On 2016/7/7 19:51, Marc Zyngier wrote:
> On 07/07/16 12:37, Ding Tianhong wrote:
>> On 2016/7/7 17:49, Marc Zyngier wrote:
>>> On 07/07/16 10:34, Ding Tianhong wrote:
>>>> On 2016/7/2 6:41, Scott Wood wrote:
>>>>> Erratum A-008585 says that the ARM generic timer counter "has the
>>>>> potential to contain an erroneous value for a small number of core
>>>>> clock cycles every time the timer value changes". Accesses to TVAL
>>>>> (both read and write) are also affected due to the implicit counter
>>>>> read. Accesses to CVAL are not affected.
>>>>>
>>>>> The workaround is to reread TVAL and count registers until successive reads
>>>>> return the same value, and when writing TVAL to retry until counter
>>>>> reads before and after the write return the same value.
>>>>>
>>>>> This erratum can be found on LS1043A and LS2080A.
>>>>>
>>>>> Signed-off-by: Scott Wood <oss@xxxxxxxxxxxx>
>>>>> ---
>>>>> v3:
>>>>> - Used cval rather than a loop for the write side of the erratum
>>>>> - Added a Kconfig control
>>>>> - Moved the device tree binding into its own patch
>>>>> - Added erratum to silicon-errata.txt
>>>>> - Changed function names to contain the erratum name
>>>>> - Factored out the setting of erratum versions of set_next_event
>>>>> to improve readability
>>>>> - Added a comment clarifying that the timeout is arbitrary
>>>>>
>>>>> v2:
>>>>> Significant rework based on feedback, including using static_key,
>>>>> disabling VDSO counter access rather than adding the workaround to the
>>>>> VDSO, and uninlining the loops.
>>>>>
>>>>> Dropped the separate property for indicating that writes to TVAL are
>>>>> affected, as I believe that's just a side effect of the implicit
>>>>> counter read being corrupted, and thus a chip that is affected by one
>>>>> will always be affected by the other.
>>>>>
>>>>> Dropped the arm32 portion as it seems there was confusion about whether
>>>>> LS1021A is affected. Currently I am being told that it is not
>>>>> affected.
>>>>>
>>>>> I considered writing to CVAL rather than looping on TVAL writes, but
>>>>> that would still have required separate set_next_event() code for the
>>>>> erratum, and adding CVAL to the enum would have required a bunch of
>>>>> extra handlers in switch statements (even where unused, due to compiler
>>>>> warnings about unhandled enum values) including in an arm32 header. It
>>>>> seemed better to avoid the arm32 interaction and new untested
>>>>> accessors.
>>>>> ---
>>>>> Documentation/arm64/silicon-errata.txt | 2 +
>>>>> arch/arm64/include/asm/arch_timer.h | 48 ++++++++++++---
>>>>> drivers/clocksource/Kconfig | 10 ++++
>>>>> drivers/clocksource/arm_arch_timer.c | 103 +++++++++++++++++++++++++++++++++
>>>>> 4 files changed, 154 insertions(+), 9 deletions(-)
>>>>>
>>>>> diff --git a/Documentation/arm64/silicon-errata.txt b/Documentation/arm64/silicon-errata.txt
>>>>> index ba4b6ac..5778f62 100644
>>>>> --- a/Documentation/arm64/silicon-errata.txt
>>>>> +++ b/Documentation/arm64/silicon-errata.txt
>>>>> @@ -57,3 +57,5 @@ stable kernels.
