Skip to contentSkip to conversation

x86-64: Use normal ptregs stub for execve

PATCH In review Brian Gerst 3 participants 5 messages Jan 25, 2010 – Jan 27, 2010 x86

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86-64: Use normal ptregs stub for execve

Execve historically passed regs by value, so it had a seperate stub. This is no longer true, so change it to use a normal PTREGSCALL stub.

---
 arch/x86/kernel/entry_64.S |   17 +----------------
 1 files changed, 1 insertions(+), 16 deletions(-)
arch/x86/kernel/entry_64.S +1 -16
index 0697ff1..0c31502 100644
@@ -696,6 +696,7 @@ END(\label)
 	PTREGSCALL stub_clone, sys_clone, %r8
 	PTREGSCALL stub_fork, sys_fork, %rdi
 	PTREGSCALL stub_vfork, sys_vfork, %rdi
+	PTREGSCALL stub_execve, sys_execve, %rcx
 	PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
 	PTREGSCALL stub_iopl, sys_iopl, %rsi
 
@@ -712,22 +713,6 @@ ENTRY(ptregscall_common)
 	CFI_ENDPROC
 END(ptregscall_common)
 
-ENTRY(stub_execve)
-	CFI_STARTPROC
-	popq %r11
-	CFI_ADJUST_CFA_OFFSET -8
-	CFI_REGISTER rip, r11
-	SAVE_REST
-	FIXUP_TOP_OF_STACK %r11
-	movq %rsp, %rcx
-	call sys_execve
-	RESTORE_TOP_OF_STACK %r11
-	movq %rax,RAX(%rsp)
-	RESTORE_REST
-	jmp int_ret_from_sys_call
-	CFI_ENDPROC
-END(stub_execve)
-
 /*
  * sigreturn is special because it needs to restore all registers on return.
  * This cannot be done with SYSRET, so use the IRET return path instead.
Signature · 1 line
1.6.6
#1 lore raw
Commit-ID:  f19a25c872b20ae4357f72687cf3e7b0da220ee2
Gitweb:     http://git.kernel.org/tip/f19a25c872b20ae4357f72687cf3e7b0da220ee2
Author:     Brian Gerst <[email protected]>
AuthorDate: Mon, 25 Jan 2010 10:47:33 -0500
Committer:  H. Peter Anvin <[email protected]>
CommitDate: Tue, 26 Jan 2010 23:38:14 -0800

x86-64: Use normal ptregs stub for execve

Execve historically passed regs by value, so it had a seperate stub. This is no longer true, so change it to use a normal PTREGSCALL stub.

LKML-Reference: <[email protected]>

---
 arch/x86/kernel/entry_64.S |   17 +----------------
 1 files changed, 1 insertions(+), 16 deletions(-)
arch/x86/kernel/entry_64.S +1 -16
index 0697ff1..0c31502 100644
@@ -696,6 +696,7 @@ END(\label)
 	PTREGSCALL stub_clone, sys_clone, %r8
 	PTREGSCALL stub_fork, sys_fork, %rdi
 	PTREGSCALL stub_vfork, sys_vfork, %rdi
+	PTREGSCALL stub_execve, sys_execve, %rcx
 	PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
 	PTREGSCALL stub_iopl, sys_iopl, %rsi
 
@@ -712,22 +713,6 @@ ENTRY(ptregscall_common)
 	CFI_ENDPROC
 END(ptregscall_common)
 
-ENTRY(stub_execve)
-	CFI_STARTPROC
-	popq %r11
-	CFI_ADJUST_CFA_OFFSET -8
-	CFI_REGISTER rip, r11
-	SAVE_REST
-	FIXUP_TOP_OF_STACK %r11
-	movq %rsp, %rcx
-	call sys_execve
-	RESTORE_TOP_OF_STACK %r11
-	movq %rax,RAX(%rsp)
-	RESTORE_REST
-	jmp int_ret_from_sys_call
-	CFI_ENDPROC
-END(stub_execve)
-
 /*
  * sigreturn is special because it needs to restore all registers on return.
  * This cannot be done with SYSRET, so use the IRET return path instead.
#2 lore raw
Quoted 12 lines · tip-bot for Brian Gerst
Commit-ID:  f19a25c872b20ae4357f72687cf3e7b0da220ee2
Gitweb:     http://git.kernel.org/tip/f19a25c872b20ae4357f72687cf3e7b0da220ee2
Author:     Brian Gerst <[email protected]>
AuthorDate: Mon, 25 Jan 2010 10:47:33 -0500
Committer:  H. Peter Anvin <[email protected]>
CommitDate: Tue, 26 Jan 2010 23:38:14 -0800

x86-64: Use normal ptregs stub for execve

Execve historically passed regs by value, so it had a seperate stub. This is no longer true, so change it to use a normal PTREGSCALL stub.

Quoted 6 lines

LKML-Reference: <[email protected]>

---
 arch/x86/kernel/entry_64.S |   17 +----------------
 1 files changed, 1 insertions(+), 16 deletions(-)

FYI, -tip testing found that this commit breaks the grub binary on 64-bit, simply running /sbin/grub causes it to segfault:

[ 3145.999766] grub[6706] general protection ip:805c481 sp:ff95629c error:0 in 

grub[8048000+8c000]

i've excluded this commit for now.

Ingo

#3 lore raw
Quoted 25 lines · Ingo Molnar
Quoted 12 lines · tip-bot for Brian Gerst

Commit-ID:  f19a25c872b20ae4357f72687cf3e7b0da220ee2
Gitweb:     http://git.kernel.org/tip/f19a25c872b20ae4357f72687cf3e7b0da220ee2
Author:     Brian Gerst <[email protected]>
AuthorDate: Mon, 25 Jan 2010 10:47:33 -0500
Committer:  H. Peter Anvin <[email protected]>
CommitDate: Tue, 26 Jan 2010 23:38:14 -0800

x86-64: Use normal ptregs stub for execve

Execve historically passed regs by value, so it had a seperate stub. This is no longer true, so change it to use a normal PTREGSCALL stub.

Quoted 6 lines

LKML-Reference: <[email protected]>

---
 arch/x86/kernel/entry_64.S |   17 +----------------

 1 files changed, 1 insertions(+), 16 deletions(-)

FYI, -tip testing found that this commit breaks the grub binary on 64-bit, simply running /sbin/grub causes it to segfault:

Quoted 8 lines
[ 3145.999766] grub[6706] general protection ip:805c481 sp:ff95629c error:0 in

grub[8048000+8c000]

i've excluded this commit for now.

       Ingo

I see what is going on. Grub is a 32-bit statically linked executable even on a 64-bit distro. The old execve stub forced return to userspace via iret, but PTREGSCALL apparently doesn't. This means that the 32-bit %cs isn't getting set. Drop this patch for now.

--
Brian Gerst

#4 lore raw
Quoted 6 lines · Brian Gerst

I see what is going on. Grub is a 32-bit statically linked executable even on a 64-bit distro. The old execve stub forced return to userspace via iret, but PTREGSCALL apparently doesn't. This means that the 32-bit %cs isn't getting set. Drop this patch for now.

OK, I presume that breaks execing any 32-bit app, not just statically linked (eww...)

Yet one more case of Grub making me shudder, though :-/

-hpa

Signature · 2 lines
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.