[PATCH v2] x86/asm/msr: Make wrmsrl() a function

[PATCH v2] x86/asm/msr: Make wrmsrl() a function

[Andy Lutomirski]

As of cf991de2f614 ("x86/asm/msr: Make wrmsrl_safe() a function"),
wrmsrl_safe is a function, but wrmsrl is still a macro.  The wrmsrl
macro performs invalid shifts if the value argument is 32 bits.
This makes it unnecessarily awkward to write code that puts an
unsigned long into an MSR.

To make this work, syscall_init needs tweaking to stop passing
a function pointer to wrmsrl.

Signed-off-by: Andy Lutomirski <luto@kernel.org>
---

Changes since v1: Fix one more warning.

arch/x86/include/asm/msr.h      | 6 ++++--
 arch/x86/include/asm/paravirt.h | 6 +++++-
 arch/x86/kernel/cpu/common.c    | 6 +++---
 3 files changed, 12 insertions(+), 6 deletions(-)

diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h
index 131eec2ca137..714c80755dae 100644
--- a/arch/x86/include/asm/msr.h
+++ b/arch/x86/include/asm/msr.h
@@ -185,8 +185,10 @@ static inline void wrmsr(unsigned msr, unsigned low, unsigned high)
 #define rdmsrl(msr, val)			\
 	((val) = native_read_msr((msr)))
 
-#define wrmsrl(msr, val)						\
-	native_write_msr((msr), (u32)((u64)(val)), (u32)((u64)(val) >> 32))
+static inline void wrmsrl(unsigned msr, u64 val)
+{
+	native_write_msr(msr, (u32)val, (u32)(val >> 32));
+}
 
 /* wrmsr with exception handling */
 static inline int wrmsr_safe(unsigned msr, unsigned low, unsigned high)
diff --git a/arch/x86/include/asm/paravirt.h b/arch/x86/include/asm/paravirt.h
index c2be0375bcad..10d0596433f8 100644
--- a/arch/x86/include/asm/paravirt.h
+++ b/arch/x86/include/asm/paravirt.h
@@ -153,7 +153,11 @@ do {						\
 	val = paravirt_read_msr(msr, &_err);	\
 } while (0)
 
-#define wrmsrl(msr, val)	wrmsr(msr, (u32)((u64)(val)), ((u64)(val))>>32)
+static inline void wrmsrl(unsigned msr, u64 val)
+{
+	wrmsr(msr, (u32)val, (u32)(val>>32));
+}
+
 #define wrmsr_safe(msr, a, b)	paravirt_write_msr(msr, a, b)
 
 /* rdmsr with exception handling */
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index ffb0020ada5f..e2ed2513a51e 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -1184,13 +1184,13 @@ void syscall_init(void)
 	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
 	 */
 	wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
-	wrmsrl(MSR_LSTAR, entry_SYSCALL_64);
+	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
 
 #ifdef CONFIG_IA32_EMULATION
-	wrmsrl(MSR_CSTAR, entry_SYSCALL_compat);
+	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
 	enable_sep_cpu();
 #else
-	wrmsrl(MSR_CSTAR, ignore_sysret);
+	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
-- 
2.4.3

[PATCH v2] x86/entry/32: Remove duplicate initialization of tss.ss1

[Andy Lutomirski]

It's statically initialized, so we don't need to dynamically
initialize it too.

Reported-by: Brian Gerst <brgerst@gmail.com>
Signed-off-by: Andy Lutomirski <luto@kernel.org>
---

Changes since v1: Delete the code :)

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

diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index e2ed2513a51e..e08eee98a5f8 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -1005,14 +1005,6 @@ 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

Re: [PATCH v2] x86/entry/32: Remove duplicate initialization of tss.ss1

[Ingo Molnar]

* Andy Lutomirski <luto@kernel.org> wrote:

> It's statically initialized, so we don't need to dynamically
> initialize it too.
> 
> Reported-by: Brian Gerst <brgerst@gmail.com>
> Signed-off-by: Andy Lutomirski <luto@kernel.org>
> ---
> 
> Changes since v1: Delete the code :)
> 
> arch/x86/kernel/cpu/common.c | 8 --------
>  1 file changed, 8 deletions(-)
> 
> diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
> index e2ed2513a51e..e08eee98a5f8 100644
> --- a/arch/x86/kernel/cpu/common.c
> +++ b/arch/x86/kernel/cpu/common.c
> @@ -1005,14 +1005,6 @@ 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 +

So this code changed substantially in tip:x86/asm - do we still need this patch?

