netdev - Re: [PATCH net-next v14 06/13] rtase: Implement .ndo_start

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date: Tue, 12 Dec 2023 11:32:12 -0800
From: Jakub Kicinski <kuba@...nel.org>
To: Justin Lai <justinlai0215@...ltek.com>
Cc: <davem@...emloft.net>, <edumazet@...gle.com>, <pabeni@...hat.com>,
 <linux-kernel@...r.kernel.org>, <netdev@...r.kernel.org>, <andrew@...n.ch>,
 <pkshih@...ltek.com>, <larry.chiu@...ltek.com>
Subject: Re: [PATCH net-next v14 06/13] rtase: Implement .ndo_start_xmit
 function

On Fri, 8 Dec 2023 17:47:26 +0800 Justin Lai wrote:
> +static int tx_handler(struct rtase_ring *ring, int budget)

I don't see how this is called, the way you split the submission makes
it a bit hard to review, oh well. Anyway - if you pass the NAPI budget
here - that's not right, it may be 0, and you'd loop forever.
For Tx - you should try to reap some fixed number of packets, say 128,
the budget is for Rx, not for Tx.

> +	const struct rtase_private *tp = ring->ivec->tp;
> +	struct net_device *dev = tp->dev;
> +	int workdone = 0;
> +	u32 dirty_tx;
> +	u32 tx_left;
> +
> +	dirty_tx = ring->dirty_idx;
> +	tx_left = READ_ONCE(ring->cur_idx) - dirty_tx;
> +
> +	while (tx_left > 0) {
> +		u32 entry = dirty_tx % NUM_DESC;
> +		struct tx_desc *desc = ring->desc +
> +				       sizeof(struct tx_desc) * entry;
> +		u32 len = ring->mis.len[entry];
> +		u32 status;
> +
> +		status = le32_to_cpu(desc->opts1);
> +
> +		if (status & DESC_OWN)
> +			break;
> +
> +		rtase_unmap_tx_skb(tp->pdev, len, desc);
> +		ring->mis.len[entry] = 0;
> +		if (ring->skbuff[entry]) {
> +			dev_consume_skb_any(ring->skbuff[entry]);

napi_consume_skb, assuming you call this from NAPI

> +			ring->skbuff[entry] = NULL;
> +		}
> +
> +		dev->stats.tx_bytes += len;
> +		dev->stats.tx_packets++;
> +		dirty_tx++;
> +		tx_left--;
> +		workdone++;
> +
> +		if (workdone == budget)
> +			break;
> +	}
> +
> +	if (ring->dirty_idx != dirty_tx) {
> +		WRITE_ONCE(ring->dirty_idx, dirty_tx);
> +
> +		if (__netif_subqueue_stopped(dev, ring->index) &&
> +		    rtase_tx_avail(ring))
> +			netif_start_subqueue(dev, ring->index);

Please use the start / stop macros from include/net/netdev_queues.h
I'm pretty sure the current code is racy.

> +		if (ring->cur_idx != dirty_tx)
> +			rtase_w8(tp, RTASE_TPPOLL, BIT(ring->index));
> +	}
> +
> +	return workdone;
> +}

> +	/* multiqueues */
> +	q_idx = skb_get_queue_mapping(skb);
> +	ring = &tp->tx_ring[q_idx];

As Paolo pointed out elsewhere you seem to only support one queue.
Remove this indirection, please, and always use queue 0, otherwise
it's a bit confusing.