[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <b686e429-b46e-4190-a5b0-23b0bf185768@uwaterloo.ca>
Date: Fri, 23 Aug 2024 16:21:19 -0400
From: Martin Karsten <mkarsten@...terloo.ca>
To: Eric Dumazet <edumazet@...gle.com>, Joe Damato <jdamato@...tly.com>
Cc: netdev@...r.kernel.org, amritha.nambiar@...el.com,
sridhar.samudrala@...el.com, sdf@...ichev.me, peter@...eblog.net,
m2shafiei@...terloo.ca, bjorn@...osinc.com, hch@...radead.org,
willy@...radead.org, willemdebruijn.kernel@...il.com, skhawaja@...gle.com,
kuba@...nel.org, "David S. Miller" <davem@...emloft.net>,
Paolo Abeni <pabeni@...hat.com>, Jiri Pirko <jiri@...nulli.us>,
Sebastian Andrzej Siewior <bigeasy@...utronix.de>,
Lorenzo Bianconi <lorenzo@...nel.org>, Daniel Borkmann
<daniel@...earbox.net>, open list <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH net-next 3/6] net: Add control functions for irq
suspension
On 2024-08-23 14:14, Eric Dumazet wrote:
> On Fri, Aug 23, 2024 at 7:31 PM Joe Damato <jdamato@...tly.com> wrote:
>>
>> From: Martin Karsten <mkarsten@...terloo.ca>
>>
>> The napi_suspend_irqs routine bootstraps irq suspension by elongating
>> the defer timeout to irq_suspend_timeout.
>>
>> The napi_resume_irqs routine effectly cancels irq suspension by forcing
>> the napi to be scheduled immediately.
>>
>> Signed-off-by: Martin Karsten <mkarsten@...terloo.ca>
>> Co-developed-by: Joe Damato <jdamato@...tly.com>
>> Signed-off-by: Joe Damato <jdamato@...tly.com>
>> Tested-by: Joe Damato <jdamato@...tly.com>
>> Tested-by: Martin Karsten <mkarsten@...terloo.ca>
>> ---
>> include/net/busy_poll.h | 3 +++
>> net/core/dev.c | 33 +++++++++++++++++++++++++++++++++
>> 2 files changed, 36 insertions(+)
>>
>> diff --git a/include/net/busy_poll.h b/include/net/busy_poll.h
>> index 9b09acac538e..f095b2bdeee1 100644
>> --- a/include/net/busy_poll.h
>> +++ b/include/net/busy_poll.h
>> @@ -52,6 +52,9 @@ void napi_busy_loop_rcu(unsigned int napi_id,
>> bool (*loop_end)(void *, unsigned long),
>> void *loop_end_arg, bool prefer_busy_poll, u16 budget);
>>
>> +void napi_suspend_irqs(unsigned int napi_id);
>> +void napi_resume_irqs(unsigned int napi_id);
>> +
>> #else /* CONFIG_NET_RX_BUSY_POLL */
>> static inline unsigned long net_busy_loop_on(void)
>> {
>> diff --git a/net/core/dev.c b/net/core/dev.c
>> index 74060ba866d4..4de0dfc86e21 100644
>> --- a/net/core/dev.c
>> +++ b/net/core/dev.c
>> @@ -6507,6 +6507,39 @@ void napi_busy_loop(unsigned int napi_id,
>> }
>> EXPORT_SYMBOL(napi_busy_loop);
>>
>> +void napi_suspend_irqs(unsigned int napi_id)
>> +{
>> + struct napi_struct *napi;
>> +
>> + rcu_read_lock();
>> + napi = napi_by_id(napi_id);
>> + if (napi) {
>> + unsigned long timeout = READ_ONCE(napi->dev->irq_suspend_timeout);
>> +
>> + if (timeout)
>> + hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
>> + }
>> + rcu_read_unlock();
>> +}
>> +EXPORT_SYMBOL(napi_suspend_irqs);
>> +
>> +void napi_resume_irqs(unsigned int napi_id)
>> +{
>> + struct napi_struct *napi;
>> +
>> + rcu_read_lock();
>> + napi = napi_by_id(napi_id);
>> + if (napi) {
>> + if (READ_ONCE(napi->dev->irq_suspend_timeout)) {
>
>
> Since we'll read irq_suspend_timeout twice, we could have a situation
> where the napi_schedule() will not be done
> if another thread changes irq_suspend_timeout ?
>
> If this is fine, a comment would be nice :)
>
> The thing is that the kernel can not trust the user (think of syzbot)
Yes, this should be fine. Calling napi_resume_irqs is done to restart
irq processing right away. In case irq_suspend_timeout is set to 0
between suspend and resume, the original value of irq_suspend_timeout
(when napi_suspend_irqs was called) determines the safety timeout as
intended and the watchdog will restart irq processing. We will a add
comment to make this clear.
Thanks,
Martin
Powered by blists - more mailing lists