FW: spinlocks dont work

FW: spinlocks dont work

[RANDAZZO]

On a uniprocessor system, with config_smp NOT Defined...

Note the following example:

driver 'A' calls spin_lock_irqsave and gets through (but does not call
..unlock).
driver 'B' calls spin_lock_irqsave and gets through???

How can B get through if A did not unlock yet?

-Mike

-----Original Message-----
From: Richard B. Johnson [mailto:root@chaos.analogic.com]
Sent: Friday, February 13, 2004 3:25 PM
To: RANDAZZO@ddc-web.com
Subject: RE: spinlocks dont work


On Fri, 13 Feb 2004 RANDAZZO@ddc-web.com wrote:

> if I need to lock out in the ISR, and I haven't
> compiled for SMP, what can I use instead?
>
> Mike

spinlock_t mylock = SPIN_LOCK_UNLOCKED;

You put:
        unsigned int flags;

	spin_lock_irqsave(&mylock, flags);
	//
        // Critical section code
	//
        spin_unlock_irqrestore(&mylock flags);

...this construct around any critical section. If you
don't know how to write interrupt service routines and
insist upon enabling interrupts within the ISR, you
need the same thing in the ISR.


Whether or not it's compiled for SMP should not then make
any difference because the 'cli' in the macro will still
exist.

Cheers,
Dick Johnson
Penguin : Linux version 2.4.24 on an i686 machine (797.90 BogoMips).
            Note 96.31% of all statistics are fiction.

 
"This message may contain company proprietary information. If you are not
the intended recipient, any disclosure, copying, distribution or reliance on
the contents of this message is prohibited. If you received this message in
error, please delete and notify me."

Re: FW: spinlocks dont work

[Tim Hockin]

On Fri, Feb 13, 2004 at 04:12:08PM -0500, RANDAZZO@ddc-web.com wrote:
> On a uniprocessor system, with config_smp NOT Defined...
> 
> Note the following example:
> 
> driver 'A' calls spin_lock_irqsave and gets through (but does not call
> ..unlock).
> driver 'B' calls spin_lock_irqsave and gets through???
> 
> How can B get through if A did not unlock yet?

Because actual mutexing is a no-op without CONFIG_SMP.  What you have is a BUG.
Don't do that.

Re: FW: spinlocks dont work

[Brian Gerst]

RANDAZZO@ddc-web.com wrote:
> On a uniprocessor system, with config_smp NOT Defined...
> 
> Note the following example:
> 
> driver 'A' calls spin_lock_irqsave and gets through (but does not call
> ..unlock).
> driver 'B' calls spin_lock_irqsave and gets through???
> 
> How can B get through if A did not unlock yet?
> 

On UP, spinlocks are no-ops.

--
				Brian Gerst

Re: FW: spinlocks dont work

[Jamie Lokier]

Brian Gerst wrote:
> On UP, spinlocks are no-ops.

No they aren't.  They disable preemption.

-- Jamie

Re: FW: spinlocks dont work

[Jamie Lokier]

RANDAZZO@ddc-web.com wrote:
> Note the following example:
> 
> driver 'A' calls spin_lock_irqsave and gets through (but does not call
> ..unlock).
> driver 'B' calls spin_lock_irqsave and gets through???
> 
> How can B get through if A did not unlock yet?

Is that code from B an interrupt handler?
If yes, then A should have called spin_lock_irqsave()

Otherwise, is that code from B a softirq or tasklet or timer handler?
If yes, then A should have called spin_lock_bh()

Otherwise, that code from B cannot run unless you are calling
schedule() from A after taking the lock, which is a bug in your code.

-- Jamie