[PATCH] Fix COMPAT_VDSO regression bug

[PATCH] Fix COMPAT_VDSO regression bug

[Leroy van Logchem]

    Revert "[PATCH] Fix CONFIG_COMPAT_VDSO"
    This reverts commit a1f3bb9ae4497a2ed3eac773fd7798ac33a0371f.

    Several systems couldnt boot using CONFIG_HIGHMEM64G=y as
    reported in bug #8040. Reverting the above patch solved the problem.


    Cc: Randy Dunlap <rdunlap@xenotime.net>
    Cc: Ingo Molnar <mingo@elte.hu>
    Cc: Roland McGrath <roland@redhat.com>
    Bisected-by: Leroy Raymond van Logchem <leroy.vanlogchem@gmail.com>


arch/i386/kernel/entry.S    |    4 ----
arch/i386/kernel/sysenter.c |    2 --
include/asm-i386/elf.h      |    7 ++++---
include/asm-i386/fixmap.h   |    2 --
include/asm-i386/page.h     |    2 --
5 files changed, 4 insertions(+), 13 deletions(-)

diff --git a/arch/i386/kernel/entry.S b/arch/i386/kernel/entry.S
index 5e47683..06461b8 100644
--- a/arch/i386/kernel/entry.S
+++ b/arch/i386/kernel/entry.S
@@ -302,16 +302,12 @@ sysenter_past_esp:
        pushl $(__USER_CS)
        CFI_ADJUST_CFA_OFFSET 4
        /*CFI_REL_OFFSET cs, 0*/
-#ifndef CONFIG_COMPAT_VDSO
        /*
         * Push current_thread_info()->sysenter_return to the stack.
         * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
         * pushed above; +8 corresponds to copy_thread's esp0 setting.
         */
        pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp)
-#else
-       pushl $SYSENTER_RETURN
-#endif
        CFI_ADJUST_CFA_OFFSET 4
        CFI_REL_OFFSET eip, 0

diff --git a/arch/i386/kernel/sysenter.c b/arch/i386/kernel/sysenter.c
index 666f70d..a1090e1 100644
--- a/arch/i386/kernel/sysenter.c
+++ b/arch/i386/kernel/sysenter.c
@@ -95,7 +95,6 @@ int __init sysenter_setup(void)
        return 0;
}

-#ifndef CONFIG_COMPAT_VDSO
static struct page *syscall_nopage(struct vm_area_struct *vma,
                                unsigned long adr, int *type)
{
@@ -190,4 +189,3 @@ int in_gate_area_no_task(unsigned long addr)
{
        return 0;
}
-#endif
diff --git a/include/asm-i386/elf.h b/include/asm-i386/elf.h
index 369035d..157bb7a 100644
--- a/include/asm-i386/elf.h
+++ b/include/asm-i386/elf.h
@@ -143,9 +143,12 @@ extern int dump_task_extended_fpu (struct
task_struct *, struct user_fxsr_struct
# define VDSO_PRELINK          0
#endif

-#define VDSO_SYM(x) \
+#define VDSO_COMPAT_SYM(x) \
                (VDSO_COMPAT_BASE + (unsigned long)(x) - VDSO_PRELINK)

+#define VDSO_SYM(x) \
+               (VDSO_BASE + (unsigned long)(x) - VDSO_PRELINK)
+
#define VDSO_HIGH_EHDR         ((const struct elfhdr *) VDSO_HIGH_BASE)
#define VDSO_EHDR              ((const struct elfhdr *) VDSO_COMPAT_BASE)

@@ -153,12 +156,10 @@ extern void __kernel_vsyscall;

#define VDSO_ENTRY             VDSO_SYM(&__kernel_vsyscall)

-#ifndef CONFIG_COMPAT_VDSO
#define ARCH_HAS_SETUP_ADDITIONAL_PAGES
struct linux_binprm;
extern int arch_setup_additional_pages(struct linux_binprm *bprm,
                                        int executable_stack);
-#endif

extern unsigned int vdso_enabled;

diff --git a/include/asm-i386/fixmap.h b/include/asm-i386/fixmap.h
index 3e9f610..02428cb 100644
--- a/include/asm-i386/fixmap.h
+++ b/include/asm-i386/fixmap.h
@@ -23,8 +23,6 @@
extern unsigned long __FIXADDR_TOP;
#else
#define __FIXADDR_TOP  0xfffff000
-#define FIXADDR_USER_START     __fix_to_virt(FIX_VDSO)
-#define FIXADDR_USER_END       __fix_to_virt(FIX_VDSO - 1)
#endif

#ifndef __ASSEMBLY__
diff --git a/include/asm-i386/page.h b/include/asm-i386/page.h
index 7b19f45..fd3f64a 100644
--- a/include/asm-i386/page.h
+++ b/include/asm-i386/page.h
@@ -143,9 +143,7 @@ extern int page_is_ram(unsigned long pagenr);
#include <asm-generic/memory_model.h>
#include <asm-generic/page.h>

-#ifndef CONFIG_COMPAT_VDSO
#define __HAVE_ARCH_GATE_AREA 1
-#endif
#endif /* __KERNEL__ */

#endif /* _I386_PAGE_H */

Re: [PATCH] Fix COMPAT_VDSO regression bug

[Roland McGrath]

I built a CONFIG_COMPAT_VDSO=y, CONFIG_HIGHMEM64G=y kernel and it has no
problems with FC-6 userland.  Everything looks fine with the vDSO.  
So either some more details of your kernel config are relevant, or
something about the userland usage pattern.


