On Tue, Mar 17, 2026 at 09:27:13AM +0100, Uros Bizjak wrote: > On Tue, Mar 17, 2026 at 9:12 AM Ewanhaioc <[email protected]> wrote: > > > > From: Ewan <[email protected]> > > > > The __get_cpuid() and __get_cpuid_count() functions use > > "__leaf & 0x80000000" to determine the CPUID range base, which only > > distinguishes between Basic (0x0) and Extended (0x80000000) ranges. > > This causes queries to the Centaur/Zhaoxin range (0xC0000000) to be > > incorrectly mapped to the Extended range, and always return 0 (failure) > > on Centaur/Zhaoxin processors that support CPUID 0xC000xxxx > > leaves (e.g. PadLock feature detection via leaf 0xC0000001). > > > > Replace the bitmask with explicit range checks that correctly identify > > all four CPUID ranges: Basic (0x0), Hypervisor (0x40000000), Extended > > (0x80000000), and Centaur/Zhaoxin (0xC0000000). > > > > Additionally, the Hypervisor range (0x40000000) is also handled as an > > extra improvement, since the original bitmask incorrectly mapped it to > > the Basic range (0x0). > > > > Also update the __get_cpuid_max() comment to document the newly > > supported range base values. > > > > gcc/ChangeLog: > > > > * config/i386/cpuid.h (__get_cpuid_max): Update comment to > > document 0x40000000 and 0xC0000000 as valid ext values. > > (__get_cpuid): Replace "__leaf & 0x80000000" with explicit > > range checks for Basic, Hypervisor, Extended, and > > Centaur/Zhaoxin CPUID ranges. > > (__get_cpuid_count): Likewise. > > OK. > > Thanks, > Uros. >
Ping. This patch was approved on March 17, but it does not appear to have reached master yet. It still applies cleanly to current master. Could someone please commit it?
