Re: [PATCH v8 15/26] arm64: alternative: Apply alternatives early in boot process

[Suzuki K Poulose <suzuki.poulose@arm.com>]

On 08/01/2019 15:20, Julien Thierry wrote:
> Hi Suzuki,
> 
> On 08/01/2019 14:51, Suzuki K Poulose wrote:
>> Hi Julien,
>>
>> On 08/01/2019 14:07, Julien Thierry wrote:
>>> From: Daniel Thompson <daniel.thompson@linaro.org>
>>>
>>> Currently alternatives are applied very late in the boot process (and
>>> a long time after we enable scheduling). Some alternative sequences,
>>> such as those that alter the way CPU context is stored, must be applied
>>> much earlier in the boot sequence.
>>>
>>> Introduce apply_boot_alternatives() to allow some alternatives to be
>>> applied immediately after we detect the CPU features of the boot CPU.
>>>
>>> Signed-off-by: Daniel Thompson <daniel.thompson@linaro.org>
>>> [julien.thierry@arm.com: rename to fit new cpufeature framework better,
>>>               apply BOOT_SCOPE feature early in boot]
>>> Signed-off-by: Julien Thierry <julien.thierry@arm.com>
>>> Cc: Catalin Marinas <catalin.marinas@arm.com>
>>> Cc: Will Deacon <will.deacon@arm.com>
>>> Cc: Christoffer Dall <christoffer.dall@arm.com>
>>> Cc: Suzuki K Poulose <suzuki.poulose@arm.com>
>>> ---
>>>    arch/arm64/include/asm/alternative.h |  1 +
>>>    arch/arm64/include/asm/cpufeature.h  |  4 ++++
>>>    arch/arm64/kernel/alternative.c      | 43
>>> +++++++++++++++++++++++++++++++-----
>>>    arch/arm64/kernel/cpufeature.c       |  6 +++++
>>>    arch/arm64/kernel/smp.c              |  7 ++++++
>>>    5 files changed, 56 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/arch/arm64/include/asm/alternative.h
>>> b/arch/arm64/include/asm/alternative.h
>>> index 9806a23..b9f8d78 100644
>>> --- a/arch/arm64/include/asm/alternative.h
>>> +++ b/arch/arm64/include/asm/alternative.h
>>> @@ -25,6 +25,7 @@ struct alt_instr {
>>>    typedef void (*alternative_cb_t)(struct alt_instr *alt,
>>>                     __le32 *origptr, __le32 *updptr, int nr_inst);
>>>    +void __init apply_boot_alternatives(void);
>>>    void __init apply_alternatives_all(void);
>>>    bool alternative_is_applied(u16 cpufeature);
>>>    diff --git a/arch/arm64/include/asm/cpufeature.h
>>> b/arch/arm64/include/asm/cpufeature.h
>>> index 89c3f31..e505e1f 100644
>>> --- a/arch/arm64/include/asm/cpufeature.h
>>> +++ b/arch/arm64/include/asm/cpufeature.h
>>> @@ -391,6 +391,10 @@ static inline int cpucap_default_scope(const
>>> struct arm64_cpu_capabilities *cap)
>>>    extern struct static_key_false cpu_hwcap_keys[ARM64_NCAPS];
>>>    extern struct static_key_false arm64_const_caps_ready;
>>>    +/* ARM64 CAPS + alternative_cb */
>>> +#define ARM64_NPATCHABLE (ARM64_NCAPS + 1)
>>> +extern DECLARE_BITMAP(boot_capabilities, ARM64_NPATCHABLE);
>>> +
>>>    #define for_each_available_cap(cap)        \
>>>        for_each_set_bit(cap, cpu_hwcaps, ARM64_NCAPS)
>>>    diff --git a/arch/arm64/kernel/alternative.c
>>> b/arch/arm64/kernel/alternative.c
>>> index c947d22..a9b4677 100644
>>> --- a/arch/arm64/kernel/alternative.c
>>> +++ b/arch/arm64/kernel/alternative.c
>>> @@ -155,7 +155,8 @@ static void clean_dcache_range_nopatch(u64 start,
>>> u64 end)
>>>        } while (cur += d_size, cur < end);
>>>    }
>>>    -static void __apply_alternatives(void *alt_region, bool is_module)
>>> +static void __apply_alternatives(void *alt_region,  bool is_module,
>>> +                 unsigned long *feature_mask)
>>>    {
>>>        struct alt_instr *alt;
>>>        struct alt_region *region = alt_region;
>>> @@ -165,6 +166,9 @@ static void __apply_alternatives(void *alt_region,
>>> bool is_module)
>>>        for (alt = region->begin; alt < region->end; alt++) {
>>>            int nr_inst;
>>>    +        if (!test_bit(alt->cpufeature, feature_mask))
>>> +            continue;
>>> +
>>>            /* Use ARM64_CB_PATCH as an unconditional patch */
>>>            if (alt->cpufeature < ARM64_CB_PATCH &&
>>>                !cpus_have_cap(alt->cpufeature))
>>> @@ -203,8 +207,11 @@ static void __apply_alternatives(void
>>> *alt_region, bool is_module)
>>>            __flush_icache_all();
>>>            isb();
>>>    -        /* We applied all that was available */
>>> -        bitmap_copy(applied_alternatives, cpu_hwcaps, ARM64_NCAPS);
>>> +        /* Ignore ARM64_CB bit from feature mask */
>>> +        bitmap_or(applied_alternatives, applied_alternatives,
>>> +              feature_mask, ARM64_NCAPS);
