[PATCH 1/2] drivers/mtd/devices/m25p80.c: Fix return code of read/write mtd callbacks

[Kevin Cernekee]

On Mon, Sep 3, 2012 at 8:58 AM, Cédric Cano <ccano at interfaceconcept.com> wrote:
> For SPI Flash devices, MTD read and write functions returns "1" if Flash is
> busy (this status is returned by "wait_till_ready" fnuction). This return

"function"

It is usually a good idea to wrap the commit text around 72-74
characters so that it is readable in "git log" on an 80-column
terminal.

> code is not an error code: if device is busy, MTD read or write are
> successful with 1 byte length.

Looking at mtd_read() - typically the transfer length is passed back
through the "size_t *retlen" pointer, not the function's return value.

In Linux 3.5, if m25p80_read() returns 1, mtd_read() will interpret it
as a bitflip count not a byte count.  That isn't the correct behavior
for a NOR device, but it probably wouldn't cause the symptom you are
reporting.

In Linux 3.4 and older, some functions like mtdchar_read() could
interpret the '1' as an error code and pass it back as-is to the
caller (who is expecting either a negative error code or a positive
length).  Is this the case you are running into?

Side note - it would be helpful to add kerneldoc markup to
include/linux/mtd/mtd.h for struct mtd_info and the new mtd_* wrapper
functions, so we have a clear definition of what the parameters and
return values should look like.

> --- linux-3.5.3/drivers/mtd/devices/m25p80.c    2012-08-26
> 04:32:13.000000000 +0200
> +++ linux-3.5.3/drivers/mtd/devices/m25p80.c    2012-09-03
> 17:35:16.159741656 +0200
> @@ -199,7 +199,7 @@

Could you please use "git format-patch" for your submissions, and base
them on the current head of tree?

Thanks.