[PATCH] arm64: Minor refactoring of cpu_switch_to() to fix build breakage

[PATCH] arm64: Minor refactoring of cpu_switch_to() to fix build breakage

[Olof Johansson]

Commit 0c8c0f03e3a2 ("x86/fpu, sched: Dynamically allocate 'struct fpu'")
moved the thread_struct to the bottom of task_struct. As a result, the
offset is now too large to be used in an immediate add on arm64 with
some kernel configs:

arch/arm64/kernel/entry.S: Assembler messages:
arch/arm64/kernel/entry.S:588: Error: immediate out of range
arch/arm64/kernel/entry.S:597: Error: immediate out of range

There's really no reason for cpu_switch_to to take a task_struct pointer
in the first place, since all it does is access the thread.cpu_context
member. So, just pass that in directly.

Fixes: 0c8c0f03e3a2 ("x86/fpu, sched: Dynamically allocate 'struct fpu'")
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Signed-off-by: Olof Johansson <olof@lixom.net>
---
 arch/arm64/include/asm/processor.h |    4 ++--
 arch/arm64/kernel/asm-offsets.c    |    2 --
 arch/arm64/kernel/entry.S          |   34 ++++++++++++++++------------------
 arch/arm64/kernel/process.c        |    3 ++-
 4 files changed, 20 insertions(+), 23 deletions(-)

diff --git a/arch/arm64/include/asm/processor.h b/arch/arm64/include/asm/processor.h
index e4c893e..ba90764 100644
--- a/arch/arm64/include/asm/processor.h
+++ b/arch/arm64/include/asm/processor.h
@@ -152,8 +152,8 @@ static inline void cpu_relax(void)
 #define cpu_relax_lowlatency()                cpu_relax()
 
 /* Thread switching */
-extern struct task_struct *cpu_switch_to(struct task_struct *prev,
-					 struct task_struct *next);
+extern struct task_struct *cpu_switch_to(struct cpu_context *prev,
+					 struct cpu_context *next);
 
 #define task_pt_regs(p) \
 	((struct pt_regs *)(THREAD_START_SP + task_stack_page(p)) - 1)
diff --git a/arch/arm64/kernel/asm-offsets.c b/arch/arm64/kernel/asm-offsets.c
index c99701a..c9e13f6 100644
--- a/arch/arm64/kernel/asm-offsets.c
+++ b/arch/arm64/kernel/asm-offsets.c
@@ -39,8 +39,6 @@ int main(void)
   DEFINE(TI_TASK,		offsetof(struct thread_info, task));
   DEFINE(TI_CPU,		offsetof(struct thread_info, cpu));
   BLANK();
-  DEFINE(THREAD_CPU_CONTEXT,	offsetof(struct task_struct, thread.cpu_context));
-  BLANK();
   DEFINE(S_X0,			offsetof(struct pt_regs, regs[0]));
   DEFINE(S_X1,			offsetof(struct pt_regs, regs[1]));
   DEFINE(S_X2,			offsetof(struct pt_regs, regs[2]));
diff --git a/arch/arm64/kernel/entry.S b/arch/arm64/kernel/entry.S
index f860bfd..2216326 100644
--- a/arch/arm64/kernel/entry.S
+++ b/arch/arm64/kernel/entry.S
@@ -579,29 +579,27 @@ ENDPROC(el0_irq)
 /*
  * Register switch for AArch64. The callee-saved registers need to be saved
  * and restored. On entry:
- *   x0 = previous task_struct (must be preserved across the switch)
- *   x1 = next task_struct
+ *   x0 = previous cpu_context (must be preserved across the switch)
+ *   x1 = next cpu_context
  * Previous and next are guaranteed not to be the same.
  *
  */
 ENTRY(cpu_switch_to)
-	add	x8, x0, #THREAD_CPU_CONTEXT
 	mov	x9, sp
-	stp	x19, x20, [x8], #16		// store callee-saved registers
-	stp	x21, x22, [x8], #16
-	stp	x23, x24, [x8], #16
-	stp	x25, x26, [x8], #16
-	stp	x27, x28, [x8], #16
-	stp	x29, x9, [x8], #16
-	str	lr, [x8]
-	add	x8, x1, #THREAD_CPU_CONTEXT
-	ldp	x19, x20, [x8], #16		// restore callee-saved registers
-	ldp	x21, x22, [x8], #16
-	ldp	x23, x24, [x8], #16
-	ldp	x25, x26, [x8], #16
-	ldp	x27, x28, [x8], #16
-	ldp	x29, x9, [x8], #16
-	ldr	lr, [x8]
+	stp	x19, x20, [x0], #16		// store callee-saved registers
+	stp	x21, x22, [x0], #16
+	stp	x23, x24, [x0], #16
+	stp	x25, x26, [x0], #16
+	stp	x27, x28, [x0], #16
+	stp	x29, x9, [x0], #16
+	str	lr, [x0]
+	ldp	x19, x20, [x1], #16		// restore callee-saved registers
+	ldp	x21, x22, [x1], #16
+	ldp	x23, x24, [x1], #16
+	ldp	x25, x26, [x1], #16
+	ldp	x27, x28, [x1], #16
+	ldp	x29, x9, [x1], #16
+	ldr	lr, [x1]
 	mov	sp, x9
 	ret
 ENDPROC(cpu_switch_to)
diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c
index 223b093..6b9a09c 100644
--- a/arch/arm64/kernel/process.c
+++ b/arch/arm64/kernel/process.c
@@ -325,7 +325,8 @@ struct task_struct *__switch_to(struct task_struct *prev,
 	dsb(ish);
 
