BUG: KCSAN: data-race in rtl8169_poll

Hi all,

In the vanilla torvalds tree kernel on Ubuntu 22.04, commit 6.6.0-rc1-kcsan-00269-ge789286468a9,
KCSAN discovered a data-race in rtl8169_poll():

[ 9591.740976] ==================================================================
[ 9591.740990] BUG: KCSAN: data-race in rtl8169_poll (drivers/net/ethernet/realtek/r8169_main.c:4430 drivers/net/ethernet/realtek/r8169_main.c:4583) r8169

[ 9591.741060] race at unknown origin, with read to 0xffff888109773130 of 4 bytes by interrupt on cpu 21:
[ 9591.741073] rtl8169_poll (drivers/net/ethernet/realtek/r8169_main.c:4430 drivers/net/ethernet/realtek/r8169_main.c:4583) r8169
[ 9591.741135] __napi_poll (net/core/dev.c:6527)
[ 9591.741149] net_rx_action (net/core/dev.c:6596 net/core/dev.c:6727)
[ 9591.741161] __do_softirq (kernel/softirq.c:553)
[ 9591.741175] __irq_exit_rcu (kernel/softirq.c:427 kernel/softirq.c:632)
[ 9591.741185] irq_exit_rcu (kernel/softirq.c:647)
[ 9591.741194] common_interrupt (arch/x86/kernel/irq.c:247 (discriminator 14))
[ 9591.741206] asm_common_interrupt (./arch/x86/include/asm/idtentry.h:636)
[ 9591.741217] cpuidle_enter_state (drivers/cpuidle/cpuidle.c:291)
[ 9591.741227] cpuidle_enter (drivers/cpuidle/cpuidle.c:390)
[ 9591.741237] call_cpuidle (kernel/sched/idle.c:135)
[ 9591.741249] do_idle (kernel/sched/idle.c:219 kernel/sched/idle.c:282)
[ 9591.741259] cpu_startup_entry (kernel/sched/idle.c:378 (discriminator 1))
[ 9591.741268] start_secondary (arch/x86/kernel/smpboot.c:210 arch/x86/kernel/smpboot.c:294)
[ 9591.741281] secondary_startup_64_no_verify (arch/x86/kernel/head_64.S:433)

[ 9591.741300] value changed: 0x80003fff -> 0x34044510

[ 9591.741314] Reported by Kernel Concurrency Sanitizer on:
[ 9591.741322] CPU: 21 PID: 0 Comm: swapper/21 Tainted: G             L     6.6.0-rc1-kcsan-00269-ge789286468a9-dirty #4
[ 9591.741334] Hardware name: ASRock X670E PG Lightning/X670E PG Lightning, BIOS 1.21 04/26/2023
[ 9591.741343] ==================================================================

(The taint is not from the proprietary module, but triggered from the previous reported and unfixed bug.)

Apparently, it is this code:

static int rtl8169_poll(struct napi_struct *napi, int budget)
{
	struct rtl8169_private *tp = container_of(napi, struct rtl8169_private, napi);
	struct net_device *dev = tp->dev;
	int work_done;

	rtl_tx(dev, tp, budget);

→	work_done = rtl_rx(dev, tp, budget);

	if (work_done < budget && napi_complete_done(napi, work_done))
		rtl_irq_enable(tp);

	return work_done;
}

and

static int rtl_rx(struct net_device *dev, struct rtl8169_private *tp, int budget)
{
	struct device *d = tp_to_dev(tp);
	int count;

	for (count = 0; count < budget; count++, tp->cur_rx++) {
		unsigned int pkt_size, entry = tp->cur_rx % NUM_RX_DESC;
		struct RxDesc *desc = tp->RxDescArray + entry;
		struct sk_buff *skb;
		const void *rx_buf;
		dma_addr_t addr;
		u32 status;

→		status = le32_to_cpu(desc->opts1);
		if (status & DescOwn)
			break;

		/* This barrier is needed to keep us from reading
		 * any other fields out of the Rx descriptor until
		 * we know the status of DescOwn
		 */
		dma_rmb();

		if (unlikely(status & RxRES)) {
.
.
.

The reason isn't obvious, so it might be interesting if this is a valid report and whether it caused spurious corruption
of the network data on Realtek 8169 compatible cards ...

Hope this helps.

Best regards,
Mirsad Todorovac