On Wed, Jul 31, 2019 at 9:06 AM Willem de Bruijn
[off-list ref] wrote:

On Wed, Jul 31, 2019 at 8:22 AM Chuhong Yuan [off-list ref] wrote:

quoted

refcount_t is better for reference counters since its
implementation can prevent overflows.
So convert atomic_t ref counters to refcount_t.

Signed-off-by: Chuhong Yuan <redacted>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c | 8 ++++----
 drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.h | 2 +-
 2 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c
index fc77caf0a076..eb7ed34639e2 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c

@@ -49,7 +49,7 @@ static int bnxt_register_dev(struct bnxt_en_dev *edev, int ulp_id,
                        return -ENOMEM;
        }

-       atomic_set(&ulp->ref_count, 0);
+       refcount_set(&ulp->ref_count, 0);

One feature of refcount_t is that it warns on refcount_inc from 0 to
detect possible use-after_free. It appears that that can trigger here?

I think that's right.  We need to change the driver to start counting
from 1 instead of 0 if we convert to refcount.