Thanks,
Roland

Re: [stable] [PATCH] Fix COMPAT_VDSO regression bug

[Greg KH]

On Thu, Mar 15, 2007 at 12:38:40AM +0100, Leroy van Logchem wrote:
> 
>     Revert "[PATCH] Fix CONFIG_COMPAT_VDSO"
>     This reverts commit a1f3bb9ae4497a2ed3eac773fd7798ac33a0371f.
> 
>     Several systems couldnt boot using CONFIG_HIGHMEM64G=y as
>     reported in bug #8040. Reverting the above patch solved the problem.

What stable version did you revert this in that solved your problem?

thanks,

greg k-h

Re: [stable] [PATCH] Fix COMPAT_VDSO regression bug

[Leroy van Logchem]

Greg KH wrote:
> On Thu, Mar 15, 2007 at 12:38:40AM +0100, Leroy van Logchem wrote:
>   
>>     Revert "[PATCH] Fix CONFIG_COMPAT_VDSO"
>>     This reverts commit a1f3bb9ae4497a2ed3eac773fd7798ac33a0371f.
>>
>>     Several systems couldnt boot using CONFIG_HIGHMEM64G=y as
>>     reported in bug #8040. Reverting the above patch solved the problem.
>>     
>
> What stable version did you revert this in that solved your problem?
>   
v2.6.20.3

--
Leroy

Re: [stable] [PATCH] Fix COMPAT_VDSO regression bug

[Leroy van Logchem]

Leroy van Logchem <leroy.vanlogchem <at> wldelft.nl> writes:

> 
> Greg KH wrote:
> > On Thu, Mar 15, 2007 at 12:38:40AM +0100, Leroy van Logchem wrote:
> >   
> >>     Revert "[PATCH] Fix CONFIG_COMPAT_VDSO"
> >>     This reverts commit a1f3bb9ae4497a2ed3eac773fd7798ac33a0371f.
> >>
> >>     Several systems couldnt boot using CONFIG_HIGHMEM64G=y as
> >>     reported in bug #8040. Reverting the above patch solved the problem.
> >>     
> >
> > What stable version did you revert this in that solved your problem?
> >   
> v2.6.20.3

From: Nilshar <nilshar <at> gmail.com>
Subject: Re: [Bug 8040] Hang before INIT when CONFIG_HIGHMEM4G=y
Date: 2007-03-15 10:02:14 GMT (4 minutes ago)

Hello,

I can confirm that the problem disappear when I disable Compat VDSO
support by setting CONFIG_COMPAT_VDSO=n.

Re: [PATCH] Fix COMPAT_VDSO regression bug

[Andrew Morton]

> On Thu, 15 Mar 2007 00:38:40 +0100 Leroy van Logchem <leroy.vanlogchem@wldelft.nl> wrote:
>     Revert "[PATCH] Fix CONFIG_COMPAT_VDSO"
>     This reverts commit a1f3bb9ae4497a2ed3eac773fd7798ac33a0371f.
> 
>     Several systems couldnt boot using CONFIG_HIGHMEM64G=y as
>     reported in bug #8040. Reverting the above patch solved the problem.

I think reverting it is probably the right thing to do, unless we can fix it for
real quite promptly.

But nobody will be able to apply this patch - your email client as really gone
to town on it - the results are phenomenally ugly.

And the patch wouldn't have applied to mainline anyway - there have been other changes.

Re: [PATCH] Fix COMPAT_VDSO regression bug

[Leroy van Logchem]

Andrew Morton <akpm <at> linux-foundation.org> writes:

> >     Revert "[PATCH] Fix CONFIG_COMPAT_VDSO"
> >     This reverts commit a1f3bb9ae4497a2ed3eac773fd7798ac33a0371f.
> > 
> >     Several systems couldnt boot using CONFIG_HIGHMEM64G=y as
> >     reported in bug #8040. Reverting the above patch solved the problem.

> I think reverting it is probably the right thing to do, unless we can fix
> it for real quite promptly.

Chuck Ebbert at redhat.com asked:

> Can you please double check this by trying with/without again -- sometimes
bisects go bad.

As requested I started to redo the test but now without git
using kernel.org tars. The results now are, still using the same .config:
linux-2.6.20.tar.gz  : bad
linux-2.6.20.1.tar.gz: bad (boot log equal)
linux-2.6.20.2.tar.gz: good
linux-2.6.20.3.tar.gz: good
(triple checked)

Chuck is right, the bisect gone bad.
I asked Nilshar to try these kernels too with:
COMPAT_VDSO=y
CONFIG_HIGHMEM64G=y

He did and says 2.6.20.3 works fine. So only 2.6.20 and 2.6.20.1 had
this 'hang' at boot behavior on my Supermicro 7044 while Nilshar's
machine started working with 2.6.20.3

Reverting avoided imo. I hope more people cheer up who reported bug #8040
and confirm it's fine with the latest stable.