arm64: GICv3 support
Subsystems (7)
cover arm64: GICv3 support
GICv3 is the base for a new generation of interrupt controllers designed to overcome some of the most glaring limitations of GICv2.
Some of the features are:
- Support for more than 8 CPUs (a lot more...)
- System registers for CPU interface access (GICC, GICV, GICH)
- Message based interrupts
This patch series currently support:
- Affinity Routing
- System Registers
- Non-Secure Group-1 interrupts only
- KVM support (GICv3 host, GICv2 guest)
What is *not yet* supported in this series:
- LPI/ITS/MSI (patches are available in the below branch, though still in development)
- KVM GICv3 guest support (requires GICv3 distributor emulation)
- Any form of power management
- 32bit systems
To be built, this code requires a fairly recent compiler/binutils combo. Linaro 13.06 seems to do the trick. This has been tested on the ARM FVP model, with non-regressions run on a VExpress TC-2.
The code is also available at the following location:git://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms.git kvm-arm64/gicv3
Marc Zyngier (18):
arm64: initial support for GICv3
arm64: GICv3 device tree binding documentation
arm64: boot protocol documentation update for GICv3
KVM: arm/arm64: vgic: move GICv2 registers to their own structure
KVM: ARM: vgic: introduce vgic_ops and LR manipulation primitives
KVM: ARM: vgic: abstract access to the ELRSR bitmap
KVM: ARM: vgic: abstract EISR bitmap access
KVM: ARM: vgic: abstract MISR decoding
KVM: ARM: vgic: move underflow handling to vgic_ops
KVM: ARM: vgic: abstract VMCR access
KVM: ARM: vgic: introduce vgic_enable
KVM: ARM: introduce vgic_params structure
KVM: ARM: vgic: split GICv2 backend from the main vgic code
arm64: KVM: remove __kvm_hyp_code_{start,end} from hyp.S
arm64: KVM: split GICv2 world switch from hyp code
arm64: KVM: move hcr_el2 setting into vgic-v2-switch.S
KVM: ARM: vgic: add the GICv3 backend
arm64: KVM: vgic: add GICv3 world switch
Documentation/arm64/booting.txt | 7 + Documentation/devicetree/bindings/arm/gic-v3.txt | 81 +++ arch/arm/include/asm/kvm_host.h | 5 + arch/arm/kernel/asm-offsets.c | 14 +- arch/arm/kvm/Makefile | 1 + arch/arm/kvm/interrupts_head.S | 26 +- arch/arm64/Kconfig | 1 + arch/arm64/include/asm/kvm_asm.h | 18 +- arch/arm64/include/asm/kvm_host.h | 21 + arch/arm64/include/asm/virt.h | 4 + arch/arm64/kernel/asm-offsets.c | 22 +- arch/arm64/kernel/head.S | 14 + arch/arm64/kernel/hyp-stub.S | 41 +- arch/arm64/kvm/Makefile | 4 + arch/arm64/kvm/hyp.S | 132 +---- arch/arm64/kvm/vgic-v2-switch.S | 149 +++++ arch/arm64/kvm/vgic-v3-switch.S | 275 ++++++++++ drivers/irqchip/Kconfig | 5 + drivers/irqchip/Makefile | 1 + drivers/irqchip/irq-gic-v3.c | 665 +++++++++++++++++++++++ include/kvm/arm_vgic.h | 103 +++- include/linux/irqchip/arm-gic-v3.h | 190 +++++++ virt/kvm/arm/vgic-v2.c | 228 ++++++++ virt/kvm/arm/vgic-v3.c | 220 ++++++++ virt/kvm/arm/vgic.c | 330 ++++++----- 25 files changed, 2238 insertions(+), 319 deletions(-) create mode 100644 Documentation/devicetree/bindings/arm/gic-v3.txt create mode 100644 arch/arm64/kvm/vgic-v2-switch.S create mode 100644 arch/arm64/kvm/vgic-v3-switch.S create mode 100644 drivers/irqchip/irq-gic-v3.c create mode 100644 include/linux/irqchip/arm-gic-v3.h create mode 100644 virt/kvm/arm/vgic-v2.c create mode 100644 virt/kvm/arm/vgic-v3.c
Signature · 1 line
1.8.3.4
PATCH 1/18 arm64: initial support for GICv3
The Generic Interrupt Controller (version 3) offers services that are similar to GICv2, with a number of additional features:
- Affinity routing based on the CPU MPIDR (ARE)
- System register for the CPU interfaces (SRE)
- Support for more that 8 CPUs
- Locality-specific Peripheral Interrupts (LPIs)
- Interrupt Translation Services (ITS)
This patch adds preliminary support for GICv3 with ARE and SRE, non-secure mode only. It relies on higher exception levels to grant ARE and SRE access.
Support for LPI and ITS will be added at a later time.
--- arch/arm64/Kconfig | 1 + arch/arm64/kernel/head.S | 14 + arch/arm64/kernel/hyp-stub.S | 41 ++- drivers/irqchip/Kconfig | 5 + drivers/irqchip/Makefile | 1 + drivers/irqchip/irq-gic-v3.c | 665 +++++++++++++++++++++++++++++++++++++ include/linux/irqchip/arm-gic-v3.h | 190 +++++++++++ 7 files changed, 915 insertions(+), 2 deletions(-) create mode 100644 drivers/irqchip/irq-gic-v3.c create mode 100644 include/linux/irqchip/arm-gic-v3.h
arch/arm64/Kconfig +1 -0
@@ -9,6 +9,7 @@ config ARM64 select ARM_AMBA select ARM_ARCH_TIMER select ARM_GIC + select ARM_GIC_V3 select BUILDTIME_EXTABLE_SORT select CLONE_BACKWARDS select COMMON_CLK
arch/arm64/kernel/head.S +14 -0
@@ -22,6 +22,7 @@ #include <linux/linkage.h> #include <linux/init.h> +#include <linux/irqchip/arm-gic-v3.h> #include <asm/assembler.h> #include <asm/ptrace.h>
@@ -183,6 +184,19 @@ CPU_LE( bic x0, x0, #(3 << 24) ) // Clear the EE and E0E bits for EL1 msr cnthctl_el2, x0 msr cntvoff_el2, xzr // Clear virtual offset + /* GICv3 system register access */ + mrs x0, id_aa64pfr0_el1 + ubfx x0, x0, #24, #4 + cmp x0, #1 + b.ne 3f + + mrs x0, ICC_SRE_EL2 // Set ICC_SRE_EL2.SRE==1 + orr x0, x0, #1 + msr ICC_SRE_EL2, x0 + isb // Make sure SRE is now 1 + msr ICH_HCR_EL2, xzr // Reset ICC_HCR_EL2 to defaults + +3: /* Populate ID registers. */ mrs x0, midr_el1 mrs x1, mpidr_el1
arch/arm64/kernel/hyp-stub.S +39 -2
61 lines not on this page · raw message
drivers/irqchip/Kconfig +5 -0
@@ -10,6 +10,11 @@ config ARM_GIC config GIC_NON_BANKED bool +config ARM_GIC_V3 + bool + select IRQ_DOMAIN + select MULTI_IRQ_HANDLER + config ARM_NVIC bool select IRQ_DOMAIN
drivers/irqchip/Makefile +1 -0
@@ -14,6 +14,7 @@ obj-$(CONFIG_ORION_IRQCHIP) += irq-orion.o obj-$(CONFIG_ARCH_SUNXI) += irq-sun4i.o obj-$(CONFIG_ARCH_SPEAR3XX) += spear-shirq.o obj-$(CONFIG_ARM_GIC) += irq-gic.o +obj-$(CONFIG_ARM_GIC_V3) += irq-gic-v3.o obj-$(CONFIG_ARM_NVIC) += irq-nvic.o obj-$(CONFIG_ARM_VIC) += irq-vic.o obj-$(CONFIG_IMGPDC_IRQ) += irq-imgpdc.o
drivers/irqchip/irq-gic-v3.c new file +665 -0
665 lines not on this page · raw message
include/linux/irqchip/arm-gic-v3.h new file +190 -0
190 lines not on this page · raw message
Signature · 1 line
1.8.3.4
PATCH 1/18 arm64: initial support for GICv3
Hi Marc
Review
Quoted 2 lines · Marc Zyngier <marc.zyngier <at> arm.com> writes:
(quoted diff) +2
-0+ +static inline void __iomem *gic_dist_base(struct irq_data *d)
I would suggest that this function be renamed (something like gic_base_for_irq?) since it returns dist or redist sgi base address. The name suggests it always returns the dist base address.
Quoted 14 lines
drivers/irqchip/irq-gic-v3.c new file +14
-0+{ +if (d->hwirq <32 )/* SGI+PPI -> SGI_base for this CPU */ +return gic_data_rdist_sgi_base (); + +if (d->hwirq <=1023 )/* SPI -> dist_base */ +return gic_data.dist_base; + +if (d->hwirq >=8192 ) +BUG ();/* LPI Detected!!! */ + +return NULL ; +} + +static inline unsigned int gic_irq (struct irq_data *d)
drivers/irqchip/irq-gic-v3.c line 77
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+{ +return d->hwirq; +} + +static void gic_do_wait_for_rwp (void __iomem *base) +{ +u32 val; + + do { + val = readl_relaxed(base + GICD_CTLR);
Maybe rename GICD_CTLR to GICx_CTLR since it is being used for both the distributor and the redistributor.
drivers/irqchip/irq-gic-v3.c line 79
Quoted 2 lines
drivers/irqchip/irq-gic-v3.c new file +2
-0+cpu_relax (); + }while (val &GICD_CTLR_RWP );
Similar to above GICx_CTLR_RWP
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c new file +8
-0+ +static int gic_irq_domain_xlate (struct irq_domain *d, +struct device_node *controller, +const u32 *intspec,unsigned int intsize, +unsigned long *out_hwirq,unsigned int *out_type) +{ +if (d->of_node != controller) +return -EINVAL ;
drivers/irqchip/irq-gic-v3.c line 559
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+if (intsize <3 ) +return -EINVAL ; + +switch (intspec[0 ]) { +case 0 :/* SPI */ + *out_hwirq = intspec[1 ] +32 ; +break ; +case 1 :/* PPI */ + *out_hwirq = intspec[1 ] +16 ; +break ;
I wonder if there is any value to syncing these with the defines in "include/dt-bindings/interrupt-controller/arm-gic.h" somehow.
Thanks
Arnab
PATCH 1/18 arm64: initial support for GICv3
Quoted 30 lines · Arnab Basu
Hi Marc
Review
Quoted 2 lines · Marc Zyngier <marc.zyngier <at> arm.com> writes:
(quoted diff) +2
-0+ +static inline void __iomem *gic_dist_base(struct irq_data *d)I would suggest that this function be renamed (something like
gic_base_for_irq?) since it returns dist or redist sgi base address. The name suggests it always returns the dist base address.Quoted 19 lines
drivers/irqchip/irq-gic-v3.c new file +19
-0+{ +if (d->hwirq <32 )/* SGI+PPI -> SGI_base for this CPU */ +return gic_data_rdist_sgi_base (); + +if (d->hwirq <=1023 )/* SPI -> dist_base */ +return gic_data.dist_base; + +if (d->hwirq >=8192 ) +BUG ();/* LPI Detected!!! */ + +return NULL ; +} + +static inline unsigned int gic_irq (struct irq_data *d) +{ +return d->hwirq; +} + +static void gic_do_wait_for_rwp (void __iomem *base)
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 77
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+{ +u32 val; + + do { + val = readl_relaxed(base + GICD_CTLR);Maybe rename
GICD_CTLRtoGICx_CTLRsince it is being used for both the distributor and the redistributor.
GICx_ makes it sound relevant for GIC CPU interfaces (GICC) as well. Unless it is, maybe something more like GICDR_ for Generic Interrupt Controller Distributor/Redistributor would be best?
Christopher
Signature · 3 lines
Employee of Qualcomm Innovation Center, Inc. Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by the Linux Foundation.
PATCH 1/18 arm64: initial support for GICv3
In reply to Arnab BasuHi Arnab,
Please do not trim the CC list. I missed this email (lost in the traffic), and contributors to the other MLs may also have comments.
Quoted 10 lines · Arnab Basu
Hi Marc
Review
Quoted 2 lines · Marc Zyngier <marc.zyngier <at> arm.com> writes:
(quoted diff) +2
-0+ +static inline void __iomem *gic_dist_base(struct irq_data *d)I would suggest that this function be renamed (something like
gic_base_for_irq?) since it returns dist or redist sgi base address. The name suggests it always returns the dist base address.
I may rename it to gic_get_irq_base.
Quoted 19 lines
Quoted 19 lines
drivers/irqchip/irq-gic-v3.c new file +19
-0+{ +if (d->hwirq <32 )/* SGI+PPI -> SGI_base for this CPU */ +return gic_data_rdist_sgi_base (); + +if (d->hwirq <=1023 )/* SPI -> dist_base */ +return gic_data.dist_base; + +if (d->hwirq >=8192 ) +BUG ();/* LPI Detected!!! */ + +return NULL ; +} + +static inline unsigned int gic_irq (struct irq_data *d) +{ +return d->hwirq; +} + +static void gic_do_wait_for_rwp (void __iomem *base)
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 77
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+{ +u32 val; + + do { + val = readl_relaxed(base + GICD_CTLR);Maybe rename
GICD_CTLRtoGICx_CTLRsince it is being used for both the distributor and the redistributor.
I really hate the "GICx" bit. I think instead, I'll add a comment so people don't get confuse about the name.
Quoted 19 lines
drivers/irqchip/irq-gic-v3.c line 79
Quoted 2 lines
drivers/irqchip/irq-gic-v3.c new file +2
-0+cpu_relax (); + }while (val &GICD_CTLR_RWP );Similar to above
GICx_CTLR_RWPQuoted 13 lines
drivers/irqchip/irq-gic-v3.c new file +13
-0+ +static int gic_irq_domain_xlate (struct irq_domain *d, +struct device_node *controller, +const u32 *intspec,unsigned int intsize, +unsigned long *out_hwirq,unsigned int *out_type) +{ +if (d->of_node != controller) +return -EINVAL ; +if (intsize <3 ) +return -EINVAL ; + +switch (intspec[0 ]) { +case 0 :/* SPI */
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 559
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+ *out_hwirq = intspec[1 ] +32 ; +break ; +case 1 :/* PPI */ + *out_hwirq = intspec[1 ] +16 ; +break ;I wonder if there is any value to syncing these with the defines in "
include/dt-bindings/interrupt-controller/arm-gic.h" somehow.
I'd expect the whole DTS stuff to vanish from the kernel some day, so I'm keen on not depending on it.
Thanks for the review,
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
PATCH 1/18 arm64: initial support for GICv3
In reply to Marc ZyngierQuoted 57 lines · Marc Zyngier
The Generic Interrupt Controller (version 3) offers services that are similar to GICv2, with a number of additional features:
- Affinity routing based on the CPU MPIDR (ARE)
- System register for the CPU interfaces (SRE)
- Support for more that 8 CPUs
- Locality-specific Peripheral Interrupts (LPIs)
- Interrupt Translation Services (ITS)This patch adds preliminary support for GICv3 with ARE and SRE, non-secure mode only. It relies on higher exception levels to grant ARE and SRE access.
Support for LPI and ITS will be added at a later time.
Acked-by Catalin Marinas <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- arch/arm64/Kconfig | 1 + arch/arm64/kernel/head.S | 14 + arch/arm64/kernel/hyp-stub.S | 41 ++- drivers/irqchip/Kconfig | 5 + drivers/irqchip/Makefile | 1 + drivers/irqchip/irq-gic-v3.c | 665 +++++++++++++++++++++++++++++++++++++ include/linux/irqchip/arm-gic-v3.h | 190 +++++++++++ 7 files changed, 915 insertions(+), 2 deletions(-) create mode 100644 drivers/irqchip/irq-gic-v3.c create mode 100644 include/linux/irqchip/arm-gic-v3.harch/arm64/Kconfig +1
-0@@ -9,6 +9,7 @@ config ARM64 select ARM_AMBA select ARM_ARCH_TIMER select ARM_GIC + select ARM_GIC_V3 select BUILDTIME_EXTABLE_SORT select CLONE_BACKWARDS select COMMON_CLKarch/arm64/kernel/head.S +2
-0@@ -22,6 +22,7 @@ #include <linux/linkage.h> #include <linux/init.h> +#include <linux/irqchip/arm-gic-v3.h> #include <asm/assembler.h> #include <asm/ptrace.h>@@ -183,6 +184,19 @@ CPU_LE( bic x0, x0, #(3 << 24) ) // Clear the EE and E0E bits for EL1 msr cnthctl_el2, x0 msr cntvoff_el2, xzr // Clear virtual offset + /* GICv3 system register access */
arch/arm64/kernel/head.S CPU_LE( bic x0, x0, #(3 << 24) ) // Clear the EE and E0E bits for EL1 · line 197
Quoted 10 lines
arch/arm64/kernel/head.S +10
-0+ mrs x0, id_aa64pfr0_el1 + ubfx x0, x0, #24, #4 + cmp x0, #1 + b.ne 3f + + mrs x0, ICC_SRE_EL2 // Set ICC_SRE_EL2.SRE==1 + orr x0, x0, #1 + msr ICC_SRE_EL2, x0 + isb // Make sure SRE is now 1 + msr ICH_HCR_EL2, xzr // Reset ICC_HCR_EL2 to defaults
Why do we not just set ICC_SRE_EL2.Enable here, but instead do the whole exception dance below?
Quoted 22 lines
arch/arm64/kernel/head.S +2
-0+ +3: /* Populate ID registers. */ mrs x0, midr_el1 mrs x1, mpidr_el1arch/arm64/kernel/hyp-stub.S +2
-0@@ -19,6 +19,7 @@ #include <linux/init.h> #include <linux/linkage.h> +#include <linux/irqchip/arm-gic-v3.h> #include <asm/assembler.h> #include <asm/ptrace.h>@@ -51,16 +52,52 @@ ENDPROC(__hyp_stub_vectors) .align 11 +#define sysreg_to_iss(op0, op1, crn, crm, op2) \
arch/arm64/kernel/hyp-stub.S ENDPROC(__hyp_stub_vectors) · line 65
Quoted 10 lines
arch/arm64/kernel/hyp-stub.S +9
-0+ (((op0 & 3) << 20) | ((op2 & 7 ) << 17) | \ + ((op1 & 7 ) << 14) | ((crn & 15) << 10) | \ + ((crm & 15) << 1)) + el1_sync: + // Hack alert: we don't have a stack, and we can't use the one + // from EL1 because our MMU is off. Thankfully, we have + // tpidr_el2, which is under our complete control, and + // tpidr_el0, which can be used as long as we haven't reached + // userspace. Oh hum...
Hmmm. I feel like either we should tone down the hack alert claim, because this is after all safe and we just explain what's going on, or we should just allocate a small stack space following el1_sync and set the SP upon entry to this handler?
Hopefully none of the code that will ever call into the sysregs trap ever makes the same assumption that user space is not called yet ;)
Quoted 18 lines
arch/arm64/kernel/hyp-stub.S +9
-2+ msr tpidr_el2, x1 // Preserve x1 + mrs x1, esr_el2 lsr x1, x1, #26 cmp x1, #0x16 - b.ne 2f // Not an HVC trap + b.ne 3f // Not an HVC trap + + // Handle getting/setting vbar_el2 cbz x0, 1f msr vbar_el2, x0 // Set vbar_el2 b 2f 1: mrs x0, vbar_el2 // Return vbar_el2 -2: eret +2: mrs x1, tpidr_el2 // Restore x1 + eret + +3: cmp x1, #0x18 // MSR/MRS?
arch/arm64/kernel/hyp-stub.S ENDPROC(__hyp_stub_vectors) · line 91
Quoted 10 lines
arch/arm64/kernel/hyp-stub.S +10
-0+ b.ne 2b + + // Handle trapped sys reg access + msr tpidr_el0, x0 // Preserve x0 + mrs x1, esr_el2 + ldr x0, =sysreg_to_iss(3, 7, 15, 15, 7) + and x1, x1, x0 + ldr x0, =sysreg_to_iss(3, 0, 12, 12, 5) + eor x0, x0, x1 // Check for ICC_SRE_EL1 + cbnz x0, 4f // Something bad happened
Shouldn't we try to do something here, like go to panic on the exception return here, or not worth it?
Quoted 335 lines
arch/arm64/kernel/hyp-stub.S +9
-0+ + // Set ICC_SRE_EL1 access enable + mrs x1, ICC_SRE_EL2 + orr x1, x1, #(1 << 3) + msr ICC_SRE_EL2, x1 + +4: mrs x0, tpidr_el0 // Restore x0 + mrs x1, tpidr_el2 // Restore x1 + eret ENDPROC(el1_sync) .macro invalid_vector labeldrivers/irqchip/Kconfig +5
-0@@ -10,6 +10,11 @@ config ARM_GIC config GIC_NON_BANKED bool +config ARM_GIC_V3 + bool + select IRQ_DOMAIN + select MULTI_IRQ_HANDLER + config ARM_NVIC bool select IRQ_DOMAINdrivers/irqchip/Makefile +1
-0@@ -14,6 +14,7 @@ obj-$(CONFIG_ORION_IRQCHIP) += irq-orion.o obj-$(CONFIG_ARCH_SUNXI) += irq-sun4i.o obj-$(CONFIG_ARCH_SPEAR3XX) += spear-shirq.o obj-$(CONFIG_ARM_GIC) += irq-gic.o +obj-$(CONFIG_ARM_GIC_V3) += irq-gic-v3.o obj-$(CONFIG_ARM_NVIC) += irq-nvic.o obj-$(CONFIG_ARM_VIC) += irq-vic.o obj-$(CONFIG_IMGPDC_IRQ) += irq-imgpdc.odrivers/irqchip/irq-gic-v3.c new file +289
-0@@ -0,0 +1,665 @@ +/* +* Copyright (C) 2013 ARM Limited, All Rights Reserved. +* Author: Marc Zyngier <[email protected]> +* +* This program is free software; you can redistribute it and/or modify +* it under the terms of the GNU General Public License version 2 as +* published by the Free Software Foundation. +* +* This program is distributed in the hope that it will be useful, +* but WITHOUT ANY WARRANTY; without even the implied warranty of +* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +* GNU General Public License for more details. +* +* You should have received a copy of the GNU General Public License +* along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +#include <linux/cpu.h> +#include <linux/interrupt.h> +#include <linux/of.h> +#include <linux/of_address.h> +#include <linux/of_irq.h> +#include <linux/percpu.h> +#include <linux/slab.h> + +#include <linux/irqchip/arm-gic-v3.h> + +#include <asm/cputype.h> +#include <asm/exception.h> +#include <asm/smp_plat.h> + +#include "irqchip.h" + +struct gic_chip_data { +void __iomem *dist_base; +void __iomem **redist_base; +void __percpu __iomem **rdist; +struct irq_domain *domain; +u64 redist_stride; +u32 redist_regions; +unsigned int irq_nr; +}; + +static struct gic_chip_data gic_data__read_mostly ; + +#define gic_data_rdist_rd_base () (*__this_cpu_ptr(gic_data.rdist)) +#define gic_data_rdist_sgi_base () (gic_data_rdist_rd_base() + SZ_64K) + +static DEFINE_RAW_SPINLOCK (dist_lock); + +#define DEFAULT_PMR_VALUE 0xf0 + +static inline void __iomem *gic_dist_base (struct irq_data *d) +{ +if (d->hwirq <32 )/* SGI+PPI -> SGI_base for this CPU */ +return gic_data_rdist_sgi_base (); + +if (d->hwirq <=1023 )/* SPI -> dist_base */ +return gic_data.dist_base; + +if (d->hwirq >=8192 ) +BUG ();/* LPI Detected!!! */ + +return NULL ; +} + +static inline unsigned int gic_irq (struct irq_data *d) +{ +return d->hwirq; +} + +static void gic_do_wait_for_rwp (void __iomem *base) +{ +u32 val; + +do { + val =readl_relaxed (base +GICD_CTLR ); +cpu_relax (); + }while (val &GICD_CTLR_RWP ); +} + +/* Wait for completion of a distributor change */ +static void gic_dist_wait_for_rwp (void ) +{ +gic_do_wait_for_rwp (gic_data.dist_base); +} + +/* Wait for completion of a redistributor change */ +static void gic_redist_wait_for_rwp (void ) +{ +gic_do_wait_for_rwp (gic_data_rdist_rd_base ()); +} + +static void gic_wait_for_rwp (int irq) +{ +if (irq <32 ) +gic_redist_wait_for_rwp (); +else +gic_dist_wait_for_rwp (); +} + +/* Low level accessors */ +static u64 gic_read_iar (void ) +{ +u64 irqstat; + + asm volatile("mrs %0, " reg(ICC_IAR1_EL1) : "=r" (irqstat)); +return irqstat; +} + +static void gic_write_pmr (u64 val) +{ + asm volatile("msr " reg(ICC_PMR_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_ctlr (u64 val) +{ + asm volatile("msr " reg(ICC_CTLR_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_grpen1 (u64 val) +{ + asm volatile("msr " reg(ICC_GRPEN1_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_sgi1r (u64 val) +{ + asm volatile("msr " reg(ICC_SGI1R_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_enable_sre (void ) +{ +u64 val; + +/* This access is likely to trap into EL2 */ + asm volatile("mrs %0, " reg(ICC_SRE_EL1) : "=r" (val)); + val |=GICC_SRE_EL1_SRE ; + asm volatile("msr " reg(ICC_SRE_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_enable_redist (void ) +{ +void __iomem *rbase; +u32 val; + + rbase =gic_data_rdist_rd_base (); + +/* Wake up this CPU redistributor */ + val =readl_relaxed (rbase +GICR_WAKER ); + val &= ~GICR_WAKER_ProcessorSleep; +writel_relaxed (val, rbase +GICR_WAKER ); + +do { + val =readl_relaxed (rbase +GICR_WAKER ); +cpu_relax (); + }while (val & GICR_WAKER_ChildrenAsleep); +} + +/* +* Routines to acknowledge, disable and enable interrupts +*/ +static void gic_mask_irq (struct irq_data *d) +{ +u32 mask =1 << (gic_irq (d) %32 ); + +raw_spin_lock (&dist_lock); +writel_relaxed (mask,gic_dist_base (d) +GICD_ICENABLER + (gic_irq (d) /32 ) *4 ); +gic_wait_for_rwp (gic_irq (d)); +raw_spin_unlock (&dist_lock); +} + +static void gic_unmask_irq (struct irq_data *d) +{ +u32 mask =1 << (gic_irq (d) %32 ); + +raw_spin_lock (&dist_lock); +writel_relaxed (mask,gic_dist_base (d) +GICD_ISENABLER + (gic_irq (d) /32 ) *4 ); +gic_wait_for_rwp (gic_irq (d)); +raw_spin_unlock (&dist_lock); +} + +static void gic_eoi_irq (struct irq_data *d) +{ +gic_write_eoir (gic_irq (d)); +} + +static int gic_set_type (struct irq_data *d,unsigned int type) +{ +void __iomem *base =gic_dist_base (d); +unsigned int irq =gic_irq (d); +u32 enablemask =1 << (irq %32 ); +u32 enableoff = (irq /32 ) *4 ; +u32 confmask =0x2 << ((irq %16 ) *2 ); +u32 confoff = (irq /16 ) *4 ; +bool enabled = false; +u32 val; + +/* Interrupt configuration for SGIs can't be changed */ +if (irq <16 ) +return -EINVAL ; + +if (type !=IRQ_TYPE_LEVEL_HIGH && type !=IRQ_TYPE_EDGE_RISING ) +return -EINVAL ; + +raw_spin_lock (&dist_lock); + + val =readl_relaxed (base +GICD_ICFGR + confoff); +if (type ==IRQ_TYPE_LEVEL_HIGH ) + val &= ~confmask; +else if (type ==IRQ_TYPE_EDGE_RISING ) + val |= confmask; + +/* +* As recommended by the spec, disable the interrupt before changing +* the configuration +*/ +if (readl_relaxed (base +GICD_ISENABLER + enableoff) & enablemask) { +writel_relaxed (enablemask, base +GICD_ICENABLER + enableoff); +gic_wait_for_rwp (irq); + enabled = true; + } + +writel_relaxed (val, base +GICD_ICFGR + confoff); + +if (enabled) { +writel_relaxed (enablemask, base +GICD_ISENABLER + enableoff); +gic_wait_for_rwp (irq); + } + +raw_spin_unlock (&dist_lock); + +return 0 ; +} + +static int gic_retrigger (struct irq_data *d) +{ +return -ENXIO ; +} + +static u64 gic_mpidr_to_affinity (u64 mpidr) +{ +/* Make sure we don't broadcast the interrupt */ +return mpidr & ~GICD_IROUTER_SPI_MODE_ANY ; +} + +static asmlinkage void __exception_irq_entry gic_handle_irq(struct pt_regs *regs) +{ +u64 irqnr; + +do { + irqnr =gic_read_iar (); + +if (likely (irqnr >15 && irqnr <1021 )) { + irqnr =irq_find_mapping (gic_data.domain, irqnr); +handle_IRQ (irqnr, regs); +continue ; + } +if (irqnr <16 ) { +gic_write_eoir (irqnr); +#ifdef CONFIG_SMP +handle_IPI (irqnr, regs); +#else +WARN_ONCE (true,"Unexpected SGI received!\n" ); +#endif +continue ; + } + }while (irqnr !=0x3ff ); +} + +static void __init gic_dist_init (void ) +{ +unsigned int i; +u64 affinity; +int gic_irqs = gic_data.irq_nr; +void __iomem *base = gic_data.dist_base; + +/* Disable the distributor */ +writel_relaxed (0 , base +GICD_CTLR ); + +/* +* Set all global interrupts to be level triggered, active low. +*/ +for (i =32 ; i < gic_data.irq_nr; i +=16 ) +writel_relaxed (0 , base +GICD_ICFGR + i /4 );
drivers/irqchip/irq-gic-v3.c line 299
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+ +/* +* Set priority on all global interrupts. +*/ +for (i =32 ; i < gic_irqs; i +=4 ) +writel_relaxed (0xa0a0a0a0 , base +GICD_IPRIORITYR + i); + +/* +* Disable all interrupts. Leave the PPI and SGIs alone +* as these enables are banked registers.
as these are enabled through banked registers?
also, should banked registers be redistributor registers?
drivers/irqchip/irq-gic-v3.c line 307
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c new file +8
-0+*/ +for (i =32 ; i < gic_irqs; i +=32 ) +writel_relaxed (0xffffffff , base +GICD_ICENABLER + i /8 ); + +gic_dist_wait_for_rwp (); + +writel_relaxed (GICD_CTLR_ARE_NS |GICD_CTLR_ENABLE_G1A |GICD_CTLR_ENABLE_G1 , + base +GICD_CTLR );
why set GICD_CTLR_ENABLE_G1 ?
