2009-10-28 07:03:32

by Michael Cree

[permalink] [raw]
Subject: Alpha: suspicious compiler warning in entry.S

Compiler warning when compiling 2.6.32-rc5 as follows:

AS arch/alpha/kernel/entry.o
arch/alpha/kernel/entry.S: Assembler messages:
arch/alpha/kernel/entry.S:326: Warning: operand out of range
(0x0000000000000406 is not between 0x0000000000000000 and
0x00000000000000ff)
arch/alpha/kernel/entry.S:388: Warning: operand out of range
(0x0000000000000406 is not between 0x0000000000000000 and
0x00000000000000ff)

One offending line (326) is:

and $5, _TIF_WORK_MASK, $2

The warning occurs because commit d0420c83f39f "KEYS: Extend
TIF_NOTIFY_RESUME to (almost) all architectures [try #6]" introduces
TIF_NOTIFIY_RESUME which sets bit 10 in _TIF_WORK_MASK (see
arch/alpha/include/asm/thread_info.h) but _TIF_WORK_MASK is used in the
immediate addressing mode in the assembler instruction above. The 'and'
instruction can only take an immediate datum in the range 0 to 255. With
the addition of the TIF_NOTIFY_RESUME bit, _TIF_WORK_MASK is no longer in
the valid range for immediate addressing in the 'and' instruction.

On the assumption that it is important to include the TIF_NOTIFY_RESUME bit
one might be tempted to modify the code to:

lda $2, _TIF_WORK_MASK
and $5, $2, $2

but this is time critical code. I am wondering whether it might be better
to rearrange the bits in the thread information flags so that _TIF_WORK_MASK
has a value less than 255 and the single instruction to perform an and
operation can be maintained.

Comments?

Cheers
Michael.


2009-10-28 23:20:42

by Richard Henderson

[permalink] [raw]
Subject: Re: Alpha: suspicious compiler warning in entry.S

On 10/28/2009 12:03 AM, Michael Cree wrote:
> I am wondering whether it might be better
> to rearrange the bits in the thread information flags so that
> _TIF_WORK_MASK
> has a value less than 255 and the single instruction to perform an and
> operation can be maintained.

It would be better to rearrange TIF_NODIFY_RESUME down to 3.

In fact, you might as well rearrange the ones currently occupying
3-7 up to 8 and leave a hole for _TIF_WORK_MASK expansion, and
add a nice comment there at the same time.


r~

2009-10-30 13:17:07

by David Howells

[permalink] [raw]
Subject: Re: Alpha: suspicious compiler warning in entry.S

Michael Cree <[email protected]> wrote:

> On the assumption that it is important to include the TIF_NOTIFY_RESUME bit
> one might be tempted to modify the code to:
>
> lda $2, _TIF_WORK_MASK
> and $5, $2, $2
>
> but this is time critical code. I am wondering whether it might be better
> to rearrange the bits in the thread information flags so that _TIF_WORK_MASK
> has a value less than 255 and the single instruction to perform an and
> operation can be maintained.

Unless the LDA can be interleaved around some memory accesses, then yes,
compressing _TIF_WORK_MASK would be good.

David

2009-10-31 02:33:54

by Michael Cree

[permalink] [raw]
Subject: [PATCH] Alpha: Rearrange thread info flags fixing two regressions

The removal of the TIF_NOTIFY_RESUME flag, commit a583f1b54249b
"remove unused TIF_NOTIFY_RESUME flag," resulted in incorrect
setting of the unaligned access control flags by the prctl syscall.

The re-addition of the TIF_NOTIFY_RESUME flag, commit d0420c83f39f
"KEYS: Extend TIF_NOTIFY_RESUME to (almost) all architectures [try #6]"
further caused problems, namely incorrect operands to assembler code
as evidenced by:

AS arch/alpha/kernel/entry.o
arch/alpha/kernel/entry.S: Assembler messages:
arch/alpha/kernel/entry.S:326: Warning: operand out of range (0x0000000000000406 is not between 0x0000000000000000 and 0x00000000000000ff)

Both regressions fixed by (1) rearranging TIF_NOTIFY_RESUME flag to be
in lower 8 bits of the thread info flags, and (2) making sure that
ALPHA_UAC_SHIFT matches the rearrangement of the thread info flags.