 	/* the actual thread switch */
-	last = cpu_switch_to(prev, next);
+	last = cpu_switch_to(&prev->thread.cpu_context,
+			     &next->thread.cpu_context);
 
 	return last;
 }
-- 
1.7.10.4

Re: [PATCH] arm64: Minor refactoring of cpu_switch_to() to fix build breakage

[Ingo Molnar]

* Olof Johansson <olof@lixom.net> wrote:

> Commit 0c8c0f03e3a2 ("x86/fpu, sched: Dynamically allocate 'struct fpu'")
> moved the thread_struct to the bottom of task_struct. As a result, the
> offset is now too large to be used in an immediate add on arm64 with
> some kernel configs:
> 
> arch/arm64/kernel/entry.S: Assembler messages:
> arch/arm64/kernel/entry.S:588: Error: immediate out of range
> arch/arm64/kernel/entry.S:597: Error: immediate out of range
> 
> There's really no reason for cpu_switch_to to take a task_struct pointer
> in the first place, since all it does is access the thread.cpu_context
> member. So, just pass that in directly.
> 
> Fixes: 0c8c0f03e3a2 ("x86/fpu, sched: Dynamically allocate 'struct fpu'")
> Cc: Dave Hansen <dave.hansen@linux.intel.com>
> Signed-off-by: Olof Johansson <olof@lixom.net>
> ---
>  arch/arm64/include/asm/processor.h |    4 ++--
>  arch/arm64/kernel/asm-offsets.c    |    2 --
>  arch/arm64/kernel/entry.S          |   34 ++++++++++++++++------------------
>  arch/arm64/kernel/process.c        |    3 ++-
>  4 files changed, 20 insertions(+), 23 deletions(-)

So why not pass in 'thread_struct' as the patch below does - it looks much simpler 
to me. This way the assembly doesn't have to be changed at all.

Thanks,

	Ingo

=====================================>

* Guenter <linux@roeck-us.net> wrote:

> On Sat, Jul 18, 2015 at 04:27:17PM -0700, Guenter wrote:
> > Hi,
> > 
> > Commit 0c8c0f03e3a2 ("x86/fpu, sched: Dynamically allocate 'struct fpu'")
> > causes s390 builds in mainline to fail as follows.
> > 
> > arch/s390/kernel/traps.c: Assembler messages:
> > arch/s390/kernel/traps.c:262: Error: operand out of range
> > 	(0x00000000000023e8 is not between 0x0000000000000000 and 0x0000000000000fff)
> > arch/s390/kernel/traps.c:300: Error: operand out of range
> > 	(0x00000000000023e8 is not between 0x0000000000000000 and 0x0000000000000fff)
> > 
> 
> Also:
> 
> arm64:allmodconfig:
> 
> arch/arm64/kernel/entry.S: Assembler messages:
> arch/arm64/kernel/entry.S:588: Error: immediate out of range
> arch/arm64/kernel/entry.S:597: Error: immediate out of range
> make[1]: *** [arch/arm64/kernel/entry.o] Error 1
> 
> I didn't bisect that one, but it looks like the cause is the same.

