Hi Arnd,
On Thursday 20 April 2017 02:53 PM, Arnd Bergmann wrote:
>> davinci_all_defconfig (arm) — PASS, 0 errors, 0 warnings, 0 section
>> mismatches
>>
>> Section mismatches:
>> WARNING: modpost: Found 1 section mismatch(es).
>> WARNING: modpost: Found 1 section mismatch(es).
> The 'section mismatches' detection in kernelci appears to be broken, so we
> don't actually see what happened. I can't reproduce it at the moment,
> so it's likely that this is fixed by an older patch of mine:
I cannot reproduce this as well.
>
> commit aae89d7dddb831aece31997cdc1c5014fb5a51c1
> Author: Arnd Bergmann <arnd@xxxxxxxx>
> Date: Sat Oct 10 21:19:48 2015 +0200
>
> [WRONG] davinci_mmc: remove incorrect __exit annotation
>
> WARNING: drivers/built-in.o(.exit.text+0x28ec): Section mismatch
> in reference from the function davinci_mmcsd_remove() to the function
> .init.text:davinci_release_dma_channels()
> The function __exit davinci_mmcsd_remove() references
> a function __init davinci_release_dma_channels().
> This is often seen when error handling in the exit function
> uses functionality in the init path.
> The fix is often to remove the __init annotation of
> davinci_release_dma_channels() so it may be used outside an init section.
>
> Signed-off-by: Arnd Bergmann <arnd@xxxxxxxx>
>
> diff --git a/drivers/mmc/host/davinci_mmc.c b/drivers/mmc/host/davinci_mmc.c
> index 621ce47e0e4a..9758d42f19a1 100644
> --- a/drivers/mmc/host/davinci_mmc.c
> +++ b/drivers/mmc/host/davinci_mmc.c
> @@ -496,7 +496,7 @@ static int mmc_davinci_start_dma_transfer(struct
> mmc_davinci_host *host,
> return ret;
> }
>
> -static void __init_or_module
> +static void
> davinci_release_dma_channels(struct mmc_davinci_host *host)
> {
> if (!host->use_dma)
> @@ -1361,7 +1361,7 @@ static int __init davinci_mmcsd_probe(struct
> platform_device *pdev)
> return ret;
> }
>
> -static int __exit davinci_mmcsd_remove(struct platform_device *pdev)
> +static int davinci_mmcsd_remove(struct platform_device *pdev)
> {
> struct mmc_davinci_host *host = platform_get_drvdata(pdev);
>
> @@ -1414,7 +1414,7 @@ static struct platform_driver davinci_mmcsd_driver = {
> .pm = davinci_mmcsd_pm_ops,
> .of_match_table = davinci_mmc_dt_ids,
> },
> - .remove = __exit_p(davinci_mmcsd_remove),
> + .remove = davinci_mmcsd_remove,
> .id_table = davinci_mmc_devtype,
> };
>
> This is a very old patch and I don't remember why I never submitted
> it or marked it as [WRONG], and I don't see why it only now showed
> up in kernelci.
I quite don't see how the existing code is wrong.
davinci_release_dma_channels() is marked __init_or_module and is
accessed in the .remove() routine. If modules are enabled,
davinci_release_dma_channels() is available at the time of module
removal since __init_or_module evaluates to nothing. And if modules are
disabled, then davinci_release_dma_channels() is marked as __init but
davinci_mmcsd_remove() is never called.
Thanks,
Sekhar
