On Wed, Mar 10, 2021 at 04:41:54PM +1100, Andrew James wrote:
> Hi everyone,
>
> I noticed the following potential unaligned memory access in
> tpm-interface.c, and I was wondering how it would be handled on
> architectures that don't support unaligned accesses. Is this TPM code
> expected to work on all architectures?
>
> ssize_t tpm_transmit_cmd(...)
> {
> const struct tpm_header *header = (struct tpm_header *)buf->data;
> int err;
> ssize_t len;
>
> len = tpm_transmit(chip, buf->data, PAGE_SIZE);
> if (len < 0)
> return len;
>
> err = be32_to_cpu(header->return_code);
> ...
> }
>
> I'm referring to the line at the bottom, before the ellipsis where we
> read 'return_code'.
>
> struct tpm_header has a __be16 tag followed by a __be32 return code.
> If we are reading 'return_code', is this an unaligned access?
> Similarly this would apply to the 'length' member too?
>
> Documentation/unaligned-memory-access.txt recommends going through the
> kernel API get_unaligned() and put_unaligned() in <asm/unaligned.h> to
> avoid unaligned accesses, but I don't see this anywhere in the TPM
> code. I'm just trying to wrap my head around this.
There's a patch for this in existence:
https://lore.kernel.org/tpmdd-devel/1479899094-9486-1-git-send-email-tomas.winkler@xxxxxxxxx/
It's years since this was sent but I recall the issues were fairly
easy to fix.
/Jarkko
