Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: [PATCH] net: stmmac: Don't call _irqoff() with hardirqs enabled
To:     Jakub Kicinski <kuba@kernel.org>
Cc:     John Keeping <john@metanate.com>, netdev@vger.kernel.org,
        Giuseppe Cavallaro <peppe.cavallaro@st.com>,
        Alexandre Torgue <alexandre.torgue@st.com>,
        Jose Abreu <joabreu@synopsys.com>,
        "David S. Miller" <davem@davemloft.net>,
        Maxime Coquelin <mcoquelin.stm32@gmail.com>,
        linux-stm32@st-md-mailman.stormreply.com,
        linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org,
        Eric Dumazet <edumazet@google.com>
References: <20201008162749.860521-1-john@metanate.com>
 <8036d473-68bd-7ee7-e2e9-677ff4060bd3@gmail.com>
 <20201009085805.65f9877a@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
 <725ba7ca-0818-074b-c380-15abaa5d037b@gmail.com>
 <070b2b87-f38c-088d-4aaf-12045dbd92f7@gmail.com>
 <20201010082248.22cc7656@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
From:   Heiner Kallweit <hkallweit1@gmail.com>
Message-ID: <c73866a9-2ee8-b549-f578-75d62b9263b4@gmail.com>
Date:   Sun, 11 Oct 2020 11:24:41 +0200
User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:68.0) Gecko/20100101
 Thunderbird/68.12.1
MIME-Version: 1.0
In-Reply-To: <20201010082248.22cc7656@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 10.10.2020 17:22, Jakub Kicinski wrote:
> On Sat, 10 Oct 2020 15:08:15 +0200 Heiner Kallweit wrote:
>> On 09.10.2020 18:06, Heiner Kallweit wrote:
>>> On 09.10.2020 17:58, Jakub Kicinski wrote:  
>>>> On Fri, 9 Oct 2020 16:54:06 +0200 Heiner Kallweit wrote:  
>>>>> I'm thinking about a __napi_schedule version that disables hard irq's
>>>>> conditionally, based on variable force_irqthreads, exported by the irq
>>>>> subsystem. This would allow to behave correctly with threadirqs set,
>>>>> whilst not loosing the _irqoff benefit with threadirqs unset.
>>>>> Let me come up with a proposal.  
>>>>
>>>> I think you'd need to make napi_schedule_irqoff() behave like that,
>>>> right?  Are there any uses of napi_schedule_irqoff() that are disabling
>>>> irqs and not just running from an irq handler?
>>>>  
>>> Right, the best approach depends on the answer to the latter question.
>>> I didn't check this yet, therefore I described the least intrusive approach.
>>>   
>>
>> With some help from coccinelle I identified the following functions that
>> call napi_schedule_irqoff() or __napi_schedule_irqoff() and do not run
>> from an irq handler (at least not at the first glance).
>>
>> dpaa2_caam_fqdan_cb
>> qede_simd_fp_handler
>> mlx4_en_rx_irq
>> mlx4_en_tx_irq
> 
> Don't know the others but FWIW the mlx4 ones run from an IRQ handler,
> AFAICT:
> 
> static irqreturn_t mlx4_interrupt(int irq, void *dev_ptr)
> static irqreturn_t mlx4_msi_x_interrupt(int irq, void *eq_ptr)
>   mlx4_eq_int()
>     mlx4_cq_completion
>       cq->comp()
> 
>> qeth_qdio_poll
>> netvsc_channel_cb
>> napi_watchdog
> 
> This one runs from a hrtimer, which I believe will be a hard irq
> context on anything but RT. I could be wrong.
> 

A similar discussion can be found e.g. here:
https://lore.kernel.org/netdev/20191126222013.1904785-1-bigeasy@linutronix.de/
However I don't see any actual outcome.