x86-64.net

Processor vulnerability

CVE-2025-62626

RDSEED with a 16- or 32-bit destination on Zen 5 returns 0 too often while setting CF, so the failure looks like a random value.

Name
RDSEED Failure on Zen 5
Kind
Other
CVE published
2025-11-21 CVE record
Advisories
AMD-SB-7055
RDSEED sets CF when it returns a random value and clears it when it has none. On Zen 5 the 16- and 32-bit forms return 0 at a rate random data would not, with CF set, so software takes the zero as a seed. The 64-bit form is not affected. The problem was first reported on the Linux kernel mailing list. Until the microcode is in place, Linux 7.3 hides RDSEED from CPUID on Zen 5 processors whose microcode is older than the fixed revisions it lists.

Affected processors

1 of the 1 microarchitectures listed are affected in all or some of their models.

What the sources say of each microarchitecture.
MicroarchitectureStatusDetailSource
Zen 5AffectedEPYC 9005 and 4005; Ryzen 9000, 9000HX, AI 300 and AI Max 300; Threadripper 9000 and PRO 9000 WX; Ryzen Z2 Extreme; embedded 9000, 9005 and 4005amd.com

Mitigations

  • Microcode: Microcode, or AGESA firmware, AMD released from October 2025 (Turin C1 0x0B00215A, Turin Dense B0 0x0B101054). Instructions: RDSEED. Source.
  • Kernel: Linux 7.3 clears the RDSEED bit of CPUID on Zen 5 processors with older microcode; clearcpuid=rdseed does the same on any kernel. Instructions: RDSEED, CPUID. Source.
  • Programs: Using the 64-bit form, or treating a returned 0 as a failure (CF=0) and retrying. Instructions: RDSEED. Source.

Sources

  1. AMD-SB-7055: RDSEED Failure on AMD "Zen 5" Processors
  2. Linux 7.3 arch/x86/kernel/cpu/amd.c: zen5_rdseed_microcode