> > > msg_perr("%s: failed to set SPI justification to %u: %s\n",
> > > 	 __func__, lsb, strerror(errno));
> >
> > This message would benefit from an explanation what SPI justification
> > is. Suggestion:
> > msg_perr("%s: failed to set SPI bit order to %s first: %s\n", __func_,
> > lsb ? "LSB" : "MSB", strerror(errno));
>
> right, was too lazy to think about a better term/solution at the
> time; fixed.
> > > close(fd);
> > > return 1;
> > > }
> > > if (ioctl(fd, SPI_IOC_WR_BITS_PER_WORD, &bits) == -1) {
> > > msg_perr("%s: failed to set the number of bits in an SPI word to %u: %s\n",
> > > 	 __func__, bits, strerror(errno));
> >
> > bits is 0. The error message would suggest that we tried to set the
> > number of bits to 0. Does 0 also mean 8 bits, or would we have to set 8
> > bits with bits=8?
>
> bits = 0 is the only defined value in the documentation and is actually
> the only one implemented in the code and means 8 bits per word.

The documentation is explicit that bits is the number of bits, unless
bits=0, in which case it is treated as 8 bits.
I also have checked the code in the Linux kernel and it agrees with me.

Note that retval in ioctl handling is _not_ the value entered by the
user, but an error/success code for accessing user data.
http://tomoyo.sourceforge.jp/cgi-bin/lxr/source/drivers/spi/spidev.c#L408
http://tomoyo.sourceforge.jp/cgi-bin/lxr/source/drivers/spi/spi.c#L727
> i have changed the message to this:
>    	msg_perr("%s: failed to set the number of bits per SPI word to %s: %s\n",
>    		 __func__, bits == 0 ? "8" : "<undef>", strerror(errno));
> > > close(fd);
> > > return 1;
> > > }
> > > if (register_shutdown(linux_spi_shutdown, NULL))
> > >   return 1;
> >
> > As an alternative, we could avoid the whole close(fd) dance by calling
> > register_shutdown() first, and then letting it do the work for us
> > automatically after we return 1.
>
> how do we do it in other programmers? we should probably define and
> document a single suggested way so that we dont have to discuss this
> every time. :)

A single suggested way (or maybe two ways, depending on the hardware
interface design we're using) is a good idea as long as it's not a hard
unbreakable rule afterwards.
The following two ways seem to make sense for linux_spi IMHO:
ret = 1; goto out_shutdown;
or
setting everything after register_shutdown.

If we want to restore old settings on shutdown, it gets a bit hackier.
One possible way would be:
int restore_settings;
init (){
open fd else return 1;
restore_settings=0;
register_shutdown() else return 1;
get all settings and store them in file-level variables or a file-level
struct else return 1;
restore_settings=1;
set all settings else return 1;
return 0;
}
shutdown() {
if (restore_settings){
  restore settings;
}
close(fd);
}
> in this particular case i think it makes sense. in general relying on
> the shutdown function only may be a bit hard to grasp/implement for
> complicated init functions that allocate/manipulate lots of stuff (e.g.
> serprog).

Agreed.

Side note: How should we handle a failed register_shutdown()? Call the
shutdown function manually and return 1? Or is there any other good way
to clean up?

Regards,
Carl-Daniel