Signed-off-by: Michael Cree <[email protected]>
---
arch/alpha/include/asm/thread_info.h | 27 +++++++++++++++------------
1 files changed, 15 insertions(+), 12 deletions(-)

diff --git a/arch/alpha/include/asm/thread_info.h b/arch/alpha/include/asm/thread_info.h
index e7a07f6..3ba4ded 100644
--- a/arch/alpha/include/asm/thread_info.h
+++ b/arch/alpha/include/asm/thread_info.h
@@ -61,21 +61,24 @@ register struct thread_info *__current_thread_info __asm__("$8");
/*
* Thread information flags:
* - these are process state flags and used from assembly
- * - pending work-to-be-done flags come first to fit in and immediate operand.
+ * - pending work-to-be-done flags come first and must be assigned to be
+ * within bits 0 to 7 to fit in and immediate operand.
+ * - ALPHA_UAC_SHIFT below must be kept consistent with the unaligned
+ * control flags.
*
* TIF_SYSCALL_TRACE is known to be 0 via blbs.
*/
#define TIF_SYSCALL_TRACE 0 /* syscall trace active */
-#define TIF_SIGPENDING 1 /* signal pending */
-#define TIF_NEED_RESCHED 2 /* rescheduling necessary */
-#define TIF_POLLING_NRFLAG 3 /* poll_idle is polling NEED_RESCHED */
-#define TIF_DIE_IF_KERNEL 4 /* dik recursion lock */
-#define TIF_UAC_NOPRINT 5 /* see sysinfo.h */
-#define TIF_UAC_NOFIX 6
-#define TIF_UAC_SIGBUS 7
-#define TIF_MEMDIE 8
-#define TIF_RESTORE_SIGMASK 9 /* restore signal mask in do_signal */
-#define TIF_NOTIFY_RESUME 10 /* callback before returning to user */
+#define TIF_NOTIFY_RESUME 1 /* callback before returning to user */
+#define TIF_SIGPENDING 2 /* signal pending */
+#define TIF_NEED_RESCHED 3 /* rescheduling necessary */
+#define TIF_POLLING_NRFLAG 8 /* poll_idle is polling NEED_RESCHED */
+#define TIF_DIE_IF_KERNEL 9 /* dik recursion lock */
+#define TIF_UAC_NOPRINT 10 /* see sysinfo.h */
+#define TIF_UAC_NOFIX 11
+#define TIF_UAC_SIGBUS 12
+#define TIF_MEMDIE 13
+#define TIF_RESTORE_SIGMASK 14 /* restore signal mask in do_signal */
#define TIF_FREEZE 16 /* is freezing for suspend */

#define _TIF_SYSCALL_TRACE (1<<TIF_SYSCALL_TRACE)
@@ -94,7 +97,7 @@ register struct thread_info *__current_thread_info __asm__("$8");
#define _TIF_ALLWORK_MASK (_TIF_WORK_MASK \
| _TIF_SYSCALL_TRACE)

-#define ALPHA_UAC_SHIFT 6
+#define ALPHA_UAC_SHIFT 10
#define ALPHA_UAC_MASK (1 << TIF_UAC_NOPRINT | 1 << TIF_UAC_NOFIX | \
1 << TIF_UAC_SIGBUS)

--
1.6.3.3

2009-12-01 04:06:48

by Matt Turner

[permalink] [raw]
Subject: Re: [PATCH] Alpha: Rearrange thread info flags fixing two regressions

