[PATCH v2] irqchip/gicv3-its: Add workaround for hip09 ITS erratum 162100801

Zhou Wang wangzhou1 at hisilicon.com
Tue Nov 12 19:37:27 PST 2024


On 2024/11/12 23:29, Marc Zyngier wrote:
> On Tue, 12 Nov 2024 09:04:26 +0000,
> Zhou Wang <wangzhou1 at hisilicon.com> wrote:
>>
>> When enabling GICv4.1 in hip09, VMAPP will fail to clear some caches
>> during unmapping operation, which will cause some vSGIs lost.
>>
>> To fix the issue, it needs to send VINVALL command after VMOVP.
>>
>> Signed-off-by: Nianyao Tang <tangnianyao at huawei.com>
>> Signed-off-by: Zhou Wang <wangzhou1 at hisilicon.com>
>> ---
>>  Documentation/arch/arm64/silicon-errata.rst |  2 ++
>>  arch/arm64/Kconfig                          | 10 +++++++
>>  drivers/irqchip/irq-gic-v3-its.c            | 29 +++++++++++++++++++++
>>  3 files changed, 41 insertions(+)
>>
>> diff --git a/Documentation/arch/arm64/silicon-errata.rst b/Documentation/arch/arm64/silicon-errata.rst
>> index 65bfab1b1861..77db10e944f0 100644
>> --- a/Documentation/arch/arm64/silicon-errata.rst
>> +++ b/Documentation/arch/arm64/silicon-errata.rst
>> @@ -258,6 +258,8 @@ stable kernels.
>>  | Hisilicon      | Hip{08,09,10,10C| #162001900      | N/A                         |
>>  |                | ,11} SMMU PMCG  |                 |                             |
>>  +----------------+-----------------+-----------------+-----------------------------+
>> +| Hisilicon      | Hip09           | #162100801      | HISILICON_ERRATUM_162100801 |
>> ++----------------+-----------------+-----------------+-----------------------------+
>>  +----------------+-----------------+-----------------+-----------------------------+
>>  | Qualcomm Tech. | Kryo/Falkor v1  | E1003           | QCOM_FALKOR_ERRATUM_1003    |
>>  +----------------+-----------------+-----------------+-----------------------------+
>> diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
>> index 70d7f4f20225..d41cf6bf1458 100644
>> --- a/arch/arm64/Kconfig
>> +++ b/arch/arm64/Kconfig
>> @@ -1232,6 +1232,16 @@ config HISILICON_ERRATUM_161600802
>>  
>>  	  If unsure, say Y.
>>  
>> +config HISILICON_ERRATUM_162100801
>> +	bool "Hip09 162100801 erratum support"
>> +	default y
>> +	help
>> +	  When enabling GICv4.1 in hip09, VMAPP will fail to clear some caches
>> +	  during unmapping operation, which will cause some vSGIs lost.
>> +	  So fix it by sending VINVALL command after VMOVP.
>> +
>> +	  If unsure, say Y.
>> +
>>  config QCOM_FALKOR_ERRATUM_1003
>>  	bool "Falkor E1003: Incorrect translation due to ASID change"
>>  	default y
>> diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
>> index 52f625e07658..e98d13d0adf9 100644
>> --- a/drivers/irqchip/irq-gic-v3-its.c
>> +++ b/drivers/irqchip/irq-gic-v3-its.c
>> @@ -44,6 +44,7 @@
>>  #define ITS_FLAGS_WORKAROUND_CAVIUM_22375	(1ULL << 1)
>>  #define ITS_FLAGS_WORKAROUND_CAVIUM_23144	(1ULL << 2)
>>  #define ITS_FLAGS_FORCE_NON_SHAREABLE		(1ULL << 3)
>> +#define ITS_FLAGS_WORKAROUND_HISILICON_162100801	(1ULL << 4)
>>  
>>  #define RD_LOCAL_LPI_ENABLED                    BIT(0)
>>  #define RD_LOCAL_PENDTABLE_PREALLOCATED         BIT(1)
>> @@ -3804,6 +3805,7 @@ static int its_vpe_set_affinity(struct irq_data *d,
>>  	struct its_vpe *vpe = irq_data_get_irq_chip_data(d);
>>  	unsigned int from, cpu = nr_cpu_ids;
>>  	struct cpumask *table_mask;
>> +	struct its_node *its;
>>  	unsigned long flags;
>>  
>>  	/*
>> @@ -3866,6 +3868,17 @@ static int its_vpe_set_affinity(struct irq_data *d,
>>  	vpe->col_idx = cpu;
>>  
>>  	its_send_vmovp(vpe);
>> +
>> +	/*
>> +	 * Version of ITS is same in one system. As there is no cache in ITS,
>> +	 * and only cache in related GICR should be clean, so one VINVALL is
>> +	 * enough here.
>> +	 */
>> +	its = list_first_entry(&its_nodes, struct its_node, entry);
>> +	if ((its->flags & ITS_FLAGS_WORKAROUND_HISILICON_162100801) &&
>> +	    is_v4_1(its))
>> +		its_send_vinvall(its, vpe);
> 
> Can this be done using the GICR_INVALLR register instead? I would
> expect it to be a bit better performance wise, as it doesn't require
> to take a global lock. Something like the hack below (untested).
> 
> Thanks,
> 
> 	M.

