Question about printking

Question about printking

[Ezequiel Garcia]

Hello,

I have a quick question.
I (think I) remember Greg KH reviewing some driver patch on staging
mailing list,
suggesting not to do this sort of printking upon an allocation request failure:

ptr = kmalloc(sizeof(foo));
if (!ptr) {
        pr_err("Cannot allocate memory for foo\n");
        return -ENOMEM;
}

His argue against it was that kmalloc already takes care of reporting/printking
a good deal of interesting information when this happens.

Is my memory right?

Can someone expand a bit on this whole idea? (of abuse of printing,
or futility of printing).

I'm asking after seeing *a lot* of drivers doing exactly this.

Thanks,

--
    Ezequiel

Re: Question about printking

[Joe Perches]

On Sat, 2013-02-02 at 16:30 -0300, Ezequiel Garcia wrote:
> ptr = kmalloc(sizeof(foo));
> if (!ptr) {
>         pr_err("Cannot allocate memory for foo\n");
>         return -ENOMEM;
> }
> His argue against it was that kmalloc already takes care of reporting/printking
> a good deal of interesting information when this happens.

> Can someone expand a bit on this whole idea? (of abuse of printing,
> or futility of printing).

k.alloc() takes a GFP_ flag as an arg.

One of those GFP flags is __GFP_NOWARN.

For all failed allocs without GFP_NOWARN
a message is emitted and a dump_stack is
done.

(see: mm/page_alloc.c warn_alloc_failed())

So, most all of these printks after
k.alloc()'s are not necessary.

Re: Question about printking

[Ezequiel Garcia]

Hi Joe,

On Sat, Feb 2, 2013 at 4:43 PM, Joe Perches <joe@perches.com> wrote:
> On Sat, 2013-02-02 at 16:30 -0300, Ezequiel Garcia wrote:
>> ptr = kmalloc(sizeof(foo));
>> if (!ptr) {
>>         pr_err("Cannot allocate memory for foo\n");
>>         return -ENOMEM;
>> }
>> His argue against it was that kmalloc already takes care of reporting/printking
>> a good deal of interesting information when this happens.
>
>> Can someone expand a bit on this whole idea? (of abuse of printing,
>> or futility of printing).
>
> k.alloc() takes a GFP_ flag as an arg.
>
> One of those GFP flags is __GFP_NOWARN.
>
> For all failed allocs without GFP_NOWARN
> a message is emitted and a dump_stack is
> done.
>
> (see: mm/page_alloc.c warn_alloc_failed())
>
> So, most all of these printks after
> k.alloc()'s are not necessary.
>
>

Thanks for the explanation.

BTW, I see you've made some patches to fix exactly this.
Nice job.

Regards,

-- 
    Ezequiel