Re: [PATCH RFC v2 0/3] PCI: rockchip-host: Handle unavailable downstream config space
From: Igor Velkov
Date: Mon Nov 16 2026 - 02:12:22 EST
Tested v2 on Kobol Helios64 (RK3399, JMB585 at x2), 7.4-rc2, DT without
ep-gpios, 150 unbind/bind cycles, bind pinned alternately to A53 and A72.
A53: the fixup ran 4 times (debug print on the err path); probe found no
device and the JMB585 stayed absent until the next bind; no crash.
A72: the same read raised an asynchronous SError (bf000002) on the dmb in
__io_ar() after the guarded ldr; panic in 1 of 75 cycles.
Patch 1 does not catch this case: LINK_UP and training done are set, the
endpoint does not answer.
Retrying the link bring-up avoids the abort here:
https://lore.kernel.org/r/20261116-rockchip-pcie-resume-v2-1-a633e279be97@xxxxxx
With both applied: 4 bad bring-ups in 150 cycles, no device lost.
--
Igor Velkov