Hi Marc,

Many thanks for your suggestion! It is OK in theory and it works fine in test
as well.

I will send V3 soon.

Best,
Zhou

> 
> diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
> index fdec478ba5e7..d30d6ee727c1 100644
> --- a/drivers/irqchip/irq-gic-v3-its.c
> +++ b/drivers/irqchip/irq-gic-v3-its.c
> @@ -62,6 +62,8 @@ static u32 lpi_id_bits;
>  
>  static u8 __ro_after_init lpi_prop_prio;
>  
> +static struct its_node *find_4_1_its(void);
> +
>  /*
>   * Collection structure - just an ID, and a redistributor address to
>   * ping. We use one per CPU as a bag of interrupts assigned to this
> @@ -3797,6 +3799,22 @@ static void its_vpe_db_proxy_move(struct its_vpe *vpe, int from, int to)
>  	raw_spin_unlock_irqrestore(&vpe_proxy.lock, flags);
>  }
>  
> +static void its_vpe_4_1_invall_locked(int cpu, struct its_vpe *vpe)
> +{
> +	void __iomem *rdbase;
> +	u64 val;
> +
> +	val  = GICR_INVALLR_V;
> +	val |= FIELD_PREP(GICR_INVALLR_VPEID, vpe->vpe_id);
> +
> +	raw_spin_lock(&gic_data_rdist_cpu(cpu)->rd_lock);
> +	rdbase = per_cpu_ptr(gic_rdists->rdist, cpu)->rd_base;
> +	gic_write_lpir(val, rdbase + GICR_INVALLR);
> +
> +	wait_for_syncr(rdbase);
> +	raw_spin_unlock(&gic_data_rdist_cpu(cpu)->rd_lock);
> +}
> +
>  static int its_vpe_set_affinity(struct irq_data *d,
>  				const struct cpumask *mask_val,
>  				bool force)
> @@ -3849,6 +3867,10 @@ static int its_vpe_set_affinity(struct irq_data *d,
>  	vpe->col_idx = cpu;
>  
>  	its_send_vmovp(vpe);
> +
> +	if (find_4_1_its()->flags & ITS_FLAGS_WORKAROUND_HISILICON_162100801)
> +		its_vpe_4_1_invall_locked(cpu, vpe);
> +
>  	its_vpe_db_proxy_move(vpe, from, cpu);
>  
>  out:
> @@ -4156,22 +4178,12 @@ static void its_vpe_4_1_deschedule(struct its_vpe *vpe,
>  
>  static void its_vpe_4_1_invall(struct its_vpe *vpe)
>  {
> -	void __iomem *rdbase;
>  	unsigned long flags;
> -	u64 val;
>  	int cpu;
>  
> -	val  = GICR_INVALLR_V;
> -	val |= FIELD_PREP(GICR_INVALLR_VPEID, vpe->vpe_id);
> -
>  	/* Target the redistributor this vPE is currently known on */
>  	cpu = vpe_to_cpuid_lock(vpe, &flags);
> -	raw_spin_lock(&gic_data_rdist_cpu(cpu)->rd_lock);
> -	rdbase = per_cpu_ptr(gic_rdists->rdist, cpu)->rd_base;
> -	gic_write_lpir(val, rdbase + GICR_INVALLR);
> -
> -	wait_for_syncr(rdbase);
> -	raw_spin_unlock(&gic_data_rdist_cpu(cpu)->rd_lock);
> +	its_vpe_4_1_invall_locked(cpu, vpe);
>  	vpe_to_cpuid_unlock(vpe, flags);
>  }
>  
> 



More information about the linux-arm-kernel mailing list