On Mon, Jan 18, 2021 at 10:44 PM Pavel Machek [off-list ref] wrote:

quoted

Fix this by allocating crypto requests with GFP_ATOMIC mask in
interrupt context.

...

This one is wrong.

quoted

+++ b/drivers/md/dm-crypt.c

@@ -1454,13 +1454,16 @@ static int crypt_convert_block_skcipher(
-     if (!ctx->r.req)
-             ctx->r.req = mempool_alloc(&cc->req_pool, GFP_NOIO);
+     if (!ctx->r.req) {
+             ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);

Good so far. Ugly but good.

quoted

-static void crypt_alloc_req_aead(struct crypt_config *cc,
+static int crypt_alloc_req_aead(struct crypt_config *cc,
                               struct convert_context *ctx)
 {
-     if (!ctx->r.req_aead)
-             ctx->r.req_aead = mempool_alloc(&cc->req_pool, GFP_NOIO);
+     if (!ctx->r.req) {
+             ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
+             if (!ctx->r.req)
+                     return -ENOMEM;
+     }

But this one can't be good. We are now allocating different field in
the structure!

Good catch! Sorry for the copy-paste. It is actually not a big deal,
because this is not a structure, but a union:
as long as the mempool was initialized with the correct size, it
should be no different.

Nevertheless, I'll send the patch to fix the typo.

Regards,
Ignat

                                                                Pavel

--
DENX Software Engineering GmbH,      Managing Director:    Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194    Groebenzell, Germany