[PATCH 2/2] mtd: spi-nor: Disable the flash quad mode in spi_nor_restore()

[Matthias Weißer]

Am Do., 3. Sept. 2020 um 05:04 Uhr schrieb Yicong Yang
<yangyicong at hisilicon.com>:
>
> Hi Matthias,
> On 2020/9/2 20:15, Matthias Weißer wrote:
> > Hi Vignesh
> >
> > Am Di., 1. Sept. 2020 um 16:20 Uhr schrieb Yicong Yang
> > <yangyicong at hisilicon.com>:
> >> I've tested the following patch with s25fs128s1.
> >> I left the flash quad enabled before managed by spi-nor driver,
> >> and it'll stay QE after removed. So I think it'll also address the issue
> >> mentioned. Please have a test.
> > Thanks for the patch. I can confirm that current 5.9-rc3 bricks my hardware
> > and your patch on top of it fixes that. Therefore
> >
> > Tested-by: Matthias Weisser <m.weisser.m at gmail.com>
> >
> > But I am still concerned about wearing out nonvolatile QE bits (not in my
> > case, but there may be others) which may also brick hardware after a lot of
> > boots (by set and reset QE bit on every boot).
> >
> > So, from my limited point of view, your patch fixes my problem but overall
> > I think the original approach should be thought-out a bit more.
>
> Thanks for testing the patch.
>
> The driver will try to set the QE bit when 1) in ->quad_enable() if QE bit is not set and
> 2) previously without this patch in spi_nor_restore() to try to disable the flash quad mode.
>
> With this patch, if the QE bit is originally set the driver will only read and record it and
> will not disable the quad mode in spi_nor_restore(). What the driver will do is to read
> the bit, without trying to set or reset it. So I think it will not wear out the those nonvolatile
> QE bits.

I think it will. Expect a flash with the QE bit not set. Kernel boots, set the
bit which will cause an erase cycle (setting it from 0 to 1). On shutdown the
bit is reset (transition from 1 to 0) which will cause a program cycle. So,
every boot -> shutdown cycle causes at least one program-erase cycle which will
wear out that area of the flash.

See also this quote from the datasheet of F25FL512S:
"Non-volatile bits have the same cycling (erase and program) endurance
as the main
flash array."

I still think the reset of the QE (or any other non-volatile) bit on
shutdown should
be prevented and only allowed by explicitly enable it (e.g. per system in DT or
per flash in flash_info struct).

Regards,
Matthias