Re: [PATCH 6.1 0000/1879] 6.1.191-rc2 review

[Jon Hunter]

On 01/11/2026 14:43, Sasha Levin wrote:
>> Bisecting is pointing to the below commit and reverting it fixes
>> the above issues ...
>>
>> # first bad commit: [26c5e136ba4e94ad708333c9bd34039a1de397b0] net: stmmac: protect updates of 64-bit statistics counters
> 
> Thanks for bisecting. 6.1 is missing a prerequisite for that backport.
> 
> 38cc3c6dcc09 ("net: stmmac: protect updates of 64-bit statistics
> counters") allocates priv->xstats.pcpu_stats in stmmac_dvr_probe().
> 6.1 does not have 2eb85b750512 ("net: stmmac: don't clear network
> statistics in .ndo_open()"), so the memset() of priv->xstats in
> __stmmac_open() zeroes that pointer on every ifup, and the first DMA
> interrupt on a dwmac4 device (Tegra186/194 EQOS) dereferences a NULL
> percpu pointer.
> 
> Greg has queued a revert of the backport for 6.1, so this goes away in
> the next 6.1 release. 38cc3c6dcc09 can come back to 6.1 later together
> with 2eb85b750512.
> 
> Could you test if adding in 2eb85b750512 fixes the issue?


I confirmed that that does fix it.

Thanks
Jon

-- 
nvpublic