Quoted 9 lines
drivers/irqchip/irq-gic-v3.c new file +9
-0+ +/* +* Set all global interrupts to the boot CPU only. ARE must be +* enabled. +*/ + affinity =gic_mpidr_to_affinity (read_cpuid_mpidr ()); +for (i =32 ; i < gic_data.irq_nr; i++) +writeq_relaxed (affinity, base +GICD_IROUTER + i *8 ); +}
drivers/irqchip/irq-gic-v3.c line 326
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+ +static int __init gic_populate_rdist (void ) +{ +u64 mpidr =cpu_logical_map (smp_processor_id ()); +u64 typer; +u64 aff; +int i; + + aff = mpidr & ((1 <<24 ) -1 ); + aff |= (mpidr >>8 ) & (0xffUL <<24 );
Some defines would be nice for this if it makes sense in any way.
Quoted 45 lines
drivers/irqchip/irq-gic-v3.c new file +45
-0+ +for (i =0 ; i < gic_data.redist_regions; i++) { +void __iomem *ptr = gic_data.redist_base[i]; +u32 reg; + + reg =readl_relaxed (ptr +GICR_PIDR2 ) &GIC_PIDR2_ARCH_MASK ; +if (reg !=0x30 && reg !=0x40 ) {/* We're in trouble... */ +pr_warn ("No redistributor present @%p\n" , ptr); +break ; + } + +do { + typer =readq_relaxed (ptr +GICR_TYPER ); +if ((typer >>32 ) == aff) { +gic_data_rdist_rd_base () = ptr; +pr_info ("CPU%d: found redistributor %llx @%p\n" , +smp_processor_id (), + (unsigned long long ) mpidr, ptr); +return 0 ; + } + +if (gic_data.redist_stride) { + ptr += gic_data.redist_stride; + }else { + ptr +=SZ_64K *2 ;/* Skip RD_base + SGI_base */ +if (typer &GICR_TYPER_VLPIS ) + ptr +=SZ_64K *2 ;/* Skip VLPI_base + reserved page */ + } + }while (!(typer &GICR_TYPER_LAST )); + } + +/* We couldn't even deal with ourselves... */ +WARN (true,"CPU%d: mpidr %lx has no re-distributor!\n" , +smp_processor_id (), (unsigned long )mpidr); +return -ENODEV ; +} + +static void __init gic_cpu_init (void ) +{ +void __iomem *rbase; +int i; + +/* Register ourselves with the rest of the world */ +if (gic_populate_rdist ()) +return ;
drivers/irqchip/irq-gic-v3.c line 381
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+ +gic_enable_redist (); + + rbase =gic_data_rdist_sgi_base (); + +/* +* Set priority on PPI and SGI interrupts +*/ +for (i =0 ; i <32 ; i +=4 ) +writel_relaxed (0xa0a0a0a0 , rbase +GICR_IPRIORITYR0 + i *4 /4 );
(i * 4) / 4 == i, last I checked.
drivers/irqchip/irq-gic-v3.c line 385
Quoted 4 lines
drivers/irqchip/irq-gic-v3.c new file +4
-0+ +/* +* Disable all PPI interrupts, ensure all SGI interrupts are +* enabled.
Why do we disable all PPIs?
drivers/irqchip/irq-gic-v3.c line 398
Quoted 13 lines
drivers/irqchip/irq-gic-v3.c new file +13
-0+*/ +writel_relaxed (0xffff0000 , rbase +GICR_ICENABLER0 ); +writel_relaxed (0x0000ffff , rbase +GICR_ISENABLER0 ); + +gic_redist_wait_for_rwp (); + +/* Enable system registers */ +gic_enable_sre (); + +/* Set priority mask register */ +gic_write_pmr (DEFAULT_PMR_VALUE ); + +/* EOI drops priority too (mode 0) */
you mean deactives the interrupt too, right? It always performs a priority drop doesn't it?
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c new file +8
-0+gic_write_ctlr (GICC_CTLR_EL1_EOImode_drop_dir); + +/* ... and let's hit the road... */ +gic_write_grpen1 (1 ); +} + +#ifdef CONFIG_SMP +static int __init gic_secondary_init(struct notifier_block *nfb,
drivers/irqchip/irq-gic-v3.c line 416
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+ unsigned long action, void *hcpu) +{ +if (action ==CPU_STARTING || action ==CPU_STARTING_FROZEN ) +gic_cpu_init (); +return NOTIFY_OK ; +} + +/* +* Notifier for enabling the GIC CPU interface. Set an arbitrarily high +* priority because the GIC needs to be up before the ARM generic timers.
I actually don't understand this bit, but it also sits in the original code, so I assume it's correct.
Quoted 15 lines
drivers/irqchip/irq-gic-v3.c new file +15
-0+*/ +static struct notifier_block __initdata gic_cpu_notifier = { + .notifier_call = gic_secondary_init, + .priority =100 , +}; + +static u16 gic_compute_target_list (int *base_cpu,const struct cpumask *mask, +u64 cluster_id) +{ +int cpu = *base_cpu; +u64 mpidr =cpu_logical_map (cpu); +u16 tlist =0 ; + +while (cpu < nr_cpu_ids) { +/*
drivers/irqchip/irq-gic-v3.c line 441
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+* If we ever get a cluster of more than 16 CPUs, just +* scream and skip that CPU. +*/ +if (WARN_ON ((mpidr &0xff ) >=16 )) + goto out; + + tlist |=1 << (mpidr &0xf ); + + cpu =cpumask_next (cpu, mask); + mpidr =cpu_logical_map (cpu);
Are you not accessing memory beyond the __cpu_logical_map array here when the cpu is the last one at the beginning of the loop?
Quoted 32 lines
drivers/irqchip/irq-gic-v3.c new file +32
-0+ +if (cluster_id != (mpidr & ~0xffUL )) { + cpu--; + goto out; + } + } +out: + *base_cpu = cpu; +return tlist; +} + +static void gic_send_sgi (u64 cluster_id,u16 tlist,unsigned int irq) +{ +u64 val; + + val = (cluster_id &0xff00ff0000UL ) <<16 ;/* Aff3 + Aff2 */ + val |= (cluster_id &0xff00 ) <<8 ;/* Aff1 */ + val |= irq <<24 ; + val |= tlist; + +pr_debug ("CPU%d: ICC_SGI1R_EL1 %llx\n" ,smp_processor_id (), val); +gic_write_sgi1r (val); +} + +static void gic_raise_softirq (const struct cpumask *mask,unsigned int irq) +{ +int cpu; + +if (WARN_ON (irq >=16 )) +return ; + +/*
drivers/irqchip/irq-gic-v3.c line 483
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+* Ensure that stores to Normal memory are visible to the +* other CPUs before issuing the IPI. +*/ +dsb (); + +for_each_cpu_mask (cpu, *mask) { +u64 cluster_id =cpu_logical_map (cpu) & ~0xffUL ; +u16 tlist; + + tlist =gic_compute_target_list (&cpu, mask, cluster_id);
I think it would be potentially slightly prettier if this function returned the last cpu in the cluster so the loop iterator mangling was kept local to this function and took a pointer to the tlist it was to fill instead, but, this is a minor detail.
Quoted 44 lines
drivers/irqchip/irq-gic-v3.c new file +44
-0+ gic_send_sgi(cluster_id, tlist, irq); + } +} + +static void gic_smp_init (void ) +{ +set_smp_cross_call (gic_raise_softirq); +register_cpu_notifier (&gic_cpu_notifier); +} + +static int gic_set_affinity (struct irq_data *d,const struct cpumask *mask_val, +bool force) +{ +unsigned int cpu =cpumask_any_and (mask_val, cpu_online_mask); +void __iomem *reg; +u64 val; + +if (gic_irq (d) <32 ) +return -EINVAL ; + + reg =gic_dist_base (d) +GICD_IROUTER + (gic_irq (d) *8 ); + val =gic_mpidr_to_affinity (cpu_logical_map (cpu)); + +writeq_relaxed (val, reg); + +return IRQ_SET_MASK_OK ; +} +#else +#define gic_set_affinity NULL +#define gic_smp_init () do { } while(0) +#endif + +static struct irq_chip gic_chip = { + .name ="GICv3" , + .irq_mask = gic_mask_irq, + .irq_unmask = gic_unmask_irq, + .irq_eoi = gic_eoi_irq, + .irq_set_type = gic_set_type, + .irq_retrigger = gic_retrigger, + .irq_set_affinity = gic_set_affinity, +}; + +static int gic_irq_domain_map (struct irq_domain *d,unsigned int irq, +irq_hw_number_t hw)
drivers/irqchip/irq-gic-v3.c line 537
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+{ +if (hw <32 ) { +irq_set_percpu_devid (irq); +irq_set_chip_and_handler (irq, &gic_chip, + handle_percpu_devid_irq); +set_irq_flags (irq,IRQF_VALID |IRQF_NOAUTOEN ); + }else { +irq_set_chip_and_handler (irq, &gic_chip, + handle_fasteoi_irq); +set_irq_flags (irq,IRQF_VALID |IRQF_PROBE );
will this ever be compiled for arch/arm? If not, does it make sense to keep the set_irq_flags calls?
Quoted 40 lines
drivers/irqchip/irq-gic-v3.c new file +40
-0+ } +irq_set_chip_data (irq, d->host_data); +return 0 ; +} + +static int gic_irq_domain_xlate (struct irq_domain *d, +struct device_node *controller, +const u32 *intspec,unsigned int intsize, +unsigned long *out_hwirq,unsigned int *out_type) +{ +if (d->of_node != controller) +return -EINVAL ; +if (intsize <3 ) +return -EINVAL ; + +switch (intspec[0 ]) { +case 0 :/* SPI */ + *out_hwirq = intspec[1 ] +32 ; +break ; +case 1 :/* PPI */ + *out_hwirq = intspec[1 ] +16 ; +break ; +default : +return -EINVAL ; + } + + *out_type = intspec[2 ] &IRQ_TYPE_SENSE_MASK ; +return 0 ; +} + +static const struct irq_domain_ops gic_irq_domain_ops = { + .map = gic_irq_domain_map, + .xlate = gic_irq_domain_xlate, +}; + +static int __init gic_of_init (struct device_node *node,struct device_node *parent) +{ +void __iomem *dist_base; +void __iomem **redist_base; +u64 redist_stride;
drivers/irqchip/irq-gic-v3.c line 587
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+u32 redist_regions; +u32 reg; +int gic_irqs; +int err; +int i; + + dist_base =of_iomap (node,0 ); +if (!dist_base) { +pr_warn ("%s: unable to map gic dist registers\n" , + node->full_name);
shouldn't these be pr_err?
drivers/irqchip/irq-gic-v3.c line 592
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+return -ENXIO ; + } + + reg =readl_relaxed (dist_base +GICD_PIDR2 ) &GIC_PIDR2_ARCH_MASK ; +if (reg !=0x30 && reg !=0x40 ) {
Is everything we do going to work with a GICv4? Anything else we need to initialize if it's a GICv4?
Quoted 17 lines
drivers/irqchip/irq-gic-v3.c new file +17
-0+pr_warn ("%s: no distributor detected, giving up\n" , + node->full_name); + err = -ENODEV ; + goto out_unmap_dist; + } + +if (of_property_read_u32 (node,"#redistributor-regions" , &redist_regions)) + redist_regions =1 ; + + redist_base =kzalloc (sizeof (*redist_base) * redist_regions,GFP_KERNEL ); +if (!redist_base) { + err = -ENOMEM ; + goto out_unmap_dist; + } + +for (i =0 ; i < redist_regions; i++) { + redist_base[i] =of_iomap (node,1 + i);
drivers/irqchip/irq-gic-v3.c line 619
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+if (!redist_base[i]) { +pr_warn ("%s: couldn't map region %d\n" , + node->full_name, i); + err = -ENODEV ; + goto out_unmap_rdist; + } + } + +if (of_property_read_u64 (node,"redistributor-stride" , &redist_stride)) + redist_stride =0 ;
The binding doesn't specify this as a two-cell property, so I assume this won't work if you follow the example in the bindings doc.
Quoted 133 lines
drivers/irqchip/irq-gic-v3.c new file +46
-0+ + gic_data.dist_base = dist_base; + gic_data.redist_base = redist_base; + gic_data.redist_regions = redist_regions; + gic_data.redist_stride = redist_stride; + +/* +* Find out how many interrupts are supported. +* The GIC only supports up to 1020 interrupt sources (SGI+PPI+SPI) +*/ + gic_irqs =readl_relaxed (gic_data.dist_base +GICD_TYPER ) &0x1f ; + gic_irqs = (gic_irqs +1 ) *32 ; +if (gic_irqs >1020 ) + gic_irqs =1020 ; + gic_data.irq_nr = gic_irqs; + + gic_data.domain =irq_domain_add_linear (node, gic_irqs -16 , + &gic_irq_domain_ops, &gic_data); + gic_data.rdist =alloc_percpu (typeof (*gic_data.rdist)); + +if (WARN_ON (!gic_data.domain) ||WARN_ON (!gic_data.rdist)) { + err = -ENOMEM ; + goto out_free; + } + +set_handle_irq (gic_handle_irq); + +gic_smp_init (); +gic_dist_init (); +gic_cpu_init (); + +return 0 ; + +out_free: +free_percpu (gic_data.rdist); +out_unmap_rdist: +for (i =0 ; i < redist_regions; i++) +if (redist_base[i]) +iounmap (redist_base[i]); +kfree (redist_base); +out_unmap_dist: +iounmap (dist_base); +return err; +} + +IRQCHIP_DECLARE (gic_v3,"arm,gic-v3" , gic_of_init);include/linux/irqchip/arm-gic-v3.h new file +81
-0@@ -0,0 +1,190 @@ +/* +* Copyright (C) 2013 ARM Limited, All Rights Reserved. +* Author: Marc Zyngier <[email protected]> +* +* +* This program is free software; you can redistribute it and/or modify +* it under the terms of the GNU General Public License version 2 as +* published by the Free Software Foundation. +* +* This program is distributed in the hope that it will be useful, +* but WITHOUT ANY WARRANTY; without even the implied warranty of +* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +* GNU General Public License for more details. +* +* You should have received a copy of the GNU General Public License +* along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ +#ifndef __LINUX_IRQCHIP_ARM_GIC_V3_H +#define __LINUX_IRQCHIP_ARM_GIC_V3_H + +/* +* Distributor registers. We assume we're running non-secure, with ARE +* being set. Secure-only and non-ARE registers are not described. +*/ +#define GICD_CTLR 0x0000 +#define GICD_TYPER 0x0004 +#define GICD_IIDR 0x0008 +#define GICD_STATUSR 0x0010 +#define GICD_SETSPI_NSR 0x0040 +#define GICD_CLRSPI_NSR 0x0048 +#define GICD_SETSPI_SR 0x0050 +#define GICD_CLRSPI_SR 0x0058 +#define GICD_SEIR 0x0068 +#define GICD_ISENABLER 0x0100 +#define GICD_ICENABLER 0x0180 +#define GICD_ISPENDR 0x0200 +#define GICD_ICPENDR 0x0280 +#define GICD_ISACTIVER 0x0300 +#define GICD_ICACTIVER 0x0380 +#define GICD_IPRIORITYR 0x0400 +#define GICD_ICFGR 0x0C00 +#define GICD_IROUTER 0x6000 +#define GICD_PIDR2 0xFFE8 + +#define GICD_CTLR_RWP (1U << 31) +#define GICD_CTLR_ARE_NS (1U << 4) +#define GICD_CTLR_ENABLE_G1A (1U << 1) +#define GICD_CTLR_ENABLE_G1 (1U << 0) + +#define GICD_IROUTER_SPI_MODE_ONE (0U << 31) +#define GICD_IROUTER_SPI_MODE_ANY (1U << 31) + +#define GIC_PIDR2_ARCH_MASK 0xf0 + +/* +* Re-Distributor registers, offsets from RD_base +*/ +#define GICR_CTLR GICD_CTLR +#define GICR_IIDR 0x0004 +#define GICR_TYPER 0x0008 +#define GICR_STATUSR GICD_STATUSR +#define GICR_WAKER 0x0014 +#define GICR_SETLPIR 0x0040 +#define GICR_CLRLPIR 0x0048 +#define GICR_SEIR GICD_SEIR +#define GICR_PROPBASER 0x0070 +#define GICR_PENDBASER 0x0078 +#define GICR_INVLPIR 0x00A0 +#define GICR_INVALLR 0x00B0 +#define GICR_SYNCR 0x00C0 +#define GICR_MOVLPIR 0x0100 +#define GICR_MOVALLR 0x0110 +#define GICR_PIDR2 GICD_PIDR2 + +#define GICR_WAKER_ProcessorSleep (1U << 1) +#define GICR_WAKER_ChildrenAsleep (1U << 2) + +/* +* Re-Distributor registers, offsets from SGI_base +*/ +#define GICR_ISENABLER0 GICD_ISENABLER
include/linux/irqchip/arm-gic-v3.h line 91
Quoted 10 lines
include/linux/irqchip/arm-gic-v3.h new file +10
-0+#define GICR_ICENABLER0 GICD_ICENABLER +#define GICR_ISPENDR0 GICD_ISPENDR +#define GICR_ICPENDR0 GICD_ICPENDR +#define GICR_ISACTIVER0 GICD_ISACTIVER +#define GICR_ICACTIVER0 GICD_ICACTIVER +#define GICR_IPRIORITYR0 GICD_IPRIORITYR +#define GICR_ICFGR0 GICD_ICFGR + +#define GICR_TYPER_VLPIS (1U << 1) +#define GICR_TYPER_LAST (1U << 4)
nit: why are these TYPER definitions down here when the TYPER register is defined above and so the WAKER bit definitions?
Quoted 73 lines
include/linux/irqchip/arm-gic-v3.h new file +73
-0+ +/* +* CPU interface registers +*/ +#define GICC_CTLR_EL1_EOImode_drop_dir (0U << 1) +#define GICC_CTLR_EL1_EOImode_drop (1U << 1) +#define GICC_SRE_EL1_SRE (1U << 0) + +/* +* Hypervisor interface registers (SRE only) +*/ +#define GICH_LR_VIRTUAL_ID_MASK ((1UL << 32) - 1) + +#define GICH_LR_EOI (1UL << 41) +#define GICH_LR_GROUP (1UL << 60) +#define GICH_LR_STATE (3UL << 62) +#define GICH_LR_PENDING_BIT (1UL << 62) +#define GICH_LR_ACTIVE_BIT (1UL << 63) + +#define GICH_MISR_EOI (1 << 0) +#define GICH_MISR_U (1 << 1) + +#define GICH_HCR_EN (1 << 0) +#define GICH_HCR_UIE (1 << 1) + +#define GICH_VMCR_CTLR_SHIFT 0 +#define GICH_VMCR_CTLR_MASK (0x21f << GICH_VMCR_CTLR_SHIFT) +#define GICH_VMCR_BPR1_SHIFT 18 +#define GICH_VMCR_BPR1_MASK (7 << GICH_VMCR_BPR1_SHIFT) +#define GICH_VMCR_BPR0_SHIFT 21 +#define GICH_VMCR_BPR0_MASK (7 << GICH_VMCR_BPR0_SHIFT) +#define GICH_VMCR_PMR_SHIFT 24 +#define GICH_VMCR_PMR_MASK (0xffUL << GICH_VMCR_PMR_SHIFT) + +#define ICC_EOIR1_EL1 S3_0_C12_C12_1 +#define ICC_IAR1_EL1 S3_0_C12_C12_0 +#define ICC_SGI1R_EL1 S3_0_C12_C11_5 +#define ICC_PMR_EL1 S3_0_C4_C6_0 +#define ICC_CTLR_EL1 S3_0_C12_C12_4 +#define ICC_SRE_EL1 S3_0_C12_C12_5 +#define ICC_GRPEN1_EL1 S3_0_C12_C12_7 + +#define ICC_SRE_EL2 S3_4_C12_C9_5 + +#define ICH_VSEIR_EL2 S3_4_C12_C9_4 +#define ICH_HCR_EL2 S3_4_C12_C11_0 +#define ICH_VTR_EL2 S3_4_C12_C11_1 +#define ICH_MISR_EL2 S3_4_C12_C11_2 +#define ICH_EISR_EL2 S3_4_C12_C11_3 +#define ICH_ELSR_EL2 S3_4_C12_C11_5 +#define ICH_VMCR_EL2 S3_4_C12_C11_7 + +#define __LR0_EL2 (x) S3_4_C12_C12_ ## x +#define __LR8_EL2 (x) S3_4_C12_C13_ ## x + +#define ICH_LR0_EL2 __LR0_EL2(0) +#define ICH_LR1_EL2 __LR0_EL2(1) +#define ICH_LR2_EL2 __LR0_EL2(2) +#define ICH_LR3_EL2 __LR0_EL2(3) +#define ICH_LR4_EL2 __LR0_EL2(4) +#define ICH_LR5_EL2 __LR0_EL2(5) +#define ICH_LR6_EL2 __LR0_EL2(6) +#define ICH_LR7_EL2 __LR0_EL2(7) +#define ICH_LR8_EL2 __LR8_EL2(0) +#define ICH_LR9_EL2 __LR8_EL2(1) +#define ICH_LR10_EL2 __LR8_EL2(2) +#define ICH_LR11_EL2 __LR8_EL2(3) +#define ICH_LR12_EL2 __LR8_EL2(4) +#define ICH_LR13_EL2 __LR8_EL2(5) +#define ICH_LR14_EL2 __LR8_EL2(6) +#define ICH_LR15_EL2 __LR8_EL2(7) + +#define __AP0Rx_EL2 (x) S3_4_C12_C8_ ## x
include/linux/irqchip/arm-gic-v3.h line 174
Quoted 10 lines
include/linux/irqchip/arm-gic-v3.h new file +10
-0+#define ICH_AP0R0_EL2 __AP0Rx_EL2(0) +#define ICH_AP0R1_EL2 __AP0Rx_EL2(1) +#define ICH_AP0R2_EL2 __AP0Rx_EL2(2) +#define ICH_AP0R3_EL2 __AP0Rx_EL2(3) + +#define __AP1Rx_EL2 (x) S3_4_C12_C9_ ## x +#define ICH_AP1R0_EL2 __AP1Rx_EL2(0) +#define ICH_AP1R1_EL2 __AP1Rx_EL2(1) +#define ICH_AP1R2_EL2 __AP1Rx_EL2(2) +#define ICH_AP1R3_EL2 __AP1Rx_EL2(3)
That was fun to review.
Quoted 19 lines
include/linux/irqchip/arm-gic-v3.h new file +16
-0+ +#ifndef __ASSEMBLY__ + +#include <linux/stringify.h> + +#define reg (x) __stringify(x) + +static inline void gic_write_eoir (u64 irq) +{ + asm volatile("msr " reg(ICC_EOIR1_EL1) ", %0" : : "r" (irq)); +isb (); +} + +#endif + +#endif Signature · 1 line
1.8.3.4
PATCH 1/18 arm64: initial support for GICv3
Hi Christoffer,
Quoted 57 lines · Christoffer Dall
Quoted 56 lines · Marc Zyngier
The Generic Interrupt Controller (version 3) offers services that are similar to GICv2, with a number of additional features:
- Affinity routing based on the CPU MPIDR (ARE)
- System register for the CPU interfaces (SRE)
- Support for more that 8 CPUs
- Locality-specific Peripheral Interrupts (LPIs)
- Interrupt Translation Services (ITS)This patch adds preliminary support for GICv3 with ARE and SRE, non-secure mode only. It relies on higher exception levels to grant ARE and SRE access.
Support for LPI and ITS will be added at a later time.