Hm, it looks like the new, increased offset of 'thread_struct' within 
'task_struct' goes over a limit that these instructions are able to support on 
arm64:

 arch/arm64/kernel/asm-offsets.c:  DEFINE(THREAD_CPU_CONTEXT,    offsetof(struct task_struct, thread.cpu_context));
 arch/arm64/kernel/entry.S:      add     x8, x0, #THREAD_CPU_CONTEXT
 arch/arm64/kernel/entry.S:      add     x8, x1, #THREAD_CPU_CONTEXT

If there's no instruction that can support such offset sizes then I suspect the 
straightforward fix would be to pass in thread_struct instead - like the patch 
below. That's a tiny bit cleaner for type encapsulation anyway.

Warning: it's not even build tested, but in case it works:

  Signed-off-by: Ingo Molnar <mingo@kernel.org>

Thanks,

	Ingo

================

 arch/arm64/include/asm/processor.h | 4 ++--
 arch/arm64/kernel/asm-offsets.c    | 2 +-
 arch/arm64/kernel/process.c        | 2 +-
 3 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/arch/arm64/include/asm/processor.h b/arch/arm64/include/asm/processor.h
index e4c893e54f01..890f84bb3b8c 100644
--- a/arch/arm64/include/asm/processor.h
+++ b/arch/arm64/include/asm/processor.h
@@ -152,8 +152,8 @@ static inline void cpu_relax(void)
 #define cpu_relax_lowlatency()                cpu_relax()
 
 /* Thread switching */
-extern struct task_struct *cpu_switch_to(struct task_struct *prev,
-					 struct task_struct *next);
+extern struct task_struct *cpu_switch_to(struct thread_struct *prev,
+					 struct thread_struct *next);
 
 #define task_pt_regs(p) \
 	((struct pt_regs *)(THREAD_START_SP + task_stack_page(p)) - 1)
diff --git a/arch/arm64/kernel/asm-offsets.c b/arch/arm64/kernel/asm-offsets.c
index c99701a34d7b..3785373c2369 100644
--- a/arch/arm64/kernel/asm-offsets.c
+++ b/arch/arm64/kernel/asm-offsets.c
@@ -39,7 +39,7 @@ int main(void)
   DEFINE(TI_TASK,		offsetof(struct thread_info, task));
   DEFINE(TI_CPU,		offsetof(struct thread_info, cpu));
   BLANK();
-  DEFINE(THREAD_CPU_CONTEXT,	offsetof(struct task_struct, thread.cpu_context));
+  DEFINE(THREAD_CPU_CONTEXT,	offsetof(struct thread_struct, cpu_context));
   BLANK();
   DEFINE(S_X0,			offsetof(struct pt_regs, regs[0]));
   DEFINE(S_X1,			offsetof(struct pt_regs, regs[1]));
diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c
index 223b093c9440..436e95bda1b2 100644
--- a/arch/arm64/kernel/process.c
+++ b/arch/arm64/kernel/process.c
@@ -325,7 +325,7 @@ struct task_struct *__switch_to(struct task_struct *prev,
 	dsb(ish);
 
 	/* the actual thread switch */
-	last = cpu_switch_to(prev, next);
+	last = cpu_switch_to(&prev.thread, &next.thread);
 
 	return last;
 }

Re: [PATCH] arm64: Minor refactoring of cpu_switch_to() to fix build breakage

[Will Deacon]

On Mon, Jul 20, 2015 at 08:36:47AM +0100, Ingo Molnar wrote:
> * Olof Johansson <olof@lixom.net> wrote:
> 
> > Commit 0c8c0f03e3a2 ("x86/fpu, sched: Dynamically allocate 'struct fpu'")
> > moved the thread_struct to the bottom of task_struct. As a result, the
> > offset is now too large to be used in an immediate add on arm64 with
> > some kernel configs:
> > 
> > arch/arm64/kernel/entry.S: Assembler messages:
> > arch/arm64/kernel/entry.S:588: Error: immediate out of range
> > arch/arm64/kernel/entry.S:597: Error: immediate out of range
> > 
> > There's really no reason for cpu_switch_to to take a task_struct pointer
> > in the first place, since all it does is access the thread.cpu_context
> > member. So, just pass that in directly.
> > 
> > Fixes: 0c8c0f03e3a2 ("x86/fpu, sched: Dynamically allocate 'struct fpu'")
> > Cc: Dave Hansen <dave.hansen@linux.intel.com>
> > Signed-off-by: Olof Johansson <olof@lixom.net>
> > ---
> >  arch/arm64/include/asm/processor.h |    4 ++--
> >  arch/arm64/kernel/asm-offsets.c    |    2 --
> >  arch/arm64/kernel/entry.S          |   34 ++++++++++++++++------------------
> >  arch/arm64/kernel/process.c        |    3 ++-
> >  4 files changed, 20 insertions(+), 23 deletions(-)
> 
> So why not pass in 'thread_struct' as the patch below does - it looks much
> simpler to me. This way the assembly doesn't have to be changed at all.

