Re: [PATCH v4 3/3] irqchip/gic-v3: Enable non-coherent redistributors/ITSes ACPI probing

[Marc Zyngier]

On Wed, 27 Dec 2023 11:00:38 +0000,
Lorenzo Pieralisi <lpieralisi@xxxxxxxxxx> wrote:
> 
> The GIC architecture specification defines a set of registers
> for redistributors and ITSes that control the sharebility and
> cacheability attributes of redistributors/ITSes initiator ports
> on the interconnect (GICR_[V]PROPBASER, GICR_[V]PENDBASER,
> GITS_BASER<n>).
> 
> Architecturally the GIC provides a means to drive shareability
> and cacheability attributes signals and related IWB/OWB/ISH barriers

IWB/OWB *barriers*? Unless you're talking about something else,
IWB/OWB refers to cacheability, and only that.

> but it is not mandatory for designs to wire up the corresponding
> interconnect signals that control the cacheability/shareability
> of transactions.
> 
> Redistributors and ITSes interconnect ports can be connected to
> non-coherent interconnects that are not able to manage the
> shareability/cacheability attributes; this implicitly makes
> the redistributors and ITSes non-coherent observers.
> 
> So far, the GIC driver on probe executes a write to "probe" for
> the redistributors and ITSes registers shareability bitfields
> by writing a value (ie InnerShareable - the shareability domain the
> CPUs are in) and check it back to detect whether the value sticks or
> not; this hinges on a GIC programming model behaviour that predates the
> current specifications, that just define shareability bits as writeable
> but do not guarantee that writing certain shareability values
> enable the expected behaviour for the redistributors/ITSes
> memory interconnect ports.
> 
> To enable non-coherent GIC designs on ACPI based systems, parse the MADT
> GICC/GICR/ITS subtables non-coherent flags to determine whether the
> respective components are non-coherent observers and force the shareability
> attributes to be programmed into the redistributors and ITSes registers.
> 
> An ACPI global function (acpi_get_madt_revision()) is added to retrieve
> the MADT revision, in that it is essential to check the MADT revision
> before checking for flags that were added with MADT revision 7 so that
> if the kernel is booted with ACPI tables (MADT rev < 7) it skips parsing
> the newly added flags (that should be zeroed reserved values for MADT
> versions < 7 but they could turn out to be buggy and should be ignored).
> 
> Signed-off-by: Lorenzo Pieralisi <lpieralisi@xxxxxxxxxx>
> Cc: Robin Murphy <robin.murphy@xxxxxxx>
> Cc: Mark Rutland <mark.rutland@xxxxxxx>
> Cc: "Rafael J. Wysocki" <rafael@xxxxxxxxxx>
> Cc: Marc Zyngier <maz@xxxxxxxxxx>
> ---
>  drivers/acpi/processor_core.c    | 21 +++++++++++++++++++++
>  drivers/irqchip/irq-gic-common.h |  8 ++++++++
>  drivers/irqchip/irq-gic-v3-its.c |  4 ++++
>  drivers/irqchip/irq-gic-v3.c     |  9 +++++++++
>  include/linux/acpi.h             |  3 +++
>  5 files changed, 45 insertions(+)
> 
> diff --git a/drivers/acpi/processor_core.c b/drivers/acpi/processor_core.c
> index b203cfe28550..c253d151275e 100644
> --- a/drivers/acpi/processor_core.c
> +++ b/drivers/acpi/processor_core.c
> @@ -215,6 +215,27 @@ phys_cpuid_t __init acpi_map_madt_entry(u32 acpi_id)
>  	return rv;
>  }
>  
> +u8 __init acpi_get_madt_revision(void)
> +{
> +	static u8 madt_revision __initdata;
> +	static bool madt_read __initdata;
> +	struct acpi_table_header *madt = NULL;
> +
> +	if (!madt_read) {
> +		madt_read = true;

Huh. Why do we need this hack? What's the issue with accessing the
MADT? Can it disappear from under our feet? While we're walking it?

> +
> +		acpi_get_table(ACPI_SIG_MADT, 0, &madt);
> +		if (!madt)
> +			return madt_revision;

What does this mean? Can we have a revision 0 of MADT?

