Re: [RFC PATCH] mtd: spi-nor: add support to non-uniform SPI NOR flash memories

[Tudor Ambarus]

Hi, Marek,

On 05/23/2018 03:54 PM, Marek Vasut wrote:
> On 05/23/2018 02:52 PM, Tudor Ambarus wrote:
> > Hi, Marek,
>
> Hi,
>
> > On 05/23/2018 12:56 PM, Marek Vasut wrote:
> > [...]
> > > > > [...]
> > > > >
> > > > > > > > +    while (len) {
> > > > > > > > +        cmd = spi_nor_find_best_erase_cmd(map, region, addr, len);
> > > > > > > > +        if (!cmd)
> > > > > > > > +            return -EINVAL;
> > > > > > > What would happen if you realize mid-way that you cannot erase some
> > > > > > > sector , do you end up with partial erase ?
> > > > > > Is this possible? In non-overlaid regions, the address is aligned with
> > > > > > at least one of the erase commands, else -EINVAL. For overlaid regions
> > > > > > alignment doesn't matter. But yes, if this is possible, in this case,
> > > > > > this proposal will do a partial erase.
> > > > > Shouldn't we fail up front instead ?
> > > > It will be great if we can do this without having performance penalties.
> > > > Can we loose the conditions for the last erase command? If one wants to
> > > > erase 80k chunk starting from offset 0 and only 32k and 64k erase type
> > > > are supported, can we erase 96k?
> > > No. But can you maybe build a list of erase commands to be executed once
> > > you validate that the erase can be performed for example ?
> >
> > My second choice was an array witch saves u8 opcode and u32 erasesize.
> > There are flashes of 256MB, in the worst case scenario with 4k erase
> > type, we will end up with 64K entries.
>
> Some RLE encoding might help here ?

Nice.

> > How about enforcing the length to be multiple of mtd->erasesize, like we
> > do in uniform_erase? With this, the problem disappears.
>
> What is the erase size for the 4k-sector 256MiB flash ?

S70FS01GS[1] is a 128 MByte flash with non-uniform erase support. It
supports 4k and 256k erase types. I would have to enforce the address
and the length to be multiple of 256k in order to vanish the issue. But
the whole point of non-uniform erase will vanish too, I guess.

I don't have any other good :) idea, so I'll implement your suggestion
with the list of erase commands and RLE encoding.

Thanks,
ta

[1] http://www.cypress.com/file/215911/download