Problem when disabled interrupt in system call (ppc8270)

Problem when disabled interrupt in system call (ppc8270)

[wilbur.chan]

ppc 8270, kernel 2.6.21.7

I took the following steps:


In a system call function , say , sys_reboot, interrupt was disabled
by local_irq_disable.

Then ,  value at the address of 0xc000050  was set to a value , say ,
0x1234.  Code was like this :

sys_reboot()
{
local_irq_disable();
*(volatile unsigned long * )0xc1000050 = 0x1234;
 while(1)
   {
    ;
   }
}

Finally, I reset the board (with power still on)  into uboot,using
'md 0x1000050' to display the content at physical address 0x1000050,
and found that , it was not 0x1234.


However, if I delete the local_irq_disable() in sys_reboot, everything
went well---After I reset the board, 'md 0x1000050' return the value
0x1234.


So, this really puzzled me , could someone explain why this happed?  Thank you.



PS:

static inline unsigned long local_irq_disable(void)
{
	unsigned long flags, zero;

	__asm__ __volatile__("li %1,0; lbz %0,%2(13); stb %1,%2(13)"
	: "=r" (flags), "=&r" (zero)
	: "i" (offsetof(struct paca_struct, soft_enabled))
	: "memory");

	return flags;
}

Re: Problem when disabled interrupt in system call (ppc8270)

[Benjamin Herrenschmidt]

On Fri, 2009-10-16 at 09:12 +0800, wilbur.chan wrote:

> static inline unsigned long local_irq_disable(void)
> {
> 	unsigned long flags, zero;
> 
> 	__asm__ __volatile__("li %1,0; lbz %0,%2(13); stb %1,%2(13)"
> 	: "=r" (flags), "=&r" (zero)
> 	: "i" (offsetof(struct paca_struct, soft_enabled))
> 	: "memory");
> 
> 	return flags;
> }

This is not the variant of local_irq_disable() used on that machine :-)

The above is the 64-bit version.

Ben.

Re: Problem when disabled interrupt in system call (ppc8270)

[wilbur.chan]

2009/10/16, Benjamin Herrenschmidt <benh@kernel.crashing.org>:
> On Fri, 2009-10-16 at 09:12 +0800, wilbur.chan wrote:
>
>> static inline unsigned long local_irq_disable(void)
>> {
>> 	unsigned long flags, zero;
>>
>> 	__asm__ __volatile__("li %1,0; lbz %0,%2(13); stb %1,%2(13)"
>> 	: "=r" (flags), "=&r" (zero)
>> 	: "i" (offsetof(struct paca_struct, soft_enabled))
>> 	: "memory");
>>
>> 	return flags;
>> }
>
> This is not the variant of local_irq_disable() used on that machine :-)
>
> The above is the 64-bit version.
>
> Ben.
Hmmm...

I just called  local_irq_disable in sys_reboot, but actually, I don't
know which

local_irq_disable  function it really called , So I just paste one of
them here ...

Re: Problem when disabled interrupt in system call (ppc8270)

[Gabriel Paubert]

On Fri, Oct 16, 2009 at 09:12:34AM +0800, wilbur.chan wrote:
> ppc 8270, kernel 2.6.21.7
> 
> I took the following steps:
> 
> 
> In a system call function , say , sys_reboot, interrupt was disabled
> by local_irq_disable.
> 
> Then ,  value at the address of 0xc000050  was set to a value , say ,
> 0x1234.  Code was like this :
> 
> sys_reboot()
> {
> local_irq_disable();
> *(volatile unsigned long * )0xc1000050 = 0x1234;
>  while(1)
>    {
>     ;
>    }
> }
> 
> Finally, I reset the board (with power still on)  into uboot,using
> 'md 0x1000050' to display the content at physical address 0x1000050,
> and found that , it was not 0x1234.
> 
> 
> However, if I delete the local_irq_disable() in sys_reboot, everything
> went well---After I reset the board, 'md 0x1000050' return the value
> 0x1234.
> 
> 
> So, this really puzzled me , could someone explain why this happed?  Thank you.

The 0x1234 stays in the data cache and is not written to memory. The hard reset
invalidates the caches and you read the previous value from memory.

If you don't disable interrupts, the cache line is ultimately written
back to memory and you see it after reset.

Just try inserting a dcbst (or dcbf) instruction for the address
of the 0x1234 value just after.

Something like asm volatile("dcbst 0,%0": : "r" (address) : "memory");

(Not even tried to compile it, but you get the idea).

	Gabriel