Re: bh_lru_install

["David S. Miller" <davem@redhat.com>]

On Tue, 27 Jul 2004 16:06:17 -0700
Andrew Morton <akpm@osdl.org> wrote:

> > It shouldn't be too hard to make the code just work without
> > an on-stack copy, shuffling the lru->bh[] array entries
> > directly.
> 
> Yup, that plus making it a ringbuffer maybe.

It seems implementable using two roving indexes, one for
reading and one for writing.  So it's just like the existing
code sans the on-stack+memcpy stuff :-)

Do you see any logic errors in this new code below?

static void bh_lru_install(struct buffer_head *bh)
{
	struct buffer_head *e = NULL;
	struct bh_lru *lru;

	check_irqs_on();
	bh_lru_lock();
	lru = &__get_cpu_var(bh_lrus);
	if ((e = lru->bhs[0]) != bh) {
		int rd_idx, wr_idx;

		wr_idx = 0;

		get_bh(bh);
		lru->bhs[wr_idx++] = bh;

		for (rd_idx = 1; rd_idx < BH_LRU_SIZE; rd_idx++) {
			struct buffer_head *nxt = lru->bhs[rd_idx];

			if (e != NULL)
				lru->bhs[wr_idx++] = e;
			e = nxt;
			if (e == bh) {
				__brelse(e);
				e = NULL;
			} else if (e == NULL)
				break;
		}

		while (wr_idx < BH_LRU_SIZE)
			lru->bhs[wr_idx++] = NULL;

	}
	bh_lru_unlock();

	if (e)
		__brelse(e);
}