Re: [ros-svn] [hbirr] 14000: Use only one access to the spinlock in the assertion, because the value may change between two access' on smp machines.

[Thomas Weidenmueller]

[e-mail] wrote:
> Use only one access to the spinlock in the assertion, because the value may change between two access' on smp machines.

You implementation is better, no doubt, but it wouldn't have mattered to 
read it twice or more. I mean, it's re-read again down in the 
InterlockedExchange call, there's a possibility it might have been 
changed in the meantime as well. But it really wouldn't matter ;)

Best Regards,
Thomas

[Hartmut Birr]

Thomas Weidenmueller schrieb:
> [e-mail] wrote:
> > Use only one access to the spinlock in the assertion, because the
> > value may change between two access' on smp machines.
>
> You implementation is better, no doubt, but it wouldn't have mattered
> to read it twice or more. I mean, it's re-read again down in the
> InterlockedExchange call, there's a possibility it might have been
> changed in the meantime as well. But it really wouldn't matter ;)

I must change this because my smp machine does crash in the early boot
phase at this point. If someone releases the spinlock between the two
comparisons the assertion gets a hit.

- Hartmut

[Thomas Weidenmueller]

Hartmut Birr wrote:
> Thomas Weidenmueller schrieb:
>
> I must change this because my smp machine does crash in the early boot
> phase at this point. If someone releases the spinlock between the two
> comparisons the assertion gets a hit.
> Hartmut

Hm, but wouldn't you then rather have to assert the previous value that 
InterlockedExchange returned? This way the value would've been read 
atomically only once instead of twice as it is at the moment.

Best Regards,
Thomas