On 12 October 2017 at 08:49, Marc Zyngier <marc.zyngier@xxxxxxx> wrote:
> Hi Ard,
>
> On 12/10/17 07:58, Ard Biesheuvel wrote:
>> The Socionext Synquacer SoC's implementation of GICv3 has a so-called
>> 'pre-ITS', which maps 32-bit writes targeted at a separate window of
>> size '4 << device_id_bits' onto writes to GITS_TRANSLATER with device
>> ID taken from bits [device_id_bits + 1:2] of the window offset.
>> Writes that target GITS_TRANSLATER directly are reported as originating
>> from device ID #0.
>>
>> So add a workaround for this. Given that this breaks isolation, clear
>> the IRQ_DOMAIN_FLAG_MSI_REMAP flag as well.
>>
>> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx>
>> ---
>> v2: - use a 32-bit host address/size rather than a PCI address, to factor
>> out the involvement of an SMMU (which the platform does have, but it
>> is unclear atm if it can be exposed to the OS)
>> - add msi_domain_flags member to move the quirk flag checks out of the
>> common code path
>>
>> Documentation/devicetree/bindings/interrupt-controller/arm,gic-v3.txt | 4 ++
>> arch/arm64/Kconfig | 8 +++
>> drivers/irqchip/irq-gic-v3-its.c | 53 +++++++++++++++++---
>> 3 files changed, 59 insertions(+), 6 deletions(-)
>>
>> diff --git a/Documentation/devicetree/bindings/interrupt-controller/arm,gic-v3.txt b/Documentation/devicetree/bindings/interrupt-controller/arm,gic-v3.txt
>> index 4c29cdab0ea5..3d8b3f910aef 100644
>> --- a/Documentation/devicetree/bindings/interrupt-controller/arm,gic-v3.txt
>> +++ b/Documentation/devicetree/bindings/interrupt-controller/arm,gic-v3.txt
>> @@ -75,6 +75,10 @@ These nodes must have the following properties:
>> - reg: Specifies the base physical address and size of the ITS
>> registers.
>>
>> +Optional:
>> +- socionext,synquacer-pre-its: (u32, u32) tuple describing the host address
>> + and size of the pre-ITS window.
>> +
>> The main GIC node must contain the appropriate #address-cells,
>> #size-cells and ranges properties for the reg property of all ITS
>> nodes.
>> diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
>> index 0df64a6a56d4..c4361dff2b74 100644
>> --- a/arch/arm64/Kconfig
>> +++ b/arch/arm64/Kconfig
>> @@ -539,6 +539,14 @@ config QCOM_QDF2400_ERRATUM_0065
>>
>> If unsure, say Y.
>>
>> +config SOCIONEXT_SYNQUACER_PREITS
>> + bool "Socionext Synquacer: Workaround for GICv3 pre-ITS"
>> + default y
>> + help
>> + Socionext Synquacer SoCs implement a separate h/w block to generate
>> + MSI doorbell writes with non-zero values for the device ID.
>> +
>> + If unsure, say Y.
>> endmenu
>>
>>
>> diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
>> index e8d89343d613..7a4536ce8e72 100644
>> --- a/drivers/irqchip/irq-gic-v3-its.c
>> +++ b/drivers/irqchip/irq-gic-v3-its.c
>> @@ -46,6 +46,7 @@
>> #define ITS_FLAGS_CMDQ_NEEDS_FLUSHING (1ULL << 0)
>> #define ITS_FLAGS_WORKAROUND_CAVIUM_22375 (1ULL << 1)
>> #define ITS_FLAGS_WORKAROUND_CAVIUM_23144 (1ULL << 2)
>> +#define ITS_FLAGS_WORKAROUND_SOCIONEXT_PREITS (1ULL << 3)
>>
>> #define RDIST_FLAGS_PROPBASE_NEEDS_FLUSHING (1 << 0)
>>
>> @@ -102,6 +103,11 @@ struct its_node {
>> u32 ite_size;
>> u32 device_ids;
>> int numa_node;
>> + unsigned int msi_domain_flags;
>> +#ifdef CONFIG_SOCIONEXT_SYNQUACER_PREITS
>> + u32 pre_its_base;
>> + u32 pre_its_size;
>> +#endif
>> bool is_v4;
>> };
>>
>> @@ -1095,14 +1101,31 @@ static int its_set_affinity(struct irq_data *d, const struct cpumask *mask_val,
>> return IRQ_SET_MASK_OK_DONE;
>> }
>>
>> +static u64 its_irq_get_msi_base(struct its_device *its_dev)
>> +{
>> + struct its_node *its = its_dev->its;
>> +
>> +#ifdef CONFIG_SOCIONEXT_SYNQUACER_PREITS
>> + if (its->flags & ITS_FLAGS_WORKAROUND_SOCIONEXT_PREITS)
>> +
>> + /*
>> + * The Socionext Synquacer SoC has a so-called 'pre-ITS',
>> + * which maps 32-bit writes targeted at a separate window of
>> + * size '4 << device_id_bits' onto writes to GITS_TRANSLATER
>> + * with device ID taken from bits [device_id_bits + 1:2] of
>> + * the window offset.
