On 2022-04-13 21:45, Jaegeuk Kim wrote:>> As I explained in the v1
thread, zoned support for f2fs assumes po2 zone
>> sizes. For e.g.,
>> static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
>> struct block_device *bdev, block_t blkstart, block_t blklen)
>> {
>> sector_t sector, nr_sects;
>> block_t lblkstart = blkstart;
>> int devi = 0;
>> ...
>> ...
>> // Assumes zone sectors to be po2
s/Assumes zone sectors to be po2/Works correctly only if the base
alignment is a power of 2 value./
Taken from align.h:
/* @a is a power of 2 value */
...
#define IS_ALIGNED(x, a) (((x) & ((typeof(x))(a) - 1)) == 0)
>
> Well, I think this will be aligned to 2MB, if the device gives NPO2? IOWs,
> this
Could you elaborate what you mean by `this will be aligned to 2MB`?
> is just checking the granularity, not PO2.
Yeah, so whenever we send a discard or a zone reset, the block zoned
device expects the `sector` to be the start of a device zone. The check
below essentially checks if the `sector` aligns with the `zone size
sector` of the device. But to check for alignment using this `sector &
(bdev_zone_sectors(bdev) - 1)` will only work if
`bdev_zone_sectors(bdev)` is a power of 2.
Even if the device zone size is a multiple of 2MB, these checks will
fail for a non power of 2 zone size device. The solution is simple, it
is to make this alignment check generic but until this is done in the
kernel, mkfs.f2fs should fail to mount for a non power of 2 device.
>
>> if (sector & (bdev_zone_sectors(bdev) - 1) ||
>> nr_sects != bdev_zone_sectors(bdev)) {
>> f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted
>> (block %x + %x)",
>> devi, sbi->s_ndevs ? FDEV(devi).path : "",
>> blkstart, blklen);
I tried it locally in QEMU along with my local changes to remove power
of 2 constraint in the block layer and zns device, and indeed the
`Unaligned zone reset attempted` error popped up for a zns drive with a
zone size 96MB.
When I changed the condition from `sector & (bdev_zone_sectors(bdev) -
1)` to `sector % bdev_zone_sectors(bdev)`, the error went away because
the calculation doesn't depend on the alignment base to be a power of 2.
TL;DR until we change some calculations in f2fs for zoned device to be
generic, we need to bail out during mkfs time for non power of 2 zone
size devices.
_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