Acked-by Catalin Marinas <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- arch/arm64/Kconfig | 1 + arch/arm64/kernel/head.S | 14 + arch/arm64/kernel/hyp-stub.S | 41 ++- drivers/irqchip/Kconfig | 5 + drivers/irqchip/Makefile | 1 + drivers/irqchip/irq-gic-v3.c | 665 +++++++++++++++++++++++++++++++++++++ include/linux/irqchip/arm-gic-v3.h | 190 +++++++++++ 7 files changed, 915 insertions(+), 2 deletions(-) create mode 100644 drivers/irqchip/irq-gic-v3.c create mode 100644 include/linux/irqchip/arm-gic-v3.harch/arm64/Kconfig +1
-0@@ -9,6 +9,7 @@ config ARM64 select ARM_AMBA select ARM_ARCH_TIMER select ARM_GIC + select ARM_GIC_V3 select BUILDTIME_EXTABLE_SORT select CLONE_BACKWARDS select COMMON_CLKarch/arm64/kernel/head.S +1
-0@@ -22,6 +22,7 @@ #include <linux/linkage.h> #include <linux/init.h> +#include <linux/irqchip/arm-gic-v3.h> #include <asm/assembler.h> #include <asm/ptrace.h>@@ -183,6 +184,19 @@ CPU_LE( bic x0, x0, #(3 << 24) ) // Clear the EE and E0E bits for EL1 msr cnthctl_el2, x0 msr cntvoff_el2, xzr // Clear virtual offset
Quoted 14 lines
arch/arm64/kernel/head.S CPU_LE( bic x0, x0, #(3 << 24) ) // Clear the EE and E0E bits for EL1 · line 197
Quoted 11 lines
arch/arm64/kernel/head.S +11
-0+ /* GICv3 system register access */ + mrs x0, id_aa64pfr0_el1 + ubfx x0, x0, #24, #4 + cmp x0, #1 + b.ne 3f + + mrs x0, ICC_SRE_EL2 // Set ICC_SRE_EL2.SRE==1 + orr x0, x0, #1 + msr ICC_SRE_EL2, x0 + isb // Make sure SRE is now 1 + msr ICH_HCR_EL2, xzr // Reset ICC_HCR_EL2 to defaultsWhy do we not just set ICC_SRE_EL2.Enable here, but instead do the whole exception dance below?
That's a very good question. The more I think of it, the more I believe it to be completely unnecessary.
I'm going to remove the trapping and see what breaks. Or who screams.
Quoted 21 lines
Quoted 21 lines
arch/arm64/kernel/head.S +2
-0+ +3: /* Populate ID registers. */ mrs x0, midr_el1 mrs x1, mpidr_el1arch/arm64/kernel/hyp-stub.S +1
-0@@ -19,6 +19,7 @@ #include <linux/init.h> #include <linux/linkage.h> +#include <linux/irqchip/arm-gic-v3.h> #include <asm/assembler.h> #include <asm/ptrace.h>@@ -51,16 +52,52 @@ ENDPROC(__hyp_stub_vectors) .align 11
Quoted 16 lines
arch/arm64/kernel/hyp-stub.S ENDPROC(__hyp_stub_vectors) · line 65
Quoted 11 lines
arch/arm64/kernel/hyp-stub.S +10
-0+#define sysreg_to_iss(op0, op1, crn, crm, op2) \ + (((op0 & 3) << 20) | ((op2 & 7 ) << 17) | \ + ((op1 & 7 ) << 14) | ((crn & 15) << 10) | \ + ((crm & 15) << 1)) + el1_sync: + // Hack alert: we don't have a stack, and we can't use the one + // from EL1 because our MMU is off. Thankfully, we have + // tpidr_el2, which is under our complete control, and + // tpidr_el0, which can be used as long as we haven't reached + // userspace. Oh hum...Hmmm. I feel like either we should tone down the hack alert claim, because this is after all safe and we just explain what's going on, or we should just allocate a small stack space following
el1_syncand set the SP upon entry to this handler?
Allocating a stack here is something I'd rather avoid (how big?). But yeah, I can probably tone the horror down a notch - I got used to see that code... ;-)
Quoted 2 lines
Hopefully none of the code that will ever call into the sysregs trap ever makes the same assumption that user space is not called yet ;)
Well, I'm going to nuke that code anyway. I'm almost convinced that we don't need it.
Quoted 23 lines
Quoted 23 lines
arch/arm64/kernel/hyp-stub.S +14
-2+ msr tpidr_el2, x1 // Preserve x1 + mrs x1, esr_el2 lsr x1, x1, #26 cmp x1, #0x16 - b.ne 2f // Not an HVC trap + b.ne 3f // Not an HVC trap + + // Handle getting/setting vbar_el2 cbz x0, 1f msr vbar_el2, x0 // Set vbar_el2 b 2f 1: mrs x0, vbar_el2 // Return vbar_el2 -2: eret +2: mrs x1, tpidr_el2 // Restore x1 + eret + +3: cmp x1, #0x18 // MSR/MRS? + b.ne 2b + + // Handle trapped sys reg access + msr tpidr_el0, x0 // Preserve x0 + mrs x1, esr_el2
Quoted 8 lines
arch/arm64/kernel/hyp-stub.S ENDPROC(__hyp_stub_vectors) · line 91
Quoted 5 lines
arch/arm64/kernel/hyp-stub.S +5
-0+ ldr x0, =sysreg_to_iss(3, 7, 15, 15, 7) + and x1, x1, x0 + ldr x0, =sysreg_to_iss(3, 0, 12, 12, 5) + eor x0, x0, x1 // Check for ICC_SRE_EL1 + cbnz x0, 4f // Something bad happenedShouldn't we try to do something here, like go to panic on the exception return here, or not worth it?
Code is now gone... ;-)
Quoted 339 lines
Quoted 339 lines
arch/arm64/kernel/hyp-stub.S +9
-0+ + // Set ICC_SRE_EL1 access enable + mrs x1, ICC_SRE_EL2 + orr x1, x1, #(1 << 3) + msr ICC_SRE_EL2, x1 + +4: mrs x0, tpidr_el0 // Restore x0 + mrs x1, tpidr_el2 // Restore x1 + eret ENDPROC(el1_sync) .macro invalid_vector labeldrivers/irqchip/Kconfig +5
-0@@ -10,6 +10,11 @@ config ARM_GIC config GIC_NON_BANKED bool +config ARM_GIC_V3 + bool + select IRQ_DOMAIN + select MULTI_IRQ_HANDLER + config ARM_NVIC bool select IRQ_DOMAINdrivers/irqchip/Makefile +1
-0@@ -14,6 +14,7 @@ obj-$(CONFIG_ORION_IRQCHIP) += irq-orion.o obj-$(CONFIG_ARCH_SUNXI) += irq-sun4i.o obj-$(CONFIG_ARCH_SPEAR3XX) += spear-shirq.o obj-$(CONFIG_ARM_GIC) += irq-gic.o +obj-$(CONFIG_ARM_GIC_V3) += irq-gic-v3.o obj-$(CONFIG_ARM_NVIC) += irq-nvic.o obj-$(CONFIG_ARM_VIC) += irq-vic.o obj-$(CONFIG_IMGPDC_IRQ) += irq-imgpdc.odrivers/irqchip/irq-gic-v3.c new file +293
-0@@ -0,0 +1,665 @@ +/* +* Copyright (C) 2013 ARM Limited, All Rights Reserved. +* Author: Marc Zyngier <[email protected]> +* +* This program is free software; you can redistribute it and/or modify +* it under the terms of the GNU General Public License version 2 as +* published by the Free Software Foundation. +* +* This program is distributed in the hope that it will be useful, +* but WITHOUT ANY WARRANTY; without even the implied warranty of +* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +* GNU General Public License for more details. +* +* You should have received a copy of the GNU General Public License +* along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +#include <linux/cpu.h> +#include <linux/interrupt.h> +#include <linux/of.h> +#include <linux/of_address.h> +#include <linux/of_irq.h> +#include <linux/percpu.h> +#include <linux/slab.h> + +#include <linux/irqchip/arm-gic-v3.h> + +#include <asm/cputype.h> +#include <asm/exception.h> +#include <asm/smp_plat.h> + +#include "irqchip.h" + +struct gic_chip_data { +void __iomem *dist_base; +void __iomem **redist_base; +void __percpu __iomem **rdist; +struct irq_domain *domain; +u64 redist_stride; +u32 redist_regions; +unsigned int irq_nr; +}; + +static struct gic_chip_data gic_data__read_mostly ; + +#define gic_data_rdist_rd_base () (*__this_cpu_ptr(gic_data.rdist)) +#define gic_data_rdist_sgi_base () (gic_data_rdist_rd_base() + SZ_64K) + +static DEFINE_RAW_SPINLOCK (dist_lock); + +#define DEFAULT_PMR_VALUE 0xf0 + +static inline void __iomem *gic_dist_base (struct irq_data *d) +{ +if (d->hwirq <32 )/* SGI+PPI -> SGI_base for this CPU */ +return gic_data_rdist_sgi_base (); + +if (d->hwirq <=1023 )/* SPI -> dist_base */ +return gic_data.dist_base; + +if (d->hwirq >=8192 ) +BUG ();/* LPI Detected!!! */ + +return NULL ; +} + +static inline unsigned int gic_irq (struct irq_data *d) +{ +return d->hwirq; +} + +static void gic_do_wait_for_rwp (void __iomem *base) +{ +u32 val; + +do { + val =readl_relaxed (base +GICD_CTLR ); +cpu_relax (); + }while (val &GICD_CTLR_RWP ); +} + +/* Wait for completion of a distributor change */ +static void gic_dist_wait_for_rwp (void ) +{ +gic_do_wait_for_rwp (gic_data.dist_base); +} + +/* Wait for completion of a redistributor change */ +static void gic_redist_wait_for_rwp (void ) +{ +gic_do_wait_for_rwp (gic_data_rdist_rd_base ()); +} + +static void gic_wait_for_rwp (int irq) +{ +if (irq <32 ) +gic_redist_wait_for_rwp (); +else +gic_dist_wait_for_rwp (); +} + +/* Low level accessors */ +static u64 gic_read_iar (void ) +{ +u64 irqstat; + + asm volatile("mrs %0, " reg(ICC_IAR1_EL1) : "=r" (irqstat)); +return irqstat; +} + +static void gic_write_pmr (u64 val) +{ + asm volatile("msr " reg(ICC_PMR_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_ctlr (u64 val) +{ + asm volatile("msr " reg(ICC_CTLR_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_grpen1 (u64 val) +{ + asm volatile("msr " reg(ICC_GRPEN1_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_sgi1r (u64 val) +{ + asm volatile("msr " reg(ICC_SGI1R_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_enable_sre (void ) +{ +u64 val; + +/* This access is likely to trap into EL2 */ + asm volatile("mrs %0, " reg(ICC_SRE_EL1) : "=r" (val)); + val |=GICC_SRE_EL1_SRE ; + asm volatile("msr " reg(ICC_SRE_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_enable_redist (void ) +{ +void __iomem *rbase; +u32 val; + + rbase =gic_data_rdist_rd_base (); + +/* Wake up this CPU redistributor */ + val =readl_relaxed (rbase +GICR_WAKER ); + val &= ~GICR_WAKER_ProcessorSleep; +writel_relaxed (val, rbase +GICR_WAKER ); + +do { + val =readl_relaxed (rbase +GICR_WAKER ); +cpu_relax (); + }while (val & GICR_WAKER_ChildrenAsleep); +} + +/* +* Routines to acknowledge, disable and enable interrupts +*/ +static void gic_mask_irq (struct irq_data *d) +{ +u32 mask =1 << (gic_irq (d) %32 ); + +raw_spin_lock (&dist_lock); +writel_relaxed (mask,gic_dist_base (d) +GICD_ICENABLER + (gic_irq (d) /32 ) *4 ); +gic_wait_for_rwp (gic_irq (d)); +raw_spin_unlock (&dist_lock); +} + +static void gic_unmask_irq (struct irq_data *d) +{ +u32 mask =1 << (gic_irq (d) %32 ); + +raw_spin_lock (&dist_lock); +writel_relaxed (mask,gic_dist_base (d) +GICD_ISENABLER + (gic_irq (d) /32 ) *4 ); +gic_wait_for_rwp (gic_irq (d)); +raw_spin_unlock (&dist_lock); +} + +static void gic_eoi_irq (struct irq_data *d) +{ +gic_write_eoir (gic_irq (d)); +} + +static int gic_set_type (struct irq_data *d,unsigned int type) +{ +void __iomem *base =gic_dist_base (d); +unsigned int irq =gic_irq (d); +u32 enablemask =1 << (irq %32 ); +u32 enableoff = (irq /32 ) *4 ; +u32 confmask =0x2 << ((irq %16 ) *2 ); +u32 confoff = (irq /16 ) *4 ; +bool enabled = false; +u32 val; + +/* Interrupt configuration for SGIs can't be changed */ +if (irq <16 ) +return -EINVAL ; + +if (type !=IRQ_TYPE_LEVEL_HIGH && type !=IRQ_TYPE_EDGE_RISING ) +return -EINVAL ; + +raw_spin_lock (&dist_lock); + + val =readl_relaxed (base +GICD_ICFGR + confoff); +if (type ==IRQ_TYPE_LEVEL_HIGH ) + val &= ~confmask; +else if (type ==IRQ_TYPE_EDGE_RISING ) + val |= confmask; + +/* +* As recommended by the spec, disable the interrupt before changing +* the configuration +*/ +if (readl_relaxed (base +GICD_ISENABLER + enableoff) & enablemask) { +writel_relaxed (enablemask, base +GICD_ICENABLER + enableoff); +gic_wait_for_rwp (irq); + enabled = true; + } + +writel_relaxed (val, base +GICD_ICFGR + confoff); + +if (enabled) { +writel_relaxed (enablemask, base +GICD_ISENABLER + enableoff); +gic_wait_for_rwp (irq); + } + +raw_spin_unlock (&dist_lock); + +return 0 ; +} + +static int gic_retrigger (struct irq_data *d) +{ +return -ENXIO ; +} + +static u64 gic_mpidr_to_affinity (u64 mpidr) +{ +/* Make sure we don't broadcast the interrupt */ +return mpidr & ~GICD_IROUTER_SPI_MODE_ANY ; +} + +static asmlinkage void __exception_irq_entry gic_handle_irq(struct pt_regs *regs) +{ +u64 irqnr; + +do { + irqnr =gic_read_iar (); + +if (likely (irqnr >15 && irqnr <1021 )) { + irqnr =irq_find_mapping (gic_data.domain, irqnr); +handle_IRQ (irqnr, regs); +continue ; + } +if (irqnr <16 ) { +gic_write_eoir (irqnr); +#ifdef CONFIG_SMP +handle_IPI (irqnr, regs); +#else +WARN_ONCE (true,"Unexpected SGI received!\n" ); +#endif +continue ; + } + }while (irqnr !=0x3ff ); +} + +static void __init gic_dist_init (void ) +{ +unsigned int i; +u64 affinity; +int gic_irqs = gic_data.irq_nr; +void __iomem *base = gic_data.dist_base; + +/* Disable the distributor */ +writel_relaxed (0 , base +GICD_CTLR ); + +/* +* Set all global interrupts to be level triggered, active low. +*/ +for (i =32 ; i < gic_data.irq_nr; i +=16 ) +writel_relaxed (0 , base +GICD_ICFGR + i /4 ); + +/* +* Set priority on all global interrupts. +*/
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 299
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+for (i =32 ; i < gic_irqs; i +=4 ) +writel_relaxed (0xa0a0a0a0 , base +GICD_IPRIORITYR + i); + +/* +* Disable all interrupts. Leave the PPI and SGIs alone +* as these enables are banked registers. as these are enabled through banked registers?
"The enable bits are banked registers", but I can see where the confision comes from. I'll rewrite this comment.
Quoted 1 line
also, should banked registers be redistributor registers?
Yup.
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c line 307
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c new file +8
-0+*/ +for (i =32 ; i < gic_irqs; i +=32 ) +writel_relaxed (0xffffffff , base +GICD_ICENABLER + i /8 ); + +gic_dist_wait_for_rwp (); + +writel_relaxed (GICD_CTLR_ARE_NS |GICD_CTLR_ENABLE_G1A |GICD_CTLR_ENABLE_G1 , + base +GICD_CTLR );why set
GICD_CTLR_ENABLE_G1?
Believe it or not, some people insist on running in secure mode.
Quoted 13 lines
Quoted 13 lines
drivers/irqchip/irq-gic-v3.c new file +13
-0+ +/* +* Set all global interrupts to the boot CPU only. ARE must be +* enabled. +*/ + affinity =gic_mpidr_to_affinity (read_cpuid_mpidr ()); +for (i =32 ; i < gic_data.irq_nr; i++) +writeq_relaxed (affinity, base +GICD_IROUTER + i *8 ); +} + +static int __init gic_populate_rdist (void ) +{ +u64 mpidr =cpu_logical_map (smp_processor_id ());
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 326
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+u64 typer; +u64 aff; +int i; + + aff = mpidr & ((1 <<24 ) -1 ); + aff |= (mpidr >>8 ) & (0xffUL <<24 );Some defines would be nice for this if it makes sense in any way.
There is at least 3 methods GICv3 uses to express affinity. At some point, I gave up and decided that I needed mental support.
I'll update it with the appropriate macros (MPIDR_AFFINITY_LEVEL and co).
Quoted 49 lines
Quoted 49 lines
drivers/irqchip/irq-gic-v3.c new file +49
-0+ +for (i =0 ; i < gic_data.redist_regions; i++) { +void __iomem *ptr = gic_data.redist_base[i]; +u32 reg; + + reg =readl_relaxed (ptr +GICR_PIDR2 ) &GIC_PIDR2_ARCH_MASK ; +if (reg !=0x30 && reg !=0x40 ) {/* We're in trouble... */ +pr_warn ("No redistributor present @%p\n" , ptr); +break ; + } + +do { + typer =readq_relaxed (ptr +GICR_TYPER ); +if ((typer >>32 ) == aff) { +gic_data_rdist_rd_base () = ptr; +pr_info ("CPU%d: found redistributor %llx @%p\n" , +smp_processor_id (), + (unsigned long long ) mpidr, ptr); +return 0 ; + } + +if (gic_data.redist_stride) { + ptr += gic_data.redist_stride; + }else { + ptr +=SZ_64K *2 ;/* Skip RD_base + SGI_base */ +if (typer &GICR_TYPER_VLPIS ) + ptr +=SZ_64K *2 ;/* Skip VLPI_base + reserved page */ + } + }while (!(typer &GICR_TYPER_LAST )); + } + +/* We couldn't even deal with ourselves... */ +WARN (true,"CPU%d: mpidr %lx has no re-distributor!\n" , +smp_processor_id (), (unsigned long )mpidr); +return -ENODEV ; +} + +static void __init gic_cpu_init (void ) +{ +void __iomem *rbase; +int i; + +/* Register ourselves with the rest of the world */ +if (gic_populate_rdist ()) +return ; + +gic_enable_redist (); + + rbase =gic_data_rdist_sgi_base ();
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 381
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+ +/* +* Set priority on PPI and SGI interrupts +*/ +for (i =0 ; i <32 ; i +=4 ) +writel_relaxed (0xa0a0a0a0 , rbase +GICR_IPRIORITYR0 + i *4 /4 );(i * 4) / 4 == i, last I checked.
Recurring comment. I decided to keep it inline with the GICv2 practice. Read it as i * increment / interrupt-per-word. Maybe not worth it if it causes more problem than anything else...
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c line 385
Quoted 4 lines
drivers/irqchip/irq-gic-v3.c new file +4
-0+ +/* +* Disable all PPI interrupts, ensure all SGI interrupts are +* enabled. Why do we disable all PPIs?
Because interrupts should default to being disabled? They will get enabled when you do a request_irq.
Quoted 8 lines
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c new file +8
-0+*/ +writel_relaxed (0xffff0000 , rbase +GICR_ICENABLER0 ); +writel_relaxed (0x0000ffff , rbase +GICR_ISENABLER0 ); + +gic_redist_wait_for_rwp (); + +/* Enable system registers */ +gic_enable_sre ();
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 398
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+ +/* Set priority mask register */ +gic_write_pmr (DEFAULT_PMR_VALUE ); + +/* EOI drops priority too (mode 0) */ you mean deactives the interrupt too, right? It always performs a priority drop doesn't it?
Yup. Just my usual brainfart.
Quoted 13 lines
Quoted 13 lines
drivers/irqchip/irq-gic-v3.c new file +13
-0+gic_write_ctlr (GICC_CTLR_EL1_EOImode_drop_dir); + +/* ... and let's hit the road... */ +gic_write_grpen1 (1 ); +} + +#ifdef CONFIG_SMP +static int __init gic_secondary_init (struct notifier_block *nfb, +unsigned long action,void *hcpu) +{ +if (action ==CPU_STARTING || action ==CPU_STARTING_FROZEN ) +gic_cpu_init (); +return NOTIFY_OK ;
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 416
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+} + +/* +* Notifier for enabling the GIC CPU interface. Set an arbitrarily high +* priority because the GIC needs to be up before the ARM generic timers. I actually don't understand this bit, but it also sits in the original code, so I assume it's correct.
GIC and timer on the secondary CPUs are intialized using the CPU notifiers. Since the need the GIC to be up and running before the timers can enable their interrupts, we use the relative priorities of the notifiers to order them.
Quoted 20 lines
Quoted 20 lines
drivers/irqchip/irq-gic-v3.c new file +20
-0+*/ +static struct notifier_block __initdata gic_cpu_notifier = { + .notifier_call = gic_secondary_init, + .priority =100 , +}; + +static u16 gic_compute_target_list (int *base_cpu,const struct cpumask *mask, +u64 cluster_id) +{ +int cpu = *base_cpu; +u64 mpidr =cpu_logical_map (cpu); +u16 tlist =0 ; + +while (cpu < nr_cpu_ids) { +/* +* If we ever get a cluster of more than 16 CPUs, just +* scream and skip that CPU. +*/ +if (WARN_ON ((mpidr &0xff ) >=16 )) + goto out;
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 441
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+ + tlist |=1 << (mpidr &0xf ); + + cpu =cpumask_next (cpu, mask); + mpidr =cpu_logical_map (cpu);Are you not accessing memory beyond the
__cpu_logical_maparray here when the cpu is the last one at the beginning of the loop?
Nice catch! I need to test cpu here before carrying on...
Quoted 39 lines
Quoted 39 lines
drivers/irqchip/irq-gic-v3.c new file +39
-0+ +if (cluster_id != (mpidr & ~0xffUL )) { + cpu--; + goto out; + } + } +out: + *base_cpu = cpu; +return tlist; +} + +static void gic_send_sgi (u64 cluster_id,u16 tlist,unsigned int irq) +{ +u64 val; + + val = (cluster_id &0xff00ff0000UL ) <<16 ;/* Aff3 + Aff2 */ + val |= (cluster_id &0xff00 ) <<8 ;/* Aff1 */ + val |= irq <<24 ; + val |= tlist; + +pr_debug ("CPU%d: ICC_SGI1R_EL1 %llx\n" ,smp_processor_id (), val); +gic_write_sgi1r (val); +} + +static void gic_raise_softirq (const struct cpumask *mask,unsigned int irq) +{ +int cpu; + +if (WARN_ON (irq >=16 )) +return ; + +/* +* Ensure that stores to Normal memory are visible to the +* other CPUs before issuing the IPI. +*/ +dsb (); + +for_each_cpu_mask (cpu, *mask) { +u64 cluster_id =cpu_logical_map (cpu) & ~0xffUL ;
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 483
Quoted 3 lines
drivers/irqchip/irq-gic-v3.c new file +3
-0+u16 tlist; + + tlist =gic_compute_target_list (&cpu, mask, cluster_id);I think it would be potentially slightly prettier if this function returned the last cpu in the cluster so the loop iterator mangling was kept local to this function and took a pointer to the tlist it was to fill instead, but, this is a minor detail.
Yeah, maybe. I'll try that and see how it looks. it's quite messy anyway...
Quoted 49 lines
Quoted 49 lines
drivers/irqchip/irq-gic-v3.c new file +49
-0+ gic_send_sgi(cluster_id, tlist, irq); + } +} + +static void gic_smp_init (void ) +{ +set_smp_cross_call (gic_raise_softirq); +register_cpu_notifier (&gic_cpu_notifier); +} + +static int gic_set_affinity (struct irq_data *d,const struct cpumask *mask_val, +bool force) +{ +unsigned int cpu =cpumask_any_and (mask_val, cpu_online_mask); +void __iomem *reg; +u64 val; + +if (gic_irq (d) <32 ) +return -EINVAL ; + + reg =gic_dist_base (d) +GICD_IROUTER + (gic_irq (d) *8 ); + val =gic_mpidr_to_affinity (cpu_logical_map (cpu)); + +writeq_relaxed (val, reg); + +return IRQ_SET_MASK_OK ; +} +#else +#define gic_set_affinity NULL +#define gic_smp_init () do { } while(0) +#endif + +static struct irq_chip gic_chip = { + .name ="GICv3" , + .irq_mask = gic_mask_irq, + .irq_unmask = gic_unmask_irq, + .irq_eoi = gic_eoi_irq, + .irq_set_type = gic_set_type, + .irq_retrigger = gic_retrigger, + .irq_set_affinity = gic_set_affinity, +}; + +static int gic_irq_domain_map (struct irq_domain *d,unsigned int irq, +irq_hw_number_t hw) +{ +if (hw <32 ) { +irq_set_percpu_devid (irq); +irq_set_chip_and_handler (irq, &gic_chip, + handle_percpu_devid_irq);
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 537
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+set_irq_flags (irq,IRQF_VALID |IRQF_NOAUTOEN ); + }else { +irq_set_chip_and_handler (irq, &gic_chip, + handle_fasteoi_irq); +set_irq_flags (irq,IRQF_VALID |IRQF_PROBE );will this ever be compiled for
arch/arm? If not, does it make sense to keep theset_irq_flagscalls?
Not impossible at all, by the look of it. That will require some effort, but if someone feels like it, I don't see the point in actuvely removing that code.
Quoted 44 lines
Quoted 44 lines
drivers/irqchip/irq-gic-v3.c new file +44
-0+ } +irq_set_chip_data (irq, d->host_data); +return 0 ; +} + +static int gic_irq_domain_xlate (struct irq_domain *d, +struct device_node *controller, +const u32 *intspec,unsigned int intsize, +unsigned long *out_hwirq,unsigned int *out_type) +{ +if (d->of_node != controller) +return -EINVAL ; +if (intsize <3 ) +return -EINVAL ; + +switch (intspec[0 ]) { +case 0 :/* SPI */ + *out_hwirq = intspec[1 ] +32 ; +break ; +case 1 :/* PPI */ + *out_hwirq = intspec[1 ] +16 ; +break ; +default : +return -EINVAL ; + } + + *out_type = intspec[2 ] &IRQ_TYPE_SENSE_MASK ; +return 0 ; +} + +static const struct irq_domain_ops gic_irq_domain_ops = { + .map = gic_irq_domain_map, + .xlate = gic_irq_domain_xlate, +}; + +static int __init gic_of_init (struct device_node *node,struct device_node *parent) +{ +void __iomem *dist_base; +void __iomem **redist_base; +u64 redist_stride; +u32 redist_regions; +u32 reg; +int gic_irqs; +int err;
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 587
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+int i; + + dist_base =of_iomap (node,0 ); +if (!dist_base) { +pr_warn ("%s: unable to map gic dist registers\n" , + node->full_name);shouldn't these be
pr_err?
