[PATCH] x86: kill handle_signal()->set_fs()

[PATCH] x86: kill handle_signal()->set_fs()

[Oleg Nesterov]

handle_signal()->set_fs() has a nice comment which explains what
set_fs() is, but it doesn't explain why it is needed and why it
depends on CONFIG_X86_64.

Afaics, the history of this confusion is:

	1. I guess today nobody can explain why it was needed
	   in arch/i386/kernel/signal.c, perhaps it was always
	   wrong. This predates 2.4.0 kernel.

	2. then it was copy-and-past'ed to the new x86_64 arch.

	3. then it was removed from i386 (but not from x86_64)
	   by b93b6ca3 "i386: remove unnecessary code".

	4. then it was reintroduced under CONFIG_X86_64 when x86
	   unified i386 and x86_64, because the patch above didn't
	   touch x86_64.

Remove it. ->addr_limit should be correct. Even if it was possible
that it is wrong, it is too late to fix it after setup_rt_frame().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---

 arch/x86/kernel/signal.c |    9 ---------
 1 file changed, 9 deletions(-)

--- ptrace/arch/x86/kernel/signal.c~1_kill_set_fs	2011-05-22 16:27:28.000000000 +0200
+++ ptrace/arch/x86/kernel/signal.c	2011-07-10 18:06:30.000000000 +0200
@@ -717,15 +717,6 @@ handle_signal(unsigned long sig, siginfo
 	if (ret)
 		return ret;
 
-#ifdef CONFIG_X86_64
-	/*
-	 * This has nothing to do with segment registers,
-	 * despite the name.  This magic affects uaccess.h
-	 * macros' behavior.  Reset it to the normal setting.
-	 */
-	set_fs(USER_DS);
-#endif
-
 	/*
 	 * Clear the direction flag as per the ABI for function entry.
 	 */

Re: [PATCH] x86: kill handle_signal()->set_fs()

[H. Peter Anvin]

On 07/10/2011 09:44 AM, Oleg Nesterov wrote:
> handle_signal()->set_fs() has a nice comment which explains what
> set_fs() is, but it doesn't explain why it is needed and why it
> depends on CONFIG_X86_64.
> 
> Afaics, the history of this confusion is:
> 
> 	1. I guess today nobody can explain why it was needed
> 	   in arch/i386/kernel/signal.c, perhaps it was always
> 	   wrong. This predates 2.4.0 kernel.
> 
> 	2. then it was copy-and-past'ed to the new x86_64 arch.
> 
> 	3. then it was removed from i386 (but not from x86_64)
> 	   by b93b6ca3 "i386: remove unnecessary code".
> 
> 	4. then it was reintroduced under CONFIG_X86_64 when x86
> 	   unified i386 and x86_64, because the patch above didn't
> 	   touch x86_64.
> 
> Remove it. ->addr_limit should be correct. Even if it was possible
> that it is wrong, it is too late to fix it after setup_rt_frame().
> 

The main reason I could think of why this would be necessary is if we
take an event while we have fs == KERNEL_DS inside the kernel which is
then promoted to a signal.  Are you absolutely sure that can't happen?

In particular, there should be a setting upstream of this, as you're
correctly pointing out that it's too late.  If not, we might actually
have a problem.

	-hpa

Re: [PATCH] x86: kill handle_signal()->set_fs()

[Oleg Nesterov]

