Re: [PATCH 1/3] dma-helpers: chunk dma_blk_cb at IOV_MAX

[Daniel Gomez]

On 29/05/2026 16.42, Keith Busch wrote:
> On Fri, May 29, 2026 at 02:36:13PM +0200, Daniel Gomez wrote:
>> From: Daniel Gomez <da.go...@samsung.com>
>>
>> dma_blk_cb() submits dbs->iov to io_func() in one shot, so callers
>> must keep dbs->iov.niov below IOV_MAX or the host preadv()/pwritev()
>> rejects the call.
>>
>> Break the accumulation loop when niov reaches IOV_MAX. The existing
>> re-entry path resumes mapping and submits a follow-up chunk under
>> the same AIOCB, so callers no longer carry that constraint.
> 
> Thanks, I like the idea. One question below.
> 
>> -        if (!mem)
>> +        if (!mem || dbs->iov.niov >= IOV_MAX) {
>>              break;
>> +        }
> 
> Won't this leak "mem" if it is not NULL, but we've hit the iov.niov
> condition? If so, I think you can move this new check to after the
> qemu_iovec_add call.

Good catch, thanks. For v2, I'll move the new break condition after
qemu_iovec_add and sg length accounting:

@@ -156,15 +156,17 @@ static void dma_blk_cb(void *opaque, int ret)
                 }
             }
         }
-        if (!mem || dbs->iov.niov >= IOV_MAX) {
+        if (!mem)
             break;
-        }
         qemu_iovec_add(&dbs->iov, mem, cur_len);
         dbs->sg_cur_byte += cur_len;
         if (dbs->sg_cur_byte == dbs->sg->sg[dbs->sg_cur_index].len) {
             dbs->sg_cur_byte = 0;
             ++dbs->sg_cur_index;
         }
+        if (dbs->iov.niov >= IOV_MAX) {
+            break;
+        }
     }

     if (dbs->iov.size == 0) {