On Tue, 13 Oct 2020 at 17:29, Jarkko Sakkinen
<jarkko.sakkinen@xxxxxxxxxxxxxxx> wrote:
>
> On Tue, Oct 13, 2020 at 04:23:36PM +0530, Sumit Garg wrote:
> > On Tue, 13 Oct 2020 at 07:13, Jarkko Sakkinen
> > <jarkko.sakkinen@xxxxxxxxxxxxxxx> wrote:
> > >
> > > On Wed, Oct 07, 2020 at 03:37:45PM +0530, Sumit Garg wrote:
> > > > Current trusted keys framework is tightly coupled to use TPM device as
> > > > an underlying implementation which makes it difficult for implementations
> > > > like Trusted Execution Environment (TEE) etc. to provide trusted keys
> > > > support in case platform doesn't posses a TPM device.
> > > >
> > > > Add a generic trusted keys framework where underlying implementations
> > > > can be easily plugged in. Create struct trusted_key_ops to achieve this,
> > > > which contains necessary functions of a backend.
> > > >
> > > > Also, add a module parameter in order to select a particular trust source
> > > > in case a platform support multiple trust sources.
> > > >
> > > > Suggested-by: Jarkko Sakkinen <jarkko.sakkinen@xxxxxxxxxxxxxxx>
> > > > Signed-off-by: Sumit Garg <sumit.garg@xxxxxxxxxx>
> > >
> > > This is exactly kind of place where I think static_call() should be
> > > taken into use, which is a v5.10 feature [1]. For background and
> > > context, I'd read [2].
> >
> > This looks like an interesting feature. But I am not sure about the
> > real benefits that it will provide in case of trusted keys. If we are
> > looking at it performance wise then I think the gain will be
> > negligible when compared with slow TPM communication interface (eg.
> > SPI, I2C) or when compared with context switching involved in TEE.
> >
> > Also, it requires arch specific support too which currently seems to
> > be limited to x86 only.
>
> Please, do not purposely add indirect calls, unless you must. Here it's
> not a must.
>
> static_call() is the correct kernel idiom to define what you are doing
> in this patch. arch's will catch up.
Okay, fair enough. I will try to use it instead.
>
> > > The other thing that I see that does not make much else than additional
> > > complexity, is trusted_tpm.ko. We can do with one trusted.ko.
> > >
> >
> > Current implementation only builds a single trusted.ko module. There
> > isn't any trusted_tpm.ko.
> > -Sumit
>
> You're right, I'm sorry. I misread this:
>
> -static void __exit cleanup_trusted(void)
> +static void __exit exit_tpm_trusted(void)
> {
> if (chip) {
> put_device(&chip->dev);
> @@ -1257,7 +1029,11 @@ static void __exit cleanup_trusted(void)
> }
> }
>
> -late_initcall(init_trusted);
> -module_exit(cleanup_trusted);
> -
> -MODULE_LICENSE("GPL");
> +struct trusted_key_ops tpm_trusted_key_ops = {
> + .migratable = 1, /* migratable by default */
> + .init = init_tpm_trusted,
> + .seal = tpm_trusted_seal,
> + .unseal = tpm_trusted_unseal,
> + .get_random = tpm_trusted_get_random,
> + .exit = exit_tpm_trusted,
> +};
>
> Please remove "__init" and "__exit" for the functions as they are used
> as fields as members of a struct that has neither life span. That messed
> up my head.
Okay.
>
> Please use a single convention for the function names. It would
> be optimal to prefix with the subsystem name because that makes easier
> to use tracing tools: trusted_tpm_<callback name> would work.
>
Okay.
-Sumit
> /Jarkko
