On Sat, Aug 26, 2023 at 12:26:25AM +0100, Conor Dooley wrote: > On Fri, Aug 25, 2023 at 04:11:38PM -0700, Evan Green wrote: > > In /proc/cpuinfo, most of the information we show for each processor is > > specific to that hart: marchid, mvendorid, mimpid, processor, hart, > > compatible, and the mmu size. But the ISA string gets filtered through a > > lowest common denominator mask, so that if one CPU is missing an ISA > > extension, no CPUs will show it. > > > > Now that we track the ISA extensions for each hart, let's report ISA > > extension info accurately per-hart in /proc/cpuinfo. We cannot change > > the "isa:" line, as usermode may be relying on that line to show only > > the common set of extensions supported across all harts. Add a new "hart > > isa" line instead, which reports the true set of extensions for that > > hart. > > > > Signed-off-by: Evan Green <evan@xxxxxxxxxxxx> > > Reviewed-by: Andrew Jones <ajones@xxxxxxxxxxxxxxxx> > > > Reviewed-by: Conor Dooley <conor.dooley@xxxxxxxxxxxxx> > > Can you drop this if you repost? > > > +"isa" vs "hart isa" lines in /proc/cpuinfo > > +------------------------------------------ > > + > > +The "isa" line in /proc/cpuinfo describes the lowest common denominator of > > +RISC-V ISA extensions recognized by the kernel and implemented on all harts. The > > +"hart isa" line, in contrast, describes the set of extensions recognized by the > > +kernel on the particular hart being described, even if those extensions may not > > +be present on all harts in the system. > > > In both cases, the presence of a feature > > +in these lines guarantees only that the hardware has the described capability. > > +Additional kernel support or policy control changes may be required before a > > +feature is fully usable by userspace programs. > > I do not think that "in both cases" matches the expectations of > userspace for the existing line. It's too late at night for me to think > properly, but I think our existing implementation does work like you > have documented for FD/V. I think I previously mentioned that it could > misreport things for vector during the review of the vector series but > forgot about it until now. I went and checked, and yes it does currently do that for vector. I don't think that that is what userspace would expect, that Google cpu_features project for example would draw incorrect conclusions.
Attachment:
signature.asc
Description: PGP signature
