2015-07-23 18:56:31

by Andy Lutomirski

Subject: [PATCH] x86/entry/32: Initialize ss1 (SYSENTER_CS shadow) even on non-SEP CPUs

native_load_sp0 relies on this. I'm not sure why we haven't seen
reports of crashes. Maybe no one tests new kernels on non-SEP CPUs.

Signed-off-by: Andy Lutomirski <[email protected]>
---

Of course, I haven't tested on a non-SEP CPU either. Where would I
find one?

arch/x86/kernel/cpu/common.c | 19 +++++++++++--------
1 file changed, 11 insertions(+), 8 deletions(-)

diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index d27ab6e5e671..2181692c1eb3 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -993,6 +993,17 @@ void enable_sep_cpu(void)
cpu = get_cpu();
tss = &per_cpu(cpu_tss, cpu);

+#ifdef CONFIG_X86_32
+ /*
+ * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1
+ * field -- see the big comment in struct x86_hw_tss's
+ * definition. We need to initialize this even on non-SEP
+ * CPUs so that comparing ss1 to thread->sysenter_cs is
+ * always valid.
+ */
+ tss->x86_tss.ss1 = __KERNEL_CS;
+#endif
+
/*
* On 64-bit CPUs, enable SEP unconditionally. On Intel CPUs,
* it works and we use it. On AMD CPUs, the MSRs exist but EIP
@@ -1005,14 +1016,6 @@ void enable_sep_cpu(void)
if (IS_ENABLED(CONFIG_X86_32) && !boot_cpu_has(X86_FEATURE_SEP))
goto out;

-#ifdef CONFIG_X86_32
- /*
- * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
- * see the big comment in struct x86_hw_tss's definition.
- */
- tss->x86_tss.ss1 = __KERNEL_CS;
-#endif
-
wrmsrl_safe(MSR_IA32_SYSENTER_CS, __KERNEL_CS);
wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
(unsigned long)tss +
--
2.4.3

2015-07-23 19:03:08

by Brian Gerst

Subject: Re: [PATCH] x86/entry/32: Initialize ss1 (SYSENTER_CS shadow) even on non-SEP CPUs

On Thu, Jul 23, 2015 at 2:56 PM, Andy Lutomirski <[email protected]> wrote:
> native_load_sp0 relies on this. I'm not sure why we haven't seen
> reports of crashes. Maybe no one tests new kernels on non-SEP CPUs.

It's already statically initialized in cpu_tss.

--
Brian Gerst

2015-07-23 19:12:27

by Andy Lutomirski

Subject: Re: [PATCH] x86/entry/32: Initialize ss1 (SYSENTER_CS shadow) even on non-SEP CPUs

On Thu, Jul 23, 2015 at 12:03 PM, Brian Gerst <[email protected]> wrote:
> On Thu, Jul 23, 2015 at 2:56 PM, Andy Lutomirski <[email protected]> wrote:
>> native_load_sp0 relies on this. I'm not sure why we haven't seen
>> reports of crashes. Maybe no one tests new kernels on non-SEP CPUs.
>
> It's already statically initialized in cpu_tss.

Indeed. I'll just delete it, then.

>
> --
> Brian Gerst



--
Andy Lutomirski
AMA Capital Management, LLC