On Fri 21-08-20 17:33:06, Matthew Wilcox wrote: > On Fri, Aug 21, 2020 at 06:07:59PM +0200, Jan Kara wrote: > > On Wed 19-08-20 16:05:51, Matthew Wilcox (Oracle) wrote: > > > This simplifies the callers and leads to a more efficient implementation > > > since the XArray has this functionality already. > > > > > > Signed-off-by: Matthew Wilcox (Oracle) <willy@xxxxxxxxxxxxx> > > > > The patch looks good to me. Just I'd note that you could drop some: > > > > if (index >= end) > > break; > > > > checks in shmem_undo_range() as well. > > Oh yes, missed a couple ;-) Thanks, I'll add. > > > In the past I was considering moving find_get_entries() to the same API as > > find_get_pages_range() has (which is essentially what you do now, but I > > also had 'start' to be a pgoff_t * so that we can return there where the > > iteration ended in the range). But in the end I've decided the churn is not > > worth the few removed lines and didn't push the patch in the end. What you > > did in this patch seems to be a reasonable middle-ground :) > > I did look at that, but since we're returning the indices, we don't _need_ > to update the index here. > > I have some other ideas for this family of interfaces, but I'm trying > to get the THP work off my plate before getting distracted by that ;-) I have one thing which I wanted to do for a long time but never got to it. IMHO the pagevec abstraction makes the loops unnecessarily complex. I'd rather have helpers like: for_each_mapping_page(mapping, page, start, end) or for_each_mapping_entry(mapping, entry, index, start, end) and hide all the pagevec magic inside those. And it's even not that hard to do including the handling of premature exit from the loop - sample userspace code is attached... Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
#include <stdio.h> #define PAGEVEC_SIZE 15 struct pagevec { unsigned char nr; int vals[PAGEVEC_SIZE]; }; struct pagevec_iter { unsigned char idx; int last_index; struct pagevec pvec; }; void pagevec_release(struct pagevec *pvec) { int i; for (i = 0; i < pvec->nr; i++) printf("Freeing val %d\n", pvec->vals[i]); pvec->nr = 0; } int pagevec_lookup(struct pagevec *pvec, void *mapping, int *start, int end) { int i; for (i = 0; i < PAGEVEC_SIZE && *start < 29; i++) { pvec->vals[pvec->nr++] = 100 + *start; (*start)++; } return pvec->nr; } static inline int get_next_pvec_val(struct pagevec_iter *pvec_i, void *mapping, int end) { if (pvec_i->idx >= pvec_i->pvec.nr) { pagevec_release(&pvec_i->pvec); if (!pagevec_lookup(&pvec_i->pvec, mapping, &pvec_i->last_index, end)) return 0; pvec_i->idx = 0; } return pvec_i->pvec.vals[pvec_i->idx++]; } void pagevec_release_iter(struct pagevec_iter *pvec_i) { pagevec_release(&pvec_i->pvec); } #define for_each_page(mapping, page, start, end) \ for (struct pagevec_iter pvec_i \ __attribute__ ((cleanup (pagevec_release_iter))) = { \ .idx = 0, .last_index = start, .pvec = { .nr = 0 } }; \ (page = get_next_pvec_val(&pvec_i, mapping, end));) int main(void) { int page; for_each_page(NULL, page, 5, 32) { printf("Seeing val %d\n", page); if (page > 122) break; } return 0; }