On 07/10, H. Peter Anvin wrote:
>
> On 07/10/2011 09:44 AM, Oleg Nesterov wrote:
> > handle_signal()->set_fs() has a nice comment which explains what
> > set_fs() is, but it doesn't explain why it is needed and why it
> > depends on CONFIG_X86_64.
> >
> > Afaics, the history of this confusion is:
> >
> > 	1. I guess today nobody can explain why it was needed
> > 	   in arch/i386/kernel/signal.c, perhaps it was always
> > 	   wrong. This predates 2.4.0 kernel.
> >
> > 	2. then it was copy-and-past'ed to the new x86_64 arch.
> >
> > 	3. then it was removed from i386 (but not from x86_64)
> > 	   by b93b6ca3 "i386: remove unnecessary code".
> >
> > 	4. then it was reintroduced under CONFIG_X86_64 when x86
> > 	   unified i386 and x86_64, because the patch above didn't
> > 	   touch x86_64.
> >
> > Remove it. ->addr_limit should be correct. Even if it was possible
> > that it is wrong, it is too late to fix it after setup_rt_frame().
> >
>
> The main reason I could think of why this would be necessary is if we
> take an event while we have fs == KERNEL_DS inside the kernel

this is possible if we are the kernel thread, or set_fs(KERNEL_DS) was
called.

> which is
> then promoted to a signal.

How? We are going to return to the user-space. Obviously this is not
possible with the kernel thread. So I think this can only happen if
we already have a bug with unbalanced set_fs().

Are you absolutely sure that can't happen?

> In particular, there should be a setting upstream of this, as you're
> correctly pointing out that it's too late.  If not, we might actually
> have a problem.

Hmm... Now I recall, this was already discussed 5 years ago. Thanks to
google, see http://lkml.org/lkml/2007/7/17/321

In particular, Linus sayd:

	Heh. I think it's entirely historical.

	Please realize that the whole reason that function is called "set_fs()" is
	that it literally used to set the %fs segment register, not
	"->addr_limit".

	So I think the "set_fs(USER_DS)" is there _only_ to match the other

		regs->xds = __USER_DS;
		regs->xes = __USER_DS;
		regs->xss = __USER_DS;
		regs->xcs = __USER_CS;

	things, and never mattered. And now it matters even less, and has been
	copied to all other architectures where it is just totally insane.

Oleg.

Re: [PATCH] x86: kill handle_signal()->set_fs()

[Oleg Nesterov]

So, Peter, do you agree we can remove this set_fs() ?

Of course, this doesn't matter from the perfomance pov.
But it is very confusing, especially with CONFIG_X86_64.


On 07/10, Oleg Nesterov wrote:
>
> On 07/10, H. Peter Anvin wrote:
> >
> > On 07/10/2011 09:44 AM, Oleg Nesterov wrote:
> > > handle_signal()->set_fs() has a nice comment which explains what
> > > set_fs() is, but it doesn't explain why it is needed and why it
> > > depends on CONFIG_X86_64.
> > >
> > > Afaics, the history of this confusion is:
> > >
> > > 	1. I guess today nobody can explain why it was needed
> > > 	   in arch/i386/kernel/signal.c, perhaps it was always
> > > 	   wrong. This predates 2.4.0 kernel.
> > >
> > > 	2. then it was copy-and-past'ed to the new x86_64 arch.
> > >
> > > 	3. then it was removed from i386 (but not from x86_64)
> > > 	   by b93b6ca3 "i386: remove unnecessary code".
> > >
> > > 	4. then it was reintroduced under CONFIG_X86_64 when x86
> > > 	   unified i386 and x86_64, because the patch above didn't
> > > 	   touch x86_64.
> > >
> > > Remove it. ->addr_limit should be correct. Even if it was possible
> > > that it is wrong, it is too late to fix it after setup_rt_frame().
> > >
> >
> > The main reason I could think of why this would be necessary is if we
> > take an event while we have fs == KERNEL_DS inside the kernel
>
> this is possible if we are the kernel thread, or set_fs(KERNEL_DS) was
> called.
>
> > which is
> > then promoted to a signal.
>
> How? We are going to return to the user-space. Obviously this is not
> possible with the kernel thread. So I think this can only happen if
> we already have a bug with unbalanced set_fs().
>
> Are you absolutely sure that can't happen?
>
> > In particular, there should be a setting upstream of this, as you're
> > correctly pointing out that it's too late.  If not, we might actually
> > have a problem.
>
> Hmm... Now I recall, this was already discussed 5 years ago. Thanks to
> google, see http://lkml.org/lkml/2007/7/17/321
>
> In particular, Linus sayd:
>
> 	Heh. I think it's entirely historical.
>
> 	Please realize that the whole reason that function is called "set_fs()" is
> 	that it literally used to set the %fs segment register, not
> 	"->addr_limit".
>
> 	So I think the "set_fs(USER_DS)" is there _only_ to match the other
>
> 		regs->xds = __USER_DS;
> 		regs->xes = __USER_DS;
> 		regs->xss = __USER_DS;
> 		regs->xcs = __USER_CS;
>
> 	things, and never mattered. And now it matters even less, and has been
> 	copied to all other architectures where it is just totally insane.
>
> Oleg.

