Linux's Performance-Boosting FSGSBASE Support Dropped For Now Over Serious Bugs
While we had been looking forward to Intel FSGSBASE support for yielding some performance benefits especially in areas impacted by Spectre / Meltdown / Foreshadow / Zombieload, after the support was queued for merging into Linux 5.3, the code has now been reverted over "serious bugs" with the implementation.
Our testing of the code queued earlier for Linux 5.3 did show some performance improvements and thus were looking forward to the addition with this next kernel, now it's not coming at least until being reworked. The FSGSBASE support has been present on Intel CPUs back to Ivy Bridge or on the AMD side with Zen version one and newer.
In the revert by Thomas Gleixner he noted, "The FSGSBASE series turned out to have serious bugs and there is still an open issue which is not fully understood yet. The confidence in those changes has become close to zero especially as the test cases which have been shipped with that series were obviously never run before sending the final series out to LKML."
The Intel developers own FSGSBASE test case was yielding a segmentation fault when running, among other problems leading to this code being deemed unready.
At least it didn't make it as far as Linus Torvalds' tree where he may have otherwise had some choice words about code quality...
Our testing of the code queued earlier for Linux 5.3 did show some performance improvements and thus were looking forward to the addition with this next kernel, now it's not coming at least until being reworked. The FSGSBASE support has been present on Intel CPUs back to Ivy Bridge or on the AMD side with Zen version one and newer.
In the revert by Thomas Gleixner he noted, "The FSGSBASE series turned out to have serious bugs and there is still an open issue which is not fully understood yet. The confidence in those changes has become close to zero especially as the test cases which have been shipped with that series were obviously never run before sending the final series out to LKML."
The Intel developers own FSGSBASE test case was yielding a segmentation fault when running, among other problems leading to this code being deemed unready.
At least it didn't make it as far as Linus Torvalds' tree where he may have otherwise had some choice words about code quality...
37 Comments