Ping-Ke Shih <pkshih@xxxxxxxxxxx> writes:
> From: Po-Hao Huang <phhuang@xxxxxxxxxxx>
>
> Allocate a per-skb completion to track those skbs we are interested in
> and wait for them to complete transmission with TX status. To avoid
> race condition between process and softirq without addtional locking,
> we use a work to free the tx_wait struct later when waiter is finished
> referencing it. This must be called in process context and with a
> timeout value greater than zero since it might sleep.
>
> We use another workqueue so works can be processed concurrently and
> when the PCI device is removed unexpectedly, all pending works can be
> flushed. This prevents some works that were scheduled but never processed
> leads to memory leak.
>
> Signed-off-by: Po-Hao Huang <phhuang@xxxxxxxxxxx>
> Signed-off-by: Ping-Ke Shih <pkshih@xxxxxxxxxxx>
[...]
> +static void rtw89_core_free_tx_wait_work(struct work_struct *work)
> +{
> + struct rtw89_tx_wait_info *wait =
> + container_of(work, struct rtw89_tx_wait_info, work);
> + struct rtw89_dev *rtwdev = wait->rtwdev;
> + int done, ret;
> +
> + ret = read_poll_timeout(atomic_read, done, done, 1000, 100000, false,
> + &wait->wait_done);
> +
> + if (ret)
> + rtw89_err(rtwdev, "tx wait timed out, stop polling\n");
> + else
> + kfree(wait);
> +}
I admit I didn't try to understand this patch in detail but this
function just looks odd to me. Why there's polling able to free
something?
--
https://patchwork.kernel.org/project/linux-wireless/list/
https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