Indeed.
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 592
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+return -ENXIO ; + } + + reg =readl_relaxed (dist_base +GICD_PIDR2 ) &GIC_PIDR2_ARCH_MASK ; +if (reg !=0x30 && reg !=0x40 ) {Is everything we do going to work with a GICv4? Anything else we need to initialize if it's a GICv4?
GICv4 is a strict superset of GICv3, and it should all just work as long as you don't try to use GICv4 specific features. *Should*.
Quoted 22 lines
Quoted 22 lines
drivers/irqchip/irq-gic-v3.c new file +22
-0+pr_warn ("%s: no distributor detected, giving up\n" , + node->full_name); + err = -ENODEV ; + goto out_unmap_dist; + } + +if (of_property_read_u32 (node,"#redistributor-regions" , &redist_regions)) + redist_regions =1 ; + + redist_base =kzalloc (sizeof (*redist_base) * redist_regions,GFP_KERNEL ); +if (!redist_base) { + err = -ENOMEM ; + goto out_unmap_dist; + } + +for (i =0 ; i < redist_regions; i++) { + redist_base[i] =of_iomap (node,1 + i); +if (!redist_base[i]) { +pr_warn ("%s: couldn't map region %d\n" , + node->full_name, i); + err = -ENODEV ; + goto out_unmap_rdist;
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 619
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+ } + } + +if (of_property_read_u64 (node,"redistributor-stride" , &redist_stride)) + redist_stride =0 ;The binding doesn't specify this as a two-cell property, so I assume this won't work if you follow the example in the bindings doc.
Yup. I'll update the DT. Nice catch.
Quoted 138 lines
Quoted 138 lines
drivers/irqchip/irq-gic-v3.c new file +46
-0+ + gic_data.dist_base = dist_base; + gic_data.redist_base = redist_base; + gic_data.redist_regions = redist_regions; + gic_data.redist_stride = redist_stride; + +/* +* Find out how many interrupts are supported. +* The GIC only supports up to 1020 interrupt sources (SGI+PPI+SPI) +*/ + gic_irqs =readl_relaxed (gic_data.dist_base +GICD_TYPER ) &0x1f ; + gic_irqs = (gic_irqs +1 ) *32 ; +if (gic_irqs >1020 ) + gic_irqs =1020 ; + gic_data.irq_nr = gic_irqs; + + gic_data.domain =irq_domain_add_linear (node, gic_irqs -16 , + &gic_irq_domain_ops, &gic_data); + gic_data.rdist =alloc_percpu (typeof (*gic_data.rdist)); + +if (WARN_ON (!gic_data.domain) ||WARN_ON (!gic_data.rdist)) { + err = -ENOMEM ; + goto out_free; + } + +set_handle_irq (gic_handle_irq); + +gic_smp_init (); +gic_dist_init (); +gic_cpu_init (); + +return 0 ; + +out_free: +free_percpu (gic_data.rdist); +out_unmap_rdist: +for (i =0 ; i < redist_regions; i++) +if (redist_base[i]) +iounmap (redist_base[i]); +kfree (redist_base); +out_unmap_dist: +iounmap (dist_base); +return err; +} + +IRQCHIP_DECLARE (gic_v3,"arm,gic-v3" , gic_of_init);include/linux/irqchip/arm-gic-v3.h new file +86
-0@@ -0,0 +1,190 @@ +/* +* Copyright (C) 2013 ARM Limited, All Rights Reserved. +* Author: Marc Zyngier <[email protected]> +* +* +* This program is free software; you can redistribute it and/or modify +* it under the terms of the GNU General Public License version 2 as +* published by the Free Software Foundation. +* +* This program is distributed in the hope that it will be useful, +* but WITHOUT ANY WARRANTY; without even the implied warranty of +* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +* GNU General Public License for more details. +* +* You should have received a copy of the GNU General Public License +* along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ +#ifndef __LINUX_IRQCHIP_ARM_GIC_V3_H +#define __LINUX_IRQCHIP_ARM_GIC_V3_H + +/* +* Distributor registers. We assume we're running non-secure, with ARE +* being set. Secure-only and non-ARE registers are not described. +*/ +#define GICD_CTLR 0x0000 +#define GICD_TYPER 0x0004 +#define GICD_IIDR 0x0008 +#define GICD_STATUSR 0x0010 +#define GICD_SETSPI_NSR 0x0040 +#define GICD_CLRSPI_NSR 0x0048 +#define GICD_SETSPI_SR 0x0050 +#define GICD_CLRSPI_SR 0x0058 +#define GICD_SEIR 0x0068 +#define GICD_ISENABLER 0x0100 +#define GICD_ICENABLER 0x0180 +#define GICD_ISPENDR 0x0200 +#define GICD_ICPENDR 0x0280 +#define GICD_ISACTIVER 0x0300 +#define GICD_ICACTIVER 0x0380 +#define GICD_IPRIORITYR 0x0400 +#define GICD_ICFGR 0x0C00 +#define GICD_IROUTER 0x6000 +#define GICD_PIDR2 0xFFE8 + +#define GICD_CTLR_RWP (1U << 31) +#define GICD_CTLR_ARE_NS (1U << 4) +#define GICD_CTLR_ENABLE_G1A (1U << 1) +#define GICD_CTLR_ENABLE_G1 (1U << 0) + +#define GICD_IROUTER_SPI_MODE_ONE (0U << 31) +#define GICD_IROUTER_SPI_MODE_ANY (1U << 31) + +#define GIC_PIDR2_ARCH_MASK 0xf0 + +/* +* Re-Distributor registers, offsets from RD_base +*/ +#define GICR_CTLR GICD_CTLR +#define GICR_IIDR 0x0004 +#define GICR_TYPER 0x0008 +#define GICR_STATUSR GICD_STATUSR +#define GICR_WAKER 0x0014 +#define GICR_SETLPIR 0x0040 +#define GICR_CLRLPIR 0x0048 +#define GICR_SEIR GICD_SEIR +#define GICR_PROPBASER 0x0070 +#define GICR_PENDBASER 0x0078 +#define GICR_INVLPIR 0x00A0 +#define GICR_INVALLR 0x00B0 +#define GICR_SYNCR 0x00C0 +#define GICR_MOVLPIR 0x0100 +#define GICR_MOVALLR 0x0110 +#define GICR_PIDR2 GICD_PIDR2 + +#define GICR_WAKER_ProcessorSleep (1U << 1) +#define GICR_WAKER_ChildrenAsleep (1U << 2) + +/* +* Re-Distributor registers, offsets from SGI_base +*/ +#define GICR_ISENABLER0 GICD_ISENABLER +#define GICR_ICENABLER0 GICD_ICENABLER +#define GICR_ISPENDR0 GICD_ISPENDR +#define GICR_ICPENDR0 GICD_ICPENDR +#define GICR_ISACTIVER0 GICD_ISACTIVER +#define GICR_ICACTIVER0 GICD_ICACTIVER
Quoted 8 lines
include/linux/irqchip/arm-gic-v3.h line 91
Quoted 5 lines
include/linux/irqchip/arm-gic-v3.h new file +5
-0+#define GICR_IPRIORITYR0 GICD_IPRIORITYR +#define GICR_ICFGR0 GICD_ICFGR + +#define GICR_TYPER_VLPIS (1U << 1) +#define GICR_TYPER_LAST (1U << 4)nit: why are these TYPER definitions down here when the TYPER register is defined above and so the WAKER bit definitions?
I'll group them in the next version.
Quoted 77 lines
Quoted 77 lines
include/linux/irqchip/arm-gic-v3.h new file +77
-0+ +/* +* CPU interface registers +*/ +#define GICC_CTLR_EL1_EOImode_drop_dir (0U << 1) +#define GICC_CTLR_EL1_EOImode_drop (1U << 1) +#define GICC_SRE_EL1_SRE (1U << 0) + +/* +* Hypervisor interface registers (SRE only) +*/ +#define GICH_LR_VIRTUAL_ID_MASK ((1UL << 32) - 1) + +#define GICH_LR_EOI (1UL << 41) +#define GICH_LR_GROUP (1UL << 60) +#define GICH_LR_STATE (3UL << 62) +#define GICH_LR_PENDING_BIT (1UL << 62) +#define GICH_LR_ACTIVE_BIT (1UL << 63) + +#define GICH_MISR_EOI (1 << 0) +#define GICH_MISR_U (1 << 1) + +#define GICH_HCR_EN (1 << 0) +#define GICH_HCR_UIE (1 << 1) + +#define GICH_VMCR_CTLR_SHIFT 0 +#define GICH_VMCR_CTLR_MASK (0x21f << GICH_VMCR_CTLR_SHIFT) +#define GICH_VMCR_BPR1_SHIFT 18 +#define GICH_VMCR_BPR1_MASK (7 << GICH_VMCR_BPR1_SHIFT) +#define GICH_VMCR_BPR0_SHIFT 21 +#define GICH_VMCR_BPR0_MASK (7 << GICH_VMCR_BPR0_SHIFT) +#define GICH_VMCR_PMR_SHIFT 24 +#define GICH_VMCR_PMR_MASK (0xffUL << GICH_VMCR_PMR_SHIFT) + +#define ICC_EOIR1_EL1 S3_0_C12_C12_1 +#define ICC_IAR1_EL1 S3_0_C12_C12_0 +#define ICC_SGI1R_EL1 S3_0_C12_C11_5 +#define ICC_PMR_EL1 S3_0_C4_C6_0 +#define ICC_CTLR_EL1 S3_0_C12_C12_4 +#define ICC_SRE_EL1 S3_0_C12_C12_5 +#define ICC_GRPEN1_EL1 S3_0_C12_C12_7 + +#define ICC_SRE_EL2 S3_4_C12_C9_5 + +#define ICH_VSEIR_EL2 S3_4_C12_C9_4 +#define ICH_HCR_EL2 S3_4_C12_C11_0 +#define ICH_VTR_EL2 S3_4_C12_C11_1 +#define ICH_MISR_EL2 S3_4_C12_C11_2 +#define ICH_EISR_EL2 S3_4_C12_C11_3 +#define ICH_ELSR_EL2 S3_4_C12_C11_5 +#define ICH_VMCR_EL2 S3_4_C12_C11_7 + +#define __LR0_EL2 (x) S3_4_C12_C12_ ## x +#define __LR8_EL2 (x) S3_4_C12_C13_ ## x + +#define ICH_LR0_EL2 __LR0_EL2(0) +#define ICH_LR1_EL2 __LR0_EL2(1) +#define ICH_LR2_EL2 __LR0_EL2(2) +#define ICH_LR3_EL2 __LR0_EL2(3) +#define ICH_LR4_EL2 __LR0_EL2(4) +#define ICH_LR5_EL2 __LR0_EL2(5) +#define ICH_LR6_EL2 __LR0_EL2(6) +#define ICH_LR7_EL2 __LR0_EL2(7) +#define ICH_LR8_EL2 __LR8_EL2(0) +#define ICH_LR9_EL2 __LR8_EL2(1) +#define ICH_LR10_EL2 __LR8_EL2(2) +#define ICH_LR11_EL2 __LR8_EL2(3) +#define ICH_LR12_EL2 __LR8_EL2(4) +#define ICH_LR13_EL2 __LR8_EL2(5) +#define ICH_LR14_EL2 __LR8_EL2(6) +#define ICH_LR15_EL2 __LR8_EL2(7) + +#define __AP0Rx_EL2 (x) S3_4_C12_C8_ ## x +#define ICH_AP0R0_EL2 __AP0Rx_EL2(0) +#define ICH_AP0R1_EL2 __AP0Rx_EL2(1) +#define ICH_AP0R2_EL2 __AP0Rx_EL2(2) +#define ICH_AP0R3_EL2 __AP0Rx_EL2(3)
Quoted 8 lines
include/linux/irqchip/arm-gic-v3.h line 174
Quoted 6 lines
include/linux/irqchip/arm-gic-v3.h new file +6
-0+ +#define __AP1Rx_EL2 (x) S3_4_C12_C9_ ## x +#define ICH_AP1R0_EL2 __AP1Rx_EL2(0) +#define ICH_AP1R1_EL2 __AP1Rx_EL2(1) +#define ICH_AP1R2_EL2 __AP1Rx_EL2(2) +#define ICH_AP1R3_EL2 __AP1Rx_EL2(3)That was fun to review.
If you think so, then you're ready to join the lucky team of madmen reviewing DT patches... ;-)
Thanks anyway!
M.
--
Jazz is not dead. It just smells funny...
PATCH 1/18 arm64: initial support for GICv3
Quoted 528 lines · Marc Zyngier
Hi Christoffer,
Quoted 57 lines · Christoffer Dall
Quoted 56 lines · Marc Zyngier
The Generic Interrupt Controller (version 3) offers services that are similar to GICv2, with a number of additional features:
- Affinity routing based on the CPU MPIDR (ARE)
- System register for the CPU interfaces (SRE)
- Support for more that 8 CPUs
- Locality-specific Peripheral Interrupts (LPIs)
- Interrupt Translation Services (ITS)This patch adds preliminary support for GICv3 with ARE and SRE, non-secure mode only. It relies on higher exception levels to grant ARE and SRE access.
Support for LPI and ITS will be added at a later time.
Acked-by Catalin Marinas <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- arch/arm64/Kconfig | 1 + arch/arm64/kernel/head.S | 14 + arch/arm64/kernel/hyp-stub.S | 41 ++- drivers/irqchip/Kconfig | 5 + drivers/irqchip/Makefile | 1 + drivers/irqchip/irq-gic-v3.c | 665 +++++++++++++++++++++++++++++++++++++ include/linux/irqchip/arm-gic-v3.h | 190 +++++++++++ 7 files changed, 915 insertions(+), 2 deletions(-) create mode 100644 drivers/irqchip/irq-gic-v3.c create mode 100644 include/linux/irqchip/arm-gic-v3.harch/arm64/Kconfig +1
-0@@ -9,6 +9,7 @@ config ARM64 select ARM_AMBA select ARM_ARCH_TIMER select ARM_GIC + select ARM_GIC_V3 select BUILDTIME_EXTABLE_SORT select CLONE_BACKWARDS select COMMON_CLKarch/arm64/kernel/head.S +1
-0@@ -22,6 +22,7 @@ #include <linux/linkage.h> #include <linux/init.h> +#include <linux/irqchip/arm-gic-v3.h> #include <asm/assembler.h> #include <asm/ptrace.h>@@ -183,6 +184,19 @@ CPU_LE( bic x0, x0, #(3 << 24) ) // Clear the EE and E0E bits for EL1 msr cnthctl_el2, x0 msr cntvoff_el2, xzr // Clear virtual offsetQuoted 14 lines
arch/arm64/kernel/head.S CPU_LE( bic x0, x0, #(3 << 24) ) // Clear the EE and E0E bits for EL1 · line 197
Quoted 11 lines
arch/arm64/kernel/head.S +11
-0+ /* GICv3 system register access */ + mrs x0, id_aa64pfr0_el1 + ubfx x0, x0, #24, #4 + cmp x0, #1 + b.ne 3f + + mrs x0, ICC_SRE_EL2 // Set ICC_SRE_EL2.SRE==1 + orr x0, x0, #1 + msr ICC_SRE_EL2, x0 + isb // Make sure SRE is now 1 + msr ICH_HCR_EL2, xzr // Reset ICC_HCR_EL2 to defaultsWhy do we not just set ICC_SRE_EL2.Enable here, but instead do the whole exception dance below?
That's a very good question. The more I think of it, the more I believe it to be completely unnecessary.
I'm going to remove the trapping and see what breaks. Or who screams.
Quoted 21 lines
Quoted 21 lines
arch/arm64/kernel/head.S +2
-0+ +3: /* Populate ID registers. */ mrs x0, midr_el1 mrs x1, mpidr_el1arch/arm64/kernel/hyp-stub.S +1
-0@@ -19,6 +19,7 @@ #include <linux/init.h> #include <linux/linkage.h> +#include <linux/irqchip/arm-gic-v3.h> #include <asm/assembler.h> #include <asm/ptrace.h>@@ -51,16 +52,52 @@ ENDPROC(__hyp_stub_vectors) .align 11Quoted 16 lines
arch/arm64/kernel/hyp-stub.S ENDPROC(__hyp_stub_vectors) · line 65
Quoted 11 lines
arch/arm64/kernel/hyp-stub.S +10
-0+#define sysreg_to_iss(op0, op1, crn, crm, op2) \ + (((op0 & 3) << 20) | ((op2 & 7 ) << 17) | \ + ((op1 & 7 ) << 14) | ((crn & 15) << 10) | \ + ((crm & 15) << 1)) + el1_sync: + // Hack alert: we don't have a stack, and we can't use the one + // from EL1 because our MMU is off. Thankfully, we have + // tpidr_el2, which is under our complete control, and + // tpidr_el0, which can be used as long as we haven't reached + // userspace. Oh hum...Hmmm. I feel like either we should tone down the hack alert claim, because this is after all safe and we just explain what's going on, or we should just allocate a small stack space following
el1_syncand set the SP upon entry to this handler?Allocating a stack here is something I'd rather avoid (how big?). But yeah, I can probably tone the horror down a notch - I got used to see that code... ;-)
Quoted 2 lines
Hopefully none of the code that will ever call into the sysregs trap ever makes the same assumption that user space is not called yet ;)
Well, I'm going to nuke that code anyway. I'm almost convinced that we don't need it.
Quoted 23 lines
Quoted 23 lines
arch/arm64/kernel/hyp-stub.S +14
-2+ msr tpidr_el2, x1 // Preserve x1 + mrs x1, esr_el2 lsr x1, x1, #26 cmp x1, #0x16 - b.ne 2f // Not an HVC trap + b.ne 3f // Not an HVC trap + + // Handle getting/setting vbar_el2 cbz x0, 1f msr vbar_el2, x0 // Set vbar_el2 b 2f 1: mrs x0, vbar_el2 // Return vbar_el2 -2: eret +2: mrs x1, tpidr_el2 // Restore x1 + eret + +3: cmp x1, #0x18 // MSR/MRS? + b.ne 2b + + // Handle trapped sys reg access + msr tpidr_el0, x0 // Preserve x0 + mrs x1, esr_el2Quoted 8 lines
arch/arm64/kernel/hyp-stub.S ENDPROC(__hyp_stub_vectors) · line 91
Quoted 5 lines
arch/arm64/kernel/hyp-stub.S +5
-0+ ldr x0, =sysreg_to_iss(3, 7, 15, 15, 7) + and x1, x1, x0 + ldr x0, =sysreg_to_iss(3, 0, 12, 12, 5) + eor x0, x0, x1 // Check for ICC_SRE_EL1 + cbnz x0, 4f // Something bad happenedShouldn't we try to do something here, like go to panic on the exception return here, or not worth it?
Code is now gone... ;-)
Quoted 339 lines
Quoted 339 lines
arch/arm64/kernel/hyp-stub.S +9
-0+ + // Set ICC_SRE_EL1 access enable + mrs x1, ICC_SRE_EL2 + orr x1, x1, #(1 << 3) + msr ICC_SRE_EL2, x1 + +4: mrs x0, tpidr_el0 // Restore x0 + mrs x1, tpidr_el2 // Restore x1 + eret ENDPROC(el1_sync) .macro invalid_vector labeldrivers/irqchip/Kconfig +5
-0@@ -10,6 +10,11 @@ config ARM_GIC config GIC_NON_BANKED bool +config ARM_GIC_V3 + bool + select IRQ_DOMAIN + select MULTI_IRQ_HANDLER + config ARM_NVIC bool select IRQ_DOMAINdrivers/irqchip/Makefile +1
-0@@ -14,6 +14,7 @@ obj-$(CONFIG_ORION_IRQCHIP) += irq-orion.o obj-$(CONFIG_ARCH_SUNXI) += irq-sun4i.o obj-$(CONFIG_ARCH_SPEAR3XX) += spear-shirq.o obj-$(CONFIG_ARM_GIC) += irq-gic.o +obj-$(CONFIG_ARM_GIC_V3) += irq-gic-v3.o obj-$(CONFIG_ARM_NVIC) += irq-nvic.o obj-$(CONFIG_ARM_VIC) += irq-vic.o obj-$(CONFIG_IMGPDC_IRQ) += irq-imgpdc.odrivers/irqchip/irq-gic-v3.c new file +293
-0@@ -0,0 +1,665 @@ +/* +* Copyright (C) 2013 ARM Limited, All Rights Reserved. +* Author: Marc Zyngier <[email protected]> +* +* This program is free software; you can redistribute it and/or modify +* it under the terms of the GNU General Public License version 2 as +* published by the Free Software Foundation. +* +* This program is distributed in the hope that it will be useful, +* but WITHOUT ANY WARRANTY; without even the implied warranty of +* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +* GNU General Public License for more details. +* +* You should have received a copy of the GNU General Public License +* along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +#include <linux/cpu.h> +#include <linux/interrupt.h> +#include <linux/of.h> +#include <linux/of_address.h> +#include <linux/of_irq.h> +#include <linux/percpu.h> +#include <linux/slab.h> + +#include <linux/irqchip/arm-gic-v3.h> + +#include <asm/cputype.h> +#include <asm/exception.h> +#include <asm/smp_plat.h> + +#include "irqchip.h" + +struct gic_chip_data { +void __iomem *dist_base; +void __iomem **redist_base; +void __percpu __iomem **rdist; +struct irq_domain *domain; +u64 redist_stride; +u32 redist_regions; +unsigned int irq_nr; +}; + +static struct gic_chip_data gic_data__read_mostly ; + +#define gic_data_rdist_rd_base () (*__this_cpu_ptr(gic_data.rdist)) +#define gic_data_rdist_sgi_base () (gic_data_rdist_rd_base() + SZ_64K) + +static DEFINE_RAW_SPINLOCK (dist_lock); + +#define DEFAULT_PMR_VALUE 0xf0 + +static inline void __iomem *gic_dist_base (struct irq_data *d) +{ +if (d->hwirq <32 )/* SGI+PPI -> SGI_base for this CPU */ +return gic_data_rdist_sgi_base (); + +if (d->hwirq <=1023 )/* SPI -> dist_base */ +return gic_data.dist_base; + +if (d->hwirq >=8192 ) +BUG ();/* LPI Detected!!! */ + +return NULL ; +} + +static inline unsigned int gic_irq (struct irq_data *d) +{ +return d->hwirq; +} + +static void gic_do_wait_for_rwp (void __iomem *base) +{ +u32 val; + +do { + val =readl_relaxed (base +GICD_CTLR ); +cpu_relax (); + }while (val &GICD_CTLR_RWP ); +} + +/* Wait for completion of a distributor change */ +static void gic_dist_wait_for_rwp (void ) +{ +gic_do_wait_for_rwp (gic_data.dist_base); +} + +/* Wait for completion of a redistributor change */ +static void gic_redist_wait_for_rwp (void ) +{ +gic_do_wait_for_rwp (gic_data_rdist_rd_base ()); +} + +static void gic_wait_for_rwp (int irq) +{ +if (irq <32 ) +gic_redist_wait_for_rwp (); +else +gic_dist_wait_for_rwp (); +} + +/* Low level accessors */ +static u64 gic_read_iar (void ) +{ +u64 irqstat; + + asm volatile("mrs %0, " reg(ICC_IAR1_EL1) : "=r" (irqstat)); +return irqstat; +} + +static void gic_write_pmr (u64 val) +{ + asm volatile("msr " reg(ICC_PMR_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_ctlr (u64 val) +{ + asm volatile("msr " reg(ICC_CTLR_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_grpen1 (u64 val) +{ + asm volatile("msr " reg(ICC_GRPEN1_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_sgi1r (u64 val) +{ + asm volatile("msr " reg(ICC_SGI1R_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_enable_sre (void ) +{ +u64 val; + +/* This access is likely to trap into EL2 */ + asm volatile("mrs %0, " reg(ICC_SRE_EL1) : "=r" (val)); + val |=GICC_SRE_EL1_SRE ; + asm volatile("msr " reg(ICC_SRE_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_enable_redist (void ) +{ +void __iomem *rbase; +u32 val; + + rbase =gic_data_rdist_rd_base (); + +/* Wake up this CPU redistributor */ + val =readl_relaxed (rbase +GICR_WAKER ); + val &= ~GICR_WAKER_ProcessorSleep; +writel_relaxed (val, rbase +GICR_WAKER ); + +do { + val =readl_relaxed (rbase +GICR_WAKER ); +cpu_relax (); + }while (val & GICR_WAKER_ChildrenAsleep); +} + +/* +* Routines to acknowledge, disable and enable interrupts +*/ +static void gic_mask_irq (struct irq_data *d) +{ +u32 mask =1 << (gic_irq (d) %32 ); + +raw_spin_lock (&dist_lock); +writel_relaxed (mask,gic_dist_base (d) +GICD_ICENABLER + (gic_irq (d) /32 ) *4 ); +gic_wait_for_rwp (gic_irq (d)); +raw_spin_unlock (&dist_lock); +} + +static void gic_unmask_irq (struct irq_data *d) +{ +u32 mask =1 << (gic_irq (d) %32 ); + +raw_spin_lock (&dist_lock); +writel_relaxed (mask,gic_dist_base (d) +GICD_ISENABLER + (gic_irq (d) /32 ) *4 ); +gic_wait_for_rwp (gic_irq (d)); +raw_spin_unlock (&dist_lock); +} + +static void gic_eoi_irq (struct irq_data *d) +{ +gic_write_eoir (gic_irq (d)); +} + +static int gic_set_type (struct irq_data *d,unsigned int type) +{ +void __iomem *base =gic_dist_base (d); +unsigned int irq =gic_irq (d); +u32 enablemask =1 << (irq %32 ); +u32 enableoff = (irq /32 ) *4 ; +u32 confmask =0x2 << ((irq %16 ) *2 ); +u32 confoff = (irq /16 ) *4 ; +bool enabled = false; +u32 val; + +/* Interrupt configuration for SGIs can't be changed */ +if (irq <16 ) +return -EINVAL ; + +if (type !=IRQ_TYPE_LEVEL_HIGH && type !=IRQ_TYPE_EDGE_RISING ) +return -EINVAL ; + +raw_spin_lock (&dist_lock); + + val =readl_relaxed (base +GICD_ICFGR + confoff); +if (type ==IRQ_TYPE_LEVEL_HIGH ) + val &= ~confmask; +else if (type ==IRQ_TYPE_EDGE_RISING ) + val |= confmask; + +/* +* As recommended by the spec, disable the interrupt before changing +* the configuration +*/ +if (readl_relaxed (base +GICD_ISENABLER + enableoff) & enablemask) { +writel_relaxed (enablemask, base +GICD_ICENABLER + enableoff); +gic_wait_for_rwp (irq); + enabled = true; + } + +writel_relaxed (val, base +GICD_ICFGR + confoff); + +if (enabled) { +writel_relaxed (enablemask, base +GICD_ISENABLER + enableoff); +gic_wait_for_rwp (irq); + } + +raw_spin_unlock (&dist_lock); + +return 0 ; +} + +static int gic_retrigger (struct irq_data *d) +{ +return -ENXIO ; +} + +static u64 gic_mpidr_to_affinity (u64 mpidr) +{ +/* Make sure we don't broadcast the interrupt */ +return mpidr & ~GICD_IROUTER_SPI_MODE_ANY ; +} + +static asmlinkage void __exception_irq_entry gic_handle_irq(struct pt_regs *regs) +{ +u64 irqnr; + +do { + irqnr =gic_read_iar (); + +if (likely (irqnr >15 && irqnr <1021 )) { + irqnr =irq_find_mapping (gic_data.domain, irqnr); +handle_IRQ (irqnr, regs); +continue ; + } +if (irqnr <16 ) { +gic_write_eoir (irqnr); +#ifdef CONFIG_SMP +handle_IPI (irqnr, regs); +#else +WARN_ONCE (true,"Unexpected SGI received!\n" ); +#endif +continue ; + } + }while (irqnr !=0x3ff ); +} + +static void __init gic_dist_init (void ) +{ +unsigned int i; +u64 affinity; +int gic_irqs = gic_data.irq_nr; +void __iomem *base = gic_data.dist_base; + +/* Disable the distributor */ +writel_relaxed (0 , base +GICD_CTLR ); + +/* +* Set all global interrupts to be level triggered, active low. +*/ +for (i =32 ; i < gic_data.irq_nr; i +=16 ) +writel_relaxed (0 , base +GICD_ICFGR + i /4 ); + +/* +* Set priority on all global interrupts. +*/ Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 299
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+for (i =32 ; i < gic_irqs; i +=4 ) +writel_relaxed (0xa0a0a0a0 , base +GICD_IPRIORITYR + i); + +/* +* Disable all interrupts. Leave the PPI and SGIs alone +* as these enables are banked registers. as these are enabled through banked registers?