> +
> +		madt_revision = madt->revision;
> +
> +		acpi_put_table(madt);
> +	}
> +
> +	return madt_revision;
> +}
> +
>  static phys_cpuid_t map_mat_entry(acpi_handle handle, int type, u32 acpi_id)
>  {
>  	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
> diff --git a/drivers/irqchip/irq-gic-common.h b/drivers/irqchip/irq-gic-common.h
> index f407cce9ecaa..8dffee95f7e8 100644
> --- a/drivers/irqchip/irq-gic-common.h
> +++ b/drivers/irqchip/irq-gic-common.h
> @@ -6,6 +6,7 @@
>  #ifndef _IRQ_GIC_COMMON_H
>  #define _IRQ_GIC_COMMON_H
>  
> +#include <linux/acpi.h>
>  #include <linux/of.h>
>  #include <linux/irqdomain.h>
>  #include <linux/irqchip/arm-gic-common.h>
> @@ -29,6 +30,13 @@ void gic_enable_quirks(u32 iidr, const struct gic_quirk *quirks,
>  void gic_enable_of_quirks(const struct device_node *np,
>  			  const struct gic_quirk *quirks, void *data);
>  
> +#ifdef CONFIG_ACPI
> +static inline bool gic_acpi_non_coherent_flag(u32 flags, u32 mask)
> +{
> +	return (acpi_get_madt_revision() >= 7) && (flags & mask);
> +}

Given that this checks *any* flag (or a combination of flags), the
name of the helper is extremely misleading. Also, GICC flags are not
necessarily tied to revision 7 of MADT.

To be honest, I don't think this helper bring much, and I'd rather see
an explicit check (or 3) for the revision in the driver code.

> +#endif
> +
>  #define RDIST_FLAGS_PROPBASE_NEEDS_FLUSHING    (1 << 0)
>  #define RDIST_FLAGS_RD_TABLES_PREALLOCATED     (1 << 1)
>  #define RDIST_FLAGS_FORCE_NON_SHAREABLE        (1 << 2)
> diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
> index 9a7a74239eab..8d088fca65a1 100644
> --- a/drivers/irqchip/irq-gic-v3-its.c
> +++ b/drivers/irqchip/irq-gic-v3-its.c
> @@ -5578,6 +5578,10 @@ static int __init gic_acpi_parse_madt_its(union acpi_subtable_headers *header,
>  		goto node_err;
>  	}
>  
> +	if (gic_acpi_non_coherent_flag(its_entry->flags,
> +				       ACPI_MADT_ITS_NON_COHERENT))
> +		its->flags |= ITS_FLAGS_FORCE_NON_SHAREABLE;
> +
>  	err = its_probe_one(its);
>  	if (!err)
>  		return 0;
> diff --git a/drivers/irqchip/irq-gic-v3.c b/drivers/irqchip/irq-gic-v3.c
> index 98b0329b7154..48e02838fdc8 100644
> --- a/drivers/irqchip/irq-gic-v3.c
> +++ b/drivers/irqchip/irq-gic-v3.c
> @@ -2356,6 +2356,11 @@ gic_acpi_parse_madt_redist(union acpi_subtable_headers *header,
>  		pr_err("Couldn't map GICR region @%llx\n", redist->base_address);
>  		return -ENOMEM;
>  	}
> +
> +	if (gic_acpi_non_coherent_flag(redist->flags,
> +				       ACPI_MADT_GICR_NON_COHERENT))
> +		gic_data.rdists.flags |= RDIST_FLAGS_FORCE_NON_SHAREABLE;
> +
>  	gic_request_region(redist->base_address, redist->length, "GICR");
>  
>  	gic_acpi_register_redist(redist->base_address, redist_base);
> @@ -2380,6 +2385,10 @@ gic_acpi_parse_madt_gicc(union acpi_subtable_headers *header,
>  		return -ENOMEM;
>  	gic_request_region(gicc->gicr_base_address, size, "GICR");
>  
> +	if (gic_acpi_non_coherent_flag(gicc->flags,
> +				       ACPI_MADT_GICC_NON_COHERENT))
> +		gic_data.rdists.flags |= RDIST_FLAGS_FORCE_NON_SHAREABLE;
> +
>  	gic_acpi_register_redist(gicc->gicr_base_address, redist_base);
>  	return 0;
>  }
> diff --git a/include/linux/acpi.h b/include/linux/acpi.h
> index 54189e0e5f41..a292f2bdb693 100644
> --- a/include/linux/acpi.h
> +++ b/include/linux/acpi.h
> @@ -283,6 +283,9 @@ static inline bool invalid_phys_cpuid(phys_cpuid_t phys_id)
>  	return phys_id == PHYS_CPUID_INVALID;
>  }
>  
> +
> +u8 __init acpi_get_madt_revision(void);
> +
>  /* Validate the processor object's proc_id */
>  bool acpi_duplicate_processor_id(int proc_id);
>  /* Processor _CTS control */

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.