On Thu, Aug 11, 2005 at 08:23:30PM +0200, Simon 'corecode' Schubert wrote: > On 11.08.2005, at 19:33, Matthew Dillon wrote: > > Same here. In this case the malloc clearly should not be > > allowed to fail as this code is called as part of the > >initialization > > phase. This is a common occurance in FreeBSD code and we've been > > changing those M_NOWAIT's to M_INTWAIT's right and left for > > initialization phase code. > > But it can fail, like other things in the initialization can fail as > well. All other code copes with failures, except cbb_insert_res.
This code must not fail if possible. One reason is that the callers might indead be allowed to block and therefore use M_WAITOK, but we can't do that here. M_INTWAIT dramatically increases the chance of success and allocating from the interrupt reserve is definitely wanted here. Joerg
