CONFIG_PREEMPT and JFFS2 oops

[Joakim Tjernlund]

Paul Walmsley <paul at pwsan.com> wrote on 2012/01/28 10:51:26:
>
> On Thu, 26 Jan 2012, Joakim Tjernlund wrote:
>
> > .. and you should delay the second allocation to when it is needed. After swapping ptrs, test if
> > tmp_buf is NULL and kmalloc if so.
>
> Hmm.  You are thinking of a GFP_ATOMIC allocation?  Seems best to avoid
> those?

No, just stick the kmalloc after the unlock:

			spin_unlock(&jffs2_compressor_list_lock);
			if (!tmp_buf) {
				tmp_buf = kmalloc(...);
				<error handling if still NULL>
			}
			*datalen  = orig_slen;
			*cdatalen = orig_dlen;
			compr_ret = this->compress(data_in, tmp_buf, datalen, cdatalen, NULL);


That way you also skip a second kmalloc if the list only holds one compressor and
you don't need GFP_ATOMIC