Re: [PATCH] tpm: Map the ACPI provided event log

[Jarkko Sakkinen]

On Sun Dec 22, 2024 at 7:41 PM EET, James Bottomley wrote:
> On Sun, 2024-12-22 at 17:33 +0200, Jarkko Sakkinen wrote:
> > On Sun Dec 22, 2024 at 5:23 PM EET, Jarkko Sakkinen wrote:
> > > On Sun Dec 22, 2024 at 5:00 PM EET, James Bottomley wrote:
> > > > If event logs grow to greater than KMALLOC_MAX_SIZE then
> > > > absolutely it makes sense to map them instead of copying them. 
> > > > But we'd have to do that for all event log locators: ACPI, EFI
> > > > and OF, because event log size should be independent of the
> > > > mechanism used to locate it.  So, even as a long term fix
> > > > (assuming we think there's a possibility of logs expanding by
> > > > 50x), this patch doesn't do the right thing because it only maps
> > > > ACPI logs.
> > > 
> > > Because we have a test target only on ACPI where this happens fix
> > > should still fix only ACPI. It's not hard to reiterate this but 
> > > precursory iteration is a bad idea.
> > 
> > Also, "event log size should be independent of the mechanism used to
> > locate it" is a sentence that is sky high too abstract to say much.
> > 
> > I don't know what it means to be frank.
>
> event log size means the number of bytes from the beginning to the end
> of the event log.  Since the event log is created by the pre-boot
> environment, there is a convention for how to communicate this
> information from pre-boot to the kernel; this is the mechanism used to
> locate it.  We decode three mechanisms: an ACPI table, an EFI table and
> an OF entry.
>
> The pre-boot environment generating the event log is supposed to
> conform to the TCG standards for what events it contains; none of the
> entries depends on the mechanism used to locate the log, which is why
> the size also can't depend on the mechanis.  There are many optional
> events, but even if the pre-boot took a maximalist approach the most it
> could contain is a couple of hundred entries.  The variable entries are
> mostly small but several types can contain device paths or
> certificates, but even if you allow a 10k size for each entry, that's
> still at most 2MB.  So I think if a pre-boot declared log area goes
> over KMALLOC_MAX_SIZE (4MB on x86) it's safe to truncate the area
> because the log will never fill all of it.
>
> The corollary is that if we ever did find an actual log over 4MB, then
> the EFI and OF mechanisms used to locate it would also fail in the
> kmalloc, which is why KMALLOC_MAX_SIZE is the correct cap for the
> declared size.

Have you verified this with the failing system?

>
> Regards,
>
> James

BR, Jarkko