On 13.05.2025 15:12, Andrew Cooper wrote: > On 03/05/2023 10:45 am, Jan Beulich wrote: >> We allow its use for writeback purposes only anyway, so let's also carry >> these out that way on capable hardware.
By implication of what you say ... >> With it now known that WBNOINVD uses the same VM exit code as WBINVD for >> both SVM and VT-x, we can now also expose the feature that way without >> further distinguishing the specific cases of those VM exits. Note that >> on SVM this builds upon INSTR_WBINVD also covering WBNOINVD, as the >> decoder won't set prefix related bits for this encoding in the resulting >> canonicalized opcode. >> >> Signed-off-by: Jan Beulich <jbeul...@suse.com> > > I realise this is an old series, Well, it's old because no-one cared to look at the v2 posting in over 2 years. > but this patch is unsafe with AMD SEV-SNP, and with CLX devices. > > Both have cases needing "genuinely evicted from the cache" semantics. ... here, are you suggesting that our present behavior is wrong? We don't have SEV-SNP support yet, so there it would only be "latently wrong" (and hence the change here would still be correct for now). What CLX is in this context I'm afraid I don't know (since surely you're not talking about vapes, the closest I can spot in search results is something gaming related, yet still not becoming very clear). I might have guessed Cascade Lake X, but then why would just that one model be a problem? Jan
