Did a quick benchmark with a Wine application (most affected by KPTI) on 4.9.75 and Cinebench under Wine used to be at 100% CPU utilization both at normal and very high process priorities with a non KPTI kernel. Now, at both priorities, some cores are given less work so CPU usage would only be around 85% on very high priority. You lose at most 100 Cinebench points on a 6 core hyperthreaded. Heavy multi-threaded workloads are seeming like the ones hit hard by this.
Needless to say, very bad news for the Wine community except on Ryzen. If anyone is crazy enough to run Photoshop CS6 on Wine, you may slow down on image previews by quite a bit.
So my windows machine has gone fubar. Intel i5 getting insane usage spikes from windows tasks and extremely noticable skipping and freezes in basic browsing and gaming. Clean reinstall and now itâs even worse.
My 4.9.75 Ubuntu machine just went fubar too. Threads literally lock up on my E5-1660 on kernel 4.9.75. All non KPTI kernels are fine, but the KPTI patches causes programs to cause the kernel to fully freeze. Itâs not panicing, cause multimedia streams still work, but itâs literally 100% frozen without panicing. A âhardâ soft lock.
Went back to 4.13.16 and all was well again. This is a direct impact of the KPTI patches.
Fedora was on the kernel just before KPTI and didnât release KPTI just yet I think for this specific reason. Also, make sure to get 384.111 Nvidia drivers for 4.14.11 or you risk not being able to boot.
If Page Table Isolation is causing soft hard locks on everything but Haswell and up, thatâs just absolutely absurd.
Regressions on updates (especially critical ones like this one that are deployed very fast) can happen with any OS, open source or not, no mater how good the coding is. That is why the forceful updates is a big issue.
If the PTI kernel on linux breaks your PC it is strait forward to roll back until the update is ironed out. Or choose different kernel temporarily that work. In windows you cannot to that. Even if you use recovery the moment you login the update will download and break things gain. That is literally insane and pretty much can trap you on a bricked machine for quite some time. It is already bad for consumer use, add critical systems to this (that often use older hardware) and things get even worse.
it is a very bad idea not to update everything immediately. It is a equally bad idea not to have the option to rollback from regressions as well.
384.111 and 4.14.12 works on my laptop as well (intel wth optimus) on Mint.
win-raid also have guides how to update your bios with new microcode (UBU) etc. No doubt a lot of motherboard vendors wonât release updated bios. Just like with the IntelME vuln.
Whatâs weird is my Core 2 Duo laptop is fine with 4.15-RC7, but my E5-1660 chokes on 4.9.75, cause I canât use 4.15-RC7 on my X79 system cause it has Nvidia GPUs.
I just installed a BIOS update on a couple Latitude 5480s at work. The update was released 22 Dec, updated yesterday (9 Jan), and includes a microcode update for CVE-2017-5715 along with some ME firmware updates.