Hi Jeff,
> songkai <songkai01@xxxxxxxxxx> writes:
>
> > In “struct io” , the "void *pdu" member points to the memory which is
> > allocated in convert_to_cpu(). We should free memory pdu point to
> > before we free "struct io", or memory leak occurred.
>
> That should be freed via io_release. Do you know of a case where that doesn't
> happen? Also, io_free_all is only called right before exit.
> I'm not sure this patch really plugs a leak that would impact normal
> operations.
>
> The patch seems harmless, but I'm not sure it's needed.
Thank you for your reminding. So far, I can't find a case either.
Freeing pdu in io_release seems like enough. My original intention is that,
when calling io_free_all before btt exits, there are still some IOs in the all_ios,
I am not sure if all the pdu is freed before, just like dip and pip.
Thanks again.
Regards,
songkai
> > Signed-off-by: songkai <songkai01@xxxxxxxxxx>
> > ---
> > btt/inlines.h | 3 +++
> > 1 file changed, 3 insertions(+)
> >
> > diff --git a/btt/inlines.h b/btt/inlines.h index 64b5a2c..fa5a546
> > 100644
> > --- a/btt/inlines.h
> > +++ b/btt/inlines.h
> > @@ -153,6 +153,9 @@ static inline void io_free_all(void)
> >
> > list_for_each_safe(p, q, &all_ios) {
> > iop = list_entry(p, struct io, a_head);
> > +
> > + if (iop->pdu)
> > + free(iop->pdu);
> > free(iop);
> > }
> > }
