Re: [PATCH 20/35] irqchip/gic-v4.1: Allow direct invalidation of VLPIs

[Zenghui Yu <yuzenghui@huawei.com>]

On 2019/9/30 17:20, Marc Zyngier wrote:
> On 2019-09-28 03:02, Zenghui Yu wrote:
>> On 2019/9/24 2:25, Marc Zyngier wrote:
>>> Just like for INVALL, GICv4.1 has grown a VPE-aware INVLPI register.
>>> Let's plumb it in and make use of the DirectLPI code in that case.
>>> Signed-off-by: Marc Zyngier <maz@kernel.org>
>>> ---
>>>   drivers/irqchip/irq-gic-v3-its.c   | 19 +++++++++++++++++--
>>>   include/linux/irqchip/arm-gic-v3.h |  1 +
>>>   2 files changed, 18 insertions(+), 2 deletions(-)
>>> diff --git a/drivers/irqchip/irq-gic-v3-its.c 
>>> b/drivers/irqchip/irq-gic-v3-its.c
>>> index b791c9beddf2..34595a7fcccb 100644
>>> --- a/drivers/irqchip/irq-gic-v3-its.c
>>> +++ b/drivers/irqchip/irq-gic-v3-its.c
>>> @@ -1200,13 +1200,27 @@ static void wait_for_syncr(void __iomem *rdbase)
>>>   static void direct_lpi_inv(struct irq_data *d)
>>>   {
>>> +    struct its_vlpi_map *map = get_vlpi_map(d);
>>>       struct its_collection *col;
>>>       void __iomem *rdbase;
>>> +    u64 val;
>>> +
>>> +    if (map) {
>>> +        struct its_device *its_dev = irq_data_get_irq_chip_data(d);
>>> +
>>> +        WARN_ON(!is_v4_1(its_dev->its));
>>> +
>>> +        val  = GICR_INVLPIR_V;
>>> +        val |= FIELD_PREP(GICR_INVLPIR_VPEID, map->vpe->vpe_id);
>>> +        val |= FIELD_PREP(GICR_INVLPIR_INTID, map->vintid);
>>> +    } else {
>>> +        val = d->hwirq;
>>> +    }
>>>
>>>       /* Target the redistributor this LPI is currently routed to */
>>>       col = irq_to_col(d);
>>
>> I think irq_to_col() may not work when GICv4.1 VLPIs are involved in.
>>
>> irq_to_col() gives us col_map[event] as the target redistributor,
>> but the correct one for VLPIs should be vlpi_maps[event]->vpe->col_idx.
>> These two are not always pointing to the same physical RD.
>> For example, if guest issues a MOVI against a VLPI, we will update the
>> corresponding vlpi_map->vpe and issue a VMOVI on ITS... but leave the
>> col_map[event] unchanged.
>>
>> col_map[event] usually describes the physical LPI's CPU affinity, but
>> when this physical LPI serves as something which the VLPI is backed by,
>> we take really little care of it.  Did I miss something here?
> 
> You didn't miss anything, and this is indeed another pretty bad bug.
> The collection mapping is completely unused when the LPI becomes a
> VLPI, and it is only the vpe->col_id that matters (which gets updated
> on VMOVP).
> 
> This shows that irq_to_col() is the wrong abstraction, and what we're
> interested is something that is more like 'irq to cpuid', allowing us
> to directly point to the right distributor.
> 
> Please see the patch I just pushed[1], which does that.
> 
> Thanks,
> 
>          M.
> 
> [1] 
> https://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms.git/commit/?h=irq/gic-v4.1-devel&id=aff363113eb26b6840136b69c2c7db2ea691db20 

Yes, I think this makes sense.


Thanks,
zenghui