On Wed, Jul 20, 2016 at 11:03:36AM -0600, Jason Gunthorpe wrote:
> On Tue, Jul 19, 2016 at 05:24:11PM -0700, Andrey Pronin wrote:
>
> > The only two things that bother me with such approach are
> > (1) whatever names I pick for the new set of functions, they
> > will be similar to and thus might be confused with the
> > original tpm_tis_read/writeXX;
>
> tpm_tis_helper_read16 ?
>
> > (2) these functions are phy-specific, so possibly it's better
> > to create tpm_tis_spi.h and put them there with proper
> > name prefixes. And then use in tpm_tis_spi and cr50_spi.
>
> No, they are generic to any tis phy that implements read only through
> read_bytes.
>
> (Honestly, I'm not sure we made the best choice here having phy
> functions for all the versions, we are not that performance
> sensitive, just getting rid of everything but read_bytes from the
> phy_ops would probably also be a reasonable thing to do.)
>
One thing we can do is re-implement functions tpm_tis_read/writeXX
to use phy-specific implementations of read16, read32, write32 if they
are provided. But if those function pointers are left NULL in phy_ops,
fallback to using read/write_bytes and byte-swapping.
I.e., instead of:
static inline int tpm_tis_read16(struct tpm_tis_data *data, u32 addr,
u16 *result)
{
return data->phy_ops->read16(data, addr, result);
}
do the following:
static inline int tpm_tis_read16(struct tpm_tis_data *data, u32 addr,
u16 *result)
{
int rc;
if (data->phy_ops->read16)
return data->phy_ops->read16(data, addr, result);
rc = data->phy_ops->read_bytes(data, addr,
sizeof(u16), (u8 *)result);
if (!rc)
*result = le16_to_cpu(*result);
return rc;
}
If you like the idea, I'll submit it as a separate patch.
Andrey
