Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Jesper Dangaard Brouer <jbrouer@redhat.com>
Message-ID: <729b360c-4d79-1025-f5be-384b17f132d3@redhat.com>
Date:   Tue, 25 Jul 2023 20:01:03 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.10.0
Cc:     brouer@redhat.com, Dexuan Cui <decui@microsoft.com>,
        KY Srinivasan <kys@microsoft.com>,
        Paul Rosswurm <paulros@microsoft.com>,
        "olaf@aepfle.de" <olaf@aepfle.de>,
        "vkuznets@redhat.com" <vkuznets@redhat.com>,
        "davem@davemloft.net" <davem@davemloft.net>,
        "wei.liu@kernel.org" <wei.liu@kernel.org>,
        "edumazet@google.com" <edumazet@google.com>,
        "kuba@kernel.org" <kuba@kernel.org>,
        "pabeni@redhat.com" <pabeni@redhat.com>,
        "leon@kernel.org" <leon@kernel.org>,
        Long Li <longli@microsoft.com>,
        "ssengar@linux.microsoft.com" <ssengar@linux.microsoft.com>,
        "linux-rdma@vger.kernel.org" <linux-rdma@vger.kernel.org>,
        "daniel@iogearbox.net" <daniel@iogearbox.net>,
        "john.fastabend@gmail.com" <john.fastabend@gmail.com>,
        "bpf@vger.kernel.org" <bpf@vger.kernel.org>,
        "ast@kernel.org" <ast@kernel.org>,
        Ajay Sharma <sharmaajay@microsoft.com>,
        "hawk@kernel.org" <hawk@kernel.org>,
        "tglx@linutronix.de" <tglx@linutronix.de>,
        "shradhagupta@linux.microsoft.com" <shradhagupta@linux.microsoft.com>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        Ilias Apalodimas <ilias.apalodimas@linaro.org>
Subject: Re: [PATCH V3,net-next] net: mana: Add page pool for RX buffers
Content-Language: en-US
To:     Haiyang Zhang <haiyangz@microsoft.com>,
        Jesper Dangaard Brouer <jbrouer@redhat.com>,
        "linux-hyperv@vger.kernel.org" <linux-hyperv@vger.kernel.org>,
        "netdev@vger.kernel.org" <netdev@vger.kernel.org>
References: <1689966321-17337-1-git-send-email-haiyangz@microsoft.com>
 <1af55bbb-7aff-e575-8dc1-8ba64b924580@redhat.com>
 <PH7PR21MB3116F8A97F3626AB04915B96CA02A@PH7PR21MB3116.namprd21.prod.outlook.com>
 <PH7PR21MB311675E57B81B49577ADE98FCA02A@PH7PR21MB3116.namprd21.prod.outlook.com>
In-Reply-To: <PH7PR21MB311675E57B81B49577ADE98FCA02A@PH7PR21MB3116.namprd21.prod.outlook.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk


