Re: [PATCH 2/3] printk: Add %ptc to safely print a task's comm

From: Joe Perches
Date: Tue May 17 2011 - 18:17:40 EST


On Wed, 2011-05-18 at 00:04 +0200, Jiri Slaby wrote:
> On 05/17/2011 11:52 PM, Joe Perches wrote:
> > On Tue, 2011-05-17 at 23:42 +0200, Jiri Slaby wrote:
> >> On 05/17/2011 10:47 PM, John Stultz wrote:
> >>> Accessing task->comm requires proper locking. However in the past
> >>> access to current->comm could be done without locking. This
> >>> is no longer the case, so all comm access needs to be done
> >>> while holding the comm_lock.
> >>> +static noinline_for_stack
> >> With my setup, the code below inlined will use 32 bytes of stack. The
> >> same as %pK case. Uninlined it obviously eats "only" 8 bytes for IP.
> > The idea is to avoid excess stack consumption for things like:
> > struct va_format vaf;
> > const char *fmt = "some format with %ptc";
> > vaf.fmt = fmt;
> > vaf.va = &va_list;
> > printk("some format with %pV\n", &vaf);
> There is no way how can noinline_for_stack for task_comm_string lower
> the stack usage here, right? Note that it adds no more requirements to
> the stack than there were before. Simply because there are no buffers on
> the stack in task_comm_string.

Isn't flags always on stack in function pointer
if task_comm_string were inlined?

I believe gcc isn't too good about reusing stack for blocks

void foo(args)
{
if (bar) {
long baz;
...
} else
int baz;
...
}

I believe gcc still creates 2 separate baz vars on
foo's stack.

> If nothing, it saves 100 bytes of .text.

Submit patches to vsprintf for all the cases you think
appropriate.

> thanks,

cheers, Joe

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/