Re: [PATCH] bcache: Revert "bcache: use bvec_virt"

[Coly Li <colyli@suse.de>]

On 11/3/21 11:46 PM, Christoph Hellwig wrote:
> On Wed, Nov 03, 2021 at 11:10:41PM +0800, Coly Li wrote:
>> diff --git a/drivers/md/bcache/btree.c b/drivers/md/bcache/btree.c
>> index 93b67b8d31c3..88c573eeb598 100644
>> --- a/drivers/md/bcache/btree.c
>> +++ b/drivers/md/bcache/btree.c
>> @@ -378,7 +378,7 @@ static void do_btree_node_write(struct btree *b)
>>   		struct bvec_iter_all iter_all;
>>   
>>   		bio_for_each_segment_all(bv, b->bio, iter_all) {
>> -			memcpy(bvec_virt(bv), addr, PAGE_SIZE);
>> +			memcpy(page_address(bv->bv_page), addr, PAGE_SIZE);
> How could there be an offset?  bch_bio_alloc_pages allocates a
> fresh page for each vec, and bio_for_each_segment_all iterates page
> by page.  IFF there is an offset there is proble in the surrounding
> code as bch_bio_alloc_pages assumes that it is called on a freshly
> allocate and initialized bio.

Yes, the offset is modified in bch_bio_alloc_pages(). Normally the 
bcache defined block size is 4KB so the issue was not triggered 
frequently. I found it during testing my nvdimm enabling code for 
bcache, where I happen to make the bcache defined block size to non-4KB. 
The offset is from the previous written bkey set, which the minimized 
unit size is 1 bcache-defined-block-size.

Coly Li