Re: [RFC part2 PATCH 9/9] ACPI / GIC: Initialize GIC using the informationin MADT

From: Hanjun Guo
Date: Thu Dec 05 2013 - 08:42:11 EST


On 2013å12æ04æ 23:50, Marc Zyngier wrote:
On 04/12/13 15:32, Hanjun Guo wrote:
On 2013å12æ04æ 01:26, Marc Zyngier wrote:
Hi Hanjun,

On 03/12/13 16:39, Hanjun Guo wrote:
In MADT table, there are GIC cpu interface base address and
GIC distributor base address, use them to convert GIC to ACPI.

Signed-off-by: Hanjun Guo <hanjun.guo@xxxxxxxxxx>
---
arch/arm64/kernel/irq.c | 5 ++++
drivers/acpi/plat/arm-core.c | 66 ++++++++++++++++++++++++++++++++++++------
include/linux/acpi.h | 6 ++++
3 files changed, 68 insertions(+), 9 deletions(-)

diff --git a/arch/arm64/kernel/irq.c b/arch/arm64/kernel/irq.c
index 473e5db..a9e68bf 100644
--- a/arch/arm64/kernel/irq.c
+++ b/arch/arm64/kernel/irq.c
@@ -25,6 +25,7 @@
#include <linux/irq.h>
#include <linux/smp.h>
#include <linux/init.h>
+#include <linux/acpi.h>
#include <linux/irqchip.h>
#include <linux/seq_file.h>
#include <linux/ratelimit.h>
@@ -78,6 +79,10 @@ void __init set_handle_irq(void (*handle_irq)(struct pt_regs *))
void __init init_IRQ(void)
{
irqchip_init();
+
+ if (!handle_arch_irq)
+ acpi_gic_init();
+
Why is the GIC hardcoded?
Very good question, thanks. I considered GIC only in my patch set.
I have no idea how to handle the GIC hardcoded problem here for
now, but I will figure it out later.

If any suggestion, I will appreciate a lot.

How are you going to support other interrupt
controllers?
ACPI 5.0 supports GICv2 only for now, if we want to
support other interrupt controller, we should introduce
some OEM table and parsing it, and it will not covered
by this patch set.

if (!handle_arch_irq)
panic("No interrupt controller found.");
}
diff --git a/drivers/acpi/plat/arm-core.c b/drivers/acpi/plat/arm-core.c
index 17c99e1..509b847 100644
--- a/drivers/acpi/plat/arm-core.c
+++ b/drivers/acpi/plat/arm-core.c
@@ -29,6 +29,7 @@
#include <linux/module.h>
#include <linux/irq.h>
#include <linux/irqdomain.h>
+#include <linux/irqchip/arm-gic.h>
#include <linux/slab.h>
#include <linux/bootmem.h>
#include <linux/ioport.h>
@@ -211,11 +212,21 @@ acpi_parse_gic(struct acpi_subtable_header *header, const unsigned long end)
return 0;
}
+#ifdef CONFIG_ARM_GIC
+/*
+ * Hard code here, we can not get memory size from MADT (but FDT does),
+ * this size is described in ARMv8 foudation model's User Guide
+ */
+#define GIC_DISTRIBUTOR_MEMORY_SIZE (SZ_8K)
+#define GIC_CPU_INTERFACE_MEMORY_SIZE (SZ_4K)
Aside from the incorrect sizes, how do you plan to address the other
regions that the GICv2 specification describes?
Did these regions have the same base address? I mean the same
as GIC distributor base address and GIC cpu interface base address.

if yes, since the base address is stored in gic_init(), it can be for
furture
use. if I misunderstood your question, please let me know.
Look at the VGIC implementation for KVM in virt/kvm/arm. It does its own
probing of the additional regions used for virtualization.

The GIC and VGIC code are completely separate, and you'll need to find
an acceptable solution for that too.

Ok, will review the VGIC code for KVM, thanks for the guidance.

static int __init
acpi_parse_gic_distributor(struct acpi_subtable_header *header,
const unsigned long end)
{
struct acpi_madt_generic_distributor *distributor = NULL;
+ void __iomem *dist_base = NULL;
+ void __iomem *cpu_base = NULL;
distributor = (struct acpi_madt_generic_distributor *)header;
@@ -224,8 +235,43 @@ acpi_parse_gic_distributor(struct acpi_subtable_header *header,
acpi_table_print_madt_entry(header);
+ /* GIC is initialised after page_init(), no need for early_ioremap */
+ dist_base = ioremap(distributor->base_address,
+ GIC_CPU_INTERFACE_MEMORY_SIZE);
+ if (!dist_base) {
+ pr_warn(PREFIX "unable to map gic dist registers\n");
+ return -ENOMEM;
+ }
+
+ /*
+ * acpi_lapic_addr is stored in acpi_parse_madt(),
+ * so we can use it here for GIC init
+ */
+ if (acpi_lapic_addr) {
+ iounmap(dist_base);
+ pr_warn(PREFIX "Invalid GIC cpu interface base address\n");
+ return -EINVAL;
+ }
+
+ cpu_base = ioremap(acpi_lapic_addr, GIC_CPU_INTERFACE_MEMORY_SIZE);
+ if (!cpu_base) {
+ iounmap(dist_base);
+ pr_warn(PREFIX "unable to map gic cpu registers\n");
+ return -ENOMEM;
+ }
+
+ gic_init(distributor->gic_id, -1, dist_base, cpu_base);
+
return 0;
}
+#else
+static int __init
+acpi_parse_gic_distributor(struct acpi_subtable_header *header,
+ const unsigned long end)
+{
+ return -ENODEV;
+}
+#endif /* CONFIG_ARM_GIC */
/*
* Parse GIC cpu interface related entries in MADT
@@ -234,7 +280,7 @@ acpi_parse_gic_distributor(struct acpi_subtable_header *header,
static int __init acpi_parse_madt_gic_entries(void)
{
int count;
-
+
/*
* do a partial walk of MADT to determine how many CPUs
* we have including disabled CPUs
@@ -468,19 +514,21 @@ static void __init acpi_process_madt(void)
* Parse MADT GIC cpu interface entries
*/
error = acpi_parse_madt_gic_entries();
- if (!error) {
- /*
- * Parse MADT GIC distributor entries
- */
- acpi_parse_madt_gic_distributor_entries();
- }
+ if (!error)
+ pr_info("Using ACPI for processor (GIC) configuration information\n");
}
- pr_info("Using ACPI for processor (GIC) configuration information\n");
-
return;
}
+int __init acpi_gic_init(void)
+{
+ /*
+ * Parse MADT GIC distributor entries
+ */
+ return acpi_parse_madt_gic_distributor_entries();
+}
+
Why can't you do the GIC init in the GIC code? We've tried hard to make
interrupt controllers discoverable and self contained.
thanks for your suggestion, Rob also had the same suggestion,
will try to update it in next version.

What are you
going to do when ACPI adds GICv3 to the mix? I don't really think this
model (shoving everything into the core ACPI code) is sustainable in the
long run...
Since GICv3 related ACPI proposal is not public and not goes into ACPI
spec, my suggestion is that we implement GICv2 only for now and post
another patches for GICv3 when the new ACPI spec is available.
Certainly. But I think you should aim for a scalable solution right
away, instead of starting with something that we already know won't work
for stuff that is already around the corner (which is what I infer from
your "non public" statement).

yes, sure I will, thanks for your comments.

Thanks
Hanjun
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/