Post B6CXaPGEEt0aMM1wES by vbabka@mastodon.social
 (DIR) More posts by vbabka@mastodon.social
 (DIR) Post #B6CEJIqOwbnoPI53s8 by piggo@piggo.space
       0 likes, 0 repeats
       
       In light of libre office calling its release channels fresh and still, I propose Linux kernel to be called spicy (latest broken) and mild (lts)
       
 (DIR) Post #B6CEO9PXRcACb8pMI4 by piggo@piggo.space
       0 likes, 0 repeats
       
       6.19 and 7.0 are both definitely spicy
       
 (DIR) Post #B6CEa4XDOqPAwx6QM4 by oleksandr@natalenko.name
       0 likes, 0 repeats
       
       @piggo Given the LTS kernel is slop-backported, the latest kernel appears to be in fact less broken.
       
 (DIR) Post #B6CEa5IiYEP1KHKMOe by piggo@piggo.space
       0 likes, 0 repeats
       
       @oleksandr I don't trust non lts since 6.19 made mongodb in docker segfault (a first time for me a kernel version caused such breakage)
       
 (DIR) Post #B6CEqQAEkYWefK9OJk by oleksandr@natalenko.name
       1 likes, 0 repeats
       
       @piggo OTOH, I trust neither mongodb nor docker, nor people who use them, fwiw
       
 (DIR) Post #B6CEqQWDQq6flUQwZk by piggo@piggo.space
       0 likes, 0 repeats
       
       @oleksandr cool, niceregrettably its for work. Anyway I don't see how host kernel should make the guest app segfault. I know its not virtualization but still, I have higher expectations of docker
       
 (DIR) Post #B6CFBcEF5aUAAG88zQ by oleksandr@natalenko.name
       0 likes, 0 repeats
       
       @piggo random googling: "Set the environment variable GLIBC_TUNABLES to glibc.cpu.hwcaps=-SHSTK before starting MongoDB"
       
 (DIR) Post #B6CFBcOAUgQQf2m4Se by piggo@piggo.space
       0 likes, 0 repeats
       
       @oleksandr set where, in the docker compose env?
       
 (DIR) Post #B6CFVQkzS0KwNQtdWS by oleksandr@natalenko.name
       0 likes, 0 repeats
       
       @piggo Yes. There are also some suggestions to try GLIBC_TUNABLES=glibc.pthread.rseq=1 instead.
       
 (DIR) Post #B6CFVQySdv713DCOWG by piggo@piggo.space
       0 likes, 0 repeats
       
       @oleksandr outstanding. I have strong feelings about these choices someone made in the kernelhttps://dev.to/seangles/the-30-second-death-a-memoir-3gc
       
 (DIR) Post #B6CXaPGEEt0aMM1wES by vbabka@mastodon.social
       0 likes, 0 repeats
       
       @piggo @oleksandr this sounds like something that would be reverted if someone reported it
       
 (DIR) Post #B6CXaPYJ9fTDGQUNPc by vbabka@mastodon.social
       0 likes, 0 repeats
       
       @piggo @oleksandr also I question that article, no other links point to shadow stacks? Oleksandr is more likely correct with rseq. Long story: https://lwn.net/SubscriberLink/1070072/528f7ce985b2d294/
       
 (DIR) Post #B6CXaPlmLaFHwCn8PQ by piggo@piggo.space
       0 likes, 0 repeats
       
       @vbabka @oleksandr so if this is the case, why is the suggestion to set GLIBC_TUNABLES=glibc.pthread.rseq=1 ?Considering the bug is a hacky api usage in TCMalloc
       
 (DIR) Post #B6CXaPuHpx3EMalvfc by vbabka@mastodon.social
       0 likes, 0 repeats
       
       @piggo @oleksandr Boris tells me shadow stacks might affect this only if there was a qemu VM involved - https://lore.kernel.org/r/20260403154915.2285621-1-david.kaplan@amd.com