On Wed, 5 Mar 2025 at 23:11, Stephen Boyd <sboyd@xxxxxxxxxx> wrote:
>
> Quoting Sasha Finkelstein via B4 Relay (2025-03-05 12:26:40)
> > + while (len_to_read < bc) {
> > + rsp = read_reg(spmi, SPMI_RSP_REG);
> > + i = 0;
> > + while ((len_to_read < bc) && (i < 4)) {
> > + __buf[len_to_read++] = ((0xff << (8 * i)) & rsp) >>
> > + (8 * i);
> > + i += 1;
> > + }
>
> Is this ioread32_rep()?
>
[...]
> > + while (i < bc) {
> > + j = 0;
> > + spmi_cmd = 0;
> > + while ((j < 4) & (i < bc))
> > + spmi_cmd |= __buf[i++] << (j++ * 8);
>
> Is this iowrite32_rep()? Perhaps unaligned sizes have to be dealt with,
> but otherwise I suspect it would be more efficient to use
> iowrite32_rep() until the number of bytes is less than 4 and then do the
> one extra pack.
I think it would be better to leave them open-coded, io{read,write}32_rep
casts the buffer to u32 and accesses it that way, probably resulting in
unaligned accesses, and we are on arm64.
