Re: [lwip-members] lightweight protection

[David Haas]

Hi,

Some History:

When I added these macros, I was being very conservative, so I only 
added them where I thought protection was missing and where I thought a 
lighter  weight protection mechanism was needed. In pbuf, protection was 
missing almost entirely. memp.c had semaphore protection, which would 
work fine, but my Coldfire driver needed to allocate memory from an ISR, 
so I wanted to replace these with the lightweight protection. I used the 
#if  LIGHTWEIGHT_PROT because I did'nt want to mess anything up that 
people were already using.

I implemented the lightweight protection on both the unix and coldfire 
ports.

SYS_ARCH_DECL_PROTECT is only supposed to declare a variable to keep the 
old protection status, not a mutex variable. The idea of the lightweight 
protection is that it is global in nature. That is what makes it light 
weight. I admit that the mutex I used to implement it on the unix port 
is only slightly lighter weight than a semephore. The disabling of 
interrupts on coldfire is really lightweight, but might not be suitable 
for all systems. Somebody (Marc Boucher?) implemented lightweight 
protection totally with macros that disabled and enabled interrupts.

At the time, I had several people comment and add suggestions for 
lightweight protection, which was why it was done with these macros.

Proposal:
I don't particularly like the #if SYS_LIGHTWEIGHT_PROT test at each of 
these places. In my opinion we could get rid of this and always use this 
protection mechanism when NO_SYS is 0. I think that it adds needed 
protection in pbuf  that is always necessary when you have multiple 
tasks. But, I did not think we really had a consensus on what was to be 
done for protection, so perhaps this needs more discussion. ;-)

I would be happy to do some implementation if we think this should 
change, but my design criteria is still to allow a lighter weight 
protection mechanism which can be implemented in some systems by 
disabling or reducing the interrupt level.

David.

Jani Monoses wrote:

> Hi
> in pbuf.c SYS_ARCH_PROTECT & co are called unconditionnally while in memp.c
> they are called only if SYS_LIGHTWEIGHT_PROT is 1.Any reason for this 
> inconsistency?
> In pbuf they seem to be an addition to semaphore locking while in memp an 
> alternative.
>
> Another question: I only see the unix arch defining these at the moment so I'm 
> not sure
> how they were intended to be used.Specifically  is sys_arch_decl_protect 
> supposed to
> declare a lock variable?In that case from the use of macros in pbuf that 
> variable seems
> to be local and thus does not protect from accesses from another function to 
> the protected data.
> Or if the lock is supposed to be global than maybe a DECL or two at the 
> beginning of the file
> and various protect/unprotect macro pairs in the functions would be a better 
> layout.
>
> Jani.
>
>
>
> _______________________________________________
> lwip-members mailing list
> address@hidden
> http://mail.nongnu.org/mailman/listinfo/lwip-members
>
>
>