Re: [PATCH net] net: stmmac: dwmac-rk: keep the GMAC powered when the PHY wakes the system

[Igor Velkov <iav@iav.lv>]

On Wed, Sep 30, 2026 at 11:51:45PM +0200, Andrew Lunn wrote:
> You have not answered the question why is hangs? Given the current
> description, this just sounds like a workaround, not a fix.

It hangs because the PHY interrupt handler accesses the powered-down MAC.

stmmac sets mac_managed_pm, so mdio_bus_phy_suspend() returns before
setting irq_suspended and phy_interrupt() runs the handler at once.
rtl8211f_handle_interrupt() reads INSR over this GMAC's MDIO bus.
From rk_gmac_suspend() to suspend_late, and from resume_early to
rk_gmac_resume(), runtime PM is still enabled and that read reaches
the MAC with aclk_mac/pclk_mac gated.

On RK3399 such a read never completes: a test readl() right after
rk_gmac_powerdown() hung the board with no soft-lockup panic until
the hardware watchdog reset it (>5 min with the watchdog off).
Refusing MDIO access to the powered-down MAC: 10/10 suspends without
a hang; the same cycle script without it hung in 4 of 5 runs, each
time within three cycles. With the GMAC kept powered (v1): 45/45.

v2 will use device_wakeup_path(), and its Fixes tag will be
d65cb2e27e6e: that commit, not 6911308d7d11, stopped setting the MAC
wakeup flag.

pw-bot: cr

--
Igor Velkov