Skip to contentSkip to conversation

fully honor vdso_enabled [i386, sh; x86_64?]

PATCH John Reiser 4 participants 4 messages Feb 22, 2007 – Feb 28, 2007 SuperH

lore
Share & tools
lore
#0 lore raw

Architectures such as i386, sh, x86_64 have a flag /proc/sys/vm/vdso_enabled to choose whether the kernel should setup a process to use vdso after execve(). Informing the user code via AT_SYSINFO* is controlled by macro ARCH_DLINFO in fs/binfmt_elf.c and include/asm-$ARCH/elf.h, but the vdso page is established always via arch_setup_additonal_pages() called from load_elf_binary(). If vdso_enabled is off, then current code wastes kernel time during execve() and fragments the address space unnecessarily.

This patch changes arch_setup_additonal_pages() to honor vdso_enabled. For i386 it also allows the option of a fixed addresss to avoid fragmenting the address space. Compiles and runs on i386. x86_64 [IA32 support] and sh maintainers also please comment.

For some related history, including interaction with exec-shield, see:
http://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=229304
and also 207020 and 162797.

arch/i386/kernel/sysenter.c +18 -1
index 13ca54a..f8c4d76 100644
@@ -22,6 +22,8 @@
 #include <asm/msr.h>
 #include <asm/pgtable.h>
 #include <asm/unistd.h>
+#include <asm/a.out.h>
+#include <asm/mman.h>
 
 /*
  * Should the kernel map a VDSO page into processes and pass its
@@ -105,10 +107,25 @@ int arch_setup_additional_pages(struct linux_binprm *bprm, int exstack)
 {
 	struct mm_struct *mm = current->mm;
 	unsigned long addr;
+	unsigned long flags;
 	int ret;
 
+	switch (vdso_enabled) {
+	case 0:  /* none */
+		return 0;
+	default:
+	case 1:  /* vdso in random available page */
+		addr = 0ul;
+		flags = 0ul;
+		break;
+	case 2:  /* out of user's way */
+		addr = STACK_TOP;
+		flags = MAP_FIXED;
+		break;
+	}
+
 	down_write(&mm->mmap_sem);
-	addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
+	addr = get_unmapped_area(NULL, addr, PAGE_SIZE, 0, flags);
 	if (IS_ERR_VALUE(addr)) {
 		ret = addr;
 		goto up_fail;
arch/sh/kernel/vsyscall/vsyscall.c +3 -0
index 7b0f66f..2b789fb 100644
@@ -64,6 +64,9 @@ int arch_setup_additional_pages(struct linux_binprm *bprm,
 	unsigned long addr;
 	int ret;
 
+	if (!vdso_enabled)
+		return 0;
+
 	down_write(&mm->mmap_sem);
 	addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
 	if (IS_ERR_VALUE(addr)) {
include/asm-i386/a.out.h +1 -1
index ab17bb8..9894f73 100644
@@ -19,7 +19,7 @@ struct exec
 
 #ifdef __KERNEL__
 
-#define STACK_TOP	TASK_SIZE
+#define STACK_TOP	(TASK_SIZE - PAGE_SIZE)  /* 1 page optional for vdso */
 
 #endif
 
Acked-by
#1 lore raw
Quoted 5 lines · John Reiser

This patch changes arch_setup_additonal_pages() to honor vdso_enabled. For i386 it also allows the option of a fixed addresss to avoid fragmenting the address space. Compiles and runs on i386. x86_64 [IA32 support] and sh maintainers also please comment.

We didn't actually have the sysctl entry wired up on SH, but once that's done, this patch works fine there too.

Andrew, do you want a separate patch for the vdso_enabled sysctl or is it more convenient through my git tree?

#3 lore raw
In reply to John Reiser

[adding Andi and Ingo]

Quoted 87 lines · John Reiser

Architectures such as i386, sh, x86_64 have a flag /proc/sys/vm/vdso_enabled to choose whether the kernel should setup a process to use vdso after execve(). Informing the user code via AT_SYSINFO* is controlled by macro ARCH_DLINFO in fs/binfmt_elf.c and include/asm-$ARCH/elf.h, but the vdso page is established always via arch_setup_additonal_pages() called from load_elf_binary(). If vdso_enabled is off, then current code wastes kernel time during execve() and fragments the address space unnecessarily.

This patch changes arch_setup_additonal_pages() to honor vdso_enabled. For i386 it also allows the option of a fixed addresss to avoid fragmenting the address space. Compiles and runs on i386. x86_64 [IA32 support] and sh maintainers also please comment.

For some related history, including interaction with exec-shield, see:
http://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=229304
and also 207020 and 162797.

arch/i386/kernel/sysenter.c +18 -1
index 13ca54a..f8c4d76 100644
@@ -22,6 +22,8 @@
 #include <asm/msr.h>
 #include <asm/pgtable.h>
 #include <asm/unistd.h>
+#include <asm/a.out.h>
+#include <asm/mman.h>
 
 /*
  * Should the kernel map a VDSO page into processes and pass its
@@ -105,10 +107,25 @@ int arch_setup_additional_pages(struct linux_binprm *bprm, int exstack)
 {
 	struct mm_struct *mm = current->mm;
 	unsigned long addr;
+	unsigned long flags;
 	int ret;
 
+	switch (vdso_enabled) {
+	case 0:  /* none */
+		return 0;
+	default:
+	case 1:  /* vdso in random available page */
+		addr = 0ul;
+		flags = 0ul;
+		break;
+	case 2:  /* out of user's way */
+		addr = STACK_TOP;
+		flags = MAP_FIXED;
+		break;
+	}
+
 	down_write(&mm->mmap_sem);
-	addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
+	addr = get_unmapped_area(NULL, addr, PAGE_SIZE, 0, flags);
 	if (IS_ERR_VALUE(addr)) {
 		ret = addr;
 		goto up_fail;
arch/sh/kernel/vsyscall/vsyscall.c +3 -0
index 7b0f66f..2b789fb 100644
@@ -64,6 +64,9 @@ int arch_setup_additional_pages(struct linux_binprm *bprm,
 	unsigned long addr;
 	int ret;
 
+	if (!vdso_enabled)
+		return 0;
+
 	down_write(&mm->mmap_sem);
 	addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
 	if (IS_ERR_VALUE(addr)) {
include/asm-i386/a.out.h +1 -1
index ab17bb8..9894f73 100644
@@ -19,7 +19,7 @@ struct exec
 
 #ifdef __KERNEL__
 
-#define STACK_TOP	TASK_SIZE
+#define STACK_TOP	(TASK_SIZE - PAGE_SIZE)  /* 1 page optional for vdso */
 
 #endif
 
#2 lore raw
In reply to Paul Mundt
Quoted 6 lines · On Wed, 28 Feb 2007 18:11:11 +0900 Paul Mundt <[email protected]> wrote:
Quoted 5 lines · John Reiser

This patch changes arch_setup_additonal_pages() to honor vdso_enabled. For i386 it also allows the option of a fixed addresss to avoid fragmenting the address space. Compiles and runs on i386. x86_64 [IA32 support] and sh maintainers also please comment.

Quoted 6 lines

We didn't actually have the sysctl entry wired up on SH, but once that's done, this patch works fine there too.

Andrew, do you want a separate patch for the vdso_enabled sysctl or is it more convenient through my git tree?

If it's an sh-only thing then through your tree is fine, thanks.