2015-04-21 8:41 GMT-07:00 Guenter Roeck <linux@xxxxxxxxxxxx>:
> On Tue, Apr 21, 2015 at 01:45:35PM +0930, Rusty Russell wrote:
>> Aaro Koskinen <aaro.koskinen@xxxxxx> writes:
>> > Hi,
>> >
>> > On Mon, Apr 20, 2015 at 12:40:28PM -0700, Guenter Roeck wrote:
>> >> the upstream kernel fails to build mips:nlm_xlp_defconfig,
>> >> mips:nlm_xlp_defconfig, mips:cavium_octeon_defconfig, and possibly
>> >> other targets, with errors such as
>> >>
>> >> arch/mips/kernel/smp.c:211:2: error:
>> >> passing argument 2 of 'cpumask_set_cpu' discards 'volatile' qualifier
>> >> from pointer target type
>> >> arch/mips/kernel/process.c:52:2: error:
>> >> passing argument 2 of 'cpumask_test_cpu' discards 'volatile' qualifier
>> >> from pointer target type
>> >> arch/mips/cavium-octeon/smp.c:242:2: error:
>> >> passing argument 2 of 'cpumask_clear_cpu' discards 'volatile' qualifier
>> >> from pointer target type
>> >>
>> >> The problem was introduced with commit 8dd928915a73 (" mips: fix up
>> >> obsolete cpu function usage"). I would send a patch to fix it, but I
>> >> am not sure if removing 'volatile' from the variable declaration(s)
>> >> would be a good idea.
>> >
>> > I think removing volatile from cpu_callin_map declaration should be OK,
>> > since test_cpu (only reader) uses test_bit which takes care of it:
>> >
>> > static inline int test_bit(int nr, const volatile unsigned long *addr)
>>
>> No, that got replaced too, with cpumask_test_cpu AFAICT.
>>
>> You can open-code it, like so:
>>
>> test_bit(0, cpumask_bits(cpu_callin_map));
>>
>> But you probably want to put a barrier in that loop instead of relying
>> on volatile.
>>
> The following might do it. Note that I can not really test it since I don't have
> a real mips system, and qemu gets rcu hangs if I enable more than one CPU (I see
> that with older kernels as well, so it is not a new problem). Someone will have
> to test the patch on a real multi-core system.
Would be nice to get this merged ASAP as a build failure is not great.
See below for the tested tag, thanks for the fix!
>
> Guenter
>
> ---
> From 94026cc98a6b7b3567780a5443674c71202e2497 Mon Sep 17 00:00:00 2001
> From: Guenter Roeck <linux@xxxxxxxxxxxx>
> Date: Tue, 21 Apr 2015 08:31:01 -0700
> Subject: [PATCH] mips: Fix SMP builds
>
> Mips SMP builds fail with error messages similar to the following.
>
> arch/mips/kernel/smp.c:211:2: error:
> passing argument 2 of 'cpumask_set_cpu' discards 'volatile'
> qualifier from pointer target type
> arch/mips/kernel/process.c:52:2: error:
> passing argument 2 of 'cpumask_test_cpu' discards 'volatile'
> qualifier from pointer target type
> arch/mips/cavium-octeon/smp.c:242:2: error:
> passing argument 2 of 'cpumask_clear_cpu' discards 'volatile'
> qualifier from pointer target type
>
> cpu_callin_map is declared as volatile variable, but passed to various
> functions with non-volatile arguments. Make it non-volatile and add a
> memory barrier at the one location where volatile might be needed.
>
> Fixes: 8dd928915a73 ("mips: fix up obsolete cpu function usage")
> Cc: Rusty Russell <rusty@xxxxxxxxxxxxxxx>
> Signed-off-by: Guenter Roeck <linux@xxxxxxxxxxxx>
On Netlogic XLP-FVP (8 cores):
Tested-by: Florian Fainelli <f.fainelli@xxxxxxxxx>
> ---
> arch/mips/include/asm/smp.h | 2 +-
> arch/mips/kernel/smp.c | 6 ++++--
> 2 files changed, 5 insertions(+), 3 deletions(-)
>
> diff --git a/arch/mips/include/asm/smp.h b/arch/mips/include/asm/smp.h
> index bb02fac9b4fa..2b25d1ba1ea0 100644
> --- a/arch/mips/include/asm/smp.h
> +++ b/arch/mips/include/asm/smp.h
> @@ -45,7 +45,7 @@ extern int __cpu_logical_map[NR_CPUS];
> #define SMP_DUMP 0x8
> #define SMP_ASK_C0COUNT 0x10
>
> -extern volatile cpumask_t cpu_callin_map;
> +extern cpumask_t cpu_callin_map;
>
> /* Mask of CPUs which are currently definitely operating coherently */
> extern cpumask_t cpu_coherent_mask;
> diff --git a/arch/mips/kernel/smp.c b/arch/mips/kernel/smp.c
> index 193ace7955fb..158191394770 100644
> --- a/arch/mips/kernel/smp.c
> +++ b/arch/mips/kernel/smp.c
> @@ -43,7 +43,7 @@
> #include <asm/time.h>
> #include <asm/setup.h>
>
> -volatile cpumask_t cpu_callin_map; /* Bitmask of started secondaries */
> +cpumask_t cpu_callin_map; /* Bitmask of started secondaries */
>
> int __cpu_number_map[NR_CPUS]; /* Map physical to logical */
> EXPORT_SYMBOL(__cpu_number_map);
> @@ -218,8 +218,10 @@ int __cpu_up(unsigned int cpu, struct task_struct *tidle)
> /*
> * Trust is futile. We should really have timeouts ...
> */
> - while (!cpumask_test_cpu(cpu, &cpu_callin_map))
> + while (!cpumask_test_cpu(cpu, &cpu_callin_map)) {
> udelay(100);
> + mb();
> + }
>
> synchronise_count_master(cpu);
> return 0;
> --
> 2.1.0
>
--
Florian
