On 2024-11-05 15:09, Mina Almasry wrote:
> On Fri, Nov 1, 2024 at 2:16 PM Pavel Begunkov <asml.silence@xxxxxxxxx> wrote:
>>
>> On 11/1/24 20:11, Mina Almasry wrote:
>>> On Tue, Oct 29, 2024 at 4:06 PM David Wei <dw@xxxxxxxxxxx> wrote:
>>>>
>>> ...
>>>> +static void io_zcrx_get_buf_uref(struct net_iov *niov)
>>>> +{
>>>> + atomic_long_add(IO_ZC_RX_UREF, &niov->pp_ref_count);
>>>> +}
>>>> +
>>>
>>> This is not specific to io_rcrx I think. Please rename this and put it
>>> somewhere generic, like netmem.h.
>>>
>>> Then tcp_recvmsg_dmabuf can use the same helper instead of the very
>>> ugly call it currently does:
>>>
>>> - atomic_long_inc(&niov->pp_ref_count);
>>> + net_iov_pp_ref_get(niov, 1);
>>>
>>> Or something.
>>>
>>> In general I think io_uring code can do whatever it wants with the
>>> io_uring specific bits in net_iov (everything under net_area_owner I
>>> think), but please lets try to keep any code touching the generic
>>> net_iov fields (pp_pagic, pp_ref_count, and others) in generic
>>> helpers.
>>
>> I'm getting confused, io_uring shouldn't be touching these
>> fields, but on the other hand should export net/ private
>> netmem_priv.h and page_pool_priv.h and directly hard code a bunch
>> of low level setup io_uring that is currently in page_pool.c
>>
>
> The only thing requested from this patch is to turn
> io_zcrx_get_buf_uref into something more generic. I'm guessing your
> confusion is following my other comments in "[PATCH v7 06/15] net:
> page pool: add helper creating area from pages". Let me take a closer
> look at my feedback there.
>
Sounds good, I'll rename io_zcrx_get_buf_uref() to something more
generic. But I'll leave changing the existing calls for a future patch.