Unfortunately, neither of these approaches really work:

  - We need to return last from __switch_to, which means not corrupting
    x0 in cpu_switch_to and then having an ugly container_of to get back
    at the task_struct

  - ret_from_fork needs to pass the task_struct of prev to schedule_tail,
    so we have the same issue there

Patch below fixes things, but it's a shame we have to use an extra register
like this.

Will

--->8

diff --git a/arch/arm64/kernel/entry.S b/arch/arm64/kernel/entry.S
index f860bfda454a..e16351819fed 100644
--- a/arch/arm64/kernel/entry.S
+++ b/arch/arm64/kernel/entry.S
@@ -585,7 +585,8 @@ ENDPROC(el0_irq)
  *
  */
 ENTRY(cpu_switch_to)
-	add	x8, x0, #THREAD_CPU_CONTEXT
+	mov	x10, #THREAD_CPU_CONTEXT
+	add	x8, x0, x10
 	mov	x9, sp
 	stp	x19, x20, [x8], #16		// store callee-saved registers
 	stp	x21, x22, [x8], #16
@@ -594,7 +595,7 @@ ENTRY(cpu_switch_to)
 	stp	x27, x28, [x8], #16
 	stp	x29, x9, [x8], #16
 	str	lr, [x8]
-	add	x8, x1, #THREAD_CPU_CONTEXT
+	add	x8, x1, x10
 	ldp	x19, x20, [x8], #16		// restore callee-saved registers
 	ldp	x21, x22, [x8], #16
 	ldp	x23, x24, [x8], #16

Re: [PATCH] arm64: Minor refactoring of cpu_switch_to() to fix build breakage

[Guenter Roeck]

On Mon, Jul 20, 2015 at 11:53:45AM +0100, Will Deacon wrote:
> On Mon, Jul 20, 2015 at 08:36:47AM +0100, Ingo Molnar wrote:
> > * Olof Johansson <olof@lixom.net> wrote:
> > 
> > > Commit 0c8c0f03e3a2 ("x86/fpu, sched: Dynamically allocate 'struct fpu'")
> > > moved the thread_struct to the bottom of task_struct. As a result, the
> > > offset is now too large to be used in an immediate add on arm64 with
> > > some kernel configs:
> > > 
> > > arch/arm64/kernel/entry.S: Assembler messages:
> > > arch/arm64/kernel/entry.S:588: Error: immediate out of range
> > > arch/arm64/kernel/entry.S:597: Error: immediate out of range
> > > 
> > > There's really no reason for cpu_switch_to to take a task_struct pointer
> > > in the first place, since all it does is access the thread.cpu_context
> > > member. So, just pass that in directly.
> > > 
> > > Fixes: 0c8c0f03e3a2 ("x86/fpu, sched: Dynamically allocate 'struct fpu'")
> > > Cc: Dave Hansen <dave.hansen@linux.intel.com>
> > > Signed-off-by: Olof Johansson <olof@lixom.net>
> > > ---
> > >  arch/arm64/include/asm/processor.h |    4 ++--
> > >  arch/arm64/kernel/asm-offsets.c    |    2 --
> > >  arch/arm64/kernel/entry.S          |   34 ++++++++++++++++------------------
> > >  arch/arm64/kernel/process.c        |    3 ++-
> > >  4 files changed, 20 insertions(+), 23 deletions(-)
> > 
> > So why not pass in 'thread_struct' as the patch below does - it looks much
> > simpler to me. This way the assembly doesn't have to be changed at all.
> 
> Unfortunately, neither of these approaches really work:
> 
>   - We need to return last from __switch_to, which means not corrupting
>     x0 in cpu_switch_to and then having an ugly container_of to get back
>     at the task_struct
> 
>   - ret_from_fork needs to pass the task_struct of prev to schedule_tail,
>     so we have the same issue there
> 
Confirmed; both Ingo's patch (after fixing it up) and Olof's patch
fail my qemu tests (qemu hangs with both patches and does not produce
any console output).

