Skip to contentSkip to conversation

x86: Let AS_WRUSS depend on X86_64

PATCH In review Jiri Slaby (SUSE) 4 participants 9 messages Oct 31, 2023

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86: Let AS_WRUSS depend on X86_64

Since commit 18e66b695e78 ("x86/shstk: Add Kconfig option for shadow stack"), AS_WRUSS is set even in 32-bit .configs. It is due to how Kbuild works. .config is not considered during make oldconfig (and other make *config), so standard (64-bit) gcc is invoked from 'as-instr' Kbuild tests. And such gcc indeed reports that wruss is supported, so AS_WRUSS=y is set.

Provided the wruss instruction is 64-bit only (and used in pure 64-bit X86_USER_SHADOW_STACK), it has little sense to have AS_WRUSS=y set on 32-bit.

Therefore, make the whole test dependent on X86_64 to ensure it's set only on 64-bit.

Cc Yu-cheng Yu <[email protected]>
Cc Rick Edgecombe <[email protected]>
Cc Dave Hansen <[email protected]>
Cc Borislav Petkov (AMD) <[email protected]>
Cc Kees Cook <[email protected]>
Cc Mike Rapoport (IBM) <[email protected]>
Cc Pengfei Xu <[email protected]>
Cc John Allen <[email protected]>
Cc Kees Cook <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
---
 arch/x86/Kconfig.assembler | 1 +
 1 file changed, 1 insertion(+)
arch/x86/Kconfig.assembler +1 -0
index 8ad41da301e5..a5b5241711e3 100644
@@ -27,5 +27,6 @@ config AS_GFNI
 
 config AS_WRUSS
 	def_bool $(as-instr,wrussq %rax$(comma)(%rbx))
+	depends on X86_64
 	help
 	  Supported by binutils >= 2.31 and LLVM integrated assembler
Signature · 1 line
2.42.0
maintainer
#1 lore raw

+ [email protected]

Quoted 46 lines · Jiri Slaby (SUSE)

Since commit 18e66b695e78 ("x86/shstk: Add Kconfig option for shadow stack"), AS_WRUSS is set even in 32-bit .configs. It is due to how Kbuild works. .config is not considered during make oldconfig (and other make *config), so standard (64-bit) gcc is invoked from 'as-instr' Kbuild tests. And such gcc indeed reports that wruss is supported, so AS_WRUSS=y is set.

Provided the wruss instruction is 64-bit only (and used in pure 64-bit X86_USER_SHADOW_STACK), it has little sense to have AS_WRUSS=y set on 32-bit.

Therefore, make the whole test dependent on X86_64 to ensure it's set only on 64-bit.

Cc Yu-cheng Yu <[email protected]>
Cc Rick Edgecombe <[email protected]>
Cc Dave Hansen <[email protected]>
Cc Borislav Petkov (AMD) <[email protected]>
Cc Kees Cook <[email protected]>
Cc Mike Rapoport (IBM) <[email protected]>
Cc Pengfei Xu <[email protected]>
Cc John Allen <[email protected]>
Cc Kees Cook <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
---
 arch/x86/Kconfig.assembler | 1 +
 1 file changed, 1 insertion(+)
arch/x86/Kconfig.assembler +1 -0
index 8ad41da301e5..a5b5241711e3 100644
@@ -27,5 +27,6 @@ config AS_GFNI
 
 config AS_WRUSS
 	def_bool $(as-instr,wrussq %rax$(comma)(%rbx))
+	depends on X86_64
 	help
 	  Supported by binutils >= 2.31 and LLVM integrated assembler
Signature · 1 line
2.42.0
Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#2 lore raw
Quoted 9 lines · Borislav Petkov

+ [email protected]

Quoted 5 lines · Jiri Slaby (SUSE)

Since commit 18e66b695e78 ("x86/shstk: Add Kconfig option for shadow stack"), AS_WRUSS is set even in 32-bit .configs. It is due to how Kbuild works. .config is not considered during make oldconfig (and other make *config), so standard (64-bit) gcc is invoked from 'as-instr' Kbuild tests.

I do not mind either way.

Please note "depends on X86_64" cannot prevent gcc from running here.

