fs/select.c [patch > to >=]

fs/select.c (from 2.4.19 but present in 2.5.x also)

asmlinkage long sys_poll(struct pollfd * ufds, unsigned int nfds, long timeout)

{
        int i, j, fdcount, err;
        struct pollfd **fds;
        poll_table table, *wait;
        int nchunks, nleft;

[ skip ]

        nchunks = 0;
        nleft = nfds;
        while (nleft > POLLFD_PER_PAGE) { /* allocate complete PAGE_SIZE chunks
*/
                fds[nchunks] = (struct pollfd *)__get_free_page(GFP_KERNEL);
                if (fds[nchunks] == NULL)
                        goto out_fds;
                nchunks++;
                nleft -= POLLFD_PER_PAGE;
        }
        if (nleft) { /* allocate last PAGE_SIZE chunk, only nleft elements used

*/

why not -->

        while (nleft >= POLLFD_PER_PAGE) {

in this case we will use only nchunks in the case that this will end up page aligned.. the current way of using only > POLLFD_PER_PAGE will end up with nchunks, and nleft == PAGE_SIZE / sizeof(pollfd) etc..

silly optimisation by changing > to >=, but still an optimisation.. if there is something obviously wrong with changing > to >=, ignore me since I probably should get more sleep.. but I dont see any problem though I havent really looked at it too much.

attachment has 1 line (1 character) patch ;-)

--
Silvio

diff -u fs/select.c.2.4.19 fs/select.c 
@@ -445,7 +445,7 @@
 
 	nchunks = 0;
 	nleft = nfds;
-	while (nleft > POLLFD_PER_PAGE) { /* allocate complete PAGE_SIZE chunks */
+	while (nleft >= POLLFD_PER_PAGE) { /* allocate complete PAGE_SIZE chunks */
 		fds[nchunks] = (struct pollfd *)__get_free_page(GFP_KERNEL);
 		if (fds[nchunks] == NULL)
 			goto out_fds;