>>>>> | Cavium | ThunderX ITS | #22375, #24313 | CAVIUM_ERRATUM_22375 |
>>>>> | Cavium | ThunderX GICv3 | #23154 | CAVIUM_ERRATUM_23154 |
>>>>> | Cavium | ThunderX Core | #27456 | CAVIUM_ERRATUM_27456 |
>>>>> +| | | | |
>>>>> +| Freescale/NXP | LS2080A/LS1043A | A-008585 | FSL_ERRATUM_A008585 |
>>>>> diff --git a/arch/arm64/include/asm/arch_timer.h b/arch/arm64/include/asm/arch_timer.h
>>>>> index fbe0ca3..70fbad9 100644
>>>>> --- a/arch/arm64/include/asm/arch_timer.h
>>>>> +++ b/arch/arm64/include/asm/arch_timer.h
>>>>> @@ -23,10 +23,34 @@
>>>>>
>>>>> #include <linux/bug.h>
>>>>> #include <linux/init.h>
>>>>> +#include <linux/jump_label.h>
>>>>> #include <linux/types.h>
>>>>>
>>>>> #include <clocksource/arm_arch_timer.h>
>>>>>
>>>>> +extern struct static_key_false arch_timer_read_ool_enabled;
>>>>> +
>>>>> +#define ARCH_TIMER_REG_READ(reg, func) \
>>>>> +extern u64 func##_ool(void); \
>>>>> +static inline u64 __##func(void) \
>>>>> +{ \
>>>>> + u64 val; \
>>>>> + asm volatile("mrs %0, " reg : "=r" (val)); \
>>>>> + return val; \
>>>>> +} \
>>>>> +static inline u64 _##func(void) \
>>>>> +{ \
>>>>> + if (IS_ENABLED(CONFIG_FSL_ERRATUM_A008585) && \
>>>>> + static_branch_unlikely(&arch_timer_read_ool_enabled)) \
>>>>> + return func##_ool(); \
>>>>> + else \
>>>>> + return __##func(); \
>>>>> +}
>>>>> +
>>>>> +ARCH_TIMER_REG_READ("cntp_tval_el0", arch_timer_get_ptval)
>>>>> +ARCH_TIMER_REG_READ("cntv_tval_el0", arch_timer_get_vtval)
>>>>> +ARCH_TIMER_REG_READ("cntvct_el0", arch_counter_get_cntvct)
>>>>> +
>>>>> /*
>>>>> * These register accessors are marked inline so the compiler can
>>>>> * nicely work out which register we want, and chuck away the rest of
>>>>> @@ -58,6 +82,16 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u32 val)
>>>>> isb();
>>>>> }
>>>>>
>>>>> +static __always_inline void arch_timer_cval_write_cp15(int access, u64 val)
>>>>> +{
>>>>> + if (access == ARCH_TIMER_PHYS_ACCESS)
>>>>> + asm volatile("msr cntp_cval_el0, %0" : : "r" (val));
>>>>> + else if (access == ARCH_TIMER_VIRT_ACCESS)
>>>>> + asm volatile("msr cntv_cval_el0, %0" : : "r" (val));
>>>>> +
>>>>> + isb();
>>>>> +}
>>>>> +
>>>>> static __always_inline
>>>>> u32 arch_timer_reg_read_cp15(int access, enum arch_timer_reg reg)
>>>>> {
>>>>> @@ -66,19 +100,19 @@ u32 arch_timer_reg_read_cp15(int access, enum arch_timer_reg reg)
>>>>> if (access == ARCH_TIMER_PHYS_ACCESS) {
>>>>> switch (reg) {
>>>>> case ARCH_TIMER_REG_CTRL:
>>>>> - asm volatile("mrs %0, cntp_ctl_el0" : "=r" (val));
>>>>> + asm volatile("mrs %0, cntp_ctl_el0" : "=r" (val));
>>>>> break;
>>>>> case ARCH_TIMER_REG_TVAL:
>>>>> - asm volatile("mrs %0, cntp_tval_el0" : "=r" (val));
>>>>> + val = _arch_timer_get_ptval();
>>>>> break;
>>>>> }
>>>>> } else if (access == ARCH_TIMER_VIRT_ACCESS) {
>>>>> switch (reg) {
>>>>> case ARCH_TIMER_REG_CTRL:
>>>>> - asm volatile("mrs %0, cntv_ctl_el0" : "=r" (val));
>>>>> + asm volatile("mrs %0, cntv_ctl_el0" : "=r" (val));
>>>>> break;
>>>>> case ARCH_TIMER_REG_TVAL:
>>>>> - asm volatile("mrs %0, cntv_tval_el0" : "=r" (val));
>>>>> + val = _arch_timer_get_vtval();
>>>>> break;
>>>>> }
>>>>> }
>>>>> @@ -116,12 +150,8 @@ static inline u64 arch_counter_get_cntpct(void)
>>>>>
>>>>> static inline u64 arch_counter_get_cntvct(void)
>>>>> {
>>>>> - u64 cval;
>>>>> -
>>>>> isb();
>>>>> - asm volatile("mrs %0, cntvct_el0" : "=r" (cval));
>>>>> -
>>>>> - return cval;
>>>>> + return _arch_counter_get_cntvct();
>>>>> }
>>>>>
>>>>> static inline int arch_timer_arch_init(void)
>>>>> diff --git a/drivers/clocksource/Kconfig b/drivers/clocksource/Kconfig
>>>>> index c346be6..672ddc3 100644
>>>>> --- a/drivers/clocksource/Kconfig
>>>>> +++ b/drivers/clocksource/Kconfig
>>>>> @@ -207,6 +207,16 @@ config ARM_ARCH_TIMER_EVTSTREAM
>>>>> This must be disabled for hardware validation purposes to detect any
>>>>> hardware anomalies of missing events.