Thanks,

	Ingo

Re: [PATCH v2] x86/entry/32: Remove duplicate initialization of tss.ss1

[Andy Lutomirski]

On Sun, Aug 23, 2015 at 4:26 AM, Ingo Molnar <mingo@kernel.org> wrote:
>
> * Andy Lutomirski <luto@kernel.org> wrote:
>
>> It's statically initialized, so we don't need to dynamically
>> initialize it too.
>>
>> Reported-by: Brian Gerst <brgerst@gmail.com>
>> Signed-off-by: Andy Lutomirski <luto@kernel.org>
>> ---
>>
>> Changes since v1: Delete the code :)
>>
>> arch/x86/kernel/cpu/common.c | 8 --------
>>  1 file changed, 8 deletions(-)
>>
>> diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
>> index e2ed2513a51e..e08eee98a5f8 100644
>> --- a/arch/x86/kernel/cpu/common.c
>> +++ b/arch/x86/kernel/cpu/common.c
>> @@ -1005,14 +1005,6 @@ 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 +
>
> So this code changed substantially in tip:x86/asm - do we still need this patch?
>

Yes, although I think it's actually the other way around -- I think
this patch may have applied on top of something that never made it
into tip/x86/asm.  I can re-check or I could just rebase the patch (or
you could apply it with the obvious fixup).

It's obviously not a critical fix.

--Andy

Re: [PATCH v2] x86/entry/32: Remove duplicate initialization of tss.ss1

[Ingo Molnar]

* Andy Lutomirski <luto@amacapital.net> wrote:

> On Sun, Aug 23, 2015 at 4:26 AM, Ingo Molnar <mingo@kernel.org> wrote:
> >
> > * Andy Lutomirski <luto@kernel.org> wrote:
> >
> >> It's statically initialized, so we don't need to dynamically
> >> initialize it too.
> >>
> >> Reported-by: Brian Gerst <brgerst@gmail.com>
> >> Signed-off-by: Andy Lutomirski <luto@kernel.org>
> >> ---
> >>
> >> Changes since v1: Delete the code :)
> >>
> >> arch/x86/kernel/cpu/common.c | 8 --------
> >>  1 file changed, 8 deletions(-)
> >>
> >> diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
> >> index e2ed2513a51e..e08eee98a5f8 100644
> >> --- a/arch/x86/kernel/cpu/common.c
> >> +++ b/arch/x86/kernel/cpu/common.c
> >> @@ -1005,14 +1005,6 @@ 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 +
> >
> > So this code changed substantially in tip:x86/asm - do we still need this patch?
> >
> 
> Yes, although I think it's actually the other way around -- I think
> this patch may have applied on top of something that never made it
> into tip/x86/asm.  I can re-check or I could just rebase the patch (or
> you could apply it with the obvious fixup).

Yeah, so I noticed that it fell amongst the cracks - please incorporate it into 
your next (v4.4 targeted) series once we get to that, so it doesn't get lost.

> It's obviously not a critical fix.

Yeah.

Thanks,

	Ingo

[tip:x86/asm] x86/asm/msr: Make wrmsrl() a function

[tip-bot for Andy Lutomirski]

Commit-ID:  47edb65178cb7056c2eea0b6c41a7d8c84547192
Gitweb:     http://git.kernel.org/tip/47edb65178cb7056c2eea0b6c41a7d8c84547192
Author:     Andy Lutomirski <luto@kernel.org>
AuthorDate: Thu, 23 Jul 2015 12:14:40 -0700
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Sun, 23 Aug 2015 13:25:38 +0200

x86/asm/msr: Make wrmsrl() a function

As of cf991de2f614 ("x86/asm/msr: Make wrmsrl_safe() a
function"), wrmsrl_safe is a function, but wrmsrl is still a
macro.  The wrmsrl macro performs invalid shifts if the value
argument is 32 bits. This makes it unnecessarily awkward to
write code that puts an unsigned long into an MSR.

To make this work, syscall_init needs tweaking to stop passing
a function pointer to wrmsrl.

Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Willy Tarreau <w@1wt.eu>
Link: http://lkml.kernel.org/r/690f0c629a1085d054e2d1ef3da073cfb3f7db92.1437678821.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 arch/x86/include/asm/msr.h      | 6 ++++--
 arch/x86/include/asm/paravirt.h | 6 +++++-
 arch/x86/kernel/cpu/common.c    | 6 +++---
 3 files changed, 12 insertions(+), 6 deletions(-)

diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h
index 54e9f08..77d8b28 100644
--- a/arch/x86/include/asm/msr.h
+++ b/arch/x86/include/asm/msr.h
@@ -188,8 +188,10 @@ static inline void wrmsr(unsigned msr, unsigned low, unsigned high)
 #define rdmsrl(msr, val)			\
 	((val) = native_read_msr((msr)))
 
