In May - April of this year, Intel
updated the documentation for its processors. It became known why - a description of the new error appeared. According to a
document published by Debian, Skylake and Kaby Lake microarchitecture chips, as well as Xeon v5 and v6 server processors and some Pentium processors can behave unpredictably with active Hyper-Threading.
/ Photo Ultra Mendoza PDThe Intel documentation
describes the error as follows: “Short cycles of less than 64 instructions using registers AH, BH, CH, or DH, as well as their corresponding registers of higher capacity (for example, RAX, EAX, or AX for AH), can cause unpredictable behavior. system. The problem is observed only if both logical processors are active on the same physical processor. ”
')
At the same time, there were “victims” in the community. At the beginning of this year, the developer of the OCaml tool, who observed the strange behavior of the compiler,
ran into trouble. He was able to isolate the error only now, when its description appeared in the Intel documentation.
According to Debian, “unpredictable system behavior” refers to a wide range of problems, from application malfunctioning to damage and data loss. Therefore, project participants
urge owners of computers based on Skylake and Kaby Lake microarchitectures to disable Hyper-Threading in the BIOS or UEFI. It is emphasized that the problem concerns not only Debian or Linux, and can manifest itself in any operating systems, including Windows.
Since Intel is aware of the problem, it should be fixed soon. The microcode with error correction is currently
released only for processors with Skylake (0xB9, 0xBA and later). In addition, in the new processors on Kaby Lake-X, the bug was fixed initially in the B0 core stepping. End users will get the necessary patch through BIOS updates for motherboards.
This whole situation gave rise to a rather heated discussion in the comments to the news. One of the residents of the Hacker News platform
noted that due to the frequent
occurrence of bugs in processors, manufacturers should more often conduct regression testing and other types of tests, even related to the use of malware, if this allows to exclude such situations. At the same time, HN users named such bugs as extremely unpleasant.
“When you work with CPU and memory, you always think that you made a mistake in the code. You think that this is the reason -
says one of the developers. “So when you run into a hardware problem, it drives you crazy.”
Note that the detected error is not the first “nuisance” of this nature, which hit the Skylake microarchitecture. At the beginning of last year, it
revealed another critical problem for end users, which led to processor freezes and crashes under high load.
PS A few more materials from our blog:
PPS In our blog on Habré:
100 practical materials on security, economics and IaaS tools.