[RFC] ath10k: change to do napi_enable and napi_disable when insmod and rmmod for sdio

[Krishna Chaitanya]

On Thu, Aug 20, 2020 at 3:45 PM Wen Gong <wgong at codeaurora.org> wrote:
>
> On 2020-08-20 17:19, Krishna Chaitanya wrote:
> ...
> >> > diff --git a/drivers/net/wireless/ath/ath10k/sdio.c b/drivers/net/wireless/ath/ath10k/sdio.c
> >> > index 7b894dcaad2e..b71499b171c6 100644
> >> > --- a/drivers/net/wireless/ath/ath10k/sdio.c
> >> > +++ b/drivers/net/wireless/ath/ath10k/sdio.c
> >> > @@ -1756,8 +1756,6 @@ static int ath10k_sdio_hif_start(struct ath10k *ar)
> >> >       struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
> >> >       int ret;
> >> >
> >> > -     napi_enable(&ar->napi);
> >> > -
> >> >       /* Sleep 20 ms before HIF interrupts are disabled.
> >> >        * This will give target plenty of time to process the BMI done
> >> >        * request before interrupts are disabled.
> >> > @@ -1884,7 +1882,6 @@ static void ath10k_sdio_hif_stop(struct ath10k *ar)
> >> >       spin_unlock_bh(&ar_sdio->wr_async_lock);
> >> >
> >> >       napi_synchronize(&ar->napi);
> >> > -     napi_disable(&ar->napi);
> >> >  }
> >> >
> >> >  #ifdef CONFIG_PM
> >> > @@ -2121,6 +2118,7 @@ static int ath10k_sdio_probe(struct sdio_func *func,
> >> >
> >> >       netif_napi_add(&ar->napi_dev, &ar->napi, ath10k_sdio_napi_poll,
> >> >                      ATH10K_NAPI_BUDGET);
> >> > +     napi_enable(&ar->napi);
> >> >
> >> >       ath10k_dbg(ar, ATH10K_DBG_BOOT,
> >> >                  "sdio new func %d vendor 0x%x device 0x%x block 0x%x/0x%x\n",
> >> > @@ -2235,6 +2233,7 @@ static void ath10k_sdio_remove(struct sdio_func *func)
> >> >
> >> >       ath10k_core_unregister(ar);
> >> >
> >> > +     napi_disable(&ar->napi);
> >> >       netif_napi_del(&ar->napi);
> >> >
> >> >       ath10k_core_destroy(ar);
> >>
> >> I'm not really convinced that this is the right fix, but I'm no NAPI
> >> expert. Can anyone else help?
> > Calling napi_disable() twice can lead to hangs, but moving NAPI from
> > start/stop to
> > the probe isn't the right approach as the datapath is tied to
> > start/stop.
> >
> > Maybe check the state of NAPI before disable?
> >
> >  if (test_bit(NAPI_STATE_SCHED, &ar->napi.napi.state))
> >   napi_disable(&ar->napi)
> > or maintain napi_state like this
> > https://patchwork.kernel.org/patch/10249365/
> it is better to use above link's patch.
> napi.state is controlled by napi API, it is better ath10k not know it.
Sure, but IMHO just canceling the async rx work should solve the issue.

> > Also, the most common cause for such issues (1st
> > napi_synchronize/napi_disable hang)
> > is that napi_poll is being scheduled, so, you might want to check that
> > napi_schedule isn't
> > called after stop.
> >
> > cd ath10k; git log --grep=napi shows plenty of such issues. the one
> > that matches closest is
> > c2cac2f74ab4bcf0db0dcf3a612f1e5b52d145c8, so, it could just be a
> > regression.
> This above commit's scene is not same with this patch.
> It is hang for only do 1 simulate crash of the commit, this patch is
> doing simulate crash and rmmod meanwhile.
yes, it is the closest one I could find.