$(as-instr,wrussq %rax$(comma)(%rbx)) is replaced with 'y' while parsing the Kconfig files.

I want to change it in the future, but that is how Kconfig works now.

You don't save the cost of running the compiler.

Quoted 48 lines

And such gcc indeed reports that wruss is supported, so

Quoted 39 lines

AS_WRUSS=y is set.

Provided the wruss instruction is 64-bit only (and used in pure 64-bit X86_USER_SHADOW_STACK), it has little sense to have AS_WRUSS=y set on 32-bit.

Therefore, make the whole test dependent on X86_64 to ensure it's set only on 64-bit.

Cc Yu-cheng Yu <[email protected]>
Cc Rick Edgecombe <[email protected]>
Cc Dave Hansen <[email protected]>
Cc Borislav Petkov (AMD) <[email protected]>
Cc Kees Cook <[email protected]>
Cc Mike Rapoport (IBM) <[email protected]>
Cc Pengfei Xu <[email protected]>
Cc John Allen <[email protected]>
Cc Kees Cook <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
---
 arch/x86/Kconfig.assembler | 1 +
 1 file changed, 1 insertion(+)
arch/x86/Kconfig.assembler +1 -1
index 8ad41da301e5..a5b5241711e3 100644
@@ -27,5 +27,6 @@ config AS_GFNI
 
 config AS_WRUSS
      def_bool $(as-instr,wrussq %rax$(comma)(%rbx))
+     depends on X86_64
      help
        Supported by binutils >= 2.31 and LLVM integrated assembler
--
Quoted 2 lines

2.42.0

--
Regards/Gruss,
Boris.

https://people.kernel.org/tglx/notes-about-netiquette

Signature · 2 lines
Best Regards
Masahiro Yamada
maintainer
#3 lore raw
Quoted 6 lines · Masahiro Yamada

Please note "depends on X86_64" cannot prevent gcc from running here.

$(as-instr,wrussq %rax$(comma)(%rbx)) is replaced with 'y' while parsing the Kconfig files.

Then what is this fixing?

Jiri, what openSUSE kernel build issues were you referring to on IRC?

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#7 lore raw
In reply to Jiri Slaby (SUSE)

On 10/31/23 03:21, Jiri Slaby (SUSE) wrote:
...

Quoted 6 lines

Provided the wruss instruction is 64-bit only (and used in pure 64-bit X86_USER_SHADOW_STACK), it has little sense to have AS_WRUSS=y set on 32-bit.

Therefore, make the whole test dependent on X86_64 to ensure it's set only on 64-bit.

...

arch/x86/Kconfig.assembler config AS_GFNI · line 32

Quoted 5 lines

config AS_WRUSS

arch/x86/Kconfig.assembler +1 -0
 	def_bool $(as-instr,wrussq %rax$(comma)(%rbx))
+	depends on X86_64
 	help
 	  Supported by binutils >= 2.31 and LLVM integrated assembler

What's the downside to just leaving this alone?

This patch just seems wrong logically. Suppose some deranged person wanted 32-bit shadow stack support. They'd have to go hunt this down via trial and error instead of just enabling X86_USER_SHADOW_STACK.

Granted, that would take one crazy person five minutes to figure out why their .config is broken, but it still seems wrong. It's especially wrong without a comment because it logically reads something along the lines of "WRUSS is only available on x86_64 configs".

A better way to do this would be:

config HAS_SHADOW_STACKS

	depends on X86_64

config AS_WRUSS
...
# Avoid setting AS_WRUSS on configs that don't need it:
depends on HAS_SHADOW_STACKS

config X86_USER_SHADOW_STACK
bool "X86 userspace shadow stack"
depends on AS_WRUSS
depends on HAS_SHADOW_STACKS

But that honestly doesn't seem worth it because (circling back to the first thing I wrote...) I don't really know what the benefit is to doing this in the first place.

author
#4 lore raw
In reply to Borislav Petkov
Quoted 9 lines · Borislav Petkov
Quoted 6 lines · Masahiro Yamada

Please note "depends on X86_64" cannot prevent gcc from running here.

