2001-10-06 03:50:11

by Thomas Hood

[permalink] [raw]
Subject: Question about rtc_lock

The file arch/i386/kernel/bootflag.c contains this:

-------------------------------------------------
static void __init sbf_write(u8 v)
{
if(sbf_port != -1)
{
v &= ~(1<<7);
if(!parity(v))
v|=1<<7;

spin_lock(&rtc_lock);
CMOS_WRITE(v, sbf_port);
spin_unlock(&rtc_lock);
}
}
--------------------------------------------------

Does this code run with irqs disabled, or should these
spinlocks be _irq ?

--
Thomas


2001-10-06 12:55:49

by Alan

[permalink] [raw]
Subject: Re: Question about rtc_lock

> spin_lock(&rtc_lock);
> CMOS_WRITE(v, sbf_port);
> spin_unlock(&rtc_lock);
>
> Does this code run with irqs disabled, or should these
> spinlocks be _irq ?

The CMOS isnt accessed from IRQ handlers

2001-10-06 13:06:50

by Thomas Hood

[permalink] [raw]
Subject: Re: Question about rtc_lock

On Sat, 2001-10-06 at 09:01, Alan Cox wrote:
> > spin_lock(&rtc_lock);
> > CMOS_WRITE(v, sbf_port);
> > spin_unlock(&rtc_lock);
> >
> > Does this code run with irqs disabled, or should these
> > spinlocks be _irq ?
>
> The CMOS isnt accessed from IRQ handlers

No, but what if the rtc interrupts while the lock is held in this
bit of code?

Thomas


2001-10-06 13:08:11

by Alan

[permalink] [raw]
Subject: Re: Question about rtc_lock

> > The CMOS isnt accessed from IRQ handlers
>
> No, but what if the rtc interrupts while the lock is held in this
> bit of code?

Thats fine. It wont take the lock

2001-10-06 14:41:08

by Thomas Hood

[permalink] [raw]
Subject: Re: Question about rtc_lock

On Sat, 2001-10-06 at 09:13, Alan Cox wrote:
> > No, but what if the rtc interrupts while the lock is held in this
> > bit of code?
>
> Thats fine. It wont take the lock

But the first line of irq_interrupt() is:
spin_lock (&rtc_lock);

If one has a multi-processor machine, and CPUx is going through
the bootflag code, which takes the rtc_lock, and that CPU is
interrupted and enters rtc_interrupt(), which tries to take the
rtc_lock, won't it deadlock?

If not, then I'm missing some clue about how these spinlocks work.

Thomas

2001-10-06 15:24:23

by Jonathan Lundell

[permalink] [raw]
Subject: Re: Question about rtc_lock

At 10:40 AM -0400 2001-10-06, Thomas Hood wrote:
>On Sat, 2001-10-06 at 09:13, Alan Cox wrote:
>> > No, but what if the rtc interrupts while the lock is held in this
>> > bit of code?
>>
>> Thats fine. It wont take the lock
>
>But the first line of irq_interrupt() is:
> spin_lock (&rtc_lock);
>
>If one has a multi-processor machine, and CPUx is going through
>the bootflag code, which takes the rtc_lock, and that CPU is
>interrupted and enters rtc_interrupt(), which tries to take the
>rtc_lock, won't it deadlock?
>
>If not, then I'm missing some clue about how these spinlocks work.

rtc_interrupt(), you mean.

Even if there weren't current interrupt code doing CMOS accesses, it
would seem prudent to assume that there might be eventually, the
RTC/NVRAM being a multi-purpose shared resource.
--
/Jonathan Lundell.

2001-10-07 03:24:22

by Thomas Hood

[permalink] [raw]
Subject: Re: Question about rtc_lock

On Sat, 2001-10-06 at 11:24, Jonathan Lundell wrote:
> rtc_interrupt(), you mean.

Right.

> Even if there weren't current interrupt code doing CMOS accesses, it
> would seem prudent to assume that there might be eventually, the
> RTC/NVRAM being a multi-purpose shared resource.

I'm not concerned about an irq handler (present or future)
interfering with us as we write to the CMOS RAM. What I'm
concerned about is getting a rtc interrupt while we hold rtc_lock,
with deadlock being the result (since rtc_interrupt will spin on
the lock).

Either (1) we need to change these spinlocks to _irq, or (2) we
need to know that this bit of code runs only with irqs disabled.
My question is: Is it (1) or (2)?

Or is it (3) Thomas Hood is failing to understand something here?

Assuming the answer is (1), I append a patch that changes the
spinlock calls to _irqsave versions.

Cheers,
Thomas

The patch:
--- linux-2.4.10-ac5-fix/arch/i386/kernel/bootflag.c_PREV Fri Oct 5 23:20:43 2001
+++ linux-2.4.10-ac5-fix/arch/i386/kernel/bootflag.c Sat Oct 6 23:15:33 2001
@@ -81,26 +81,30 @@

static void __init sbf_write(u8 v)
{
+ unsigned long flags;
+
if(sbf_port != -1)
{
v &= ~(1<<7);
if(!parity(v))
v|=1<<7;

- spin_lock(&rtc_lock);
+ spin_lock_irqsave(&rtc_lock, flags);
CMOS_WRITE(v, sbf_port);
- spin_unlock(&rtc_lock);
+ spin_unlock_irqrestore(&rtc_lock, flags);
}
}

static u8 __init sbf_read(void)
{
u8 v;
+ unsigned long flags;
+
if(sbf_port == -1)
return 0;
- spin_lock(&rtc_lock);
+ spin_lock_irqsave(&rtc_lock, flags);
v = CMOS_READ(sbf_port);
- spin_unlock(&rtc_lock);
+ spin_unlock_irqrestore(&rtc_lock, flags);
return v;
}