"The enable bits are banked registers", but I can see where the confision comes from. I'll rewrite this comment.
Quoted 1 line
also, should banked registers be redistributor registers?
Yup.
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c line 307
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c new file +8
-0+*/ +for (i =32 ; i < gic_irqs; i +=32 ) +writel_relaxed (0xffffffff , base +GICD_ICENABLER + i /8 ); + +gic_dist_wait_for_rwp (); + +writel_relaxed (GICD_CTLR_ARE_NS |GICD_CTLR_ENABLE_G1A |GICD_CTLR_ENABLE_G1 , + base +GICD_CTLR );why set
GICD_CTLR_ENABLE_G1?
Quoted 2 lines
Believe it or not, some people insist on running in secure mode.
Ah, didn't consider this. Maybe because the cover-letter ssays this patch only supports "Non-Secure Group-1 interrupts only".
Quoted 22 lines
Quoted 13 lines
Quoted 13 lines
drivers/irqchip/irq-gic-v3.c new file +13
-0+ +/* +* Set all global interrupts to the boot CPU only. ARE must be +* enabled. +*/ + affinity =gic_mpidr_to_affinity (read_cpuid_mpidr ()); +for (i =32 ; i < gic_data.irq_nr; i++) +writeq_relaxed (affinity, base +GICD_IROUTER + i *8 ); +} + +static int __init gic_populate_rdist (void ) +{ +u64 mpidr =cpu_logical_map (smp_processor_id ());Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 326
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+u64 typer; +u64 aff; +int i; + + aff = mpidr & ((1 <<24 ) -1 ); + aff |= (mpidr >>8 ) & (0xffUL <<24 );Some defines would be nice for this if it makes sense in any way.
Quoted 5 lines
There is at least 3 methods GICv3 uses to express affinity. At some point, I gave up and decided that I needed mental support.
I'll update it with the appropriate macros (
MPIDR_AFFINITY_LEVELand co).
I don't feel strongly about it, maybe we need giant diagrams on A1 paper instead :)
Quoted 58 lines
Quoted 49 lines
Quoted 49 lines
drivers/irqchip/irq-gic-v3.c new file +49
-0+ +for (i =0 ; i < gic_data.redist_regions; i++) { +void __iomem *ptr = gic_data.redist_base[i]; +u32 reg; + + reg =readl_relaxed (ptr +GICR_PIDR2 ) &GIC_PIDR2_ARCH_MASK ; +if (reg !=0x30 && reg !=0x40 ) {/* We're in trouble... */ +pr_warn ("No redistributor present @%p\n" , ptr); +break ; + } + +do { + typer =readq_relaxed (ptr +GICR_TYPER ); +if ((typer >>32 ) == aff) { +gic_data_rdist_rd_base () = ptr; +pr_info ("CPU%d: found redistributor %llx @%p\n" , +smp_processor_id (), + (unsigned long long ) mpidr, ptr); +return 0 ; + } + +if (gic_data.redist_stride) { + ptr += gic_data.redist_stride; + }else { + ptr +=SZ_64K *2 ;/* Skip RD_base + SGI_base */ +if (typer &GICR_TYPER_VLPIS ) + ptr +=SZ_64K *2 ;/* Skip VLPI_base + reserved page */ + } + }while (!(typer &GICR_TYPER_LAST )); + } + +/* We couldn't even deal with ourselves... */ +WARN (true,"CPU%d: mpidr %lx has no re-distributor!\n" , +smp_processor_id (), (unsigned long )mpidr); +return -ENODEV ; +} + +static void __init gic_cpu_init (void ) +{ +void __iomem *rbase; +int i; + +/* Register ourselves with the rest of the world */ +if (gic_populate_rdist ()) +return ; + +gic_enable_redist (); + + rbase =gic_data_rdist_sgi_base ();Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 381
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+ +/* +* Set priority on PPI and SGI interrupts +*/ +for (i =0 ; i <32 ; i +=4 ) +writel_relaxed (0xa0a0a0a0 , rbase +GICR_IPRIORITYR0 + i *4 /4 );(i * 4) / 4 == i, last I checked.
Quoted 4 lines
Recurring comment. I decided to keep it inline with the GICv2 practice. Read it as i * increment / interrupt-per-word. Maybe not worth it if it causes more problem than anything else...
Yeah, I noticed it in the GICv2 code afterwards, but I don't think you're consistent with this approach throughout the file. I personaly find this notation mind-boggingly-difficult, but that is a personal observation.
Quoted 10 lines
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c line 385
Quoted 4 lines
drivers/irqchip/irq-gic-v3.c new file +4
-0+ +/* +* Disable all PPI interrupts, ensure all SGI interrupts are +* enabled. Why do we disable all PPIs?
Because interrupts should default to being disabled? They will get enabled when you do a
request_irq.
Duh, yeah, and SGI don't need to be disabled because we're in control of generating them I guess. Thanks for the explanation.
Quoted 41 lines
Quoted 8 lines
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c new file +8
-0+*/ +writel_relaxed (0xffff0000 , rbase +GICR_ICENABLER0 ); +writel_relaxed (0x0000ffff , rbase +GICR_ISENABLER0 ); + +gic_redist_wait_for_rwp (); + +/* Enable system registers */ +gic_enable_sre ();Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 398
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+ +/* Set priority mask register */ +gic_write_pmr (DEFAULT_PMR_VALUE ); + +/* EOI drops priority too (mode 0) */ you mean deactives the interrupt too, right? It always performs a priority drop doesn't it?
Yup. Just my usual brainfart.
Quoted 13 lines
Quoted 13 lines
drivers/irqchip/irq-gic-v3.c new file +13
-0+gic_write_ctlr (GICC_CTLR_EL1_EOImode_drop_dir); + +/* ... and let's hit the road... */ +gic_write_grpen1 (1 ); +} + +#ifdef CONFIG_SMP +static int __init gic_secondary_init (struct notifier_block *nfb, +unsigned long action,void *hcpu) +{ +if (action ==CPU_STARTING || action ==CPU_STARTING_FROZEN ) +gic_cpu_init (); +return NOTIFY_OK ;Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 416
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+} + +/* +* Notifier for enabling the GIC CPU interface. Set an arbitrarily high +* priority because the GIC needs to be up before the ARM generic timers. I actually don't understand this bit, but it also sits in the original code, so I assume it's correct.
Quoted 5 lines
GIC and timer on the secondary CPUs are intialized using the CPU notifiers. Since the need the GIC to be up and running before the timers can enable their interrupts, we use the relative priorities of the notifiers to order them.
thanks again for the explanation.
Quoted 29 lines
Quoted 20 lines
Quoted 20 lines
drivers/irqchip/irq-gic-v3.c new file +20
-0+*/ +static struct notifier_block __initdata gic_cpu_notifier = { + .notifier_call = gic_secondary_init, + .priority =100 , +}; + +static u16 gic_compute_target_list (int *base_cpu,const struct cpumask *mask, +u64 cluster_id) +{ +int cpu = *base_cpu; +u64 mpidr =cpu_logical_map (cpu); +u16 tlist =0 ; + +while (cpu < nr_cpu_ids) { +/* +* If we ever get a cluster of more than 16 CPUs, just +* scream and skip that CPU. +*/ +if (WARN_ON ((mpidr &0xff ) >=16 )) + goto out;Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 441
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+ + tlist |=1 << (mpidr &0xf ); + + cpu =cpumask_next (cpu, mask); + mpidr =cpu_logical_map (cpu);Are you not accessing memory beyond the
__cpu_logical_maparray here when the cpu is the last one at the beginning of the loop?
Quoted 2 lines
Nice catch! I need to test cpu here before carrying on...
You could also simply use for_each_cpu_mask here and in the beginning of the loop increment the first cpu to *base_cpu if it makes the code cleaner. Not sure.
Quoted 48 lines
Quoted 39 lines
Quoted 39 lines
drivers/irqchip/irq-gic-v3.c new file +39
-0+ +if (cluster_id != (mpidr & ~0xffUL )) { + cpu--; + goto out; + } + } +out: + *base_cpu = cpu; +return tlist; +} + +static void gic_send_sgi (u64 cluster_id,u16 tlist,unsigned int irq) +{ +u64 val; + + val = (cluster_id &0xff00ff0000UL ) <<16 ;/* Aff3 + Aff2 */ + val |= (cluster_id &0xff00 ) <<8 ;/* Aff1 */ + val |= irq <<24 ; + val |= tlist; + +pr_debug ("CPU%d: ICC_SGI1R_EL1 %llx\n" ,smp_processor_id (), val); +gic_write_sgi1r (val); +} + +static void gic_raise_softirq (const struct cpumask *mask,unsigned int irq) +{ +int cpu; + +if (WARN_ON (irq >=16 )) +return ; + +/* +* Ensure that stores to Normal memory are visible to the +* other CPUs before issuing the IPI. +*/ +dsb (); + +for_each_cpu_mask (cpu, *mask) { +u64 cluster_id =cpu_logical_map (cpu) & ~0xffUL ;Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 483
Quoted 3 lines
drivers/irqchip/irq-gic-v3.c new file +3
-0+u16 tlist; + + tlist =gic_compute_target_list (&cpu, mask, cluster_id);I think it would be potentially slightly prettier if this function returned the last cpu in the cluster so the loop iterator mangling was kept local to this function and took a pointer to the tlist it was to fill instead, but, this is a minor detail.
Quoted 2 lines
Yeah, maybe. I'll try that and see how it looks. it's quite messy anyway...
yeah, nothing I feel strongly about anyway.
Quoted 58 lines
Quoted 49 lines
Quoted 49 lines
drivers/irqchip/irq-gic-v3.c new file +49
-0+ gic_send_sgi(cluster_id, tlist, irq); + } +} + +static void gic_smp_init (void ) +{ +set_smp_cross_call (gic_raise_softirq); +register_cpu_notifier (&gic_cpu_notifier); +} + +static int gic_set_affinity (struct irq_data *d,const struct cpumask *mask_val, +bool force) +{ +unsigned int cpu =cpumask_any_and (mask_val, cpu_online_mask); +void __iomem *reg; +u64 val; + +if (gic_irq (d) <32 ) +return -EINVAL ; + + reg =gic_dist_base (d) +GICD_IROUTER + (gic_irq (d) *8 ); + val =gic_mpidr_to_affinity (cpu_logical_map (cpu)); + +writeq_relaxed (val, reg); + +return IRQ_SET_MASK_OK ; +} +#else +#define gic_set_affinity NULL +#define gic_smp_init () do { } while(0) +#endif + +static struct irq_chip gic_chip = { + .name ="GICv3" , + .irq_mask = gic_mask_irq, + .irq_unmask = gic_unmask_irq, + .irq_eoi = gic_eoi_irq, + .irq_set_type = gic_set_type, + .irq_retrigger = gic_retrigger, + .irq_set_affinity = gic_set_affinity, +}; + +static int gic_irq_domain_map (struct irq_domain *d,unsigned int irq, +irq_hw_number_t hw) +{ +if (hw <32 ) { +irq_set_percpu_devid (irq); +irq_set_chip_and_handler (irq, &gic_chip, + handle_percpu_devid_irq);Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 537
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+set_irq_flags (irq,IRQF_VALID |IRQF_NOAUTOEN ); + }else { +irq_set_chip_and_handler (irq, &gic_chip, + handle_fasteoi_irq); +set_irq_flags (irq,IRQF_VALID |IRQF_PROBE );will this ever be compiled for
arch/arm? If not, does it make sense to keep theset_irq_flagscalls?
Quoted 4 lines
Not impossible at all, by the look of it. That will require some effort, but if someone feels like it, I don't see the point in actuvely removing that code.
ok
Quoted 61 lines
Quoted 44 lines
Quoted 44 lines
drivers/irqchip/irq-gic-v3.c new file +44
-0+ } +irq_set_chip_data (irq, d->host_data); +return 0 ; +} + +static int gic_irq_domain_xlate (struct irq_domain *d, +struct device_node *controller, +const u32 *intspec,unsigned int intsize, +unsigned long *out_hwirq,unsigned int *out_type) +{ +if (d->of_node != controller) +return -EINVAL ; +if (intsize <3 ) +return -EINVAL ; + +switch (intspec[0 ]) { +case 0 :/* SPI */ + *out_hwirq = intspec[1 ] +32 ; +break ; +case 1 :/* PPI */ + *out_hwirq = intspec[1 ] +16 ; +break ; +default : +return -EINVAL ; + } + + *out_type = intspec[2 ] &IRQ_TYPE_SENSE_MASK ; +return 0 ; +} + +static const struct irq_domain_ops gic_irq_domain_ops = { + .map = gic_irq_domain_map, + .xlate = gic_irq_domain_xlate, +}; + +static int __init gic_of_init (struct device_node *node,struct device_node *parent) +{ +void __iomem *dist_base; +void __iomem **redist_base; +u64 redist_stride; +u32 redist_regions; +u32 reg; +int gic_irqs; +int err;Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 587
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+int i; + + dist_base =of_iomap (node,0 ); +if (!dist_base) { +pr_warn ("%s: unable to map gic dist registers\n" , + node->full_name);shouldn't these be
pr_err?Indeed.
Quoted 6 lines
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+return -ENXIO ; + } + + reg =readl_relaxed (dist_base +GICD_PIDR2 ) &GIC_PIDR2_ARCH_MASK ; +if (reg !=0x30 && reg !=0x40 ) {
Quoted 6 lines
Quoted 2 lines
Is everything we do going to work with a GICv4? Anything else we need to initialize if it's a GICv4?
GICv4 is a strict superset of GICv3, and it should all just work as long as you don't try to use GICv4 specific features. *Should*.
cool
Quoted 268 lines
Quoted 22 lines
Quoted 22 lines
drivers/irqchip/irq-gic-v3.c new file +22
-0+pr_warn ("%s: no distributor detected, giving up\n" , + node->full_name); + err = -ENODEV ; + goto out_unmap_dist; + } + +if (of_property_read_u32 (node,"#redistributor-regions" , &redist_regions)) + redist_regions =1 ; + + redist_base =kzalloc (sizeof (*redist_base) * redist_regions,GFP_KERNEL ); +if (!redist_base) { + err = -ENOMEM ; + goto out_unmap_dist; + } + +for (i =0 ; i < redist_regions; i++) { + redist_base[i] =of_iomap (node,1 + i); +if (!redist_base[i]) { +pr_warn ("%s: couldn't map region %d\n" , + node->full_name, i); + err = -ENODEV ; + goto out_unmap_rdist;Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 619
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+ } + } + +if (of_property_read_u64 (node,"redistributor-stride" , &redist_stride)) + redist_stride =0 ;The binding doesn't specify this as a two-cell property, so I assume this won't work if you follow the example in the bindings doc.
Yup. I'll update the DT. Nice catch.
Quoted 138 lines
Quoted 138 lines
drivers/irqchip/irq-gic-v3.c new file +46
-0+ + gic_data.dist_base = dist_base; + gic_data.redist_base = redist_base; + gic_data.redist_regions = redist_regions; + gic_data.redist_stride = redist_stride; + +/* +* Find out how many interrupts are supported. +* The GIC only supports up to 1020 interrupt sources (SGI+PPI+SPI) +*/ + gic_irqs =readl_relaxed (gic_data.dist_base +GICD_TYPER ) &0x1f ; + gic_irqs = (gic_irqs +1 ) *32 ; +if (gic_irqs >1020 ) + gic_irqs =1020 ; + gic_data.irq_nr = gic_irqs; + + gic_data.domain =irq_domain_add_linear (node, gic_irqs -16 , + &gic_irq_domain_ops, &gic_data); + gic_data.rdist =alloc_percpu (typeof (*gic_data.rdist)); + +if (WARN_ON (!gic_data.domain) ||WARN_ON (!gic_data.rdist)) { + err = -ENOMEM ; + goto out_free; + } + +set_handle_irq (gic_handle_irq); + +gic_smp_init (); +gic_dist_init (); +gic_cpu_init (); + +return 0 ; + +out_free: +free_percpu (gic_data.rdist); +out_unmap_rdist: +for (i =0 ; i < redist_regions; i++) +if (redist_base[i]) +iounmap (redist_base[i]); +kfree (redist_base); +out_unmap_dist: +iounmap (dist_base); +return err; +} + +IRQCHIP_DECLARE (gic_v3,"arm,gic-v3" , gic_of_init);include/linux/irqchip/arm-gic-v3.h new file +86
-0@@ -0,0 +1,190 @@ +/* +* Copyright (C) 2013 ARM Limited, All Rights Reserved. +* Author: Marc Zyngier <[email protected]> +* +* +* This program is free software; you can redistribute it and/or modify +* it under the terms of the GNU General Public License version 2 as +* published by the Free Software Foundation. +* +* This program is distributed in the hope that it will be useful, +* but WITHOUT ANY WARRANTY; without even the implied warranty of +* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +* GNU General Public License for more details. +* +* You should have received a copy of the GNU General Public License +* along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ +#ifndef __LINUX_IRQCHIP_ARM_GIC_V3_H +#define __LINUX_IRQCHIP_ARM_GIC_V3_H + +/* +* Distributor registers. We assume we're running non-secure, with ARE +* being set. Secure-only and non-ARE registers are not described. +*/ +#define GICD_CTLR 0x0000 +#define GICD_TYPER 0x0004 +#define GICD_IIDR 0x0008 +#define GICD_STATUSR 0x0010 +#define GICD_SETSPI_NSR 0x0040 +#define GICD_CLRSPI_NSR 0x0048 +#define GICD_SETSPI_SR 0x0050 +#define GICD_CLRSPI_SR 0x0058 +#define GICD_SEIR 0x0068 +#define GICD_ISENABLER 0x0100 +#define GICD_ICENABLER 0x0180 +#define GICD_ISPENDR 0x0200 +#define GICD_ICPENDR 0x0280 +#define GICD_ISACTIVER 0x0300 +#define GICD_ICACTIVER 0x0380 +#define GICD_IPRIORITYR 0x0400 +#define GICD_ICFGR 0x0C00 +#define GICD_IROUTER 0x6000 +#define GICD_PIDR2 0xFFE8 + +#define GICD_CTLR_RWP (1U << 31) +#define GICD_CTLR_ARE_NS (1U << 4) +#define GICD_CTLR_ENABLE_G1A (1U << 1) +#define GICD_CTLR_ENABLE_G1 (1U << 0) + +#define GICD_IROUTER_SPI_MODE_ONE (0U << 31) +#define GICD_IROUTER_SPI_MODE_ANY (1U << 31) + +#define GIC_PIDR2_ARCH_MASK 0xf0 + +/* +* Re-Distributor registers, offsets from RD_base +*/ +#define GICR_CTLR GICD_CTLR +#define GICR_IIDR 0x0004 +#define GICR_TYPER 0x0008 +#define GICR_STATUSR GICD_STATUSR +#define GICR_WAKER 0x0014 +#define GICR_SETLPIR 0x0040 +#define GICR_CLRLPIR 0x0048 +#define GICR_SEIR GICD_SEIR +#define GICR_PROPBASER 0x0070 +#define GICR_PENDBASER 0x0078 +#define GICR_INVLPIR 0x00A0 +#define GICR_INVALLR 0x00B0 +#define GICR_SYNCR 0x00C0 +#define GICR_MOVLPIR 0x0100 +#define GICR_MOVALLR 0x0110 +#define GICR_PIDR2 GICD_PIDR2 + +#define GICR_WAKER_ProcessorSleep (1U << 1) +#define GICR_WAKER_ChildrenAsleep (1U << 2) + +/* +* Re-Distributor registers, offsets from SGI_base +*/ +#define GICR_ISENABLER0 GICD_ISENABLER +#define GICR_ICENABLER0 GICD_ICENABLER +#define GICR_ISPENDR0 GICD_ISPENDR +#define GICR_ICPENDR0 GICD_ICPENDR +#define GICR_ISACTIVER0 GICD_ISACTIVER +#define GICR_ICACTIVER0 GICD_ICACTIVERQuoted 8 lines
include/linux/irqchip/arm-gic-v3.h line 91
Quoted 5 lines
include/linux/irqchip/arm-gic-v3.h new file +5
-0+#define GICR_IPRIORITYR0 GICD_IPRIORITYR +#define GICR_ICFGR0 GICD_ICFGR + +#define GICR_TYPER_VLPIS (1U << 1) +#define GICR_TYPER_LAST (1U << 4)nit: why are these TYPER definitions down here when the TYPER register is defined above and so the WAKER bit definitions?
I'll group them in the next version.
Quoted 77 lines
Quoted 77 lines
include/linux/irqchip/arm-gic-v3.h new file +77
-0+ +/* +* CPU interface registers +*/ +#define GICC_CTLR_EL1_EOImode_drop_dir (0U << 1) +#define GICC_CTLR_EL1_EOImode_drop (1U << 1) +#define GICC_SRE_EL1_SRE (1U << 0) + +/* +* Hypervisor interface registers (SRE only) +*/ +#define GICH_LR_VIRTUAL_ID_MASK ((1UL << 32) - 1) + +#define GICH_LR_EOI (1UL << 41) +#define GICH_LR_GROUP (1UL << 60) +#define GICH_LR_STATE (3UL << 62) +#define GICH_LR_PENDING_BIT (1UL << 62) +#define GICH_LR_ACTIVE_BIT (1UL << 63) + +#define GICH_MISR_EOI (1 << 0) +#define GICH_MISR_U (1 << 1) + +#define GICH_HCR_EN (1 << 0) +#define GICH_HCR_UIE (1 << 1) + +#define GICH_VMCR_CTLR_SHIFT 0 +#define GICH_VMCR_CTLR_MASK (0x21f << GICH_VMCR_CTLR_SHIFT) +#define GICH_VMCR_BPR1_SHIFT 18 +#define GICH_VMCR_BPR1_MASK (7 << GICH_VMCR_BPR1_SHIFT) +#define GICH_VMCR_BPR0_SHIFT 21 +#define GICH_VMCR_BPR0_MASK (7 << GICH_VMCR_BPR0_SHIFT) +#define GICH_VMCR_PMR_SHIFT 24 +#define GICH_VMCR_PMR_MASK (0xffUL << GICH_VMCR_PMR_SHIFT) + +#define ICC_EOIR1_EL1 S3_0_C12_C12_1 +#define ICC_IAR1_EL1 S3_0_C12_C12_0 +#define ICC_SGI1R_EL1 S3_0_C12_C11_5 +#define ICC_PMR_EL1 S3_0_C4_C6_0 +#define ICC_CTLR_EL1 S3_0_C12_C12_4 +#define ICC_SRE_EL1 S3_0_C12_C12_5 +#define ICC_GRPEN1_EL1 S3_0_C12_C12_7 + +#define ICC_SRE_EL2 S3_4_C12_C9_5 + +#define ICH_VSEIR_EL2 S3_4_C12_C9_4 +#define ICH_HCR_EL2 S3_4_C12_C11_0 +#define ICH_VTR_EL2 S3_4_C12_C11_1 +#define ICH_MISR_EL2 S3_4_C12_C11_2 +#define ICH_EISR_EL2 S3_4_C12_C11_3 +#define ICH_ELSR_EL2 S3_4_C12_C11_5 +#define ICH_VMCR_EL2 S3_4_C12_C11_7 + +#define __LR0_EL2 (x) S3_4_C12_C12_ ## x +#define __LR8_EL2 (x) S3_4_C12_C13_ ## x + +#define ICH_LR0_EL2 __LR0_EL2(0) +#define ICH_LR1_EL2 __LR0_EL2(1) +#define ICH_LR2_EL2 __LR0_EL2(2) +#define ICH_LR3_EL2 __LR0_EL2(3) +#define ICH_LR4_EL2 __LR0_EL2(4) +#define ICH_LR5_EL2 __LR0_EL2(5) +#define ICH_LR6_EL2 __LR0_EL2(6) +#define ICH_LR7_EL2 __LR0_EL2(7) +#define ICH_LR8_EL2 __LR8_EL2(0) +#define ICH_LR9_EL2 __LR8_EL2(1) +#define ICH_LR10_EL2 __LR8_EL2(2) +#define ICH_LR11_EL2 __LR8_EL2(3) +#define ICH_LR12_EL2 __LR8_EL2(4) +#define ICH_LR13_EL2 __LR8_EL2(5) +#define ICH_LR14_EL2 __LR8_EL2(6) +#define ICH_LR15_EL2 __LR8_EL2(7) + +#define __AP0Rx_EL2 (x) S3_4_C12_C8_ ## x +#define ICH_AP0R0_EL2 __AP0Rx_EL2(0) +#define ICH_AP0R1_EL2 __AP0Rx_EL2(1) +#define ICH_AP0R2_EL2 __AP0Rx_EL2(2) +#define ICH_AP0R3_EL2 __AP0Rx_EL2(3)Quoted 8 lines
include/linux/irqchip/arm-gic-v3.h line 174
Quoted 6 lines
include/linux/irqchip/arm-gic-v3.h new file +6
-0+ +#define __AP1Rx_EL2 (x) S3_4_C12_C9_ ## x +#define ICH_AP1R0_EL2 __AP1Rx_EL2(0) +#define ICH_AP1R1_EL2 __AP1Rx_EL2(1) +#define ICH_AP1R2_EL2 __AP1Rx_EL2(2) +#define ICH_AP1R3_EL2 __AP1Rx_EL2(3)That was fun to review.
Quoted 3 lines
If you think so, then you're ready to join the lucky team of madmen reviewing DT patches... ;-)
I could not have been more sarcastic in that last comment. It required a full liter of Haagen-Dazs to get through.
