Hi guys,
I found an issue in managed version of dma_declare_coherent_memory,
i.e. in dmam_declare_coherent_memory. It looks like the return value
of dma_declare_coherent_memory is zero in case of error and a
requested flag on success, which dmam_* version doesn't take into
account and this leads to leaking of resources.

---
Yours sincerely,
Vyacheslav V. Yurkov

P.S.: I'm not subscribed to the maillist, so please include me in CC
when responding to this Email.

---
--- linux-4.5.2/drivers/base/dma-mapping.c.orig 2016-04-20
08:46:35.000000000 +0200
+++ linux-4.5.2/drivers/base/dma-mapping.c      2016-04-28 10:15:34.295080491
+0200
@@ -198,10 +198,15 @@ int dmam_declare_coherent_memory(struct

        rc = dma_declare_coherent_memory(dev, phys_addr, device_addr, size,
                                         flags);
-       if (rc == 0)
+
+       if ((flags & DMA_MEMORY_MAP) == DMA_MEMORY_MAP ||
+           (flags & DMA_MEMORY_IO) == DMA_MEMORY_IO) {
                devres_add(dev, res);
-       else
+               rc = 0;
+       } else {
                devres_free(res);
+               rc = -ENOMEM;
+       }

        return rc;
 }

Signed-off-by: Vyacheslav Yurkov <uvv.m...@gmail.com>

Reply via email to