[tip:x86/signal] x86: Kill handle_signal()->set_fs()

[tip-bot for Oleg Nesterov]

Commit-ID:  73d382deccac186d103496bf10388bc2432a8384
Gitweb:     http://git.kernel.org/tip/73d382deccac186d103496bf10388bc2432a8384
Author:     Oleg Nesterov <oleg@redhat.com>
AuthorDate: Sun, 10 Jul 2011 18:44:24 +0200
Committer:  H. Peter Anvin <hpa@linux.intel.com>
CommitDate: Thu, 14 Jul 2011 21:46:20 -0700

x86: Kill handle_signal()->set_fs()

handle_signal()->set_fs() has a nice comment which explains what
set_fs() is, but it doesn't explain why it is needed and why it
depends on CONFIG_X86_64.

Afaics, the history of this confusion is:

	1. I guess today nobody can explain why it was needed
	   in arch/i386/kernel/signal.c, perhaps it was always
	   wrong. This predates 2.4.0 kernel.

	2. then it was copy-and-past'ed to the new x86_64 arch.

	3. then it was removed from i386 (but not from x86_64)
	   by b93b6ca3 "i386: remove unnecessary code".

	4. then it was reintroduced under CONFIG_X86_64 when x86
	   unified i386 and x86_64, because the patch above didn't
	   touch x86_64.

Remove it. ->addr_limit should be correct. Even if it was possible
that it is wrong, it is too late to fix it after setup_rt_frame().

Linus commented in:
http://lkml.kernel.org/r/alpine.LFD.0.999.0707170902570.19166@woody.linux-foundation.org

... about the equivalent bit from i386:

Heh. I think it's entirely historical.

Please realize that the whole reason that function is called "set_fs()" is 
that it literally used to set the %fs segment register, not 
"->addr_limit".

So I think the "set_fs(USER_DS)" is there _only_ to match the other

        regs->xds = __USER_DS;
        regs->xes = __USER_DS;
        regs->xss = __USER_DS;
        regs->xcs = __USER_CS;

things, and never mattered. And now it matters even less, and has been 
copied to all other architectures where it is just totally insane.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Link: http://lkml.kernel.org/r/20110710164424.GA20261@redhat.com
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
---
 arch/x86/kernel/signal.c |    9 ---------
 1 files changed, 0 insertions(+), 9 deletions(-)

diff --git a/arch/x86/kernel/signal.c b/arch/x86/kernel/signal.c
index 8c55f97..54ddaeb2 100644
--- a/arch/x86/kernel/signal.c
+++ b/arch/x86/kernel/signal.c
@@ -720,15 +720,6 @@ handle_signal(unsigned long sig, siginfo_t *info, struct k_sigaction *ka,
 	if (ret)
 		return ret;
 
-#ifdef CONFIG_X86_64
-	/*
-	 * This has nothing to do with segment registers,
-	 * despite the name.  This magic affects uaccess.h
-	 * macros' behavior.  Reset it to the normal setting.
-	 */
-	set_fs(USER_DS);
-#endif
-
 	/*
 	 * Clear the direction flag as per the ABI for function entry.
 	 */