Re: [PATCH] wifi: brcmfmac: keep ARP and ND offload off on the BCM4364

From: Aditya Garg

Date: Wed Nov 18 2026 - 02:09:04 EST




On 17/11/26 02:07, Arend van Spriel wrote:
> On Wed, 4 Nov 2026 16:16:20 +0100, Andre Eikmeyer <dev@deq.rocks> wrote:
>> brcmfmac enables the firmware's ARP and IPv6 ND offload (ndoe) whenever
>> the interface is up and not promiscuous. On the Apple firmware for the
>> BCM4364, ndoe makes IPv6 downloads collapse.
>
> [...]
>
>> ---
>> .../net/wireless/broadcom/brcm80211/brcmfmac/core.c | 10 ++++++++++
>> 1 file changed, 10 insertions(+)
>
>> @@ -103,6 +104,15 @@ void brcmf_configure_arp_nd_offload(struct brcmf_if *ifp, bool enable)
>> return;
>> }
>>
>> + /* With ND offload on while awake, the Apple firmware for the BCM4364
>> + * makes IPv6 downloads drop to low rates. macOS runs ARP and ND
>> + * offload only in sleep, keep both off on this chip.
>> + */
>> + if (enable && ifp->drvr->bus_if->chip == BRCM_CC_4364_CHIP_ID) {
>> + brcmf_dbg(TRACE, "Skip ARP/ND offload enable on BCM4364\n");
>> + enable = false;
>> + }
>> +
>> if (enable)
>> mode = BRCMF_ARP_OL_AGENT | BRCMF_ARP_OL_PEER_AUTO_REPLY;
>> else
>
> A few comments:
>
> 1) The measurements only implicate ndoe. Do you have data showing arpoe
> also needs to be off? Otherwise please limit this to ND offload.

Limiting this to ND offload works for my MacBook Pro 16,1. We have also got a bug report stating BCM4355 is also affected.>
> 2) The commit message says this is an issue of the Apple firmware, but the
> check is on the chip ID only. The same chip IDs are also used with other
> firmware builds and boards (WCC/non-Apple), which would lose the feature
> needlessly. Please restrict the quirk to the affected platform, e.g. with
> the Apple board type (settings->board_type) or a per-firmware quirk, or
> move it to the vendor feat_attach callback instead of a chip check in
> this helper. The same applies to the other chip ID quirks of this batch
> (txbf and FWAUTH).
>
> 3) This looks like a bug fix. Please add a Fixes: tag and Cc: stable (as
> Aditya suggested), and mark the target tree in the subject.
>
> [...]
>
> Regards,
> Arend

Arend and Andre, could you see if this works:

-->8--