On Wed, 10 Apr 2019 16:01:02 +0800
masonccyang@xxxxxxxxxxx wrote:
> Hi,
>
> > Re: [PATCH v9 2/3] spi: Add Renesas R-Car Gen3 RPC-IF SPI controller
> driver
> >
> > On 4/10/19 3:33 AM, masonccyang@xxxxxxxxxxx wrote:
> > > Hi Sergei,
> > >
> > >> > +static ssize_t rpc_spi_mem_dirmap_write(struct spi_mem_dirmap_desc
> > > *desc,
> > >> > + u64 offs, size_t len, const void *buf)
> > >> > +{
> > >> > + struct rpc_spi *rpc =
> > >> > + spi_controller_get_devdata(desc->mem->spi->controller);
> > >> > + int ret;
> > >> > +
> > >> > + if (offs + desc->info.offset + len > U32_MAX)
> > >> > + return -EINVAL;
> > >> > +
> > >> > + if (len > RPC_WBUF_SIZE)
> > >> > + len = RPC_WBUF_SIZE;
> > >> > +
> > >> > + ret = rpc_spi_set_freq(rpc, desc->mem->spi->max_speed_hz);
> > >> > + if (ret)
> > >> > + return ret;
> > >> > +
> > >> > + rpc_spi_mem_set_prep_op_cfg(desc->mem->spi,
> > >> > + &desc->info.op_tmpl, &offs, &len);
> > >> > +
> > >> > + regmap_update_bits(rpc->regmap, RPC_CMNCR, RPC_CMNCR_MD,
> > > RPC_CMNCR_MD);
> > >> > +
> > >> > + regmap_write(rpc->regmap, RPC_SMDRENR, 0);
> > >> > +
> > >> > + regmap_update_bits(rpc->regmap, RPC_PHYCNT,
> RPC_PHYCNT_STRTIM(7) |
> > >> > + RPC_PHYCNT_WBUF2 | RPC_PHYCNT_WBUF,
> > >> > + RPC_PHYCNT_WBUF2 | RPC_PHYCNT_WBUF);
> > >> > +
> > >> > + memcpy_toio(rpc->wbuf, buf, len);
> > >>
> > >> Wait, doesn't the manual say that the whole 256-byte buffer should
> be
> > >> filled?
>
> it could be less than 256 bytes, i.e., 128 bytes to rpc->wbuf !
>
> I think that short chunks have to be written w/o WBUF (done,
> > > in fact,
> > >> by the HF driver).
> > >>
> > >
> > > From spi-nor.c layer always transfer 256 bytes data with page program
> > > command.
> >
> > Does that apply even for flashes with not-256-byte pages ?
> >
>
> I think it needs to patch in case of nor->page_size = 512 bytes.
I think the main problem here is that you assume the memory is a
NOR :-). Just do what the spi-mem user asks: if he asks you to write
X bytes, then write no more than X bytes. Use whatever trick you have
to make sure this is always true, and if this requires using a slow
path for non-aligned accesses, then do it, because it's better to have
a slow+working memory than a fast+non-working one :P.
