[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <39e701b4-0992-2c96-67b3-38c341c77af5@linux.dev>
Date: Tue, 15 Aug 2023 19:38:41 +0100
From: Vadim Fedorenko <vadim.fedorenko@...ux.dev>
To: Jakub Kicinski <kuba@...nel.org>
Cc: Jiri Pirko <jiri@...nulli.us>,
Arkadiusz Kubalewski <arkadiusz.kubalewski@...el.com>,
Jonathan Lemon <jonathan.lemon@...il.com>, Paolo Abeni <pabeni@...hat.com>,
Milena Olech <milena.olech@...el.com>,
Michal Michalik <michal.michalik@...el.com>,
linux-arm-kernel@...ts.infradead.org, poros@...hat.com, mschmidt@...hat.com,
netdev@...r.kernel.org, linux-clk@...r.kernel.org,
Bart Van Assche <bvanassche@....org>, intel-wired-lan@...ts.osuosl.org,
Jiri Pirko <jiri@...dia.com>
Subject: Re: [PATCH net-next v4 3/9] dpll: core: Add DPLL framework base
functions
On 15/08/2023 19:28, Jakub Kicinski wrote:
> On Tue, 15 Aug 2023 19:20:31 +0100 Vadim Fedorenko wrote:
>>>> + ret = xa_alloc(&dpll_device_xa, &dpll->id, dpll, xa_limit_16b,
>>>> + GFP_KERNEL);
>>>
>>> Why only 16b and why not _cyclic?
>>
>> I cannot image systems with more than 65k of DPLL devices. We don't
>> store any id's of last used DPLL device, so there is no easy way to
>> restart the search from the last point. And it's not a hot path to
>> optimize it.
>
> I think this gets used under the xa_lock() so you can just add a static
> variable inside the function to remember previous allocation.
>
> I don't expect >64k devices either, obviously, but what are we saving
> by not allowing the full u32 range?
I don't see any benefits for either _cyclic or u32 range, but if you
insist I can change it.
Powered by blists - more mailing lists