-#define wrmsrl(msr, val)						\
-	native_write_msr((msr), (u32)((u64)(val)), (u32)((u64)(val) >> 32))
+static inline void wrmsrl(unsigned msr, u64 val)
+{
+	native_write_msr(msr, (u32)val, (u32)(val >> 32));
+}
 
 /* wrmsr with exception handling */
 static inline int wrmsr_safe(unsigned msr, unsigned low, unsigned high)
diff --git a/arch/x86/include/asm/paravirt.h b/arch/x86/include/asm/paravirt.h
index c2be037..10d0596 100644
--- a/arch/x86/include/asm/paravirt.h
+++ b/arch/x86/include/asm/paravirt.h
@@ -153,7 +153,11 @@ do {						\
 	val = paravirt_read_msr(msr, &_err);	\
 } while (0)
 
-#define wrmsrl(msr, val)	wrmsr(msr, (u32)((u64)(val)), ((u64)(val))>>32)
+static inline void wrmsrl(unsigned msr, u64 val)
+{
+	wrmsr(msr, (u32)val, (u32)(val>>32));
+}
+
 #define wrmsr_safe(msr, a, b)	paravirt_write_msr(msr, a, b)
 
 /* rdmsr with exception handling */
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index cb9e5df..b128808 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -1185,10 +1185,10 @@ void syscall_init(void)
 	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
 	 */
 	wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
-	wrmsrl(MSR_LSTAR, entry_SYSCALL_64);
+	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
 
 #ifdef CONFIG_IA32_EMULATION
-	wrmsrl(MSR_CSTAR, entry_SYSCALL_compat);
+	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
 	/*
 	 * This only works on Intel CPUs.
 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
@@ -1199,7 +1199,7 @@ void syscall_init(void)
 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
 #else
-	wrmsrl(MSR_CSTAR, ignore_sysret);
+	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);

Re: [tip:x86/asm] x86/asm/msr: Make wrmsrl() a function

[H. Peter Anvin]

On 08/23/2015 04:45 AM, tip-bot for Andy Lutomirski wrote:
> Commit-ID:  47edb65178cb7056c2eea0b6c41a7d8c84547192
> Gitweb:     http://git.kernel.org/tip/47edb65178cb7056c2eea0b6c41a7d8c84547192
> Author:     Andy Lutomirski <luto@kernel.org>
> AuthorDate: Thu, 23 Jul 2015 12:14:40 -0700
> Committer:  Ingo Molnar <mingo@kernel.org>
> CommitDate: Sun, 23 Aug 2015 13:25:38 +0200
> 
> x86/asm/msr: Make wrmsrl() a function
> 
> As of cf991de2f614 ("x86/asm/msr: Make wrmsrl_safe() a
> function"), wrmsrl_safe is a function, but wrmsrl is still a
> macro.  The wrmsrl macro performs invalid shifts if the value
> argument is 32 bits. This makes it unnecessarily awkward to
> write code that puts an unsigned long into an MSR.
> 

Looking at this: where do you see an invalid shift?  Everywhere I can
see we do the proper casting.  Still not side effect free, though.

	-hpa

Re: [tip:x86/asm] x86/asm/msr: Make wrmsrl() a function

[Andy Lutomirski]

On Sun, Aug 23, 2015 at 12:23 PM, H. Peter Anvin <hpa@zytor.com> wrote:
> On 08/23/2015 04:45 AM, tip-bot for Andy Lutomirski wrote:
>> Commit-ID:  47edb65178cb7056c2eea0b6c41a7d8c84547192
>> Gitweb:     http://git.kernel.org/tip/47edb65178cb7056c2eea0b6c41a7d8c84547192
>> Author:     Andy Lutomirski <luto@kernel.org>
>> AuthorDate: Thu, 23 Jul 2015 12:14:40 -0700
>> Committer:  Ingo Molnar <mingo@kernel.org>
>> CommitDate: Sun, 23 Aug 2015 13:25:38 +0200
>>
>> x86/asm/msr: Make wrmsrl() a function
>>
>> As of cf991de2f614 ("x86/asm/msr: Make wrmsrl_safe() a
>> function"), wrmsrl_safe is a function, but wrmsrl is still a
>> macro.  The wrmsrl macro performs invalid shifts if the value
>> argument is 32 bits. This makes it unnecessarily awkward to
>> write code that puts an unsigned long into an MSR.
>>
>
> Looking at this: where do you see an invalid shift?  Everywhere I can
> see we do the proper casting.  Still not side effect free, though.

I'm having trouble reproducing it right now, but IIRC at one point I
got wrmsrl to warn that a constant turned into zero as a result of a
shift.

--Andy

>
>         -hpa
>



-- 
Andy Lutomirski
AMA Capital Management, LLC