-Christoffer
PATCH 1/18 arm64: initial support for GICv3
In reply to Marc ZyngierHi Marc,
Quoted 14 lines · Marc Zyngier
arch/arm64/kernel/head.S +1
-0@@ -22,6 +22,7 @@ #include <linux/linkage.h> #include <linux/init.h> +#include <linux/irqchip/arm-gic-v3.h> #include <asm/assembler.h> #include <asm/ptrace.h>@@ -183,6 +184,19 @@ CPU_LE( bic x0, x0, #(3 << 24) ) // Clear the EE and E0E bits for EL1 msr cnthctl_el2, x0
arch/arm64/kernel/head.S CPU_LE( bic x0, x0, #(3 << 24) ) // Clear the EE and E0E bits for EL1 · line 194
Quoted 10 lines
arch/arm64/kernel/head.S +8
-0msr cntvoff_el2, xzr // Clear virtual offset + /* GICv3 system register access */ + mrs x0, id_aa64pfr0_el1 + ubfx x0, x0, #24, #4 + cmp x0, #1 + b.ne 3f + + mrs x0, ICC_SRE_EL2 // Set ICC_SRE_EL2.SRE==1 + orr x0, x0, #1
Don't you need to set the Enable bit (bit 3) too? Also, does setting ICC_SRE_EL2.SRE force KVM to use the ICH_* registers? If so, that seems like a pretty big decision to take this early on (there's not even an #ifdef CONFIG_ARM_GICV3).
Quoted 25 lines
arch/arm64/kernel/head.S +5
-0+ msr ICC_SRE_EL2, x0 + isb // Make sure SRE is now 1 + msr ICH_HCR_EL2, xzr // Reset ICC_HCR_EL2 to defaults + +3: /* Populate ID registers. */ mrs x0, midr_el1 mrs x1, mpidr_el1arch/arm64/kernel/hyp-stub.S +2
-0@@ -19,6 +19,7 @@ #include <linux/init.h> #include <linux/linkage.h> +#include <linux/irqchip/arm-gic-v3.h> #include <asm/assembler.h> #include <asm/ptrace.h>@@ -51,16 +52,52 @@ ENDPROC(__hyp_stub_vectors) .align 11 +#define sysreg_to_iss(op0, op1, crn, crm, op2) \
arch/arm64/kernel/hyp-stub.S ENDPROC(__hyp_stub_vectors) · line 65
Quoted 10 lines
arch/arm64/kernel/hyp-stub.S +9
-0+ (((op0 & 3) << 20) | ((op2 & 7 ) << 17) | \ + ((op1 & 7 ) << 14) | ((crn & 15) << 10) | \ + ((crm & 15) << 1)) + el1_sync: + // Hack alert: we don't have a stack, and we can't use the one + // from EL1 because our MMU is off. Thankfully, we have + // tpidr_el2, which is under our complete control, and + // tpidr_el0, which can be used as long as we haven't reached + // userspace. Oh hum...
Yuck. Why can't we create a minimal stack when we initialise EL2 when booting the host? Even simpler, just use some .quads in here (you're already using a literal pool).
Quoted 16 lines
arch/arm64/kernel/hyp-stub.S +7
-2+ msr tpidr_el2, x1 // Preserve x1 + mrs x1, esr_el2 lsr x1, x1, #26 cmp x1, #0x16 - b.ne 2f // Not an HVC trap + b.ne 3f // Not an HVC trap + + // Handle getting/setting vbar_el2 cbz x0, 1f msr vbar_el2, x0 // Set vbar_el2 b 2f 1: mrs x0, vbar_el2 // Return vbar_el2 -2: eret +2: mrs x1, tpidr_el2 // Restore x1 + eret
arch/arm64/kernel/hyp-stub.S ENDPROC(__hyp_stub_vectors) · line 89
Quoted 10 lines
arch/arm64/kernel/hyp-stub.S +10
-0+ +3: cmp x1, #0x18 // MSR/MRS? + b.ne 2b + + // Handle trapped sys reg access + msr tpidr_el0, x0 // Preserve x0 + mrs x1, esr_el2 + ldr x0, =sysreg_to_iss(3, 7, 15, 15, 7) + and x1, x1, x0 + ldr x0, =sysreg_to_iss(3, 0, 12, 12, 5)
I think sysreg_to_iss actually makes this *less* readable! Why not simply #define the mask to extract OPx, CRx then compare with another #define for ICC_SRE_EL1? If we can represent those as immediates, you can avoid the literal loads too.
arch/arm64/kernel/hyp-stub.S ENDPROC(__hyp_stub_vectors) · line 96
Quoted 7 lines
arch/arm64/kernel/hyp-stub.S +7
-0+ eor x0, x0, x1 // Check for ICC_SRE_EL1 + cbnz x0, 4f // Something bad happened + + // Set ICC_SRE_EL1 access enable + mrs x1, ICC_SRE_EL2 + orr x1, x1, #(1 << 3) + msr ICC_SRE_EL2, x1
Ah, ok, you do this part lazily. Does that make sense for the host? (i.e. is it possible for KVM to use SRE at EL2 but not the host at EL1?).
Quoted 49 lines
arch/arm64/kernel/hyp-stub.S +4
-0+ +4: mrs x0, tpidr_el0 // Restore x0 + mrs x1, tpidr_el2 // Restore x1 + eret ENDPROC(el1_sync) .macro invalid_vector labeldrivers/irqchip/irq-gic-v3.c new file +36
-0@@ -0,0 +1,665 @@ +/* +* Copyright (C) 2013 ARM Limited, All Rights Reserved. +* Author: Marc Zyngier <[email protected]> +* +* This program is free software; you can redistribute it and/or modify +* it under the terms of the GNU General Public License version 2 as +* published by the Free Software Foundation. +* +* This program is distributed in the hope that it will be useful, +* but WITHOUT ANY WARRANTY; without even the implied warranty of +* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +* GNU General Public License for more details. +* +* You should have received a copy of the GNU General Public License +* along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +#include <linux/cpu.h> +#include <linux/interrupt.h> +#include <linux/of.h> +#include <linux/of_address.h> +#include <linux/of_irq.h> +#include <linux/percpu.h> +#include <linux/slab.h> + +#include <linux/irqchip/arm-gic-v3.h> + +#include <asm/cputype.h> +#include <asm/exception.h> +#include <asm/smp_plat.h> + +#include "irqchip.h" + +struct gic_chip_data { +void __iomem *dist_base; +void __iomem **redist_base;
drivers/irqchip/irq-gic-v3.c line 46
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+void __percpu __iomem **rdist; +struct irq_domain *domain; +u64 redist_stride; +u32 redist_regions; +unsigned int irq_nr; +}; + +static struct gic_chip_data gic_data__read_mostly ; + +#define gic_data_rdist_rd_base () (*__this_cpu_ptr(gic_data.rdist))
You should avoid __this_cpu_ptr (it's being removed afaik). Take a look at the patches from Christoph Lameter, e.g. 170d800af83f ("block: Replace __get_cpu_var uses").
Quoted 23 lines
drivers/irqchip/irq-gic-v3.c new file +23
-0+#define gic_data_rdist_sgi_base () (gic_data_rdist_rd_base() + SZ_64K) + +static DEFINE_RAW_SPINLOCK (dist_lock); + +#define DEFAULT_PMR_VALUE 0xf0 + +static inline void __iomem *gic_dist_base (struct irq_data *d) +{ +if (d->hwirq <32 )/* SGI+PPI -> SGI_base for this CPU */ +return gic_data_rdist_sgi_base (); + +if (d->hwirq <=1023 )/* SPI -> dist_base */ +return gic_data.dist_base; + +if (d->hwirq >=8192 ) +BUG ();/* LPI Detected!!! */ + +return NULL ; +} + +static inline unsigned int gic_irq (struct irq_data *d) +{ +return d->hwirq;
drivers/irqchip/irq-gic-v3.c line 79
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+} + +static void gic_do_wait_for_rwp (void __iomem *base) +{ +u32 val; + +do { + val =readl_relaxed (base +GICD_CTLR ); +cpu_relax (); + }while (val &GICD_CTLR_RWP );
You should add a timeout in here to save us in the face of broken hardware (especially as you hold the distributor lock). See, for example, arm_smmu_tlb_sync(...) in the arm-smmu driver.
Thinking about it, is this the only point of the dist_lock (to ensure you wait only for your request)? Do we actually need locking for that?
Quoted 7 lines
drivers/irqchip/irq-gic-v3.c new file +7
-0+} + +/* Wait for completion of a distributor change */ +static void gic_dist_wait_for_rwp (void ) +{ +gic_do_wait_for_rwp (gic_data.dist_base); +}
drivers/irqchip/irq-gic-v3.c line 96
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+ +/* Wait for completion of a redistributor change */ +static void gic_redist_wait_for_rwp (void ) +{ +gic_do_wait_for_rwp (gic_data_rdist_rd_base ()); +} + +static void gic_wait_for_rwp (int irq) +{ +if (irq <32 )
Maybe we should just bite the bullet and have helpers for gic_irq_is_sgi(irq) etc. so we can avoid these magic numbers all over the place.
drivers/irqchip/irq-gic-v3.c line 100
Quoted 4 lines
drivers/irqchip/irq-gic-v3.c new file +4
-0+gic_redist_wait_for_rwp (); + else +gic_dist_wait_for_rwp (); +}
Saying that, I question the value of mixing this into a single function. gic_redist_wait_for_rwp absolutely *must* be called with preemption disabled and on the appropriate CPU for the IRQ in question, whilst the latter can be called anywhere. So even though you can call this single function from a bunch of places, you're really relying on the irq parameter being correct for the context of the caller.
I reckon it would actually be clearer to call the relevant wait function explicitly in the caller. Same for gic_dist_base.
drivers/irqchip/irq-gic-v3.c line 107
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+/* Low level accessors */ +static u64 gic_read_iar (void ) +{ +u64 irqstat; + + asm volatile("mrs %0, " reg(ICC_IAR1_EL1) : "=r" (irqstat));
Hmm, reg() is a terrible name for a macro :) Can't you just use stringify directly?
Quoted 16 lines
drivers/irqchip/irq-gic-v3.c new file +16
-0+return irqstat; +} + +static void gic_write_pmr (u64 val) +{ + asm volatile("msr " reg(ICC_PMR_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_ctlr (u64 val) +{ + asm volatile("msr " reg(ICC_CTLR_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_grpen1 (u64 val)
drivers/irqchip/irq-gic-v3.c line 133
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+{ + asm volatile("msr " reg(ICC_GRPEN1_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_sgi1r (u64 val) +{ + asm volatile("msr " reg(ICC_SGI1R_EL1) ", %0" : : "r" (val)); +isb (); +}
I'm not convinced you *always* need these isbs. For example, the gic_handle_irq loop probably just needs one at the end, before returning.
In fact, looking at the spec, ICC_PMR_EL1 is actually self-synchronising so you don't need anything there. For SGI1R at least, you need a dsb to ensure write observability by the GIC (it's not clear to me whether you also need to the isb).
Quoted 17 lines
drivers/irqchip/irq-gic-v3.c new file +17
-0+static void gic_enable_sre (void ) +{ +u64 val; + +/* This access is likely to trap into EL2 */ + asm volatile("mrs %0, " reg(ICC_SRE_EL1) : "=r" (val)); + val |=GICC_SRE_EL1_SRE ; + asm volatile("msr " reg(ICC_SRE_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_enable_redist (void ) +{ +void __iomem *rbase; +u32 val; + + rbase =gic_data_rdist_rd_base ();
drivers/irqchip/irq-gic-v3.c line 161
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+ +/* Wake up this CPU redistributor */ + val =readl_relaxed (rbase +GICR_WAKER ); + val &= ~GICR_WAKER_ProcessorSleep; +writel_relaxed (val, rbase +GICR_WAKER ); + +do { + val =readl_relaxed (rbase +GICR_WAKER ); +cpu_relax (); + }while (val & GICR_WAKER_ChildrenAsleep);
Again, you want a timeout here.
Quoted 19 lines
drivers/irqchip/irq-gic-v3.c new file +19
-0+} + +/* +* Routines to acknowledge, disable and enable interrupts +*/ +static void gic_mask_irq (struct irq_data *d) +{ +u32 mask =1 << (gic_irq (d) %32 ); + +raw_spin_lock (&dist_lock); +writel_relaxed (mask,gic_dist_base (d) +GICD_ICENABLER + (gic_irq (d) /32 ) *4 ); +gic_wait_for_rwp (gic_irq (d)); +raw_spin_unlock (&dist_lock); +} + +static void gic_unmask_irq (struct irq_data *d) +{ +u32 mask =1 << (gic_irq (d) %32 ); +
drivers/irqchip/irq-gic-v3.c line 190
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+raw_spin_lock (&dist_lock); +writel_relaxed (mask,gic_dist_base (d) +GICD_ISENABLER + (gic_irq (d) /32 ) *4 ); +gic_wait_for_rwp (gic_irq (d)); +raw_spin_unlock (&dist_lock); +} + +static void gic_eoi_irq (struct irq_data *d) +{ +gic_write_eoir (gic_irq (d)); +}
Can you inline the code here (this is the only user) and move the isb out?
Quoted 37 lines
drivers/irqchip/irq-gic-v3.c new file +37
-0+static int gic_set_type (struct irq_data *d,unsigned int type) +{ +void __iomem *base =gic_dist_base (d); +unsigned int irq =gic_irq (d); +u32 enablemask =1 << (irq %32 ); +u32 enableoff = (irq /32 ) *4 ; +u32 confmask =0x2 << ((irq %16 ) *2 ); +u32 confoff = (irq /16 ) *4 ; +bool enabled = false; +u32 val; + +/* Interrupt configuration for SGIs can't be changed */ +if (irq <16 ) +return -EINVAL ; + +if (type !=IRQ_TYPE_LEVEL_HIGH && type !=IRQ_TYPE_EDGE_RISING ) +return -EINVAL ; + +raw_spin_lock (&dist_lock); + + val =readl_relaxed (base +GICD_ICFGR + confoff); +if (type ==IRQ_TYPE_LEVEL_HIGH ) + val &= ~confmask; +else if (type ==IRQ_TYPE_EDGE_RISING ) + val |= confmask; + +/* +* As recommended by the spec, disable the interrupt before changing +* the configuration +*/ +if (readl_relaxed (base +GICD_ISENABLER + enableoff) & enablemask) { +writel_relaxed (enablemask, base +GICD_ICENABLER + enableoff); +gic_wait_for_rwp (irq); + enabled = true; + } + +writel_relaxed (val, base +GICD_ICFGR + confoff);
drivers/irqchip/irq-gic-v3.c line 238
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+ +if (enabled) { +writel_relaxed (enablemask, base +GICD_ISENABLER + enableoff); +gic_wait_for_rwp (irq); + } + +raw_spin_unlock (&dist_lock); + +return 0 ; +}
I really think this should be merged with the existing GIC driver -- the code is basically identical.
drivers/irqchip/irq-gic-v3.c line 243
Quoted 4 lines
drivers/irqchip/irq-gic-v3.c new file +4
-0+static int gic_retrigger (struct irq_data *d) +{ +return -ENXIO ; +}
Why bother? (just don't assign the pointer in the chip).
drivers/irqchip/irq-gic-v3.c line 248
Quoted 4 lines
drivers/irqchip/irq-gic-v3.c new file +4
-0+static u64 gic_mpidr_to_affinity (u64 mpidr) +{ +/* Make sure we don't broadcast the interrupt */ +return mpidr & ~GICD_IROUTER_SPI_MODE_ANY ;
Yikes, mpidr meets the GIC! I'd be a lot more comfortable with this if we used our mpidr accessors to construct the value which GICD_IROUTER expects (e.g. using the MPIDR_AFFINITY_LEVEL macro).
drivers/irqchip/irq-gic-v3.c line 256
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c new file +8
-0+} + +static asmlinkage void __exception_irq_entry gic_handle_irq(struct pt_regs *regs) +{ +u64 irqnr; + + do { + irqnr = gic_read_iar();
What stops the CPU speculating this access around the loop?
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+if (likely (irqnr >15 && irqnr <1021 )) { + irqnr =irq_find_mapping (gic_data.domain, irqnr); +handle_IRQ (irqnr, regs); +continue ; + }
drivers/irqchip/irq-gic-v3.c line 272
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+if (irqnr <16 ) { +gic_write_eoir (irqnr); +#ifdef CONFIG_SMP +handle_IPI (irqnr, regs); +#else +WARN_ONCE (true,"Unexpected SGI received!\n" ); +#endif +continue ; + } + }while (irqnr !=0x3ff );
Any reason to adjust this control-flow from the existing GIC driver?
Quoted 31 lines
drivers/irqchip/irq-gic-v3.c new file +31
-0+} + +static void __init gic_dist_init (void ) +{ +unsigned int i; +u64 affinity; +int gic_irqs = gic_data.irq_nr; +void __iomem *base = gic_data.dist_base; + +/* Disable the distributor */ +writel_relaxed (0 , base +GICD_CTLR ); + +/* +* Set all global interrupts to be level triggered, active low. +*/ +for (i =32 ; i < gic_data.irq_nr; i +=16 ) +writel_relaxed (0 , base +GICD_ICFGR + i /4 ); + +/* +* Set priority on all global interrupts. +*/ +for (i =32 ; i < gic_irqs; i +=4 ) +writel_relaxed (0xa0a0a0a0 , base +GICD_IPRIORITYR + i); + +/* +* Disable all interrupts. Leave the PPI and SGIs alone +* as these enables are banked registers. +*/ +for (i =32 ; i < gic_irqs; i +=32 ) +writel_relaxed (0xffffffff , base +GICD_ICENABLER + i /8 ); +
drivers/irqchip/irq-gic-v3.c line 313
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+gic_dist_wait_for_rwp (); + +writel_relaxed (GICD_CTLR_ARE_NS |GICD_CTLR_ENABLE_G1A |GICD_CTLR_ENABLE_G1 , + base +GICD_CTLR ); + +/* +* Set all global interrupts to the boot CPU only. ARE must be +* enabled. +*/ + affinity =gic_mpidr_to_affinity (read_cpuid_mpidr ());
You can use the logical map here.
drivers/irqchip/irq-gic-v3.c line 316
Quoted 3 lines
drivers/irqchip/irq-gic-v3.c new file +3
-0+for (i =32 ; i < gic_data.irq_nr; i++) +writeq_relaxed (affinity, base +GICD_IROUTER + i *8 ); +}
Again, this looks very similar to the existing GIC driver.
drivers/irqchip/irq-gic-v3.c line 326
Quoted 9 lines
drivers/irqchip/irq-gic-v3.c new file +9
-0+static int __init gic_populate_rdist (void ) +{ +u64 mpidr =cpu_logical_map (smp_processor_id ()); +u64 typer; +u64 aff; +int i; + + aff = mpidr & ((1 <<24 ) -1 ); + aff |= (mpidr >>8 ) & (0xffUL <<24 );
We should have helpers for this in cputype.h.
Also, can we absolutely, 100%, "put-your-mother's-house-on-it" guarantee that the ITS, GICRn and the CPUs all agree on an identification scheme, as described in the mpidr? My reading of the spec doesn't fill me with confidence, and retrofitting a scheme in DT will end up with the same mess that we have for CPU topology.
drivers/irqchip/irq-gic-v3.c line 339
Quoted 13 lines
drivers/irqchip/irq-gic-v3.c new file +13
-0+ +for (i =0 ; i < gic_data.redist_regions; i++) { +void __iomem *ptr = gic_data.redist_base[i]; +u32 reg; + + reg =readl_relaxed (ptr +GICR_PIDR2 ) &GIC_PIDR2_ARCH_MASK ; +if (reg !=0x30 && reg !=0x40 ) {/* We're in trouble... */ +pr_warn ("No redistributor present @%p\n" , ptr); +break ; + } + + do { + typer = readq_relaxed(ptr + GICR_TYPER);
Not sure if you care, but readq is a show-stopper if you want to build this with arch/arm/.
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+if ((typer >>32 ) == aff) { +gic_data_rdist_rd_base () = ptr; +pr_info ("CPU%d: found redistributor %llx @%p\n" , +smp_processor_id (), + (unsigned long long ) mpidr, ptr);
drivers/irqchip/irq-gic-v3.c line 354
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+return 0 ; + } + +if (gic_data.redist_stride) { + ptr += gic_data.redist_stride; + }else { + ptr +=SZ_64K *2 ;/* Skip RD_base + SGI_base */ +if (typer &GICR_TYPER_VLPIS ) + ptr +=SZ_64K *2 ;/* Skip VLPI_base + reserved page */ + }
VLPIS is RES0 in GICv3.
drivers/irqchip/irq-gic-v3.c line 360
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+ }while (!(typer &GICR_TYPER_LAST )); + } + +/* We couldn't even deal with ourselves... */ +WARN (true,"CPU%d: mpidr %lx has no re-distributor!\n" , +smp_processor_id (), (unsigned long )mpidr);
Inconsistency with the type of mpidr (you use unsigned long long above).
Quoted 18 lines
drivers/irqchip/irq-gic-v3.c new file +18
-0+return -ENODEV ; +} + +static void __init gic_cpu_init (void ) +{ +void __iomem *rbase; +int i; + +/* Register ourselves with the rest of the world */ +if (gic_populate_rdist ()) +return ; + +gic_enable_redist (); + + rbase =gic_data_rdist_sgi_base (); + +/* +* Set priority on PPI and SGI interrupts
drivers/irqchip/irq-gic-v3.c line 388
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+*/ +for (i =0 ; i <32 ; i +=4 ) +writel_relaxed (0xa0a0a0a0 , rbase +GICR_IPRIORITYR0 + i *4 /4 ); + +/* +* Disable all PPI interrupts, ensure all SGI interrupts are +* enabled. +*/ +writel_relaxed (0xffff0000 , rbase +GICR_ICENABLER0 ); +writel_relaxed (0x0000ffff , rbase +GICR_ISENABLER0 );
Whilst there are some clear differences between this and the existing driver, you could still factor parts of this out into a common file.
Quoted 18 lines
drivers/irqchip/irq-gic-v3.c new file +18
-0+gic_redist_wait_for_rwp (); + +/* Enable system registers */ +gic_enable_sre (); + +/* Set priority mask register */ +gic_write_pmr (DEFAULT_PMR_VALUE ); + +/* EOI drops priority too (mode 0) */ +gic_write_ctlr (GICC_CTLR_EL1_EOImode_drop_dir); + +/* ... and let's hit the road... */ +gic_write_grpen1 (1 ); +} + +#ifdef CONFIG_SMP +static int __init gic_secondary_init (struct notifier_block *nfb, +unsigned long action,void *hcpu)
drivers/irqchip/irq-gic-v3.c line 417
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+{ +if (action ==CPU_STARTING || action ==CPU_STARTING_FROZEN ) +gic_cpu_init (); +return NOTIFY_OK ; +} + +/* +* Notifier for enabling the GIC CPU interface. Set an arbitrarily high +* priority because the GIC needs to be up before the ARM generic timers. +*/
Why must the GIC be up before the timers? Obviously, the timers won't be able to generate interrupts without the GIC initialised, but the comment implies that something worse will happen.
Quoted 9 lines
drivers/irqchip/irq-gic-v3.c new file +9
-0+static struct notifier_block __initdata gic_cpu_notifier = { + .notifier_call = gic_secondary_init, + .priority =100 , +}; + +static u16 gic_compute_target_list (int *base_cpu,const struct cpumask *mask, +u64 cluster_id) +{ +int cpu = *base_cpu;
drivers/irqchip/irq-gic-v3.c line 436
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+u64 mpidr =cpu_logical_map (cpu); +u16 tlist =0 ; + +while (cpu < nr_cpu_ids) { +/* +* If we ever get a cluster of more than 16 CPUs, just +* scream and skip that CPU. +*/ +if (WARN_ON ((mpidr &0xff ) >=16 )) + goto out;
I'm not comfortable with this. The mpidr in ARMv8 is just a unique identifier, so relying on aff0 to provide some sort of upper bound on the number of CPUs in a cluster feels extremely fragile.
I think you'll need to deal with SGIs to target lists that you can't represent with a single sgi1r value (and instead batch them up and have lots of fun dealing with the result).
Quoted 12 lines
drivers/irqchip/irq-gic-v3.c new file +12
-0+ + tlist |=1 << (mpidr &0xf ); + + cpu =cpumask_next (cpu, mask); + mpidr =cpu_logical_map (cpu); + +if (cluster_id != (mpidr & ~0xffUL )) { + cpu--; + goto out; + } + } +out:
drivers/irqchip/irq-gic-v3.c line 458
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+ *base_cpu = cpu; +return tlist; +} + +static void gic_send_sgi (u64 cluster_id,u16 tlist,unsigned int irq) +{ +u64 val; + + val = (cluster_id &0xff00ff0000UL ) <<16 ;/* Aff3 + Aff2 */ + val |= (cluster_id &0xff00 ) <<8 ;/* Aff1 */
Use the helpers.
Quoted 9 lines
drivers/irqchip/irq-gic-v3.c new file +9
-0+ val |= irq <<24 ; + val |= tlist; + +pr_debug ("CPU%d: ICC_SGI1R_EL1 %llx\n" ,smp_processor_id (), val); +gic_write_sgi1r (val); +} + +static void gic_raise_softirq (const struct cpumask *mask,unsigned int irq) +{
drivers/irqchip/irq-gic-v3.c line 477
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+int cpu; + +if (WARN_ON (irq >=16 )) +return ; + +/* +* Ensure that stores to Normal memory are visible to the +* other CPUs before issuing the IPI. +*/ +dsb ();
dmb(ishst), obviously (!) ;)
Quoted 20 lines
drivers/irqchip/irq-gic-v3.c new file +20
-0+ +for_each_cpu_mask (cpu, *mask) { +u64 cluster_id =cpu_logical_map (cpu) & ~0xffUL ; +u16 tlist; + + tlist =gic_compute_target_list (&cpu, mask, cluster_id); +gic_send_sgi (cluster_id, tlist, irq); + } +} + +static void gic_smp_init (void ) +{ +set_smp_cross_call (gic_raise_softirq); +register_cpu_notifier (&gic_cpu_notifier); +} + +static int gic_set_affinity (struct irq_data *d,const struct cpumask *mask_val, +bool force) +{ +unsigned int cpu =cpumask_any_and (mask_val, cpu_online_mask);
drivers/irqchip/irq-gic-v3.c line 507
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+void __iomem *reg; +u64 val; + +if (gic_irq (d) <32 ) +return -EINVAL ; + + reg =gic_dist_base (d) +GICD_IROUTER + (gic_irq (d) *8 ); + val =gic_mpidr_to_affinity (cpu_logical_map (cpu)); + +writeq_relaxed (val, reg);
How do you ensure that the interrupt has actually moved?
Quoted 23 lines
drivers/irqchip/irq-gic-v3.c new file +23
-0+return IRQ_SET_MASK_OK ; +} +#else +#define gic_set_affinity NULL +#define gic_smp_init () do { } while(0) +#endif + +static struct irq_chip gic_chip = { + .name ="GICv3" , + .irq_mask = gic_mask_irq, + .irq_unmask = gic_unmask_irq, + .irq_eoi = gic_eoi_irq, + .irq_set_type = gic_set_type, + .irq_retrigger = gic_retrigger, + .irq_set_affinity = gic_set_affinity, +}; + +static int gic_irq_domain_map (struct irq_domain *d,unsigned int irq, +irq_hw_number_t hw) +{ +if (hw <32 ) { +irq_set_percpu_devid (irq); + irq_set_chip_and_handler(irq, &gic_chip,
drivers/irqchip/irq-gic-v3.c line 541
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+ handle_percpu_devid_irq); +set_irq_flags (irq,IRQF_VALID |IRQF_NOAUTOEN ); + }else { +irq_set_chip_and_handler (irq, &gic_chip, + handle_fasteoi_irq); +set_irq_flags (irq,IRQF_VALID |IRQF_PROBE ); + } +irq_set_chip_data (irq, d->host_data); +return 0 ; +}
More copied GICv2 code...