On Fri, Oct 30, 2009 at 9:33 PM, Michael Cree <[email protected]> wrote:
> The removal of the TIF_NOTIFY_RESUME flag, commit a583f1b54249b
> "remove unused TIF_NOTIFY_RESUME flag," resulted in incorrect
> setting of the unaligned access control flags by the prctl syscall.
>
> The re-addition of the TIF_NOTIFY_RESUME flag, commit d0420c83f39f
> "KEYS: Extend TIF_NOTIFY_RESUME to (almost) all architectures [try #6]"
> further caused problems, namely incorrect operands to assembler code
> as evidenced by:
>
> AS ? ? ?arch/alpha/kernel/entry.o
> arch/alpha/kernel/entry.S: Assembler messages:
> arch/alpha/kernel/entry.S:326: Warning: operand out of range (0x0000000000000406 is not between 0x0000000000000000 and 0x00000000000000ff)
>
> Both regressions fixed by (1) rearranging TIF_NOTIFY_RESUME flag to be
> in lower 8 bits of the thread info flags, and (2) making sure that
> ALPHA_UAC_SHIFT matches the rearrangement of the thread info flags.
>
> Signed-off-by: Michael Cree <[email protected]>
> ---
> ?arch/alpha/include/asm/thread_info.h | ? 27 +++++++++++++++------------
> ?1 files changed, 15 insertions(+), 12 deletions(-)
>
> diff --git a/arch/alpha/include/asm/thread_info.h b/arch/alpha/include/asm/thread_info.h
> index e7a07f6..3ba4ded 100644
> --- a/arch/alpha/include/asm/thread_info.h
> +++ b/arch/alpha/include/asm/thread_info.h
> @@ -61,21 +61,24 @@ register struct thread_info *__current_thread_info __asm__("$8");
> ?/*
> ?* Thread information flags:
> ?* - these are process state flags and used from assembly
> - * - pending work-to-be-done flags come first to fit in and immediate operand.
> + * - pending work-to-be-done flags come first and must be assigned to be
> + * ? within bits 0 to 7 to fit in and immediate operand.
> + * - ALPHA_UAC_SHIFT below must be kept consistent with the unaligned
> + * ? control flags.
> ?*
> ?* TIF_SYSCALL_TRACE is known to be 0 via blbs.
> ?*/
> ?#define TIF_SYSCALL_TRACE ? ? ?0 ? ? ? /* syscall trace active */
> -#define TIF_SIGPENDING ? ? ? ? 1 ? ? ? /* signal pending */
> -#define TIF_NEED_RESCHED ? ? ? 2 ? ? ? /* rescheduling necessary */
> -#define TIF_POLLING_NRFLAG ? ? 3 ? ? ? /* poll_idle is polling NEED_RESCHED */
> -#define TIF_DIE_IF_KERNEL ? ? ?4 ? ? ? /* dik recursion lock */
> -#define TIF_UAC_NOPRINT ? ? ? ? ? ? ? ?5 ? ? ? /* see sysinfo.h */
> -#define TIF_UAC_NOFIX ? ? ? ? ?6
> -#define TIF_UAC_SIGBUS ? ? ? ? 7
> -#define TIF_MEMDIE ? ? ? ? ? ? 8
> -#define TIF_RESTORE_SIGMASK ? ?9 ? ? ? /* restore signal mask in do_signal */
> -#define TIF_NOTIFY_RESUME ? ? ?10 ? ? ?/* callback before returning to user */
> +#define TIF_NOTIFY_RESUME ? ? ?1 ? ? ? /* callback before returning to user */
> +#define TIF_SIGPENDING ? ? ? ? 2 ? ? ? /* signal pending */
> +#define TIF_NEED_RESCHED ? ? ? 3 ? ? ? /* rescheduling necessary */
> +#define TIF_POLLING_NRFLAG ? ? 8 ? ? ? /* poll_idle is polling NEED_RESCHED */
> +#define TIF_DIE_IF_KERNEL ? ? ?9 ? ? ? /* dik recursion lock */
> +#define TIF_UAC_NOPRINT ? ? ? ? ? ? ? ?10 ? ? ?/* see sysinfo.h */
> +#define TIF_UAC_NOFIX ? ? ? ? ?11
> +#define TIF_UAC_SIGBUS ? ? ? ? 12
> +#define TIF_MEMDIE ? ? ? ? ? ? 13
> +#define TIF_RESTORE_SIGMASK ? ?14 ? ? ?/* restore signal mask in do_signal */
> ?#define TIF_FREEZE ? ? ? ? ? ? 16 ? ? ?/* is freezing for suspend */
>
> ?#define _TIF_SYSCALL_TRACE ? ? (1<<TIF_SYSCALL_TRACE)
> @@ -94,7 +97,7 @@ register struct thread_info *__current_thread_info __asm__("$8");
> ?#define _TIF_ALLWORK_MASK ? ? ?(_TIF_WORK_MASK ? ? ? ? \
> ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? | _TIF_SYSCALL_TRACE)
>
> -#define ALPHA_UAC_SHIFT ? ? ? ? ? ? ? ?6
> +#define ALPHA_UAC_SHIFT ? ? ? ? ? ? ? ?10
> ?#define ALPHA_UAC_MASK ? ? ? ? (1 << TIF_UAC_NOPRINT | 1 << TIF_UAC_NOFIX | \
> ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? 1 << TIF_UAC_SIGBUS)
>
> --
> 1.6.3.3
>
> --

Applied to alpha-2.6.git. Thanks!

Matt

2009-12-01 12:31:28

by Christoph Hellwig

[permalink] [raw]
Subject: Re: [PATCH] Alpha: Rearrange thread info flags fixing two regressions

On Mon, Nov 30, 2009 at 11:06:52PM -0500, Matt Turner wrote:
> Applied to alpha-2.6.git. Thanks!

Can you push this to Linus ASAP so that 2.6.33 will work nicely out of
the box on alpha?

2009-12-01 15:58:43

by Matt Turner

[permalink] [raw]
Subject: Re: [PATCH] Alpha: Rearrange thread info flags fixing two regressions

On Tue, Dec 1, 2009 at 7:31 AM, Christoph Hellwig <[email protected]> wrote:
> On Mon, Nov 30, 2009 at 11:06:52PM -0500, Matt Turner wrote:
>> Applied to alpha-2.6.git. Thanks!
>
> Can you push this to Linus ASAP so that 2.6.33 will work nicely out of
> the box on alpha?

It's in mainline now. Linus just pulled my tree. :)

Matt