>>>>>
>>>>> +config FSL_ERRATUM_A008585
>>>>> + bool "Workaround for Freescale/NXP Erratum A-008585"
>>>>> + default y
>>>>> + depends on ARM_ARCH_TIMER && ARM64
>>>>> + help
>>>>> + This option enables a workaround for Freescale/NXP Erratum
>>>>> + A-008585 ("ARM generic timer may contain an erroneous
>>>>> + value"). The workaround will only be active if the
>>>>> + fsl,erratum-a008585 property is found in the timer node.
>>>>> +
>>>>> config ARM_GLOBAL_TIMER
>>>>> bool
>>>>> select CLKSRC_OF if OF
>>>>> diff --git a/drivers/clocksource/arm_arch_timer.c b/drivers/clocksource/arm_arch_timer.c
>>>>> index 5152b38..7ead4eb 100644
>>>>> --- a/drivers/clocksource/arm_arch_timer.c
>>>>> +++ b/drivers/clocksource/arm_arch_timer.c
>>>>> @@ -83,6 +83,51 @@ static bool arch_timer_mem_use_virtual;
>>>>> * Architected system timer support.
>>>>> */
>>>>>
>>>>> +#ifdef CONFIG_FSL_ERRATUM_A008585
>>>>> +DEFINE_STATIC_KEY_FALSE(arch_timer_read_ool_enabled);
>>>>> +EXPORT_SYMBOL_GPL(arch_timer_read_ool_enabled);
>>>>> +
>>>>> +/*
>>>>> + * __always_inline is used to ensure that func() is not an actual function
>>>>> + * pointer, which would result in the register accesses potentially being too
>>>>> + * far apart for the loop to work.
>>>>> + *
>>>>> + * The timeout is an arbitrary value well beyond the highest number
>>>>> + * of iterations the loop has been observed to take.
>>>>> + */
>>>>> +static __always_inline u64 fsl_a008585_reread_counter(u64 (*func)(void))
>>>>> +{
>>>>> + u64 cval_old, cval_new;
>>>>> + int timeout = 200;
>>>>> +
>>>>> + do {
>>>>> + isb();
>>>>> + cval_old = func();
>>>>> + cval_new = func();
>>>>> + timeout--;
>>>>> + } while (unlikely(cval_old != cval_new) && timeout);
>>>>> +
>>>>> + WARN_ON_ONCE(!timeout);
>>>>> + return cval_new;
>>>>> +}
>>>> Hi Scott:
>>>>
>>>> I have test this patch, this solution looks will break the performance a little more than I expected.
>>>> it will have more than 10% that the cval will read again, we could sure that the cval_old always equal to the
>>>> cval_new in the normal circumstances, so I prefer this way:
>>>>
>>>> do {
>>>> isb();
>>>> cval_old = func();
>>>> cval_new = func();
>>>> timeout--;
>>>> } while (unlikely((cval_new - cval_old) >> 2) && timeout);
>>>
>>> What makes you think that ignoring the two bottom bits is a safe thing
>>> to do? Talking about performance when the HW has such a dramatic bug is
>>> like putting a bigger engine on a car that has no brakes: you just hit
>>> the wall quicker.
>>>
>>> Thanks,
>>>
>>
>> I have a chip which has the same problem like Scott's chip, and I
>> wish to solve this problem in the same way, our chip designer told me
>> that if you got a wrong value from the cntvct_el0, you would not get
>> a wrong value until 8 cycles later, so I could ignoring the lowest 3
>> bits if I reading twice together.
>
> Is that CPU cycles? Or timer cycles? What guarantees do you have that
> the two reads are *always* done in the right timing window?
>
The timer counter only use 56 bits in aarch64, my chip would change one of the higher
bit(55 to 3) to a wrong value when occur bug, so there will be more than 8 cycles between
correct value and wrong value from the timer counter. Maybe Scott's problem is not just like
mine.
>> The key problem is the probability of this bug, my chip has 1/100000
>> chance to met this bug, so use 10% performance to fix this bug looks
>> more expensive.
>
> You care about performance, I care about correctness. If 10% of your CPU
> is wasted on a correct workaround, tough. Next time, your HW guy will
> spend more time getting it right.
>
> Anyway, what you are describing is a different bug from what FSL has, so
> don't try and shoehorn your own constraints in another workaround.
> Please implement your own.
>
OK, In deed I should, thanks.
Ding
> Thanks,
>
> M.
>
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
