Skip to contentSkip to conversation

dm: era: use do_div for 64-bit division

PATCH In review Arnd Bergmann 2 participants 2 messages Mar 11, 2014 Device mapper

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 dm: era: use do_div for 64-bit division

Commit 983043be0c1de "dm era: support non power-of-2 blocksize" introduced the use of sector_div to divide a variable of type dm_block_t by an integer. This fails on 32-bit machines when CONFIG_LBDAF is not set, because that assumes that the first argument is a 32-bit number.

To allow building this driver in all configurations, we either have to cast the dm_block_t to sector_t or use do_div(), which always takes a 64-bit argument. This patch uses the second approach in order to cope with theoretical configurations where dm_block_t contains a number that does not fit into sector_t.

Cc Mike Snitzer <[email protected]>
Cc Joe Thornber <[email protected]>
Cc Alasdair Kergon <[email protected]>

---
This currently only occurs in linux-next. Please apply.

drivers/md/dm-era-target.c +1 -1
index d7bd7b3..3383423 100644
@@ -1123,7 +1123,7 @@ static dm_block_t get_block(struct era *era, struct bio *bio)
 	dm_block_t block_nr = bio->bi_iter.bi_sector;
 
 	if (!block_size_is_power_of_two(era))
-		(void) sector_div(block_nr, era->sectors_per_block);
+		(void) do_div(block_nr, era->sectors_per_block);
 	else
 		block_nr >>= era->sectors_per_block_shift;
 
#1 lore raw
Quoted 11 lines · Arnd Bergmann

Commit 983043be0c1de "dm era: support non power-of-2 blocksize" introduced the use of sector_div to divide a variable of type dm_block_t by an integer. This fails on 32-bit machines when CONFIG_LBDAF is not set, because that assumes that the first argument is a 32-bit number.

To allow building this driver in all configurations, we either have to cast the dm_block_t to sector_t or use do_div(), which always takes a 64-bit argument. This patch uses the second approach in order to cope with theoretical configurations where dm_block_t contains a number that does not fit into sector_t.

Thanks for taking the time to look into this.

But the first approach was used on Friday (and is adequate considering we're starting with the bio's bi_sector). The new commit id is 1c5a63b138638d08f4dc65ed48d3d5d3bad1b08b.

The moral of the story is: don't make a mistake on Thursday because that last linux-next of the week will plague you for the next 5 days. I was hoping for relief with Monday's linux-next build but alas... ;)