[ANNOUNCEMENT] Updated: cpuid 20220812

Mon Aug 15 06:41:00 GMT 2022

The following package has been upgraded in the Cygwin distribution:
* cpuid		20220812
The program displays detailed information about the CPU(s) gathered from
the CPUID instruction, and also determines the exact model of CPU(s).
Whereas /proc/cpuinfo is like an abstract of the features important to
Linux in a system, cpuid is a standalone utility which writes a paper
expounding on every feature in each CPU's architecture and what it can
do, at about the one line per bit level.
It is updated and released frequently to stay current with Intel and
AMD information and supports other vendors' chips.
See the project home page for more information:
	http://etallen.com/cpuid.html
For information about changes since the previous Cygwin release,
see below or /usr/share/doc/cpuid/ChangeLog after installation.
Fri Aug 12 2022		20220812
* Corrected (synth) decoding for (0,6),(8,6) Intel Snow
 Ridge/Parker Ridge. It had been lumped in with Elkhart Lake, but only
 because that had been the only known core name for the Tremont uarch.
 These appear to be different cores. Also added steppings from SSG*.
* Added 8000000a/edx X2AVIC flag, from Linux kernel patches
 It appears to be undocumented, so far.
* Improved (synth) decoding for (0,6),(9,7),2, adding Alder Lake-HX.
* Reverted May 27 2022 split of 7/0/ebx hack to report bit 22
 as RDPID on AMD architectures. The AMD documentation is inconsistent
 on the location of this flag. In E.3.6, it claims 7/0/ebx. But in
 section 3, the RDPID instruction itself claims 7/0/ecx, as does the
 mention in Table 3-1. This also is consistent with Intel
 architectures. Thanks to Stefan Kanthak for pointing this out.
* Generalized (0,6),(8,14),9,YP stepping case to include
 Pentium 4425Y, from instlatx64 sample.
* Updated 7/0/edx comments to reflect original info source
 for SRBDS mitigation MSR available, previously just marked LX*.
* Updated 7/0/edx comments to reflect original info source
 for RTM transaction always aborts, previously just marked LX*.
* Added (vuln to branch type confusion synth) synthetic leaf
 to correct for the one known inaccuracy.
* cpuid.man: Added those two original source web pages from Intel:
 Intel Transactional Synchronization Extensions (Intel TSX) Memory and
 Performance Monitoring Update for Intel Processors (Article ID
 000059422), Special Register Buffer Data Sampling.
* Added 0x80000008/ebx not vulnerable to branch type confusion
 flag from "Technical Guidance For Mitigating Branch Type Confusion
 (White Paper)". Also added a synthetic flag to correct the special
 case for Family 0x19, where the raw flag is documented to be wrong.
* Added 7/2/edx indirect branch prediction related flags from
 Intel's "Branch History Injection and Intra-mode Branch Target
 Injection / CVE-2022-0001, CVE-2022-0002 / INTEL-SA-00598".
* Added (uarch synth) decoding for (0,6),(6,14) Cougar
 Mountain, mentioned as Airmont by Intel's "Retpoline: A Branch Target
 Injection Mitigation".
* cpuid.man: Added "Branch History Injection and Intra-mode Branch Target
 Injection / CVE-2022-0001, CVE-2022-0002 / INTEL-SA-00598" and
 "Retpoline: A Branch Target Injection Mitigation".
* Clarified (synth) for (0,6),(8,13) Tiger Lake-H from SSG*.
* Added support for hypervisor+3/ecx (Microsoft) flags.
* Added support for hypervisor+0xa/eax (Microsoft) VMCS
 GuestIa32DebugCtl support flag.
* Added support for hypervisor+0xa/ebx (Microsoft) VMCS
 HvFlushGuestPhysicalAddress* flag.
* Added (synth) for (0,6),(11,10),3 Raptor Lake-P Q0, from
 Coreboot*.
* Lionel Debroux's patch used MAX_CPUS all the time. But it
 really was meaningful only for the USE_KERNEL_SCHED_SETAFFINITY case
 (although, by happenstance, it may have been correct for all three
 cases). Replace this with an nr_cpu_ids global, determined by
 get_nr_cpu_ids(). The simplest version just returns
 sysconf(_SC_NPROCESSORS_CONF), although that could be problematic on
 systems with non-contiguous CPU numbers.
* For USE_KERNEL_SCHED_SETAFFINITY, improve this, and also
 support systems with > 1024 CPUs, by estimating nr_cpu_ids using a
 power-of-2 walk through successively larger cpu_set_t sizes until
 sched_getaffinity succeeds.
* For systems using cpu_set_t (only Cygwin?), cap the
 nr_cpu_ids to CPU_SETSIZE.
* The _SC_NPROCESSORS_CONF check in real_setup() is removed
 because it's redundant now.
* In do_real() and do_real_one(), avoid breaking out of loop
 because of downed CPUs.