On 24/07/2023 20.35, Haiyang Zhang wrote:
> 
[...]
>>> On 21/07/2023 21.05, Haiyang Zhang wrote:
>>>> Add page pool for RX buffers for faster buffer cycle and reduce CPU
>>>> usage.
>>>>
>>>> The standard page pool API is used.
>>>>
>>>> Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
>>>> ---
>>>> V3:
>>>> Update xdp mem model, pool param, alloc as suggested by Jakub Kicinski
>>>> V2:
>>>> Use the standard page pool API as suggested by Jesper Dangaard Brouer
>>>>
>>>> ---
>>>>    drivers/net/ethernet/microsoft/mana/mana_en.c | 91 +++++++++++++++--
>> --
>>>>    include/net/mana/mana.h                       |  3 +
>>>>    2 files changed, 78 insertions(+), 16 deletions(-)
>>>>
>>>> diff --git a/drivers/net/ethernet/microsoft/mana/mana_en.c
>>> b/drivers/net/ethernet/microsoft/mana/mana_en.c
>>>> index a499e460594b..4307f25f8c7a 100644
>>>> --- a/drivers/net/ethernet/microsoft/mana/mana_en.c
>>>> +++ b/drivers/net/ethernet/microsoft/mana/mana_en.c
>>> [...]
>>>> @@ -1659,6 +1679,8 @@ static void mana_poll_rx_cq(struct mana_cq *cq)
>>>>
>>>>    	if (rxq->xdp_flush)
>>>>    		xdp_do_flush();
>>>> +
>>>> +	page_pool_nid_changed(rxq->page_pool, numa_mem_id());
>>>
>>> I don't think this page_pool_nid_changed() called is needed, if you do
>>> as I suggest below (nid = NUMA_NO_NODE).
>>>
>>>
>>>>    }
>>>>
>>>>    static int mana_cq_handler(void *context, struct gdma_queue
>>> *gdma_queue)
>>> [...]
>>>
>>>> @@ -2008,6 +2041,25 @@ static int mana_push_wqe(struct mana_rxq
>> *rxq)
>>>>    	return 0;
>>>>    }
>>>>
>>>> +static int mana_create_page_pool(struct mana_rxq *rxq)
>>>> +{
>>>> +	struct page_pool_params pprm = {};
>>>
>>> You are implicitly assigning NUMA node id zero.
>>>
>>>> +	int ret;
>>>> +
>>>> +	pprm.pool_size = RX_BUFFERS_PER_QUEUE;
>>>> +	pprm.napi = &rxq->rx_cq.napi;
>>>
>>> You likely want to assign pprm.nid to NUMA_NO_NODE
>>>
>>>    pprm.nid = NUMA_NO_NODE;
>>>
>>> For most drivers it is recommended to assign ``NUMA_NO_NODE`` (value -1)
>>> as the NUMA ID to ``pp_params.nid``. When ``CONFIG_NUMA`` is enabled
>>> this setting will automatically select the (preferred) NUMA node (via
>>> ``numa_mem_id()``) based on where NAPI RX-processing is currently
>>> running. The effect is that page_pool will only use recycled memory when
>>> NUMA node match running CPU. This assumes CPU refilling driver RX-ring
>>> will also run RX-NAPI.
>>>
>>> If a driver want more control over the NUMA node memory selection,
>>> drivers can assign (``pp_params.nid``) something else than
>>> `NUMA_NO_NODE`` and runtime adjust via function
>>> ``page_pool_nid_changed()``.
>>
>> Our driver is using NUMA 0 by default, so I implicitly assign NUMA node id
>> to zero during pool init.
>>
>> And, if the IRQ/CPU affinity is changed, the page_pool_nid_changed()
>> will update the nid for the pool. Does this sound good?
>>
> 
> Also, since our driver is getting the default node from here:
> 	gc->numa_node = dev_to_node(&pdev->dev);
> I will update this patch to set the default node as above, instead of implicitly
> assigning it to 0.
> 

In that case, I agree that it make sense to use dev_to_node(&pdev->dev), 
like:
	pprm.nid = dev_to_node(&pdev->dev);

Driver must have a reason for assigning gc->numa_node for this hardware,
which is okay. That is why page_pool API allows driver to control this.

But then I don't think you should call page_pool_nid_changed() like

	page_pool_nid_changed(rxq->page_pool, numa_mem_id());

Because then you will (at first packet processing event) revert the
dev_to_node() setting to use numa_mem_id() of processing/running CPU.
(In effect this will be the same as setting NUMA_NO_NODE).

I know, mlx5 do call page_pool_nid_changed(), but they showed benchmark
numbers that this was preferred action, even-when sysadm had
"misconfigured" the default smp_affinity RX-processing to happen on a
remote NUMA node.  AFAIK mlx5 keeps the descriptor rings on the
originally configured NUMA node that corresponds to the NIC PCIe slot.

--Jesper