>> + */
>> + return its->pre_its_base + (its_dev->device_id << 2);
>> +#endif
>> + return its->phys_base + GITS_TRANSLATER;
>> +}
>> +
>> static void its_irq_compose_msi_msg(struct irq_data *d, struct msi_msg *msg)
>> {
>> struct its_device *its_dev = irq_data_get_irq_chip_data(d);
>> - struct its_node *its;
>> u64 addr;
>>
>> - its = its_dev->its;
>> - addr = its->phys_base + GITS_TRANSLATER;
>> + addr = its_irq_get_msi_base(its_dev);
>>
>> msg->address_lo = lower_32_bits(addr);
>> msg->address_hi = upper_32_bits(addr);
>> @@ -1666,6 +1689,11 @@ static int its_alloc_tables(struct its_node *its)
>> ids = 0x14; /* 20 bits, 8MB */
>> }
>>
>> +#ifdef CONFIG_SOCIONEXT_SYNQUACER_PREITS
>> + if (its->flags & ITS_FLAGS_WORKAROUND_SOCIONEXT_PREITS)
>> + ids = ilog2(its->pre_its_size) - 2;
>
> You probably want to check that this is smaller or equal to the ITS's
> own view of the DevID space.
>
>> +#endif
>> +
>> its->device_ids = ids;
>>
>> for (i = 0; i < GITS_BASER_NR_REGS; i++) {
>> @@ -2788,11 +2816,22 @@ static const struct gic_quirk its_quirks[] = {
>> }
>> };
>>
>> -static void its_enable_quirks(struct its_node *its)
>> +static void its_enable_quirks(struct its_node *its,
>> + struct fwnode_handle *handle)
>> {
>> u32 iidr = readl_relaxed(its->base + GITS_IIDR);
>>
>> gic_enable_quirks(iidr, its_quirks, its);
>> +
>> +#ifdef CONFIG_SOCIONEXT_SYNQUACER_PREITS
>> + if (!fwnode_property_read_u32_array(handle,
>> + "socionext,synquacer-pre-its",
>> + &its->pre_its_base, 2)) {
>
> Erm. Populating these two fields with a single read is pretty horrible...
>
>> + its->flags |= ITS_FLAGS_WORKAROUND_SOCIONEXT_PREITS;
>> + its->msi_domain_flags &= ~IRQ_DOMAIN_FLAG_MSI_REMAP;
>> + pr_info("ITS: enabling workaround for Socionext Synquacer pre-ITS\n");
>> + }
>> +#endif
>
> I must say I'm not fond of adding a quirk outside of the quirk
> infrastructure. Could you try this slightly different approach?
>
> - Add the handle to the struct its_node
> - Add a quirk structure to the its_quirks array, with a match on
> whatever version of the GIC this platform is using (I suppose some
> revision of GIC-500)
After spinning v3 with this change, I realised that this will result in the
GIC: enabling workaround for xxx
message to be printed for all GIC-500s, given that this is triggered
by the IIDR match itself.
I will add a patch to the series to change the quirk->init() prototype
to return a bool indicating whether it matched or not.
--
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
