Hello!
On 3/23/23 11:32 AM, Geert Uytterhoeven wrote:
[...]
>>> It had a purpose back in the days, but today we have a handy helper.
>>
>> Well, the is_opened flag doesn't get set/cleared at the same time as
>> __LINK_STATE_START. I'm not sure they are interchangeable...
>>
>>> Reported-by: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
>>> Signed-off-by: Wolfram Sang <wsa+renesas@xxxxxxxxxxxxxxxxxxxx>
>
>>> --- a/drivers/net/ethernet/renesas/sh_eth.c
>>> +++ b/drivers/net/ethernet/renesas/sh_eth.c
>>> @@ -2441,8 +2441,6 @@ static int sh_eth_open(struct net_device *ndev)
>>>
>>> netif_start_queue(ndev);
>>>
>>> - mdp->is_opened = 1;
>>> -
>>
>> __LINK_STATE_START gets set before the ndo_open() method call, so
>> before the RPM call that enbales the clocks...
>>
>>> return ret;
>>>
>>> out_free_irq:
>>> @@ -2565,7 +2563,7 @@ static struct net_device_stats *sh_eth_get_stats(struct net_device *ndev)
>>> if (mdp->cd->no_tx_cntrs)
>>> return &ndev->stats;
>>>
>>> - if (!mdp->is_opened)
>>> + if (!netif_running(ndev))
>>> return &ndev->stats;
>>
>> I guess mdp->is_opened is checked here to avoid reading the counter
>> regs when RPM hasn't been called yet to enable the clocks...
>
> Exactly, cfr. commit 7fa2955ff70ce453 ("sh_eth: Fix sleeping function
> called from invalid context").
Yeah, pm_runtime_get_sync() couldn't be called in this case as
netstat_show() invoked read_lock() that ensued calling preempt_disable()...
> So you mean sh_eth_get_stats() can now be called after setting
> __LINK_STATE_START, but before RPM has enabled the clocks?
Yes, probably...
> Is there some protection against parallel execution of ndo_open()
> and get_stats()?
Haven't seen it (yet?)...
> Gr{oetje,eeting}s,
>
> Geert
MBR, Sergey
