Hi Adrian,
On Mon, 7 Oct 2013 09:50:11 +0300, Adrian Hunter wrote:
> Before using kcore we need to check that modules are
> in memory at the same addresses that they were when
> data was recorded.
>
> This is done because, while we could remap symbols
> to different addresses, the object code linkages
> would still be different which would provide an
> erroneous view of the object code.
[SNIP]
>
> -static int kcore_mapfn(u64 start, u64 len, u64 pgoff, void *data)
> +static void add_module(struct module_info *mi, struct rb_root *modules)
> {
> - struct kcore_mapfn_data *md = data;
> - struct map *map;
> + struct rb_node **p = &modules->rb_node;
> + struct rb_node *parent = NULL;
> + struct module_info *m;
>
> - map = map__new2(start, md->dso, md->type);
> - if (map == NULL)
> + while (*p != NULL) {
> + parent = *p;
> + m = rb_entry(parent, struct module_info, rb_node);
> + if (strcmp(mi->name, m->name) < 0)
> + p = &(*p)->rb_left;
> + else
> + p = &(*p)->rb_right;
> + }
> + rb_link_node(&mi->rb_node, parent, p);
> + rb_insert_color(&mi->rb_node, modules);
> +}
[SNIP]
> +static int __read_proc_modules(void *arg, const char *name, u64 start)
> +{
> + struct rb_root *modules = arg;
> + struct module_info *mi;
> +
> + mi = zalloc(sizeof(struct module_info));
> + if (!mi)
> return -ENOMEM;
>
> - map->end = map->start + len;
> - map->pgoff = pgoff;
> + mi->name = strdup(name);
> + mi->start = start;
>
> - list_add(&map->node, &md->maps);
> + add_module(mi, modules);
> +
> + if (!mi->name)
> + return -ENOMEM;
Why did you add the mi if it failed to allocate name? I guess it'd
crash since the add_module() uses mi->name for strcmp().
Thanks,
Namhyung
> +
> + return 0;
> +}
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
