On Thu, Sep 23, 2021 at 08:21:03PM +0200, Borislav Petkov wrote:

On Thu, Sep 23, 2021 at 12:05:58AM +0300, Kirill A. Shutemov wrote:

quoted

Unless we find other way to guarantee RIP-relative access, we must use
fixup_pointer() to access any global variables.

Yah, I've asked compiler folks about any guarantees we have wrt
rip-relative addresses but it doesn't look good. Worst case, we'd have
to do the fixup_pointer() thing.

In the meantime, Tom and I did some more poking at this and here's a
diff ontop.

The direction being that we'll stick both the AMD and Intel
*cc_platform_has() call into cc_platform.c for which instrumentation
will be disabled so no issues with that.

And that will keep all that querying all together in a single file.

And still do cc_platform_has() calls in __startup_64() codepath?

It's broken.

Intel detection in cc_platform_has() relies on boot_cpu_data.x86_vendor
which is not initialized until early_cpu_init() in setup_arch(). Given
that X86_VENDOR_INTEL is 0 it leads to false-positive.

I think opencode these two calls is the way forward. Maybe also move the
check from sme_encrypt_kernel() to __startup_64().

-- 
 Kirill A. Shutemov