2009-07-30 16:37:49

by Michael S. Zick

[permalink] [raw]
Subject: [RFC] Uni-processor interrupt state preservation.

In kernel/up.c - -

int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
int wait)
{
unsigned long flags;

WARN_ON(cpu != 0);

local_irq_save(flags);
(func)(info);
local_irq_restore(flags);

return 0;
}

It **looks** reasonable - -
but:

.L2:
#APP
# ul native_save_fl(void)
pushf ; pop %ebx
# native_irq_disable(void)
cli
#NO_APP
#(insn:TI 29 27 30 (set (reg:SI 0 ax [ info ])
# (reg/v/f:SI 4 si [orig:62 info ] [62])) 34 {*movsi_1} (insn_list:REG_DEP_TRUE 27 (nil))
# (expr_list:REG_DEAD (reg/v/f:SI 4 si [orig:62 info ] [62])
# (nil)))
movl %esi, %eax # 29 *movsi_1/1 [length = 2]
#(call_insn:TI 30 29 43 (call (mem:QI (reg/v/f:SI 5 di [orig:61 func ] [61]) [0 S1 A8])
# (const_int 0 [0x0])) 386 {*call_1} (insn_list:REG_DEP_TRUE 29 (insn_list:REG_DEP_TRUE 27 (nil)))
# (expr_list:REG_DEAD (reg:SI 0 ax [ info ])
# (expr_list:REG_DEAD (reg/v/f:SI 5 di [orig:61 func ] [61])
# (nil)))
# (expr_list:REG_DEP_TRUE (use (reg:SI 0 ax [ info ]))
# (nil)))
call *%edi # 30 *call_1 [length = 2]
#APP
# native_restore_fl(ul flags)
push %ebx ; popf
#NO_APP

Just how is the interrupt state (in ebx) supposed to be preserved across the function call?
(assembly comments where added in macros, just to clarify where this is coming from.)

Mike


2009-07-30 16:56:00

by Roland Dreier

[permalink] [raw]
Subject: Re: [RFC] Uni-processor interrupt state preservation.


> Just how is the interrupt state (in ebx) supposed to be preserved across
> the function call?

Doesn't the i386 ABI say that ebx is preserved across function calls?

2009-07-30 17:02:35

by Michael S. Zick

[permalink] [raw]
Subject: Re: [RFC] Uni-processor interrupt state preservation.

On Thu July 30 2009, Roland Dreier wrote:
>
> > Just how is the interrupt state (in ebx) supposed to be preserved across
> > the function call?
>
> Doesn't the i386 ABI say that ebx is preserved across function calls?
>
>

If it does, then that is the answer to my question -
Link to the ABI document please.

Mike

2009-07-30 17:14:31

by Roland Dreier

[permalink] [raw]
Subject: Re: [RFC] Uni-processor interrupt state preservation.


> If it does, then that is the answer to my question -
> Link to the ABI document please.

http://www.sco.com/developers/devspecs/abi386-4.pdf:

Registers %ebp, %ebx, %edi, %esi, and %esp ‘‘belong’’ to the cal-
ling function. In other words, a called function must preserve these
registers’ values for its caller.

2009-07-30 17:52:48

by Michael S. Zick

[permalink] [raw]
Subject: Re: [RFC] Uni-processor interrupt state preservation.

On Thu July 30 2009, Roland Dreier wrote:
>
> > If it does, then that is the answer to my question -
> > Link to the ABI document please.
>
> http://www.sco.com/developers/devspecs/abi386-4.pdf:
>
> Registers %ebp, %ebx, %edi, %esi, and %esp ‘‘belong’’ to the cal-
> ling function. In other words, a called function must preserve these
> registers’ values for its caller.
>
>

Super!

Should not be too big a job to make sure that is indeed happening as expected.
Most uses of this function dispatcher are not built in Uni-processor builds.
Only a small handful of references.

Thanks,
Mike