Hi, On Wed, 19 Jan 2022 at 17:56, Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote: > > Hi Alexander, > > alex.aring@xxxxxxxxx wrote on Mon, 17 Jan 2022 19:36:39 -0500: > > > Hi, > > > > On Mon, 17 Jan 2022 at 19:34, Alexander Aring <alex.aring@xxxxxxxxx> wrote: > > > > > > Hi, > > > > > > On Mon, 17 Jan 2022 at 06:55, Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote: > > > > > > > > ieee802154_xmit_complete() is the right helper to call when a > > > > transmission is over. The fact that it completed or not is not really a > > > > question, but drivers must tell the core that the completion is over, > > > > even if it was canceled. Do not call ieee802154_wake_queue() manually, > > > > in order to let full control of this task to the core. > > > > > > > > By using the complete helper we also avoid leacking the skb structure. > > > > > > > > Signed-off-by: Miquel Raynal <miquel.raynal@xxxxxxxxxxx> > > > > --- > > > > drivers/net/ieee802154/at86rf230.c | 2 +- > > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > > > diff --git a/drivers/net/ieee802154/at86rf230.c b/drivers/net/ieee802154/at86rf230.c > > > > index 583f835c317a..1941e1f3d2ef 100644 > > > > --- a/drivers/net/ieee802154/at86rf230.c > > > > +++ b/drivers/net/ieee802154/at86rf230.c > > > > @@ -343,7 +343,7 @@ at86rf230_async_error_recover_complete(void *context) > > > > if (ctx->free) > > > > kfree(ctx); > > > > > > > > - ieee802154_wake_queue(lp->hw); > > > > + ieee802154_xmit_complete(lp->hw, lp->tx_skb, false); > > > > > > also this lp->tx_skb can be a dangled pointer, after xmit_complete() > > > we need to set it to NULL in a xmit_error() we can check on NULL > > > before calling kfree_skb(). > > > > > > > forget the NULL checking, it's already done by core. However in some > > cases this is called with a dangled pointer on lp->tx_skb. > > Actually I don't see why tx_skb is dangling? > > There is no function that could accesses lp->tx_skb between the free > operation and the next call to ->xmit() which does a lp->tx_skb = skb. > Am I missing something? > look into at86rf230_sync_state_change() it is a sync over async and the function "at86rf230_async_error_recover_complete()" is a generic error handling to recover from a state change. It's e.g. being used in e.g. at86rf230_start() which can occur in cases which are not xmit related. Indeed there is no dangled pointer in the irq handling, sorry. I thought maybe the receive handling but the transceiver is doing a lot of its own state change handling because of some framebuffer protection which is not the case. - Alex