> Patch below fixes things, but it's a shame we have to use an extra register
> like this.
> 
Yes, your patch works, at least with my qemu tests, and the allmodconfig build
no longer fails.

Tested-by: Guenter Roeck <linux@roeck-us.net>

> Will
> 
> --->8
> 
> diff --git a/arch/arm64/kernel/entry.S b/arch/arm64/kernel/entry.S
> index f860bfda454a..e16351819fed 100644
> --- a/arch/arm64/kernel/entry.S
> +++ b/arch/arm64/kernel/entry.S
> @@ -585,7 +585,8 @@ ENDPROC(el0_irq)
>   *
>   */
>  ENTRY(cpu_switch_to)
> -	add	x8, x0, #THREAD_CPU_CONTEXT
> +	mov	x10, #THREAD_CPU_CONTEXT
> +	add	x8, x0, x10
>  	mov	x9, sp
>  	stp	x19, x20, [x8], #16		// store callee-saved registers
>  	stp	x21, x22, [x8], #16
> @@ -594,7 +595,7 @@ ENTRY(cpu_switch_to)
>  	stp	x27, x28, [x8], #16
>  	stp	x29, x9, [x8], #16
>  	str	lr, [x8]
> -	add	x8, x1, #THREAD_CPU_CONTEXT
> +	add	x8, x1, x10
>  	ldp	x19, x20, [x8], #16		// restore callee-saved registers
>  	ldp	x21, x22, [x8], #16
>  	ldp	x23, x24, [x8], #16

Re: [PATCH] arm64: Minor refactoring of cpu_switch_to() to fix build breakage

[Olof Johansson]

On Mon, Jul 20, 2015 at 7:20 AM, Guenter Roeck <linux@roeck-us.net> wrote:
> On Mon, Jul 20, 2015 at 11:53:45AM +0100, Will Deacon wrote:
>> On Mon, Jul 20, 2015 at 08:36:47AM +0100, Ingo Molnar wrote:
>> > * Olof Johansson <olof@lixom.net> wrote:
>> >
>> > > Commit 0c8c0f03e3a2 ("x86/fpu, sched: Dynamically allocate 'struct fpu'")
>> > > moved the thread_struct to the bottom of task_struct. As a result, the
>> > > offset is now too large to be used in an immediate add on arm64 with
>> > > some kernel configs:
>> > >
>> > > arch/arm64/kernel/entry.S: Assembler messages:
>> > > arch/arm64/kernel/entry.S:588: Error: immediate out of range
>> > > arch/arm64/kernel/entry.S:597: Error: immediate out of range
>> > >
>> > > There's really no reason for cpu_switch_to to take a task_struct pointer
>> > > in the first place, since all it does is access the thread.cpu_context
>> > > member. So, just pass that in directly.
>> > >
>> > > Fixes: 0c8c0f03e3a2 ("x86/fpu, sched: Dynamically allocate 'struct fpu'")
>> > > Cc: Dave Hansen <dave.hansen@linux.intel.com>
>> > > Signed-off-by: Olof Johansson <olof@lixom.net>
>> > > ---
>> > >  arch/arm64/include/asm/processor.h |    4 ++--
>> > >  arch/arm64/kernel/asm-offsets.c    |    2 --
>> > >  arch/arm64/kernel/entry.S          |   34 ++++++++++++++++------------------
>> > >  arch/arm64/kernel/process.c        |    3 ++-
>> > >  4 files changed, 20 insertions(+), 23 deletions(-)
>> >
>> > So why not pass in 'thread_struct' as the patch below does - it looks much
>> > simpler to me. This way the assembly doesn't have to be changed at all.
>>
>> Unfortunately, neither of these approaches really work:
>>
>>   - We need to return last from __switch_to, which means not corrupting
>>     x0 in cpu_switch_to and then having an ugly container_of to get back
>>     at the task_struct
>>
>>   - ret_from_fork needs to pass the task_struct of prev to schedule_tail,
>>     so we have the same issue there
>>
> Confirmed; both Ingo's patch (after fixing it up) and Olof's patch
> fail my qemu tests (qemu hangs with both patches and does not produce
> any console output).
>
>> Patch below fixes things, but it's a shame we have to use an extra register
>> like this.
>>
> Yes, your patch works, at least with my qemu tests, and the allmodconfig build
> no longer fails.
>
> Tested-by: Guenter Roeck <linux@roeck-us.net>

Yep, my bad for not looking harder (and resurrecting my only arm64
test system that's currently down).

Thanks all. :)


-Olof