Quoted 14 lines
drivers/irqchip/irq-gic-v3.c new file +14
-0+static int gic_irq_domain_xlate (struct irq_domain *d, +struct device_node *controller, +const u32 *intspec,unsigned int intsize, +unsigned long *out_hwirq,unsigned int *out_type) +{ +if (d->of_node != controller) +return -EINVAL ; +if (intsize <3 ) +return -EINVAL ; + +switch (intspec[0 ]) { +case 0 :/* SPI */ + *out_hwirq = intspec[1 ] +32 ; +break ;
drivers/irqchip/irq-gic-v3.c line 566
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+case 1 :/* PPI */ + *out_hwirq = intspec[1 ] +16 ; +break ; +default : +return -EINVAL ; + } + + *out_type = intspec[2 ] &IRQ_TYPE_SENSE_MASK ; +return 0 ; +}
... and some slightly mutated but functionally identical GICv2 code ;)
Quoted 29 lines
drivers/irqchip/irq-gic-v3.c new file +29
-0+static const struct irq_domain_ops gic_irq_domain_ops = { + .map = gic_irq_domain_map, + .xlate = gic_irq_domain_xlate, +}; + +static int __init gic_of_init (struct device_node *node,struct device_node *parent) +{ +void __iomem *dist_base; +void __iomem **redist_base; +u64 redist_stride; +u32 redist_regions; +u32 reg; +int gic_irqs; +int err; +int i; + + dist_base =of_iomap (node,0 ); +if (!dist_base) { +pr_warn ("%s: unable to map gic dist registers\n" , + node->full_name); +return -ENXIO ; + } + + reg =readl_relaxed (dist_base +GICD_PIDR2 ) &GIC_PIDR2_ARCH_MASK ; +if (reg !=0x30 && reg !=0x40 ) { +pr_warn ("%s: no distributor detected, giving up\n" , + node->full_name); + err = -ENODEV ; + goto out_unmap_dist;
drivers/irqchip/irq-gic-v3.c line 606
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+ } + +if (of_property_read_u32 (node,"#redistributor-regions" , &redist_regions)) + redist_regions =1 ; + + redist_base =kzalloc (sizeof (*redist_base) * redist_regions,GFP_KERNEL ); +if (!redist_base) { + err = -ENOMEM ; + goto out_unmap_dist; + }
Can you use the devm_* managed resource helpers in irqchip init functions? I see a couple of drivers doing it, but it's not clear to me whether things are actually freed on failure. Anyway, if you can, then it would clean this up a bit.
Quoted 36 lines
drivers/irqchip/irq-gic-v3.c new file +36
-0+for (i =0 ; i < redist_regions; i++) { + redist_base[i] =of_iomap (node,1 + i); +if (!redist_base[i]) { +pr_warn ("%s: couldn't map region %d\n" , + node->full_name, i); + err = -ENODEV ; + goto out_unmap_rdist; + } + } + +if (of_property_read_u64 (node,"redistributor-stride" , &redist_stride)) + redist_stride =0 ; + + gic_data.dist_base = dist_base; + gic_data.redist_base = redist_base; + gic_data.redist_regions = redist_regions; + gic_data.redist_stride = redist_stride; + +/* +* Find out how many interrupts are supported. +* The GIC only supports up to 1020 interrupt sources (SGI+PPI+SPI) +*/ + gic_irqs =readl_relaxed (gic_data.dist_base +GICD_TYPER ) &0x1f ; + gic_irqs = (gic_irqs +1 ) *32 ; +if (gic_irqs >1020 ) + gic_irqs =1020 ; + gic_data.irq_nr = gic_irqs; + + gic_data.domain =irq_domain_add_linear (node, gic_irqs -16 , + &gic_irq_domain_ops, &gic_data); + gic_data.rdist =alloc_percpu (typeof (*gic_data.rdist)); + +if (WARN_ON (!gic_data.domain) ||WARN_ON (!gic_data.rdist)) { + err = -ENOMEM ; + goto out_free; + }
drivers/irqchip/irq-gic-v3.c line 653
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+ +set_handle_irq (gic_handle_irq); + +gic_smp_init (); +gic_dist_init (); +gic_cpu_init (); + +return 0 ; + +out_free:
remove the irq_domain here?
Quoted 182 lines
drivers/irqchip/irq-gic-v3.c new file +12
-0+free_percpu (gic_data.rdist); +out_unmap_rdist: +for (i =0 ; i < redist_regions; i++) +if (redist_base[i]) +iounmap (redist_base[i]); +kfree (redist_base); +out_unmap_dist: +iounmap (dist_base); +return err; +} + +IRQCHIP_DECLARE (gic_v3,"arm,gic-v3" , gic_of_init);include/linux/irqchip/arm-gic-v3.h new file +164
-0@@ -0,0 +1,190 @@ +/* +* Copyright (C) 2013 ARM Limited, All Rights Reserved. +* Author: Marc Zyngier <[email protected]> +* +* +* This program is free software; you can redistribute it and/or modify +* it under the terms of the GNU General Public License version 2 as +* published by the Free Software Foundation. +* +* This program is distributed in the hope that it will be useful, +* but WITHOUT ANY WARRANTY; without even the implied warranty of +* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +* GNU General Public License for more details. +* +* You should have received a copy of the GNU General Public License +* along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ +#ifndef __LINUX_IRQCHIP_ARM_GIC_V3_H +#define __LINUX_IRQCHIP_ARM_GIC_V3_H + +/* +* Distributor registers. We assume we're running non-secure, with ARE +* being set. Secure-only and non-ARE registers are not described. +*/ +#define GICD_CTLR 0x0000 +#define GICD_TYPER 0x0004 +#define GICD_IIDR 0x0008 +#define GICD_STATUSR 0x0010 +#define GICD_SETSPI_NSR 0x0040 +#define GICD_CLRSPI_NSR 0x0048 +#define GICD_SETSPI_SR 0x0050 +#define GICD_CLRSPI_SR 0x0058 +#define GICD_SEIR 0x0068 +#define GICD_ISENABLER 0x0100 +#define GICD_ICENABLER 0x0180 +#define GICD_ISPENDR 0x0200 +#define GICD_ICPENDR 0x0280 +#define GICD_ISACTIVER 0x0300 +#define GICD_ICACTIVER 0x0380 +#define GICD_IPRIORITYR 0x0400 +#define GICD_ICFGR 0x0C00 +#define GICD_IROUTER 0x6000 +#define GICD_PIDR2 0xFFE8 + +#define GICD_CTLR_RWP (1U << 31) +#define GICD_CTLR_ARE_NS (1U << 4) +#define GICD_CTLR_ENABLE_G1A (1U << 1) +#define GICD_CTLR_ENABLE_G1 (1U << 0) + +#define GICD_IROUTER_SPI_MODE_ONE (0U << 31) +#define GICD_IROUTER_SPI_MODE_ANY (1U << 31) + +#define GIC_PIDR2_ARCH_MASK 0xf0 + +/* +* Re-Distributor registers, offsets from RD_base +*/ +#define GICR_CTLR GICD_CTLR +#define GICR_IIDR 0x0004 +#define GICR_TYPER 0x0008 +#define GICR_STATUSR GICD_STATUSR +#define GICR_WAKER 0x0014 +#define GICR_SETLPIR 0x0040 +#define GICR_CLRLPIR 0x0048 +#define GICR_SEIR GICD_SEIR +#define GICR_PROPBASER 0x0070 +#define GICR_PENDBASER 0x0078 +#define GICR_INVLPIR 0x00A0 +#define GICR_INVALLR 0x00B0 +#define GICR_SYNCR 0x00C0 +#define GICR_MOVLPIR 0x0100 +#define GICR_MOVALLR 0x0110 +#define GICR_PIDR2 GICD_PIDR2 + +#define GICR_WAKER_ProcessorSleep (1U << 1) +#define GICR_WAKER_ChildrenAsleep (1U << 2) + +/* +* Re-Distributor registers, offsets from SGI_base +*/ +#define GICR_ISENABLER0 GICD_ISENABLER +#define GICR_ICENABLER0 GICD_ICENABLER +#define GICR_ISPENDR0 GICD_ISPENDR +#define GICR_ICPENDR0 GICD_ICPENDR +#define GICR_ISACTIVER0 GICD_ISACTIVER +#define GICR_ICACTIVER0 GICD_ICACTIVER +#define GICR_IPRIORITYR0 GICD_IPRIORITYR +#define GICR_ICFGR0 GICD_ICFGR + +#define GICR_TYPER_VLPIS (1U << 1) +#define GICR_TYPER_LAST (1U << 4) + +/* +* CPU interface registers +*/ +#define GICC_CTLR_EL1_EOImode_drop_dir (0U << 1) +#define GICC_CTLR_EL1_EOImode_drop (1U << 1) +#define GICC_SRE_EL1_SRE (1U << 0) + +/* +* Hypervisor interface registers (SRE only) +*/ +#define GICH_LR_VIRTUAL_ID_MASK ((1UL << 32) - 1) + +#define GICH_LR_EOI (1UL << 41) +#define GICH_LR_GROUP (1UL << 60) +#define GICH_LR_STATE (3UL << 62) +#define GICH_LR_PENDING_BIT (1UL << 62) +#define GICH_LR_ACTIVE_BIT (1UL << 63) + +#define GICH_MISR_EOI (1 << 0) +#define GICH_MISR_U (1 << 1) + +#define GICH_HCR_EN (1 << 0) +#define GICH_HCR_UIE (1 << 1) + +#define GICH_VMCR_CTLR_SHIFT 0 +#define GICH_VMCR_CTLR_MASK (0x21f << GICH_VMCR_CTLR_SHIFT) +#define GICH_VMCR_BPR1_SHIFT 18 +#define GICH_VMCR_BPR1_MASK (7 << GICH_VMCR_BPR1_SHIFT) +#define GICH_VMCR_BPR0_SHIFT 21 +#define GICH_VMCR_BPR0_MASK (7 << GICH_VMCR_BPR0_SHIFT) +#define GICH_VMCR_PMR_SHIFT 24 +#define GICH_VMCR_PMR_MASK (0xffUL << GICH_VMCR_PMR_SHIFT) + +#define ICC_EOIR1_EL1 S3_0_C12_C12_1 +#define ICC_IAR1_EL1 S3_0_C12_C12_0 +#define ICC_SGI1R_EL1 S3_0_C12_C11_5 +#define ICC_PMR_EL1 S3_0_C4_C6_0 +#define ICC_CTLR_EL1 S3_0_C12_C12_4 +#define ICC_SRE_EL1 S3_0_C12_C12_5 +#define ICC_GRPEN1_EL1 S3_0_C12_C12_7 + +#define ICC_SRE_EL2 S3_4_C12_C9_5 + +#define ICH_VSEIR_EL2 S3_4_C12_C9_4 +#define ICH_HCR_EL2 S3_4_C12_C11_0 +#define ICH_VTR_EL2 S3_4_C12_C11_1 +#define ICH_MISR_EL2 S3_4_C12_C11_2 +#define ICH_EISR_EL2 S3_4_C12_C11_3 +#define ICH_ELSR_EL2 S3_4_C12_C11_5 +#define ICH_VMCR_EL2 S3_4_C12_C11_7 + +#define __LR0_EL2 (x) S3_4_C12_C12_ ## x +#define __LR8_EL2 (x) S3_4_C12_C13_ ## x + +#define ICH_LR0_EL2 __LR0_EL2(0) +#define ICH_LR1_EL2 __LR0_EL2(1) +#define ICH_LR2_EL2 __LR0_EL2(2) +#define ICH_LR3_EL2 __LR0_EL2(3) +#define ICH_LR4_EL2 __LR0_EL2(4) +#define ICH_LR5_EL2 __LR0_EL2(5) +#define ICH_LR6_EL2 __LR0_EL2(6) +#define ICH_LR7_EL2 __LR0_EL2(7) +#define ICH_LR8_EL2 __LR8_EL2(0) +#define ICH_LR9_EL2 __LR8_EL2(1) +#define ICH_LR10_EL2 __LR8_EL2(2) +#define ICH_LR11_EL2 __LR8_EL2(3) +#define ICH_LR12_EL2 __LR8_EL2(4) +#define ICH_LR13_EL2 __LR8_EL2(5) +#define ICH_LR14_EL2 __LR8_EL2(6) +#define ICH_LR15_EL2 __LR8_EL2(7) + +#define __AP0Rx_EL2 (x) S3_4_C12_C8_ ## x
include/linux/irqchip/arm-gic-v3.h line 174
Quoted 10 lines
include/linux/irqchip/arm-gic-v3.h new file +10
-0+#define ICH_AP0R0_EL2 __AP0Rx_EL2(0) +#define ICH_AP0R1_EL2 __AP0Rx_EL2(1) +#define ICH_AP0R2_EL2 __AP0Rx_EL2(2) +#define ICH_AP0R3_EL2 __AP0Rx_EL2(3) + +#define __AP1Rx_EL2 (x) S3_4_C12_C9_ ## x +#define ICH_AP1R0_EL2 __AP1Rx_EL2(0) +#define ICH_AP1R1_EL2 __AP1Rx_EL2(1) +#define ICH_AP1R2_EL2 __AP1Rx_EL2(2) +#define ICH_AP1R3_EL2 __AP1Rx_EL2(3)
I'm going to rely on Christoffer having looked at all these magic numbers (he at least had ice cream!).
include/linux/irqchip/arm-gic-v3.h line 186
Quoted 11 lines
include/linux/irqchip/arm-gic-v3.h new file +11
-0+#ifndef __ASSEMBLY__ + +#include <linux/stringify.h> + +#define reg (x) __stringify(x) + +static inline void gic_write_eoir (u64 irq) +{ + asm volatile("msr " reg(ICC_EOIR1_EL1) ", %0" : : "r" (irq)); +isb (); +}
Why is this in the header file?
Will
PATCH 1/18 arm64: initial support for GICv3
Quoted 19 lines · Will Deacon
Hi Marc,
Quoted 16 lines · Marc Zyngier
arch/arm64/kernel/head.S +1
-0@@ -22,6 +22,7 @@ #include <linux/linkage.h> #include <linux/init.h> +#include <linux/irqchip/arm-gic-v3.h> #include <asm/assembler.h> #include <asm/ptrace.h>@@ -183,6 +184,19 @@ CPU_LE( bic x0, x0, #(3 << 24) ) // Clear the EE and E0E bits for EL1 msr cnthctl_el2, x0 msr cntvoff_el2, xzr // Clear virtual offset
Quoted 13 lines
arch/arm64/kernel/head.S CPU_LE( bic x0, x0, #(3 << 24) ) // Clear the EE and E0E bits for EL1 · line 194
Quoted 8 lines
arch/arm64/kernel/head.S +8
-0+ /* GICv3 system register access */ + mrs x0, id_aa64pfr0_el1 + ubfx x0, x0, #24, #4 + cmp x0, #1 + b.ne 3f + + mrs x0, ICC_SRE_EL2 // Set ICC_SRE_EL2.SRE==1 + orr x0, x0, #1Don't you need to set the Enable bit (bit 3) too? Also, does setting ICC_SRE_EL2.SRE force KVM to use the ICH_* registers? If so, that seems like a pretty big decision to take this early on (there's not even an #ifdef
CONFIG_ARM_GICV3).
The test on id_aa64pfr0_el1 is an absolute guarantee that we have the GICv3 system registers.
As for the Enable bit, this is handle by the magic trapping we have below, but I decided to get rid of that code (see the previous discussion with Christoffer).
Quoted 25 lines
Quoted 25 lines
arch/arm64/kernel/head.S +5
-0+ msr ICC_SRE_EL2, x0 + isb // Make sure SRE is now 1 + msr ICH_HCR_EL2, xzr // Reset ICC_HCR_EL2 to defaults + +3: /* Populate ID registers. */ mrs x0, midr_el1 mrs x1, mpidr_el1arch/arm64/kernel/hyp-stub.S +1
-0@@ -19,6 +19,7 @@ #include <linux/init.h> #include <linux/linkage.h> +#include <linux/irqchip/arm-gic-v3.h> #include <asm/assembler.h> #include <asm/ptrace.h>@@ -51,16 +52,52 @@ ENDPROC(__hyp_stub_vectors) .align 11
Quoted 15 lines
arch/arm64/kernel/hyp-stub.S ENDPROC(__hyp_stub_vectors) · line 65
Quoted 11 lines
arch/arm64/kernel/hyp-stub.S +10
-0+#define sysreg_to_iss(op0, op1, crn, crm, op2) \ + (((op0 & 3) << 20) | ((op2 & 7 ) << 17) | \ + ((op1 & 7 ) << 14) | ((crn & 15) << 10) | \ + ((crm & 15) << 1)) + el1_sync: + // Hack alert: we don't have a stack, and we can't use the one + // from EL1 because our MMU is off. Thankfully, we have + // tpidr_el2, which is under our complete control, and + // tpidr_el0, which can be used as long as we haven't reached + // userspace. Oh hum...Yuck. Why can't we create a minimal stack when we initialise EL2 when booting the host? Even simpler, just use some .quads in here (you're already using a literal pool).
I really liked the "funkiness" here, but that code is now gone. RIP! ;-)
Quoted 23 lines
Quoted 23 lines
arch/arm64/kernel/hyp-stub.S +14
-2+ msr tpidr_el2, x1 // Preserve x1 + mrs x1, esr_el2 lsr x1, x1, #26 cmp x1, #0x16 - b.ne 2f // Not an HVC trap + b.ne 3f // Not an HVC trap + + // Handle getting/setting vbar_el2 cbz x0, 1f msr vbar_el2, x0 // Set vbar_el2 b 2f 1: mrs x0, vbar_el2 // Return vbar_el2 -2: eret +2: mrs x1, tpidr_el2 // Restore x1 + eret + +3: cmp x1, #0x18 // MSR/MRS? + b.ne 2b + + // Handle trapped sys reg access + msr tpidr_el0, x0 // Preserve x0 + mrs x1, esr_el2
Quoted 8 lines
arch/arm64/kernel/hyp-stub.S ENDPROC(__hyp_stub_vectors) · line 89
Quoted 3 lines
arch/arm64/kernel/hyp-stub.S +3
-0+ ldr x0, =sysreg_to_iss(3, 7, 15, 15, 7) + and x1, x1, x0 + ldr x0, =sysreg_to_iss(3, 0, 12, 12, 5)I think
sysreg_to_issactually makes this *less* readable! Why not simply #define the mask to extract OPx, CRx then compare with another #define forICC_SRE_EL1? If we can represent those as immediates, you can avoid the literal loads too.
Gone.
Quoted 10 lines
arch/arm64/kernel/hyp-stub.S ENDPROC(__hyp_stub_vectors) · line 96
Quoted 7 lines
arch/arm64/kernel/hyp-stub.S +7
-0+ eor x0, x0, x1 // Check for ICC_SRE_EL1 + cbnz x0, 4f // Something bad happened + + // Set ICC_SRE_EL1 access enable + mrs x1, ICC_SRE_EL2 + orr x1, x1, #(1 << 3) + msr ICC_SRE_EL2, x1Ah, ok, you do this part lazily. Does that make sense for the host? (i.e. is it possible for KVM to use SRE at EL2 but not the host at EL1?).
You could, but I'm not sure it is worth supporting. In the interest of simplicity, I got rid of all that crap.
Quoted 55 lines
Quoted 55 lines
arch/arm64/kernel/hyp-stub.S +4
-0+ +4: mrs x0, tpidr_el0 // Restore x0 + mrs x1, tpidr_el2 // Restore x1 + eret ENDPROC(el1_sync) .macro invalid_vector labeldrivers/irqchip/irq-gic-v3.c new file +42
-0@@ -0,0 +1,665 @@ +/* +* Copyright (C) 2013 ARM Limited, All Rights Reserved. +* Author: Marc Zyngier <[email protected]> +* +* This program is free software; you can redistribute it and/or modify +* it under the terms of the GNU General Public License version 2 as +* published by the Free Software Foundation. +* +* This program is distributed in the hope that it will be useful, +* but WITHOUT ANY WARRANTY; without even the implied warranty of +* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +* GNU General Public License for more details. +* +* You should have received a copy of the GNU General Public License +* along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ + +#include <linux/cpu.h> +#include <linux/interrupt.h> +#include <linux/of.h> +#include <linux/of_address.h> +#include <linux/of_irq.h> +#include <linux/percpu.h> +#include <linux/slab.h> + +#include <linux/irqchip/arm-gic-v3.h> + +#include <asm/cputype.h> +#include <asm/exception.h> +#include <asm/smp_plat.h> + +#include "irqchip.h" + +struct gic_chip_data { +void __iomem *dist_base; +void __iomem **redist_base; +void __percpu __iomem **rdist; +struct irq_domain *domain; +u64 redist_stride; +u32 redist_regions; +unsigned int irq_nr; +};
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 46
Quoted 4 lines
drivers/irqchip/irq-gic-v3.c new file +4
-0+ +static struct gic_chip_data gic_data__read_mostly ; + +#define gic_data_rdist_rd_base () (*__this_cpu_ptr(gic_data.rdist))You should avoid
__this_cpu_ptr(it's being removed afaik). Take a look at the patches from Christoph Lameter, e.g.170d800af83f("block: Replace__get_cpu_varuses").
Ah, interesting. I'll repaint this.
Quoted 29 lines
Quoted 29 lines
drivers/irqchip/irq-gic-v3.c new file +29
-0+#define gic_data_rdist_sgi_base () (gic_data_rdist_rd_base() + SZ_64K) + +static DEFINE_RAW_SPINLOCK (dist_lock); + +#define DEFAULT_PMR_VALUE 0xf0 + +static inline void __iomem *gic_dist_base (struct irq_data *d) +{ +if (d->hwirq <32 )/* SGI+PPI -> SGI_base for this CPU */ +return gic_data_rdist_sgi_base (); + +if (d->hwirq <=1023 )/* SPI -> dist_base */ +return gic_data.dist_base; + +if (d->hwirq >=8192 ) +BUG ();/* LPI Detected!!! */ + +return NULL ; +} + +static inline unsigned int gic_irq (struct irq_data *d) +{ +return d->hwirq; +} + +static void gic_do_wait_for_rwp (void __iomem *base) +{ +u32 val; +
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 79
Quoted 4 lines
drivers/irqchip/irq-gic-v3.c new file +4
-0+do { + val =readl_relaxed (base +GICD_CTLR ); +cpu_relax (); + }while (val &GICD_CTLR_RWP );You should add a timeout in here to save us in the face of broken hardware (especially as you hold the distributor lock). See, for example,
arm_smmu_tlb_sync(...)in the arm-smmu driver.
Yup.
Quoted 2 lines
Thinking about it, is this the only point of the
dist_lock(to ensure you wait only for your request)? Do we actually need locking for that?
You need to ensure that the distributor has propagated the request. I'm not sure what the effect would be if we had several CPUs writing to the distributor at once... I'll investigate.
Quoted 34 lines
Quoted 7 lines
drivers/irqchip/irq-gic-v3.c new file +7
-0+} + +/* Wait for completion of a distributor change */ +static void gic_dist_wait_for_rwp (void ) +{ +gic_do_wait_for_rwp (gic_data.dist_base); +}drivers/irqchip/irq-gic-v3.c line 96
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+ +/* Wait for completion of a redistributor change */ +static void gic_redist_wait_for_rwp (void ) +{ +gic_do_wait_for_rwp (gic_data_rdist_rd_base ()); +} + +static void gic_wait_for_rwp (int irq) +{ +if (irq <32 )Maybe we should just bite the bullet and have helpers for
gic_irq_is_sgi(irq)etc. so we can avoid these magic numbers all over the place.drivers/irqchip/irq-gic-v3.c line 100
Quoted 4 lines
drivers/irqchip/irq-gic-v3.c new file +4
-0+gic_redist_wait_for_rwp (); + else +gic_dist_wait_for_rwp (); +}Saying that, I question the value of mixing this into a single function.
gic_redist_wait_for_rwpabsolutely *must* be called with preemption disabled and on the appropriate CPU for the IRQ in question, whilst the latter can be called anywhere. So even though you can call this single function from a bunch of places, you're really relying on the irq parameter being correct for the context of the caller.
Quoted 2 lines
I reckon it would actually be clearer to call the relevant wait function explicitly in the caller. Same for
gic_dist_base.
Fair enough. I'll try to rework that part.
Quoted 9 lines
drivers/irqchip/irq-gic-v3.c line 107
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+/* Low level accessors */ +static u64 gic_read_iar (void ) +{ +u64 irqstat; + + asm volatile("mrs %0, " reg(ICC_IAR1_EL1) : "=r" (irqstat));Hmm,
reg()is a terrible name for a macro :) Can't you just use stringify directly?
Too long? ;-)
Quoted 30 lines
Quoted 16 lines
drivers/irqchip/irq-gic-v3.c new file +16
-0+return irqstat; +} + +static void gic_write_pmr (u64 val) +{ + asm volatile("msr " reg(ICC_PMR_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_ctlr (u64 val) +{ + asm volatile("msr " reg(ICC_CTLR_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_grpen1 (u64 val)drivers/irqchip/irq-gic-v3.c line 133
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+{ + asm volatile("msr " reg(ICC_GRPEN1_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_write_sgi1r (u64 val) +{ + asm volatile("msr " reg(ICC_SGI1R_EL1) ", %0" : : "r" (val)); +isb (); +}I'm not convinced you *always* need these isbs. For example, the
gic_handle_irqloop probably just needs one at the end, before returning.
Quoted 4 lines
In fact, looking at the spec,
ICC_PMR_EL1is actually self-synchronising so you don't need anything there. For SGI1R at least, you need a dsb to ensure write observability by the GIC (it's not clear to me whether you also need to the isb).
The dsb is actually in gic_send_sgi. We may end-up doing several writes to ICC_SGI1R_EL1 if the targets are on different Aff{3,2,1} levels, so it made sense to put it there.
Quoted 21 lines
Quoted 21 lines
drivers/irqchip/irq-gic-v3.c new file +21
-0+static void gic_enable_sre (void ) +{ +u64 val; + +/* This access is likely to trap into EL2 */ + asm volatile("mrs %0, " reg(ICC_SRE_EL1) : "=r" (val)); + val |=GICC_SRE_EL1_SRE ; + asm volatile("msr " reg(ICC_SRE_EL1) ", %0" : : "r" (val)); +isb (); +} + +static void gic_enable_redist (void ) +{ +void __iomem *rbase; +u32 val; + + rbase =gic_data_rdist_rd_base (); + +/* Wake up this CPU redistributor */ + val =readl_relaxed (rbase +GICR_WAKER ); + val &= ~GICR_WAKER_ProcessorSleep;
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 161
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+writel_relaxed (val, rbase +GICR_WAKER ); + +do { + val =readl_relaxed (rbase +GICR_WAKER ); +cpu_relax (); + }while (val & GICR_WAKER_ChildrenAsleep);Again, you want a timeout here.
Agreed.