$(as-instr,wrussq %rax$(comma)(%rbx)) is replaced with 'y' while parsing the Kconfig files.

Then what is this fixing?

Quoted 1 line

Jiri, what openSUSE kernel build issues were you referring to on IRC?

As I wrote, we have a config which maintainers feed through oldconfig when updating the kernel. This time we got WRUSS=y. It was stored in git and later passed to build the kernel. And we received a different config from the build (WRUSS=n).

We diff the two and error out if the built kernel differs from the passed configuration. Which is the case here.

hth,

Signature · 1 line
js
maintainer
#5 lore raw
Quoted 7 lines · Jiri Slaby

As I wrote, we have a config which maintainers feed through oldconfig when updating the kernel. This time we got WRUSS=y. It was stored in git and later passed to build the kernel. And we received a different config from the build (WRUSS=n).

We diff the two and error out if the built kernel differs from the passed configuration. Which is the case here.

I sure hope that diffing script can ignore Kconfig items. Because this doesn't sound like a reason to "fix" the upstream kernel and as Masahiro says, it doesn't change anything anyway.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
author
#8 lore raw
In reply to Dave Hansen
Quoted 17 lines · Dave Hansen

On 10/31/23 03:21, Jiri Slaby (SUSE) wrote:
...

Quoted 6 lines

Provided the wruss instruction is 64-bit only (and used in pure 64-bit X86_USER_SHADOW_STACK), it has little sense to have AS_WRUSS=y set on 32-bit.

Therefore, make the whole test dependent on X86_64 to ensure it's set only on 64-bit.

...

arch/x86/Kconfig.assembler config AS_GFNI · line 32

Quoted 5 lines
  config AS_WRUSS
  	def_bool $(as-instr,wrussq %rax$(comma)(%rbx))
arch/x86/Kconfig.assembler +1 -0
+	depends on X86_64
  	help
  	  Supported by binutils >= 2.31 and LLVM integrated assembler

What's the downside to just leaving this alone?

Quoted 3 lines

This patch just seems wrong logically. Suppose some deranged person wanted 32-bit shadow stack support. They'd have to go hunt this down via trial and error instead of just enabling X86_USER_SHADOW_STACK.

All wrussq, rax, rbx can never be right on 32-bit anyway...

Quoted 4 lines

Granted, that would take one crazy person five minutes to figure out why their .config is broken, but it still seems wrong. It's especially wrong without a comment because it logically reads something along the lines of "WRUSS is only available on x86_64 configs".

Which is right, or what am I missing?

Quoted 10 lines

A better way to do this would be:

config HAS_SHADOW_STACKS

	depends on X86_64

config AS_WRUSS
...
# Avoid setting AS_WRUSS on configs that don't need it:
depends on HAS_SHADOW_STACKS

Quoted 8 lines

config X86_USER_SHADOW_STACK
bool "X86 userspace shadow stack"
depends on AS_WRUSS
depends on HAS_SHADOW_STACKS

But that honestly doesn't seem worth it because (circling back to the first thing I wrote...) I don't really know what the benefit is to doing this in the first place.

Again, to avoid nonsense in 32bit configs produced by oldconfig (as I noted in the commit log).

thanks,

Signature · 2 lines
js
suse labs
author
#6 lore raw
In reply to Borislav Petkov
Quoted 10 lines · Borislav Petkov
Quoted 7 lines · Jiri Slaby

As I wrote, we have a config which maintainers feed through oldconfig when updating the kernel. This time we got WRUSS=y. It was stored in git and later passed to build the kernel. And we received a different config from the build (WRUSS=n).

We diff the two and error out if the built kernel differs from the passed configuration. Which is the case here.

I sure hope that diffing script can ignore Kconfig items.

Definitely, I can add an ignore on that level.

Quoted 3 lines

Because this
doesn't sound like a reason to "fix" the upstream kernel and as Masahiro says, it doesn't change anything anyway.

It changes what's in .config and corresponds 1:1 to what is used during the build.

Now, I have a .config from "make oldconfig". But "make all" will read it and flips the value. (That is IMHO misleading, but I can live with that, provided I now know. Not sure about others though.)

thanks,

Signature · 2 lines
js
suse labs