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:   Alexander Lobakin <alexandr.lobakin@intel.com>
To:     Horatiu Vultur <horatiu.vultur@microchip.com>
Cc:     Alexander Lobakin <alexandr.lobakin@intel.com>,
        netdev@vger.kernel.org, linux-kernel@vger.kernel.org,
        bpf@vger.kernel.org, davem@davemloft.net, edumazet@google.com,
        kuba@kernel.org, pabeni@redhat.com, ast@kernel.org,
        daniel@iogearbox.net, hawk@kernel.org, john.fastabend@gmail.com,
        UNGLinuxDriver@microchip.com
Subject: Re: [PATCH net-next v3 7/7] net: lan966x: Add support for XDP_REDIRECT
Date:   Tue, 22 Nov 2022 13:04:30 +0100
Message-Id: <20221122120430.419770-1-alexandr.lobakin@intel.com>
In-Reply-To: <20221121212850.3212649-8-horatiu.vultur@microchip.com>
References: <20221121212850.3212649-1-horatiu.vultur@microchip.com> <20221121212850.3212649-8-horatiu.vultur@microchip.com>
MIME-Version: 1.0
Content-Type: text/plain
Content-Transfer-Encoding: 8bit
Precedence: bulk

From: Horatiu Vultur <horatiu.vultur@microchip.com>
Date: Mon, 21 Nov 2022 22:28:50 +0100

> Extend lan966x XDP support with the action XDP_REDIRECT. This is similar
> with the XDP_TX, so a lot of functionality can be reused.
> 
> Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
> ---
>  .../ethernet/microchip/lan966x/lan966x_fdma.c | 83 +++++++++++++++----
>  .../ethernet/microchip/lan966x/lan966x_main.c |  1 +
>  .../ethernet/microchip/lan966x/lan966x_main.h | 10 ++-
>  .../ethernet/microchip/lan966x/lan966x_xdp.c  | 31 ++++++-
>  4 files changed, 109 insertions(+), 16 deletions(-)

[...]

> @@ -558,6 +575,10 @@ static int lan966x_fdma_napi_poll(struct napi_struct *napi, int weight)
>  		case FDMA_TX:
>  			lan966x_fdma_rx_advance_dcb(rx);
>  			continue;
> +		case FDMA_REDIRECT:
> +			lan966x_fdma_rx_advance_dcb(rx);
> +			redirect = true;
> +			continue;

I think you can save a couple lines here and avoid small code dup:

+		case FDMA_REDIRECT:
+			redirect = true;
+			fallthrough;
 		case FDMA_TX:
 			lan966x_fdma_rx_advance_dcb(rx);
 			continue;

The logics stays the same.

>  		case FDMA_DROP:
>  			lan966x_fdma_rx_free_page(rx);
>  			lan966x_fdma_rx_advance_dcb(rx);

[...]

> @@ -178,6 +180,7 @@ struct lan966x_tx_dcb_buf {
>  	struct net_device *dev;
>  	struct sk_buff *skb;
>  	struct xdp_frame *xdpf;
> +	bool xdp_ndo;

I suggest carefully inspecting this struct with pahole (or by just
printkaying its layout/sizes/offsets at runtime) and see if there's
any holes and how it could be optimized.
Also, it's just my personal preference, but it's not that unpopular:
I don't trust bools inside structures as they may surprise with
their sizes or alignment depending on the architercture. Considering
all the blah I wrote, I'd define it as:

struct lan966x_tx_dcb_buf {
	dma_addr_t dma_addr;		// can be 8 bytes on 32-bit plat
	struct net_device *dev;		// ensure natural alignment
	struct sk_buff *skb;
	struct xdp_frame *xdpf;
	u32 len;
	u32 xdp_ndo:1;			// put all your booleans here in
	u32 used:1;			// one u32
	...
};

BTW, we usually do union { skb, xdpf } since they're mutually
exclusive. And to distinguish between XDP and regular Tx you can use
one more bit/bool. This can also come handy later when you add XSk
support (you will be adding it, right? Please :P).

>  	int len;
>  	dma_addr_t dma_addr;
>  	bool used;

[...]

> -- 
> 2.38.0

Thanks,
Olek