On Thu, Aug 18, 2022 at 04:19:52PM +0000, Marcel Ziswiler wrote:
> Hi there
>
> If one enables both CONFIG_IMX_BOOTAUX and CONFIG_LTO (e.g. just adding the
> later to the verdin-
> imx8mp_defconfig) one gets the following build failure:
>
> arch/arm/mach-imx/imx_bootaux.c:24:31: warning: type of ‘hostmap’ does not
> match original declaration [-Wlto-
> type-mismatch]
> 24 | const __weak struct rproc_att hostmap[] = { };
> | ^
> arch/arm/mach-imx/imx8m/soc.c:1578:24: note: array types have different bounds
> 1578 | const struct rproc_att hostmap[] = {
> | ^
> arch/arm/mach-imx/imx8m/soc.c:1578:24: note: ‘hostmap’ was previously
> declared here
>
> This is with Arm's gcc 10.3-2021.07 (arm-10.29).
>
> I tried the approach with adding pragmas around it e.g. as follows:
>
> #pragma GCC diagnostic push
> #pragma GCC diagnostic ignored "-Wlto-type-mismatch"
> const __weak struct rproc_att hostmap[] = { };
> #pragma GCC diagnostic pop
>
> But that does not seem to improve the situation. Is anybody more familiar
> with this stuff and what exactly
> needs to be done to fix this?So it looks like LTO and weak const structures don't get along? I would file a gcc bug to see if it'll get fixed, or if we'll be told to stop doing that as it's a bad idea. That said, maybe we should rework the code in question? Is there a valid case to be building arch/arm/mach-imx/imx_bootaux.c and not have a non-empty hostmap defined? -- Tom
signature.asc
Description: PGP signature
