On Fri, Oct 09, 2026 at 11:02:52AM +0200, Daniel Machon wrote:
The fix seems OK, but it could have implications on other drivers, for an issue
that is dwmac1000 specific. Did you consider a fix that would only touch
dwmac1000?
Not before you asked; I have tried it now. Reading RGSMIIIS in
dwmac1000_irq_status(), the read that 2e2c878a3141 removed, stops the
storm as well. Worst case, the resume thread and the MAC interrupt on
one CPU, Wake-on-LAN resume: without a fix a soft lockup in 3 of 3
resumes on Helios64 and 2 of 2 on ODROID-HC4; with the read 0 in 10 on
each, irq_rgmii_n going up by one per resume. I'll send a v4 that
touches dwmac1000 only.
Igor