David Brownell wrote:
> On Wednesday 07 February 2007 3:04 am, Phil Endecott wrote:
>> I won't pretend to understand all the subtleties of
>> __attribute__((packed)), but I did learn something about it when I
>> hacked USB/IP to run on ARM, for the NSLU2, last year. The issue
>> boiled down to this:
>>
>> struct S {
>> int intfield;
>> char charfield;
>> } /* maybe add: __attribute__((packed)) here */ ;
>>
>> f(struct S * s) {
>> ...
>> (s->intfield)++;
>> ...
>> }
>>
>> ...
>>
>> What happens when we add __attribute__((packed)) to struct S? ...
>> will work, but a lot more slowly. However, the ++ in f() WILL NOT
>> WORK! The compiler assumes that the pointer that is passed IS
>> NATURALLY ALIGNED, and when you pass an unaligned pointer from g() IT
>> DOES NOT ISSUE A WARNING!
>
> Looks like a bug in your version of the toolchain. I just tried
> this not a minute ago (gcc 3.4.3) and it works correctly: not only
> do both versions of the function handle the "intfield" misaligned
> for loads, they also do so for the store.
Err, maybe my example was over-simplified then, sorry. Try another
level of indirection:
void inc(int* i) {
(*i)++;
}
void f(struct S* s) {
inc(&(s->intfield);
}
Please look in the gcc-help archives for December and search for my
name; you'll find a long thread where I slowly come to understand how
this works (or rather how it doesn't work and how it doesn't warn you
that it's not going to work). I believe that the people that I was
corresponding with know what they are talking about. In particular,
here's a quote from Paul Brook
(http://gcc.gnu.org/ml/gcc-help/2006-12/msg00115.html):
" the compiler is allowed to assume that the low 2 bits of an int*
are zero. "
Phil.
-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier.
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________
linux-usb-devel@lists.sourceforge.net
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
