在 2025/3/19 17:46, Herbert Xu 写道:
On Wed, Mar 19, 2025 at 05:44:17PM +0800, Zhihao Cheng wrote:
Tested-by: Zhihao Cheng <chengzhihao1@xxxxxxxxxx> # For xfstests
Thank you for testing!
diff --git a/fs/ubifs/compress.c b/fs/ubifs/compress.c
index a241ba01c9a8..ea6f06adcd43 100644
--- a/fs/ubifs/compress.c
+++ b/fs/ubifs/compress.c
@@ -16,6 +16,7 @@
*/
#include <crypto/acompress.h>
+#include <linux/highmem.h>
#include "ubifs.h"
/* Fake description object for the "none" compressor */
@@ -126,7 +127,7 @@ void ubifs_compress(const struct ubifs_info *c, const void *in_buf,
{
ACOMP_REQUEST_ALLOC(req, compr->cc, GFP_NOFS | __GFP_NOWARN);
- acomp_request_set_src_nondma(req, in_buf, in_len);
+ acomp_request_set_src_dma(req, in_buf, in_len);
Why not merging it into patch 13?
Because it will break without this patch. If the input is a highmem
folio it cannot be directly passed over to DMA (because the virtual
address has been remapped by kmap_local).
Makes sense.
Cheers,
