Date: Tue, 3 Jul 2018 20:16:19 +0200 From: Solar Designer <solar@...nwall.com> To: Vasily Averin <vvs@...tuozzo.com> Cc: owl-dev@...ts.openwall.com Subject: Re: 32-bit syscall breakage in -431 kernel with KAISER On Tue, Jul 03, 2018 at 08:32:57PM +0300, Vasily Averin wrote: > However then node was crashed on start of microcode_ctl service on host. > > I've found that on our kernel __cpuinit moves get_scattered_cpuid_extra() into init section, > but this function can be called lagally called after initialization, > from microcode_write() -> spec_ctrl_rescan_cpuid() > > Issue was caused by CONFIG_HOTPLUG_CPU enabled on Red Hat but disabled on our kernels. > Therefore I afraid your kernel should be affected too. > > Could you please clarify this question? It's quite possible we never tested microcode_ctl on Owl. It isn't part of Owl, even though there was intent to add it at some point. I enabled CONFIG_MICROCODE=m in our kernels in late 2016: * Sat Dec 10 2016 Solar Designer <solar-at-owl.openwall.com> 2.6.18-408.el5.028stab120.1.owl5 - Merged in Red Hat's CVE-2016-5195 "Dirty COW" fix from -416 (slightly different from the fix included in OpenVZ's 120.3 released earlier) while also keeping the mitigation introduced in owl4. - In the x86_64 config, enabled CONFIG_MICROCODE=m, CONFIG_NUMA=y and many related options, CONFIG_HUGETLB_PAGE=y, CONFIG_HUGETLBFS=y, CONFIG_I2C=m and many sensors (similar to RHEL's), bumped up CONFIG_NR_CPUS from 32 to 255. Tested many of these on a Dell PowerEdge R720xd with 2x E5-2660 v2 (NUMA, huge pages, some I2C sensors, 40 logical CPUs). IIRC, I intended to test it on that Dell machine mentioned in there, but it was already at Intel's latest microcode revision for its CPUs, so I did not. I think I did test loading of the kernel module, though. Alexander
Powered by blists - more mailing lists
Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.