netdev - Re: [PATCH net-next v4] net: qualcomm: rmnet: Add side band flow control support

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <e8ca04f4-08a1-0d79-7916-fa853e9aeda6@quicinc.com>
Date: Tue, 10 Oct 2023 15:32:00 -0600
From: "Subash Abhinov Kasiviswanathan (KS)" <quic_subashab@...cinc.com>
To: Jakub Kicinski <kuba@...nel.org>
CC: <davem@...emloft.net>, <edumazet@...gle.com>, <pabeni@...hat.com>,
        <netdev@...r.kernel.org>, <vadim.fedorenko@...ux.dev>, <lkp@...el.com>,
        <horms@...nel.org>, Sean Tranchetti <quic_stranche@...cinc.com>
Subject: Re: [PATCH net-next v4] net: qualcomm: rmnet: Add side band flow
 control support



On 10/10/2023 12:21 PM, Jakub Kicinski wrote:
> On Tue, 10 Oct 2023 09:23:12 -0600 Subash Abhinov Kasiviswanathan (KS)
> wrote:
>>
>> I was wondering if it would be acceptable to add a user accessible
>> interface in core networking to stop_queue / wake_queue instead of the
>> driver.
> 
> Maybe not driver queue control but if there's no qdisc which allows
> users to pause from user space, I think that would be a much easier
> sale.
> 
> That said the flow of the whole thing seems a bit complex.
> Can't the driver somehow be notified by the device directly?
> User space will suffer from all sort of wake up / scheduling
> latencies, it'd be better if the whole sleep / wake thing was
> handled in the kernel.

Our userspace module relies on various inputs from radio hardware and 
has proprietary logic to determine when to transmit / stop sending 
packets corresponding to a specific bearer. I agree that an in kernel 
scheme might be faster than an userspace - kernel solution. However, I 
believe that this latency impact could be reduced through schemes like 
setting process priority, pinning applications in isolated cores etc.