Re: [PATCH v4] tpm_crb: fix fTPM on AMD Zen+ CPUs

From: Jarkko Sakkinen
Date: Fri Sep 20 2019 - 11:03:04 EST


On Tue, Sep 17, 2019 at 11:54:03PM +0300, Vanya Lazeev wrote:
> On Tue, Sep 17, 2019 at 10:10:13PM +0300, Jarkko Sakkinen wrote:
> > On Mon, Sep 16, 2019 at 11:00:30PM +0300, Vanya Lazeev wrote:
> > > On Mon, Sep 16, 2019 at 08:51:30AM +0300, Jarkko Sakkinen wrote:
> > > > On Sat, Sep 14, 2019 at 08:17:44PM +0300, ivan.lazeev@xxxxxxxxx wrote:
> > > > > + struct list_head acpi_resources, crb_resources;
> > > >
> > > > Please do not create crb_resources. I said this already last time.
> > >
> > > But then, if I'm not mistaken, it will be impossible to track pointers
> > > to multiple remaped regions. In this particular case, it
> > > doesn't matter, because both buffers are in different ACPI regions,
> > > and using acpi_resources only to fix buffer will be enough.
> > > However, this creates incosistency between single- and
> > > multiple-region cases: in the latter iobase field of struct crb_priv
> > > doesn't make any difference. Am I understanding the situation correctly?
> > > Will such fix be ok?
> >
> > So why you need to track pointers other than in initialization as devm
> > will take care of freeing them. Just trying to understand the problem.
> >
>
> We need to know, which ioremap'ed address assign to control area, command
> and response buffer, based on which ACPI region contains each of them.
> Is there any method of getting remapped address for the raw one after
> resouce containing it has been allocated?
> And what do you mean by initialization? crb_resources lives only in
> crb_map_io, which seems to run only once.

Aah, I see.

Well at leat we want the dynamic allocation away from the callback e.g.
use a fixed array:

#define TPM_CRB_MAX_RESOURCES 4 /* Or however many you need */

struct list_head acpi_res_list;
struct acpi_resource *acpi_res_array[TPM_CRB_MAX_RESOURCES];
void __iomem *iobase_array[TPM_CRB_MAX_RESOURCES];

If there are more resources than the constant you could issue a warning
to klog but still try top continue initialization.

PS. Use for new symbols TPM_CRB_ and tpm_crb_ prefixes. Because of
easier tracing of TPM code I want to move to this naming over time.

/Jarkko