On Fri, 9 Feb 2018 12:15:47 +0900
Namhyung Kim <namhyung@xxxxxxxxxx> wrote:
> > @@ -124,6 +128,16 @@ enum {
> > FUNC_TYPE_MAX
> > };
> >
> > +#define MAX_STR 512
> > +
> > +/* Two contexts, normal and NMI, hence the " * 2" */
> > +struct func_string {
> > + char buf[MAX_STR * 2];
> > +};
> > +
> > +static struct func_string __percpu *str_buffer;
> > +static int nr_strings;
>
> What protects it?
Grumble, I was thinking that the entire create_function_event was under
the func_event_mutex, which it is not. So nr_strings is not fully
protected. I'll fix that thanks.
As for str_buffer, I should comment this as it is rather subtle.
+static int read_string(char *str, unsigned long addr)
+{
+ unsigned long flags;
+ struct func_string *strbuf;
+ char *ptr = (void *)addr;
+ char *buf;
+ int ret;
+
+ if (!str_buffer)
+ return 0;
+
+ strbuf = this_cpu_ptr(str_buffer);
+ buf = &strbuf->buf[0];
+
+ if (in_nmi())
+ buf += MAX_STR;
+
+ local_irq_save(flags);
Like I said, this is really subtle, and desperately needs a comment.
The str_buffer is per cpu and can only be access under irqs disabled.
If we are in NMI, then we move the starting position forward by MAX_STR.
I'll add comments and protect create_function_event with the mutex.
Thanks for pointing this out!
-- Steve
+ ret = strncpy_from_unsafe(buf, ptr, MAX_STR);
+ if (ret < 0)
+ ret = 0;
+ if (ret > 0 && str)
+ memcpy(str, buf, ret);
+ local_irq_restore(flags);
+
+ return ret;
+}
--
To unsubscribe from this list: send the line "unsubscribe linux-trace-users" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
|