Re: next-20250813 s390 allyesconfig undefined reference to `stmmac_simple_pm_ops'
From: "Arnd Bergmann" <arnd@arndb.de>
Date: 2025-08-19 17:35:44
Also in:
linux-s390, lkml, regressions
On Tue, Aug 19, 2025, at 18:00, Alexander Gordeev wrote:
quoted hunk ↗ jump to hunk
On Tue, Aug 19, 2025 at 03:07:56PM +0530, Naresh Kamboju wrote: CONFIG_PM is not defined on s390 and as result stmmac_simple_pm_ops ends up in _DISCARD_PM_OPS(). The below patch fixes the linking, but it is by no means a correct solution:diff --git a/drivers/net/ethernet/stmicro/stmmac/dwmac-loongson.cb/drivers/net/ethernet/stmicro/stmmac/dwmac-loongson.c index 5769165ee5ba..d475a77e4871 100644--- a/drivers/net/ethernet/stmicro/stmmac/dwmac-loongson.c +++ b/drivers/net/ethernet/stmicro/stmmac/dwmac-loongson.c@@ -668,7 +668,7 @@ static struct pci_driver loongson_dwmac_driver = { .probe = loongson_dwmac_probe, .remove = loongson_dwmac_remove, .driver = { - .pm = &stmmac_simple_pm_ops, + .pm = &__static_stmmac_simple_pm_ops,
The correct solution is to make the PM_SLEEP versions use this
.pm = pm_sleep_ptr(stmmac_simple_pm_ops),
or the corresponding version for the PM_RUNTIME+PM_SLEEP drivers:
.pm = pm_ptr(stmmac_pltfrm_pm_ops),
By convention, the pm_ptr()/pm_sleep_ptr() macro should be
used for any driver using DEFINE_DEV_PM_OPS() or its variants,
though missing that does not produce a warning for non-exported
options and only wastes a few bytes of .data.
#define _DISCARD_PM_OPS(name, license, ns) \ - static __maybe_unused const struct dev_pm_ops __static_##name + __maybe_unused const struct dev_pm_ops __static_##name
This would cause a lot of link failures elsewhere, since _DISCARD_PM_OPS
needs to ensure the operations are discarded by the compiler, which does
not happen when they are defined as a global symbol.
The idea of making this a 'static __maybe_unused' symbol is that
the actual functions get discarded as well but don't need an individual
__maybe_unused annotation or an #ifdef around them to prevent a
warning for unused symbols.
Arnd