On Mon, 25 Jan 2021 at 18:40, Florian Fainelli [off-list ref] wrote:

+Nicolas,

On 1/25/2021 4:50 AM, Arnd Bergmann wrote:

quoted

From: Arnd Bergmann <arnd@arndb.de>

sdhci_pltfm_suspend() is only available when CONFIG_PM_SLEEP
support is built into the kernel, which caused a regression
in a recent bugfix:

ld.lld: error: undefined symbol: sdhci_pltfm_suspend

quoted

quoted

quoted

referenced by sdhci-brcmstb.c
              mmc/host/sdhci-brcmstb.o:(sdhci_brcmstb_shutdown) in archive drivers/built-in.a

Making the call conditional on the symbol fixes the link
error.

Fixes: 5b191dcba719 ("mmc: sdhci-brcmstb: Fix mmc timeout errors on S5 suspend")
Fixes: e7b5d63a82fe ("mmc: sdhci-brcmstb: Add shutdown callback")
Cc: stable@vger.kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
It would be helpful if someone could test this to ensure that the
driver works correctly even when CONFIG_PM_SLEEP is disabled

Why not create stubs for sdhci_pltfm_suspend() when CONFIG_PM_SLEEP=n? I
don't think this is going to be a functional issue given that the
purpose of having the .shutdown() function is to save power if we cannot
that is fine, too.
--
Florian

I would prefer this approach - we shouldn't leave stub functions
unimplemented, which is what looks to me.

I just posted a new patch for this, please have a look and test it.

Kind regards
Uffe