>>> +        bitmap_and(applied_alternatives, applied_alternatives,
>>> +               cpu_hwcaps, ARM64_NCAPS);
>>>        }
>>>    }
>>>    @@ -225,8 +232,13 @@ static int __apply_alternatives_multi_stop(void
>>> *unused)
>>>                cpu_relax();
>>>            isb();
>>>        } else {
>>> +        DECLARE_BITMAP(remaining_capabilities, ARM64_NPATCHABLE);
>>> +
>>> +        bitmap_complement(remaining_capabilities, boot_capabilities,
>>> +                  ARM64_NPATCHABLE);
>>> +
>>>            BUG_ON(all_alternatives_applied);
>>> -        __apply_alternatives(&region, false);
>>> +        __apply_alternatives(&region, false, remaining_capabilities);
>>>            /* Barriers provided by the cache flushing */
>>>            WRITE_ONCE(all_alternatives_applied, 1);
>>>        }
>>> @@ -240,6 +252,24 @@ void __init apply_alternatives_all(void)
>>>        stop_machine(__apply_alternatives_multi_stop, NULL,
>>> cpu_online_mask);
>>>    }
>>>    +/*
>>> + * This is called very early in the boot process (directly after we run
>>> + * a feature detect on the boot CPU). No need to worry about other CPUs
>>> + * here.
>>> + */
>>> +void __init apply_boot_alternatives(void)
>>> +{
>>> +    struct alt_region region = {
>>> +        .begin    = (struct alt_instr *)__alt_instructions,
>>> +        .end    = (struct alt_instr *)__alt_instructions_end,
>>> +    };
>>> +
>>> +    /* If called on non-boot cpu things could go wrong */
>>> +    WARN_ON(smp_processor_id() != 0);
>>> +
>>> +    __apply_alternatives(&region, false, &boot_capabilities[0]);
>>> +}
>>> +
>>>    #ifdef CONFIG_MODULES
>>>    void apply_alternatives_module(void *start, size_t length)
>>>    {
>>> @@ -247,7 +277,10 @@ void apply_alternatives_module(void *start,
>>> size_t length)
>>>            .begin    = start,
>>>            .end    = start + length,
>>>        };
>>> +    DECLARE_BITMAP(all_capabilities, ARM64_NPATCHABLE);
>>> +
>>> +    bitmap_fill(all_capabilities, ARM64_NPATCHABLE);
>>>    -    __apply_alternatives(&region, true);
>>> +    __apply_alternatives(&region, true, &all_capabilities[0]);
>>>    }
>>>    #endif
>>> diff --git a/arch/arm64/kernel/cpufeature.c
>>> b/arch/arm64/kernel/cpufeature.c
>>> index 84fa5be..71c8d4f 100644
>>> --- a/arch/arm64/kernel/cpufeature.c
>>> +++ b/arch/arm64/kernel/cpufeature.c
>>> @@ -54,6 +54,9 @@
>>>    EXPORT_SYMBOL(cpu_hwcaps);
>>>    static struct arm64_cpu_capabilities const __ro_after_init
>>> *cpu_hwcaps_ptrs[ARM64_NCAPS];
>>>    +/* Need also bit for ARM64_CB_PATCH */
>>> +DECLARE_BITMAP(boot_capabilities, ARM64_NPATCHABLE);
>>> +
>>>    /*
>>>     * Flag to indicate if we have computed the system wide
>>>     * capabilities based on the boot time active CPUs. This
>>> @@ -1672,6 +1675,9 @@ static void update_cpu_capabilities(u16 scope_mask)
>>>            if (caps->desc)
>>>                pr_info("detected: %s\n", caps->desc);
>>>            cpus_set_cap(caps->capability);
>>> +
>>> +        if (caps->type & SCOPE_BOOT_CPU)
>>
>> You may want to do :
>>          if (scope_mask & SCOPE_BOOT_CPU)
>>
>> for a tighter check to ensure this doesn't update the boot_capabilities
>> after we have applied the boot_scope alternatives and miss applying the
>> alternatives for those, should someone add a multi-scope (i.e
>> SCOPE_BOOT_CPU and
>> something else) capability (even by mistake).
>>
> 
> But a multi-scope capability containing SCOPE_BOOT_CPU should already
> get updated for setup_boot_cpu_capabilities. Capabilities marked with
> SCOPE_BOOT_CPU need to be enabled on the boot CPU or not at all.

Yes, you're right. It is not normal to have multiple SCOPE for a "capability".
But if someone comes with such a cap, we may miss handling this case. It is
always better to be safer.

> 
> Shouldn't the call to caps->matches() fail for a boot feature that was
> not found on the boot cpu?
> 
> Also, you made the opposite suggestion 4 version ago with a more
> worrying scenario :) :
> https://lkml.org/lkml/2018/5/25/208

Ah, you're right. I missed that. We need the additional check as you mention
below.

> 
> Otherwise, if my assumption above is wrong, it means the check should
> probably be:
> 	if (caps->type & SCOPE_BOOT_CPU && scope_mask & SCOPE_BOOT_CPU)

Yes, this is what we want.


> But my current understanding is that we don't need that.
> 
>> With that:
>>
>> Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com>


Cheers
Suzuki