Re: [PATCH v7 1/3] riscv: Avoid unaligned access when relocating modules

[Charlie Jenkins]

On Tue, Oct 31, 2023 at 05:35:57PM +0100, Andreas Schwab wrote:
> On Okt 31 2023, Emil Renner Berthing wrote:
> 
> >> +static int riscv_insn_rmw(void *location, u32 keep, u32 set)
> >> +{
> >> +	u16 *parcel = location;
> >> +	u32 insn = (u32)le16_to_cpu(parcel[0]) | (u32)le16_to_cpu(parcel[1]) << 16;
> >> +
> >> +	insn &= keep;
> >> +	insn |= set;
> >> +
> >> +	parcel[0] = cpu_to_le32(insn);
> >
> > Why cpu_to_le32(insn)? Unless I've misunderstood something downcasting unsigned
> > to unsigned values in C (eg. from u32 to u16) is defined to always discard the
> > most signifcant bits, so cpu_to_le16(insn) should be fine.
> 
> cpu_to_le32(insn) can't be right here anyway, since it also swaps the
> two u16 halves and would be the same as cpu_to_le16(insn >> 16) on big
> endian.

Yes, not sure why I did that... I will fix that up.

- Charlie

> 
> -- 
> Andreas Schwab, schwab@xxxxxxxxxxxxxx
> GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
> "And now for something completely different."