x86_64 has 2 x arch_vma_name() - can we drop one?

x86_64 has 2 x arch_vma_name() - can we drop one?

[Sam Ravnborg]

In include/linux/mm.h arch_vma_name() is declared __weak.
This hide the fact that x86_64 has 2 implementations of
said function.

In arch/x86_64/mm/init.c:
const char *arch_vma_name(struct vm_area_struct *vma)
{
	if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
		return "[vdso]";
	if (vma == &gate_vma)
		return "[vsyscall]";
	return NULL;
}

And in arch/x86_64/ia32/syscall32.c:
const char *arch_vma_name(struct vm_area_struct *vma)
{
	if (vma->vm_start == VSYSCALL32_BASE &&
	    vma->vm_mm && vma->vm_mm->task_size == IA32_PAGE_OFFSET)
		return "[vdso]";
	return NULL;
}

As no comment were preceeding the fucntions this seems not to be on
purpose.
If I am correct which one should die?

The reason why this popped up is that the __weak definition in mm.h
causes problems on at least powerpc (trigger a binutils bug).
A similar bug is present at ia64 - but I have not confirmed if the
same fix is needed.

	Sam

Re: x86_64 has 2 x arch_vma_name() - can we drop one?

[Sam Ravnborg]

On Wed, Jul 25, 2007 at 08:31:32PM +0200, Sam Ravnborg wrote:
> In include/linux/mm.h arch_vma_name() is declared __weak.
> This hide the fact that x86_64 has 2 implementations of
> said function.
> 
> In arch/x86_64/mm/init.c:
> const char *arch_vma_name(struct vm_area_struct *vma)
> {
> 	if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
> 		return "[vdso]";
> 	if (vma == &gate_vma)
> 		return "[vsyscall]";
> 	return NULL;
> }
> 
> And in arch/x86_64/ia32/syscall32.c:
> const char *arch_vma_name(struct vm_area_struct *vma)
> {
> 	if (vma->vm_start == VSYSCALL32_BASE &&
> 	    vma->vm_mm && vma->vm_mm->task_size == IA32_PAGE_OFFSET)
> 		return "[vdso]";
> 	return NULL;
> }
> 
> As no comment were preceeding the fucntions this seems not to be on
> purpose.
> If I am correct which one should die?
> 
> The reason why this popped up is that the __weak definition in mm.h
> causes problems on at least powerpc (trigger a binutils bug).
> A similar bug is present at ia64 - but I have not confirmed if the
> same fix is needed.

I cooked up following patch - but I do not claim I understand this
code. It was done only by merging the functionality into one function.
I need this patch or a similar one to be added so the weak declaration
of arch_vma_name in mm.h can be made a normal prototype.

	Sam


[PATCH] x86_64: merge two identical named funtion into one

The function arch_vma_name() is declared weak and thus it was
not noticed that x86_64 had two almost identical implementations.

It was introduced in syscall32.c by: c633090e3105e779c97d4978e5e3d7d66b291cfb
It was introduced in mm/init.c by: 2aae950b21e4bc789d1fc6668faf67e8748300b7

Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
Cc: Roland McGrath <roland@redhat.com>
---
diff --git a/arch/x86_64/ia32/syscall32.c b/arch/x86_64/ia32/syscall32.c
index fc4419f..15013ba 100644
--- a/arch/x86_64/ia32/syscall32.c
+++ b/arch/x86_64/ia32/syscall32.c
@@ -49,14 +49,6 @@ int syscall32_setup_pages(struct linux_binprm *bprm, int exstack)
 	return ret;
 }
 
-const char *arch_vma_name(struct vm_area_struct *vma)
-{
-	if (vma->vm_start == VSYSCALL32_BASE &&
-	    vma->vm_mm && vma->vm_mm->task_size == IA32_PAGE_OFFSET)
-		return "[vdso]";
-	return NULL;
-}
-
 static int __init init_syscall32(void)
 { 
 	char *syscall32_page = (void *)get_zeroed_page(GFP_KERNEL);
diff --git a/arch/x86_64/mm/init.c b/arch/x86_64/mm/init.c
index 38f5d63..6f608d8 100644
--- a/arch/x86_64/mm/init.c
+++ b/arch/x86_64/mm/init.c
@@ -43,6 +43,7 @@
 #include <asm/proto.h>
 #include <asm/smp.h>
 #include <asm/sections.h>
+#include <asm/vsyscall32.h>
 
