[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [Xen-devel] [PATCH 4/7] x86/ucode/intel: Reimplement get_{data, total}size() helpers
On 26.03.2020 15:35, Andrew Cooper wrote: > On 25/03/2020 13:41, Jan Beulich wrote: >> On 23.03.2020 11:17, Andrew Cooper wrote: >>> --- a/xen/arch/x86/cpu/microcode/intel.c >>> +++ b/xen/arch/x86/cpu/microcode/intel.c >>> @@ -46,9 +46,16 @@ struct microcode_header_intel { >>> unsigned int sig; >>> unsigned int cksum; >>> unsigned int ldrver; >>> + >>> + /* >>> + * Microcode for the Pentium Pro and II had all further fields in the >>> + * header reserved, had a fixed datasize of 2000 and totalsize of 2048, >>> + * and didn't use platform flags despite the availability of the MSR. >>> + */ >>> + >>> unsigned int pf; >>> - unsigned int datasize; >>> - unsigned int totalsize; >>> + unsigned int _datasize; >>> + unsigned int _totalsize; >> ... the underscores here dropped again. Or else - why did you add >> them? This (to me at least) doesn't e.g. make any more clear that >> the fields may be zero on old hardware. > > No, but it is our normal hint that you shouldn't be using the field > directly, and should be using the accessors instead. Yet in patch 5 you do. Perhaps for an understandable reason, but that way you at least partly invalidate what you say above. >> Furthermore - do we really need this PPro/PentiumII logic seeing >> that these aren't 64-bit capable CPUs? > > I did actually drop support in one version of my series, but put it back in. > > These old microcode blobs are still around, including in some versions > of microcode.dat. By dropping the ability to recognise them as > legitimate, we'd break the logic to search through a container of > multiple blobs to find the one which matches. Oh, good point. Jan
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |