RE: [EXT] Re: [PATCH nvmem 1/1] nvmem: imx: correct the fuse word index

[Andy Duan]

From: Lothar WaÃmann <LW@xxxxxxxxxxxxxxxxxxx> Sent: Friday, July 5, 2019 3:13 PM
> Hi,
> 
> On Fri, 5 Jul 2019 02:46:32 +0000 Andy Duan wrote:
> > From: Andy Duan Sent: Friday, July 5, 2019 12:08 AM
> > > From: Lothar WaÃmann <LW@xxxxxxxxxxxxxxxxxxx> Sent: Thursday, July
> > > 4,
> > > 2019 11:46 PM
> > > > Hi,
> > > >
> > > > On Thu,  4 Jul 2019 22:20:15 +0800 fugang.duan@xxxxxxx wrote:
> > > > > From: Fugang Duan <fugang.duan@xxxxxxx>
> > > > >
> > > > > iMX8 fuse word index represent as one 4-bytes word, it should
> > > > > not be divided by 4.
> > > > >
> > > > > Exp:
> > > > > - MAC0 address layout in fuse:
> > > > > offset 708: MAC[3] MAC[2] MAC[1] MAC[0]
> > > > > offset 709: XX     xx     MAC[5] MAC[4]
> > > > >
> > > > > Signed-off-by: Fugang Duan <fugang.duan@xxxxxxx>
> > > > > ---
> > > > >  drivers/nvmem/imx-ocotp-scu.c | 6 +++---
> > > > >  1 file changed, 3 insertions(+), 3 deletions(-)
> > > > >
> > > > > diff --git a/drivers/nvmem/imx-ocotp-scu.c
> > > > > b/drivers/nvmem/imx-ocotp-scu.c index d9dc482..be2f5f0 100644
> > > > > --- a/drivers/nvmem/imx-ocotp-scu.c
> > > > > +++ b/drivers/nvmem/imx-ocotp-scu.c
> > > > > @@ -71,8 +71,8 @@ static int imx_scu_ocotp_read(void *context,
> > > > > unsigned
> > > > int offset,
> > > > >       void *p;
> > > > >       int i, ret;
> > > > >
> > > > > -     index = offset >> 2;
> > > > > -     num_bytes = round_up((offset % 4) + bytes, 4);
> > > > > +     index = offset;
> > > > > +     num_bytes = round_up(bytes, 4);
> > > > >       count = num_bytes >> 2;
> > > > >
> > > > >       if (count > (priv->data->nregs - index)) @@ -100,7 +100,7
> > > > > @@ static int imx_scu_ocotp_read(void *context, unsigned int offset,
> > > > >               buf++;
> > > > >       }
> > > > >
> > > > > -     memcpy(val, (u8 *)p + offset % 4, bytes);
> > > > > +     memcpy(val, (u8 *)p, bytes);
> > > > >
> > > > >       kfree(p);
> > > > >
> > > > With these changes you could use the 'val' pointer directly as the
> > > > destination for ocotp_read() without need for an intermediate buffer.
> > > >
> > > >
> > > > Lothar WaÃmann
> > >
> > > You are right, in fact, we can remove "p" and "buf" pointer.
> > > Thanks for your review, I will send out the V2.
> >
> > Hi Lothar,
> >
> > It still need intermediate buffer to read out n words (n * 4 bytes) from
> eFuse.
> > Because 'val' buffer size is real count parsed from DT, which is not an integer
> multiple of 4.
> >
> > For example, cell->bytes is parsed from "reg" property and it is real count
> like 6.
> >                         fec_mac0: mac@2c4 {
> >                                 reg = <0x2c4 6>;
> >                         };
> >
> > So we have to use intermediate buffer here.
> >
> val is a u32 pointer, so legally it cannot point to any buffer whose size is not
> divisible by four!
> 
Yes, val is a u32 pointer, as my understand, it point to the 'buf' whose size is cell->bytes
(the size is not parsed from 'reg' property)

The piece of code: 
buf = kzalloc(cell->bytes, GFP_KERNEL);
nvmem_reg_read(nvmem, cell->offset, buf, cell->bytes);
imx_scu_ocotp_read(void *context, unsigned int offset, void *val, size_t bytes);

If so, we still need intermediate buffer to read out eFuse words, and copy 'cell->bytes'
bytes to 'val' buffer. 
May I understand wrong ? Thanks very much!