Re: [PATCH 1/4] target/i386: Introduce SierraForest-v2 model

[Tao Su <tao1.su@linux.intel.com>]

On Tue, Jan 21, 2025 at 09:34:58AM -0800, Pawan Gupta wrote:
> On Tue, Jan 21, 2025 at 10:06:47AM +0800, Tao Su wrote:
> > Update SierraForest CPU model to add LAM, 4 bits indicating certain bits
> > of IA32_SPEC_CTR are supported(intel-psfd, ipred-ctrl, rrsba-ctrl,
> > bhi-ctrl) and the missing features(ss, tsc-adjust, cldemote, movdiri,
> > movdir64b)
> > 
> > Also add GDS-NO and RFDS-NO to indicate the related vulnerabilities are
> > mitigated in stepping 3.
> 
> Does this only apply to stepping 3? I don't think Sierra Forest was ever
> vulnerable to GDS and RFDS [1].
> 

On the real machine, stepping 0 does not set GDS_NO and RFDS_NO, but
stepping 3 does.

> There are many other vulnerabilities that Sierra Forest is not vulnerable to,
> is it really necessary to add the *_NO bits to CPU definitions?
> 
> [1] https://www.intel.com/content/www/us/en/developer/topic-technology/software-security-guidance/processors-affected-consolidated-product-cpu-model.html

*_NO bits indicate processor is not affected by *, so adding these to the
CPU model will prevent the guest OS (using the CPU model) from trying to
use related software mitigation, which I think is reasonable.