 #ifndef Dprintk
 #define Dprintk(x...)
@@ -734,6 +735,9 @@ const char *arch_vma_name(struct vm_area_struct *vma)
 {
 	if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 		return "[vdso]";
+	if (vma->vm_mm && vma->vm_start == VSYSCALL32_BASE &&
+	    vma->vm_mm->task_size == IA32_PAGE_OFFSET)
+		return "[vdso]";
 	if (vma == &gate_vma)
 		return "[vsyscall]";
 	return NULL;

Re: x86_64 has 2 x arch_vma_name() - can we drop one?

[Roland McGrath]

Your version should work, but I would do it a little differently.


Thanks,
Roland

---
[PATCH] x86_64: fix arch_vma_name

The function arch_vma_name() is declared weak and thus it was
not noticed that x86_64 had two almost identical implementations.

It was introduced in syscall32.c by: c633090e3105e779c97d4978e5e3d7d66b291cfb
It was introduced in mm/init.c by: 2aae950b21e4bc789d1fc6668faf67e8748300b7

Signed-off-by: Roland McGrath <roland@redhat.com>
CC: Sam Ravnborg <sam@ravnborg.org>
---
 arch/x86_64/ia32/ia32_binfmt.c |    5 +++--
 arch/x86_64/ia32/syscall32.c   |    8 --------
 2 files changed, 3 insertions(+), 10 deletions(-)

diff --git a/arch/x86_64/ia32/ia32_binfmt.c b/arch/x86_64/ia32/ia32_binfmt.c
index b70f3e7..90b37fc 100644
--- a/arch/x86_64/ia32/ia32_binfmt.c
+++ b/arch/x86_64/ia32/ia32_binfmt.c
@@ -41,8 +41,9 @@ int sysctl_vsyscall32 = 1;
 #undef ARCH_DLINFO
 #define ARCH_DLINFO do {  \
 	if (sysctl_vsyscall32) { \
-	NEW_AUX_ENT(AT_SYSINFO, (u32)(u64)VSYSCALL32_VSYSCALL); \
-	NEW_AUX_ENT(AT_SYSINFO_EHDR, VSYSCALL32_BASE);    \
+		current->mm->context.vdso = (void *)VSYSCALL32_BASE;	\
+		NEW_AUX_ENT(AT_SYSINFO, (u32)(u64)VSYSCALL32_VSYSCALL); \
+		NEW_AUX_ENT(AT_SYSINFO_EHDR, VSYSCALL32_BASE);    \
 	}	\
 } while(0)
 
diff --git a/arch/x86_64/ia32/syscall32.c b/arch/x86_64/ia32/syscall32.c
index fc4419f..6df8ec7 100644
--- a/arch/x86_64/ia32/syscall32.c
+++ b/arch/x86_64/ia32/syscall32.c
@@ -49,14 +49,6 @@ int syscall32_setup_pages(struct linux_binprm *bprm, int exstack)
 	return ret;
 }
 
-const char *arch_vma_name(struct vm_area_struct *vma)
-{
-	if (vma->vm_start == VSYSCALL32_BASE &&
-	    vma->vm_mm && vma->vm_mm->task_size == IA32_PAGE_OFFSET)
-		return "[vdso]";
-	return NULL;
-}
-
 static int __init init_syscall32(void)
 { 
 	char *syscall32_page = (void *)get_zeroed_page(GFP_KERNEL);

Re: x86_64 has 2 x arch_vma_name() - can we drop one?

[Sam Ravnborg]

On Wed, Jul 25, 2007 at 02:23:21PM -0700, Roland McGrath wrote:
> Your version should work, but I would do it a little differently.

Thanks, I hope to get this merged soon so we can fix Powerpc build.

	Sam