[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <253jzr8juvl.fsf@nvidia.com>
Date: Fri, 27 Oct 2023 12:07:58 +0300
From: Aurelien Aptel <aaptel@...dia.com>
To: Jiri Pirko <jiri@...nulli.us>
Cc: linux-nvme@...ts.infradead.org, netdev@...r.kernel.org,
sagi@...mberg.me, hch@....de, kbusch@...nel.org, axboe@...com,
chaitanyak@...dia.com, davem@...emloft.net, kuba@...nel.org, Boris
Pismenny <borisp@...dia.com>, aurelien.aptel@...il.com, smalin@...dia.com,
malin1024@...il.com, ogerlitz@...dia.com, yorayz@...dia.com,
galshalom@...dia.com, mgurtovoy@...dia.com, edumazet@...gle.com,
pabeni@...hat.com, dsahern@...nel.org, imagedong@...cent.com,
ast@...nel.org, jacob.e.keller@...el.com
Subject: Re: [PATCH v17 01/20] net: Introduce direct data placement tcp offload
Jiri Pirko <jiri@...nulli.us> writes:
>>@@ -2134,6 +2146,9 @@ struct net_device {
>> netdev_features_t mpls_features;
>> netdev_features_t gso_partial_features;
>>
>>+#ifdef CONFIG_ULP_DDP
>>+ struct ulp_ddp_netdev_caps ulp_ddp_caps;
>
> Why can't you have this inside the driver? You have set_caps/get_stats
> ops. Try to avoid netdev struct pollution.
Ok, we will move ulp_ddp_caps to the driver and add a get_caps() operation.
>>+struct netlink_ulp_ddp_stats {
> There is nothing "netlink" about this. Just stats. Exposed over netlink,
> yes, but that does not need the prefix.
Ok, we will remove the netlink prefix.
>>+enum {
>>+ ULP_DDP_C_NVME_TCP_BIT,
>>+ ULP_DDP_C_NVME_TCP_DDGST_RX_BIT,
>>+
>>+ /*
>>+ * add capabilities above and keep in sync with
>>+ * Documentation/netlink/specs/ulp_ddp.yaml
>
> Wait what? Why do you need this at all? Just use the uapi enum.
The generated enum does not define a "count" (ULP_DDP_C_COUNT) which we
need to know how big the bitfield should be. Maybe the code generator
can be patched to add a #define with the number of values in the enum?
Powered by blists - more mailing lists