On Sun, 19 Mar 2000, Alan Cox wrote:
> Spot the problem
>
>
> np = (void *)(((long)kmalloc(sizeof(*np), GFP_KERNEL) + 31) & ~31);
The old method was to put the padding at the end of the private structure with
code like
int pad0, pad1; /* Pad for alignment. */
That ran into the problem of people not understanding the comment and adding
new structure element after this.
The pci-skeleton driver and many others use
{ /* Make certain elements e.g. descriptor lists are aligned. */
void *mem = kmalloc(sizeof(*np) + PRIV_ALIGN, GFP_KERNEL);
if ( ! mem) {
....
}
dev->priv = np = (void *)(((long)mem + PRIV_ALIGN) & ~PRIV_ALIGN);
memset(np, 0, sizeof(*np));
np->priv_addr = mem;
}
> and later
>
> kfree(dev->priv);
This must be
kfree(dev->priv_addr)
> So all the existing drivers are not remotely back compatible anyway, they
> are broken in the case of a misaligned return and have been for a long time.
> Some of the other drivers 'solve' the problem by simply forgetting to free
> the buffer.
This was a bug in some of the drivers, but has since been fixed. Well, at
least in my versions, which have been explicitly rejected because they
contain backwards-compatibility code..
Donald Becker
Scyld Computing Corporation, [EMAIL PROTECTED]
-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to [EMAIL PROTECTED]
