Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.debian.kernel > #60186 > unrolled thread

Revisiting x32 ABI state

Back to article view | Back to linux.debian.kernel

Contents

  Revisiting x32 ABI state Bastian Blank <waldi@debian.org> - 2018-02-11 11:20 +0100
    Re: Revisiting x32 ABI state Ben Hutchings <ben@decadent.org.uk> - 2018-02-11 23:30 +0100

#60186 — Revisiting x32 ABI state

Moin

After almost four years and due to recent changes in the syscall path,
I'd like to revisit our current handling of the x32 ABI.

Currently we disable support for the x32 ABI on runtime, enabling it
needs a kernel command line parameter.  This Debian specific change was
done due to concerns of ABI specific problems with the pretty new code.

In 4.15.2 the syscall path was cleaned up.  The so called "fast-path"
was dropped completely, it didn't provide much of a fast path anymore
because of the whole Spectre and Meltdown stuff.  Also our patch made
inline patches of the fast path, to remove the overhead of the checks.

I think after that four years we shoud revisit that change.  Do we still
think x32 may be a problem?  Does the patch introduce runtime latency
due to it's not longer inline patched form?

Regards,
Bastian

-- 
There is a multi-legged creature crawling on your shoulder.
		-- Spock, "A Taste of Armageddon", stardate 3193.9

[toc] | [next] | [standalone]

#60192

[Multipart message — attachments visible in raw view] — view raw

On Sun, 2018-02-11 at 10:56 +0100, Bastian Blank wrote:
> Moin
> 
> After almost four years and due to recent changes in the syscall path,
> I'd like to revisit our current handling of the x32 ABI.
> 
> Currently we disable support for the x32 ABI on runtime, enabling it
> needs a kernel command line parameter.  This Debian specific change was
> done due to concerns of ABI specific problems with the pretty new code.
> 
> In 4.15.2 the syscall path was cleaned up.  The so called "fast-path"
> was dropped completely, it didn't provide much of a fast path anymore
> because of the whole Spectre and Meltdown stuff.  Also our patch made
> inline patches of the fast path, to remove the overhead of the checks.
> 
> I think after that four years we shoud revisit that change.  Do we still
> think x32 may be a problem?

Yes, I think so.  Some people try to sandbox applications with syscall
blacklists that don't take x32 into account.  This is obviously a bad
idea since new syscalls are constantly being added, but that's the
reality we have to work with.

> Does the patch introduce runtime latency due to it's not longer
> inline patched form?

Yes, a little.  Maybe I should update the slow path to use a static
key.

Ben.

-- 
Ben Hutchings
Sturgeon's Law: Ninety percent of everything is crap.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.debian.kernel

csiph-web