Linux's Performance-Boosting FSGSBASE Support Dropped For Now Over Serious Bugs

[phoronix]

Phoronix: 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...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-FSGSBASE-Reverted

[L_A_G]

The last of Linus' rants before he went on his hiatus to think things over was because of a muppet at Intel pushing shoddy code and then insisting it was fine when people brought up serious issues with it. Judging by this it seems like Intel took that hiatus as them being in the right and deciding not to improve the standards of the code they try to push into mainline.

Methinks we may be about so see another Linus rant aimed at an Intel muppet.

[microcode]

> ...the test cases which have been shipped with that series were obviously never run before sending the final series out to LKML

This is the sort of time-wasting nonsense that led me not to blame Linus or other maintainers for seeking to publicly embarrass the perpetrators through rants.
There might be nothing else you can do to prevent them from sending you this garbage quality of patch: it's just their dayjob, so they feel safe, but they're also making a dog's breakfast of things.

[F.Ultra]

The Intel developers own FSGSBASE test case was yielding a segmentation fault when running

Wow, just wow! Have Intel outsourced all this to the their janitors or what the actual fuck?

> ...the test cases which have been shipped with that series were obviously never run before sending the final series out to LKML

This is the sort of time-wasting nonsense that led me not to blame Linus or other maintainers for seeking to publicly embarrass the perpetrators through rants.
There might be nothing else you can do to prevent them from sending you this garbage quality of patch: it's just their dayjob, so they feel safe, but they're also making a dog's breakfast of things.

While it might be "just their dayjob" I wonder for how long if they cannot get their code merged. At some point the upper management should notice that they are paying for nothing (at least one can hope/dream).

[Wojcian]

Another reason to avoid Intel at all costs.

[tuxd3v]

I don't know about implications it can have,
FSGSBASE, is present at least on AMD CPUs going back to Kaveri( Steamroller uarch 2014 ) and maybe earlier..
I have been running some code, with FSGSBASE, on Kaveri, but not Kernel related..

Maybe there are implementation differences between Intel and Amd, or differences between newer iterations of the processors..?

It would be interesting a benchmark going back to at least Steamroller, Intel and amd, like picking one of each series..
This could give a clear picture of performance gains/losses is could be in last 4-5 years..

Another interesting test,
Would be picking a compiled kernel, compiled on a determined Socket, and then switch only cpus for that socket..
In that way it could be determined if there are incompatibilities between implementations of same vendor..

[stormcrow]

Wow, just wow! Have Intel outsourced all this to the their janitors or what the actual fuck?



While it might be "just their dayjob" I wonder for how long if they cannot get their code merged. At some point the upper management should notice that they are paying for nothing (at least one can hope/dream).

I wonder if Management already noticed and that's why he's on the Linux team. Third parties are vetting his work for them. He's not in a position to seriously screw anything major up. Maybe he's already on probation? Who knows. Not us, for sure.

That reminds me of a particularly egregious case from a number of years ago where a particular hardware mfg employee tried to get a driver integrated into the Linux kernel. It had multiple copyright notices all over it, had little in the way of comments, and was in generally rather poor shape. It's rather no wonder the computing industry is in such dire straights with quality and security problems as these problems are rife and mostly hidden from direct scrutiny.

[AsuMagic]

Why did anyone accept it for merging in the first place, if this was "obviously never tested"? There should never have been anything to revert in the first place.

Edit: I didn't read the mail.

I could definitely slap myself for not noticing the test case fail when merging that series, but TBH my expectations weren't that low back then. Won't happen again.

Sounds like the one to blame is the maintainer who merged this, honestly.
When Linus rants about code quality, he's generally flaming the maintainers at least as much as the ones who wrote the code. Just saying.
But yeah, better go into conspiracies (EMHARGGD WAT IF INTEL PURPOSEFULI PUT A BAD DEVELOPR FOR LINUX TO SABOTAEG???) and personally insult people (because this is known to make you a better person).

[L_A_G]

While it might be "just their dayjob" I wonder for how long if they cannot get their code merged. At some point the upper management should notice that they are paying for nothing (at least one can hope/dream).

I seriously doubt that Intel, being primarily a hardware vendor, has the kind of people who would understand what's going on here in upper management. Instead of software people their board is bound to be a mix of hardware people and the MBA types that run pretty much all big corporations these days. More probably than not they'll just think that Linus and the other lead kernel maintainers are just being difficult because they hate the company or something even without the muppets in question trying to explain it to them as such.

Not that there aren't valid reasons to hate Intel as many of the business practices they've employed over the years are akin to the practices Monsanto has used to push out it's competitors and exploit farmers. Because of the easily drawn parallels I personally consider the to essentially be the Monsanto of the semiconductor industry.

[AsuMagic]

Not that there aren't valid reasons to hate Intel as many of the business practices they've employed over the years are akin to the practices Monsanto has used to push out it's competitors and exploit farmers. Because of the easily drawn parallels I personally consider the to essentially be the Monsanto of the semiconductor industry.

Please provide a single example that proves your point for anything intel open source.