On 8/2/20 11:24 PM, Julia Lawall wrote:
>> +@initialize:python@
>> +@@
>> +# low-level memory api
>> +filter = frozenset(['__vmalloc_area_node'])
>> +
>> +def relevant(p):
>> + return not (filter & {el.current_element for el in p})
>
> Is this used?
I'll remove it in v8. Or do you want me to add iterate_dir_item() in the list?
>
> Otherwise, I think it would be good to not warn about a use of kvfree
> if that use is reachable from a kvmalloc. There seems to be such a false
> positive in fs/btrfs/send.c, on line 1118.
I don't know how to handle this case without position filter.
It's too complex. In iterate_dir_item() there is:
buf = kmalloc(buf_len, GFP_KERNEL);
while(...) {
if (...) {
if (is_vmalloc_addr(buf)) {
vfree(buf);
...
} else {
char *tmp = krealloc(buf, ...);
if (!tmp)
kfree(buf);
...
}
if (!buf) {
buf = kvmalloc(buf_len, GFP_KERNEL);
...
}
}
}
kvfree(buf);
Adding "when != kvfree(E)" is not enough:
* E = \(kvmalloc\|kvzalloc\|kvcalloc\|kvzalloc_node\|kvmalloc_node\|
* kvmalloc_array\)(...)@k
... when != is_vmalloc_addr(E)
+ when != kvfree(E)
when any
* \(kfree\|kzfree\|vfree\|vfree_atomic\)(E)@p
>
> It also seems that when there are both a kmalloc and a vmalloc, there is
> no warning if kfree or vfree is used. Is that intentional?
>
No, I will try to address it in v8.
Regards,
Denis
