[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAK8P3a1kn-y5a+M=ByLJh17Ky6=wWKhwfUDhAa8pmKL+56xmJQ@mail.gmail.com>
Date: Mon, 18 Sep 2017 22:13:44 +0200
From: Arnd Bergmann <arnd@...db.de>
To: Lorenzo Pieralisi <lorenzo.pieralisi@....com>
Cc: "kernelci.org bot" <bot@...nelci.org>,
Kernel Build Reports Mailman List
<kernel-build-reports@...ts.linaro.org>,
Ralf Baechle <ralf@...ux-mips.org>,
Steve French <smfrench@...il.com>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: mainline/master build: 212 builds: 2 failed, 210 passed, 2
errors, 55 warnings (v4.14-rc1-9-g0666f560b71b)
On Mon, Sep 18, 2017 at 7:24 PM, Lorenzo Pieralisi
<lorenzo.pieralisi@....com> wrote:
> On Mon, Sep 18, 2017 at 05:56:02PM +0200, Arnd Bergmann wrote:
>
> [...]
>
>> > Section mismatches summary:
>> > 27 WARNING: modpost: Found 1 section mismatch(es).
>> > 6 WARNING: modpost: Found 2 section mismatch(es).
>>
>> These are all MIPS warnings caused by Lorenzo's
>>
>> 04c81c7293df ("MIPS: PCI: Replace pci_fixup_irqs() call with host
>> bridge IRQ mapping hooks")
>>
>> The problem is that now we can reference pci_fixup_irqs
>> from a function that is not marked __init, and probably
>> should not be, since pcibios_scanbus() itself can be called
>> at any time. Even before Lorenzo's change, having
>> pcibios_map_irq as an __init function was problematic AFAICT
>> on any platform that has hotplug-capable PCIe slots, or
>> that can have a PCI host bridge driver in a loadable module.
>>
>> The safe fix would be to remove all the __init annotations as well
>> as the __initdata and __initconst annotations on data referenced
>> by the map_irq functions.
>
> Right. Patch below, I will send it out tomorrow (I certainly missed
> some annotations) please let me know if that's reasonable.
I tried the same thing, but ran into a problem at this point:
> diff --git a/arch/mips/txx9/rbtx4927/setup.c b/arch/mips/txx9/rbtx4927/setup.c
> index f5b367e..29b7179 100644
> --- a/arch/mips/txx9/rbtx4927/setup.c
> +++ b/arch/mips/txx9/rbtx4927/setup.c
> @@ -354,7 +354,7 @@ static void __init rbtx4927_device_init(void)
> rbtx4927_gpioled_init();
> }
>
> -struct txx9_board_vec rbtx4927_vec __initdata = {
> +struct txx9_board_vec rbtx4927_vec = {
> .system = "Toshiba RBTX4927",
> .prom_init = rbtx4927_prom_init,
> .mem_setup = rbtx4927_mem_setup,
Now we have the next problem with the pointers inside of rbtx4927_vec
that kbuild complains about pointing to __init functions. I think the
txx9_board_vec structures all have to remain __initdata or __initconst,
with some other way to get the pointer at runtime.
something like this maybe:
diff --git a/arch/mips/txx9/generic/pci.c b/arch/mips/txx9/generic/pci.c
index 0bd2a1e1ff9a..f0b291c889c4 100644
--- a/arch/mips/txx9/generic/pci.c
+++ b/arch/mips/txx9/generic/pci.c
@@ -386,9 +386,11 @@ int pcibios_plat_dev_init(struct pci_dev *dev)
return 0;
}
-int __init pcibios_map_irq(const struct pci_dev *dev, u8 slot, u8 pin)
+static int (*txx9_pci_map_irq)(const struct pci_dev *dev, u8 slot, u8 pin);
+
+int pcibios_map_irq(const struct pci_dev *dev, u8 slot, u8 pin)
{
- return txx9_board_vec->pci_map_irq(dev, slot, pin);
+ return txx9_pci_map_irq(dev, slot, pin);
}
char * (*txx9_board_pcibios_setup)(char *str) __initdata;
@@ -424,5 +426,7 @@ char *__init txx9_pcibios_setup(char *str)
txx9_pci_err_action = TXX9_PCI_ERR_IGNORE;
return NULL;
}
+ txx9_pci_map_irq = txx9_board_vec->pci_map_irq;
+
return str;
}
Arnd
Powered by blists - more mailing lists