Quoted 23 lines
Quoted 23 lines
drivers/irqchip/irq-gic-v3.c new file +23
-0+} + +/* +* Routines to acknowledge, disable and enable interrupts +*/ +static void gic_mask_irq (struct irq_data *d) +{ +u32 mask =1 << (gic_irq (d) %32 ); + +raw_spin_lock (&dist_lock); +writel_relaxed (mask,gic_dist_base (d) +GICD_ICENABLER + (gic_irq (d) /32 ) *4 ); +gic_wait_for_rwp (gic_irq (d)); +raw_spin_unlock (&dist_lock); +} + +static void gic_unmask_irq (struct irq_data *d) +{ +u32 mask =1 << (gic_irq (d) %32 ); + +raw_spin_lock (&dist_lock); +writel_relaxed (mask,gic_dist_base (d) +GICD_ISENABLER + (gic_irq (d) /32 ) *4 ); +gic_wait_for_rwp (gic_irq (d)); +raw_spin_unlock (&dist_lock);
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 190
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+} + +static void gic_eoi_irq (struct irq_data *d) +{ +gic_write_eoir (gic_irq (d)); +}Can you inline the code here (this is the only user) and move the isb out?
Yup.
Quoted 42 lines
Quoted 42 lines
drivers/irqchip/irq-gic-v3.c new file +42
-0+static int gic_set_type (struct irq_data *d,unsigned int type) +{ +void __iomem *base =gic_dist_base (d); +unsigned int irq =gic_irq (d); +u32 enablemask =1 << (irq %32 ); +u32 enableoff = (irq /32 ) *4 ; +u32 confmask =0x2 << ((irq %16 ) *2 ); +u32 confoff = (irq /16 ) *4 ; +bool enabled = false; +u32 val; + +/* Interrupt configuration for SGIs can't be changed */ +if (irq <16 ) +return -EINVAL ; + +if (type !=IRQ_TYPE_LEVEL_HIGH && type !=IRQ_TYPE_EDGE_RISING ) +return -EINVAL ; + +raw_spin_lock (&dist_lock); + + val =readl_relaxed (base +GICD_ICFGR + confoff); +if (type ==IRQ_TYPE_LEVEL_HIGH ) + val &= ~confmask; +else if (type ==IRQ_TYPE_EDGE_RISING ) + val |= confmask; + +/* +* As recommended by the spec, disable the interrupt before changing +* the configuration +*/ +if (readl_relaxed (base +GICD_ISENABLER + enableoff) & enablemask) { +writel_relaxed (enablemask, base +GICD_ICENABLER + enableoff); +gic_wait_for_rwp (irq); + enabled = true; + } + +writel_relaxed (val, base +GICD_ICFGR + confoff); + +if (enabled) { +writel_relaxed (enablemask, base +GICD_ISENABLER + enableoff); +gic_wait_for_rwp (irq); + }
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 238
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+ +raw_spin_unlock (&dist_lock); + +return 0 ; +}I really think this should be merged with the existing GIC driver -- the code is basically identical.
Not sure. I agree that the code is extremely similar, but some things are slightly different:
- gic_dist_base can actually return a redistributor base address (which is per-cpu). That doesn't fit with with the GICv2 model very well.
- The synchronization function doesn't exist in GICv2.
The closest thing I can imagine is a primitive like:
void gic_configure_irq(unsigned int irq, unsigned int type,
void __iomem *base, void (*sync)(void));
that'd do the low level stuff.
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c line 243
Quoted 4 lines
drivers/irqchip/irq-gic-v3.c new file +4
-0+static int gic_retrigger (struct irq_data *d) +{ +return -ENXIO ; +}Why bother? (just don't assign the pointer in the chip).
Fair enough.
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 248
Quoted 4 lines
drivers/irqchip/irq-gic-v3.c new file +4
-0+static u64 gic_mpidr_to_affinity (u64 mpidr) +{ +/* Make sure we don't broadcast the interrupt */ +return mpidr & ~GICD_IROUTER_SPI_MODE_ANY ;Yikes, mpidr meets the GIC! I'd be a lot more comfortable with this if we used our mpidr accessors to construct the value which
GICD_IROUTERexpects (e.g. using theMPIDR_AFFINITY_LEVELmacro).
Sure.
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c line 256
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c new file +8
-0+} + +static asmlinkage void __exception_irq_entry gic_handle_irq(struct pt_regs *regs) +{ +u64 irqnr; + + do { + irqnr = gic_read_iar();What stops the CPU speculating this access around the loop?
Probably not much, if ICC_IAR_EL1 is not a bit "special". Speculating into this would be very dangerous though, but nothing says otherwise. I'll ask the question.
Quoted 9 lines
Quoted 9 lines
drivers/irqchip/irq-gic-v3.c new file +9
-0+if (likely (irqnr >15 && irqnr <1021 )) { + irqnr =irq_find_mapping (gic_data.domain, irqnr); +handle_IRQ (irqnr, regs); +continue ; + } +if (irqnr <16 ) { +gic_write_eoir (irqnr); +#ifdef CONFIG_SMP +handle_IPI (irqnr, regs);
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 272
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+#else +WARN_ONCE (true,"Unexpected SGI received!\n" ); +#endif +continue ; + } + }while (irqnr !=0x3ff );Any reason to adjust this control-flow from the existing GIC driver?
Mostly to accomodate the ITS code that goes right in the middle of this.
Quoted 35 lines
Quoted 35 lines
drivers/irqchip/irq-gic-v3.c new file +35
-0+} + +static void __init gic_dist_init (void ) +{ +unsigned int i; +u64 affinity; +int gic_irqs = gic_data.irq_nr; +void __iomem *base = gic_data.dist_base; + +/* Disable the distributor */ +writel_relaxed (0 , base +GICD_CTLR ); + +/* +* Set all global interrupts to be level triggered, active low. +*/ +for (i =32 ; i < gic_data.irq_nr; i +=16 ) +writel_relaxed (0 , base +GICD_ICFGR + i /4 ); + +/* +* Set priority on all global interrupts. +*/ +for (i =32 ; i < gic_irqs; i +=4 ) +writel_relaxed (0xa0a0a0a0 , base +GICD_IPRIORITYR + i); + +/* +* Disable all interrupts. Leave the PPI and SGIs alone +* as these enables are banked registers. +*/ +for (i =32 ; i < gic_irqs; i +=32 ) +writel_relaxed (0xffffffff , base +GICD_ICENABLER + i /8 ); + +gic_dist_wait_for_rwp (); + +writel_relaxed (GICD_CTLR_ARE_NS |GICD_CTLR_ENABLE_G1A |GICD_CTLR_ENABLE_G1 , + base +GICD_CTLR );
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 313
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+ +/* +* Set all global interrupts to the boot CPU only. ARE must be +* enabled. +*/ + affinity =gic_mpidr_to_affinity (read_cpuid_mpidr ());You can use the logical map here.
OK.
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c line 316
Quoted 3 lines
drivers/irqchip/irq-gic-v3.c new file +3
-0+for (i =32 ; i < gic_data.irq_nr; i++) +writeq_relaxed (affinity, base +GICD_IROUTER + i *8 ); +}Again, this looks very similar to the existing GIC driver.
I can probably do a similar thing as with gic_set_type, but I'm not sure it adds much in terms of being readable.
Quoted 11 lines
drivers/irqchip/irq-gic-v3.c line 326
Quoted 9 lines
drivers/irqchip/irq-gic-v3.c new file +9
-0+static int __init gic_populate_rdist (void ) +{ +u64 mpidr =cpu_logical_map (smp_processor_id ()); +u64 typer; +u64 aff; +int i; + + aff = mpidr & ((1 <<24 ) -1 ); + aff |= (mpidr >>8 ) & (0xffUL <<24 );We should have helpers for this in
cputype.h.
Yes, I fixed that already in the latest version of the code.
Quoted 5 lines
Also, can we absolutely, 100%, "put-your-mother's-house-on-it" guarantee that the ITS, GICRn and the CPUs all agree on an identification scheme, as described in the mpidr? My reading of the spec doesn't fill me with confidence, and retrofitting a scheme in DT will end up with the same mess that we have for CPU topology.
My discussions with the architect of this fine piece of [insert expletive here] has led me to believe that the MPIDR is the only thing that holds the thing together. Of course, that doesn't preclude an implementer from wrecking the ship, but I don't plan to support something that doesn't follow the spec (or at least its intent).
Quoted 8 lines
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c new file +8
-0+ +for (i =0 ; i < gic_data.redist_regions; i++) { +void __iomem *ptr = gic_data.redist_base[i]; +u32 reg; + + reg =readl_relaxed (ptr +GICR_PIDR2 ) &GIC_PIDR2_ARCH_MASK ; +if (reg !=0x30 && reg !=0x40 ) {/* We're in trouble... */ +pr_warn ("No redistributor present @%p\n" , ptr);
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 339
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+break ; + } + + do { + typer = readq_relaxed(ptr + GICR_TYPER);Not sure if you care, but readq is a show-stopper if you want to build this with
arch/arm/.
I realized that recently. Any reason why we don't have writeq/readq on LPAE-capable stuff (other than the obvious runtime detection madness)?
Quoted 16 lines
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+if ((typer >>32 ) == aff) { +gic_data_rdist_rd_base () = ptr; +pr_info ("CPU%d: found redistributor %llx @%p\n" , +smp_processor_id (), + (unsigned long long ) mpidr, ptr);drivers/irqchip/irq-gic-v3.c line 354
Quoted 10 lines
drivers/irqchip/irq-gic-v3.c new file +10
-0+return 0 ; + } + +if (gic_data.redist_stride) { + ptr += gic_data.redist_stride; + }else { + ptr +=SZ_64K *2 ;/* Skip RD_base + SGI_base */ +if (typer &GICR_TYPER_VLPIS ) + ptr +=SZ_64K *2 ;/* Skip VLPI_base + reserved page */ + }
Quoted 1 line
VLPIS is RES0 in GICv3.
Indeed. But we still want this driver to run on GICv4 HW, even if we don't use VLPIs just yet.
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 360
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+ }while (!(typer &GICR_TYPER_LAST )); + } + +/* We couldn't even deal with ourselves... */ +WARN (true,"CPU%d: mpidr %lx has no re-distributor!\n" , +smp_processor_id (), (unsigned long )mpidr);Inconsistency with the type of mpidr (you use unsigned long long above).
Well spotted.
Quoted 23 lines
Quoted 23 lines
drivers/irqchip/irq-gic-v3.c new file +23
-0+return -ENODEV ; +} + +static void __init gic_cpu_init (void ) +{ +void __iomem *rbase; +int i; + +/* Register ourselves with the rest of the world */ +if (gic_populate_rdist ()) +return ; + +gic_enable_redist (); + + rbase =gic_data_rdist_sgi_base (); + +/* +* Set priority on PPI and SGI interrupts +*/ +for (i =0 ; i <32 ; i +=4 ) +writel_relaxed (0xa0a0a0a0 , rbase +GICR_IPRIORITYR0 + i *4 /4 ); + +/*
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 388
Quoted 5 lines
drivers/irqchip/irq-gic-v3.c new file +5
-0+* Disable all PPI interrupts, ensure all SGI interrupts are +* enabled. +*/ +writel_relaxed (0xffff0000 , rbase +GICR_ICENABLER0 ); +writel_relaxed (0x0000ffff , rbase +GICR_ISENABLER0 );Whilst there are some clear differences between this and the existing driver, you could still factor parts of this out into a common file.
OK.
Quoted 24 lines
Quoted 24 lines
drivers/irqchip/irq-gic-v3.c new file +24
-0+gic_redist_wait_for_rwp (); + +/* Enable system registers */ +gic_enable_sre (); + +/* Set priority mask register */ +gic_write_pmr (DEFAULT_PMR_VALUE ); + +/* EOI drops priority too (mode 0) */ +gic_write_ctlr (GICC_CTLR_EL1_EOImode_drop_dir); + +/* ... and let's hit the road... */ +gic_write_grpen1 (1 ); +} + +#ifdef CONFIG_SMP +static int __init gic_secondary_init (struct notifier_block *nfb, +unsigned long action,void *hcpu) +{ +if (action ==CPU_STARTING || action ==CPU_STARTING_FROZEN ) +gic_cpu_init (); +return NOTIFY_OK ; +} +
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 417
Quoted 4 lines
drivers/irqchip/irq-gic-v3.c new file +4
-0+/* +* Notifier for enabling the GIC CPU interface. Set an arbitrarily high +* priority because the GIC needs to be up before the ARM generic timers. +*/ Why must the GIC be up before the timers? Obviously, the timers won't be able to generate interrupts without the GIC initialised, but the comment implies that something worse will happen.
They will also fail to request the IRQ (no irq domain present yet), hence they will give up, and your kernel will grind to a halt...
Quoted 15 lines
Quoted 15 lines
drivers/irqchip/irq-gic-v3.c new file +15
-0+static struct notifier_block __initdata gic_cpu_notifier = { + .notifier_call = gic_secondary_init, + .priority =100 , +}; + +static u16 gic_compute_target_list (int *base_cpu,const struct cpumask *mask, +u64 cluster_id) +{ +int cpu = *base_cpu; +u64 mpidr =cpu_logical_map (cpu); +u16 tlist =0 ; + +while (cpu < nr_cpu_ids) { +/* +* If we ever get a cluster of more than 16 CPUs, just
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 436
Quoted 4 lines
drivers/irqchip/irq-gic-v3.c new file +4
-0+* scream and skip that CPU. +*/ +if (WARN_ON ((mpidr &0xff ) >=16 )) + goto out;I'm not comfortable with this. The mpidr in ARMv8 is just a unique identifier, so relying on aff0 to provide some sort of upper bound on the number of CPUs in a cluster feels extremely fragile.
And yet this is the only thing the GICv3 architecture supports. I've brought this up ages ago, and been told that "it is all fine". See the note at the bottom of 5.7.29: "this restricts distribution of SGIs to the first 16 processors of an affinity 1 cluster".
Quoted 3 lines
I think you'll need to deal with SGIs to target lists that you can't represent with a single sgi1r value (and instead batch them up and have lots of fun dealing with the result).
That's what the code already does. It is just that some processors may be out of the reach of GICv3 altogether (those with Aff0 >= 16).
Quoted 16 lines
Quoted 16 lines
drivers/irqchip/irq-gic-v3.c new file +16
-0+ + tlist |=1 << (mpidr &0xf ); + + cpu =cpumask_next (cpu, mask); + mpidr =cpu_logical_map (cpu); + +if (cluster_id != (mpidr & ~0xffUL )) { + cpu--; + goto out; + } + } +out: + *base_cpu = cpu; +return tlist; +} +
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 458
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+static void gic_send_sgi (u64 cluster_id,u16 tlist,unsigned int irq) +{ +u64 val; + + val = (cluster_id &0xff00ff0000UL ) <<16 ;/* Aff3 + Aff2 */ + val |= (cluster_id &0xff00 ) <<8 ;/* Aff1 */ Use the helpers.
Yup.
Quoted 13 lines
Quoted 13 lines
drivers/irqchip/irq-gic-v3.c new file +13
-0+ val |= irq <<24 ; + val |= tlist; + +pr_debug ("CPU%d: ICC_SGI1R_EL1 %llx\n" ,smp_processor_id (), val); +gic_write_sgi1r (val); +} + +static void gic_raise_softirq (const struct cpumask *mask,unsigned int irq) +{ +int cpu; + +if (WARN_ON (irq >=16 )) +return ;
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 477
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+ +/* +* Ensure that stores to Normal memory are visible to the +* other CPUs before issuing the IPI. +*/ +dsb ();dmb(ishst), obviously (!) ;)
Indeed ;-).
Quoted 24 lines
Quoted 24 lines
drivers/irqchip/irq-gic-v3.c new file +24
-0+ +for_each_cpu_mask (cpu, *mask) { +u64 cluster_id =cpu_logical_map (cpu) & ~0xffUL ; +u16 tlist; + + tlist =gic_compute_target_list (&cpu, mask, cluster_id); +gic_send_sgi (cluster_id, tlist, irq); + } +} + +static void gic_smp_init (void ) +{ +set_smp_cross_call (gic_raise_softirq); +register_cpu_notifier (&gic_cpu_notifier); +} + +static int gic_set_affinity (struct irq_data *d,const struct cpumask *mask_val, +bool force) +{ +unsigned int cpu =cpumask_any_and (mask_val, cpu_online_mask); +void __iomem *reg; +u64 val; + +if (gic_irq (d) <32 )
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 507
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+return -EINVAL ; + + reg =gic_dist_base (d) +GICD_IROUTER + (gic_irq (d) *8 ); + val =gic_mpidr_to_affinity (cpu_logical_map (cpu)); + +writeq_relaxed (val, reg);How do you ensure that the interrupt has actually moved?
There is no way to ensure it actually did (other than reading back from the same register, if that's what you're aiming for?).
Quoted 27 lines
Quoted 27 lines
drivers/irqchip/irq-gic-v3.c new file +27
-0+return IRQ_SET_MASK_OK ; +} +#else +#define gic_set_affinity NULL +#define gic_smp_init () do { } while(0) +#endif + +static struct irq_chip gic_chip = { + .name ="GICv3" , + .irq_mask = gic_mask_irq, + .irq_unmask = gic_unmask_irq, + .irq_eoi = gic_eoi_irq, + .irq_set_type = gic_set_type, + .irq_retrigger = gic_retrigger, + .irq_set_affinity = gic_set_affinity, +}; + +static int gic_irq_domain_map (struct irq_domain *d,unsigned int irq, +irq_hw_number_t hw) +{ +if (hw <32 ) { +irq_set_percpu_devid (irq); +irq_set_chip_and_handler (irq, &gic_chip, + handle_percpu_devid_irq); +set_irq_flags (irq,IRQF_VALID |IRQF_NOAUTOEN ); + }else { + irq_set_chip_and_handler(irq, &gic_chip,
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 541
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+ handle_fasteoi_irq); +set_irq_flags (irq,IRQF_VALID |IRQF_PROBE ); + } +irq_set_chip_data (irq, d->host_data); +return 0 ; +}More copied GICv2 code...
I'll move it out, then.
Quoted 18 lines
Quoted 18 lines
drivers/irqchip/irq-gic-v3.c new file +18
-0+static int gic_irq_domain_xlate (struct irq_domain *d, +struct device_node *controller, +const u32 *intspec,unsigned int intsize, +unsigned long *out_hwirq,unsigned int *out_type) +{ +if (d->of_node != controller) +return -EINVAL ; +if (intsize <3 ) +return -EINVAL ; + +switch (intspec[0 ]) { +case 0 :/* SPI */ + *out_hwirq = intspec[1 ] +32 ; +break ; +case 1 :/* PPI */ + *out_hwirq = intspec[1 ] +16 ; +break ; +default :
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 566
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+return -EINVAL ; + } + + *out_type = intspec[2 ] &IRQ_TYPE_SENSE_MASK ; +return 0 ; +}... and some slightly mutated but functionally identical GICv2 code ;)
Same.
Quoted 36 lines
Quoted 36 lines
drivers/irqchip/irq-gic-v3.c new file +36
-0+static const struct irq_domain_ops gic_irq_domain_ops = { + .map = gic_irq_domain_map, + .xlate = gic_irq_domain_xlate, +}; + +static int __init gic_of_init (struct device_node *node,struct device_node *parent) +{ +void __iomem *dist_base; +void __iomem **redist_base; +u64 redist_stride; +u32 redist_regions; +u32 reg; +int gic_irqs; +int err; +int i; + + dist_base =of_iomap (node,0 ); +if (!dist_base) { +pr_warn ("%s: unable to map gic dist registers\n" , + node->full_name); +return -ENXIO ; + } + + reg =readl_relaxed (dist_base +GICD_PIDR2 ) &GIC_PIDR2_ARCH_MASK ; +if (reg !=0x30 && reg !=0x40 ) { +pr_warn ("%s: no distributor detected, giving up\n" , + node->full_name); + err = -ENODEV ; + goto out_unmap_dist; + } + +if (of_property_read_u32 (node,"#redistributor-regions" , &redist_regions)) + redist_regions =1 ; + + redist_base =kzalloc (sizeof (*redist_base) * redist_regions,GFP_KERNEL ); +if (!redist_base) {
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 606
Quoted 3 lines
drivers/irqchip/irq-gic-v3.c new file +3
-0+ err = -ENOMEM ; + goto out_unmap_dist; + }Can you use the devm_* managed resource helpers in irqchip init functions? I see a couple of drivers doing it, but it's not clear to me whether things are actually freed on failure. Anyway, if you can, then it would clean this up a bit.
The device model is not available at the stage where we probe the GIC. If it was, I'd have converted the GIC to be a proper driver a long time ago...
The only two users are secondary interrupt controllers, which are probed through DT, when the whole infrastructure is up and running.
Quoted 40 lines
Quoted 40 lines
drivers/irqchip/irq-gic-v3.c new file +40
-0+for (i =0 ; i < redist_regions; i++) { + redist_base[i] =of_iomap (node,1 + i); +if (!redist_base[i]) { +pr_warn ("%s: couldn't map region %d\n" , + node->full_name, i); + err = -ENODEV ; + goto out_unmap_rdist; + } + } + +if (of_property_read_u64 (node,"redistributor-stride" , &redist_stride)) + redist_stride =0 ; + + gic_data.dist_base = dist_base; + gic_data.redist_base = redist_base; + gic_data.redist_regions = redist_regions; + gic_data.redist_stride = redist_stride; + +/* +* Find out how many interrupts are supported. +* The GIC only supports up to 1020 interrupt sources (SGI+PPI+SPI) +*/ + gic_irqs =readl_relaxed (gic_data.dist_base +GICD_TYPER ) &0x1f ; + gic_irqs = (gic_irqs +1 ) *32 ; +if (gic_irqs >1020 ) + gic_irqs =1020 ; + gic_data.irq_nr = gic_irqs; + + gic_data.domain =irq_domain_add_linear (node, gic_irqs -16 , + &gic_irq_domain_ops, &gic_data); + gic_data.rdist =alloc_percpu (typeof (*gic_data.rdist)); + +if (WARN_ON (!gic_data.domain) ||WARN_ON (!gic_data.rdist)) { + err = -ENOMEM ; + goto out_free; + } + +set_handle_irq (gic_handle_irq); + +gic_smp_init ();
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 653
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+gic_dist_init (); +gic_cpu_init (); + +return 0 ; + +out_free:remove the
irq_domainhere?
Yup, good point.
Quoted 187 lines
Quoted 187 lines
drivers/irqchip/irq-gic-v3.c new file +12
-0+free_percpu (gic_data.rdist); +out_unmap_rdist: +for (i =0 ; i < redist_regions; i++) +if (redist_base[i]) +iounmap (redist_base[i]); +kfree (redist_base); +out_unmap_dist: +iounmap (dist_base); +return err; +} + +IRQCHIP_DECLARE (gic_v3,"arm,gic-v3" , gic_of_init);include/linux/irqchip/arm-gic-v3.h new file +169
-0@@ -0,0 +1,190 @@ +/* +* Copyright (C) 2013 ARM Limited, All Rights Reserved. +* Author: Marc Zyngier <[email protected]> +* +* +* This program is free software; you can redistribute it and/or modify +* it under the terms of the GNU General Public License version 2 as +* published by the Free Software Foundation. +* +* This program is distributed in the hope that it will be useful, +* but WITHOUT ANY WARRANTY; without even the implied warranty of +* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +* GNU General Public License for more details. +* +* You should have received a copy of the GNU General Public License +* along with this program. If not, see <http://www.gnu.org/licenses/>. +*/ +#ifndef __LINUX_IRQCHIP_ARM_GIC_V3_H +#define __LINUX_IRQCHIP_ARM_GIC_V3_H + +/* +* Distributor registers. We assume we're running non-secure, with ARE +* being set. Secure-only and non-ARE registers are not described. +*/ +#define GICD_CTLR 0x0000 +#define GICD_TYPER 0x0004 +#define GICD_IIDR 0x0008 +#define GICD_STATUSR 0x0010 +#define GICD_SETSPI_NSR 0x0040 +#define GICD_CLRSPI_NSR 0x0048 +#define GICD_SETSPI_SR 0x0050 +#define GICD_CLRSPI_SR 0x0058 +#define GICD_SEIR 0x0068 +#define GICD_ISENABLER 0x0100 +#define GICD_ICENABLER 0x0180 +#define GICD_ISPENDR 0x0200 +#define GICD_ICPENDR 0x0280 +#define GICD_ISACTIVER 0x0300 +#define GICD_ICACTIVER 0x0380 +#define GICD_IPRIORITYR 0x0400 +#define GICD_ICFGR 0x0C00 +#define GICD_IROUTER 0x6000 +#define GICD_PIDR2 0xFFE8 + +#define GICD_CTLR_RWP (1U << 31) +#define GICD_CTLR_ARE_NS (1U << 4) +#define GICD_CTLR_ENABLE_G1A (1U << 1) +#define GICD_CTLR_ENABLE_G1 (1U << 0) + +#define GICD_IROUTER_SPI_MODE_ONE (0U << 31) +#define GICD_IROUTER_SPI_MODE_ANY (1U << 31) + +#define GIC_PIDR2_ARCH_MASK 0xf0 + +/* +* Re-Distributor registers, offsets from RD_base +*/ +#define GICR_CTLR GICD_CTLR +#define GICR_IIDR 0x0004 +#define GICR_TYPER 0x0008 +#define GICR_STATUSR GICD_STATUSR +#define GICR_WAKER 0x0014 +#define GICR_SETLPIR 0x0040 +#define GICR_CLRLPIR 0x0048 +#define GICR_SEIR GICD_SEIR +#define GICR_PROPBASER 0x0070 +#define GICR_PENDBASER 0x0078 +#define GICR_INVLPIR 0x00A0 +#define GICR_INVALLR 0x00B0 +#define GICR_SYNCR 0x00C0 +#define GICR_MOVLPIR 0x0100 +#define GICR_MOVALLR 0x0110 +#define GICR_PIDR2 GICD_PIDR2 + +#define GICR_WAKER_ProcessorSleep (1U << 1) +#define GICR_WAKER_ChildrenAsleep (1U << 2) + +/* +* Re-Distributor registers, offsets from SGI_base +*/ +#define GICR_ISENABLER0 GICD_ISENABLER +#define GICR_ICENABLER0 GICD_ICENABLER +#define GICR_ISPENDR0 GICD_ISPENDR +#define GICR_ICPENDR0 GICD_ICPENDR +#define GICR_ISACTIVER0 GICD_ISACTIVER +#define GICR_ICACTIVER0 GICD_ICACTIVER +#define GICR_IPRIORITYR0 GICD_IPRIORITYR +#define GICR_ICFGR0 GICD_ICFGR + +#define GICR_TYPER_VLPIS (1U << 1) +#define GICR_TYPER_LAST (1U << 4) + +/* +* CPU interface registers +*/ +#define GICC_CTLR_EL1_EOImode_drop_dir (0U << 1) +#define GICC_CTLR_EL1_EOImode_drop (1U << 1) +#define GICC_SRE_EL1_SRE (1U << 0) + +/* +* Hypervisor interface registers (SRE only) +*/ +#define GICH_LR_VIRTUAL_ID_MASK ((1UL << 32) - 1) + +#define GICH_LR_EOI (1UL << 41) +#define GICH_LR_GROUP (1UL << 60) +#define GICH_LR_STATE (3UL << 62) +#define GICH_LR_PENDING_BIT (1UL << 62) +#define GICH_LR_ACTIVE_BIT (1UL << 63) + +#define GICH_MISR_EOI (1 << 0) +#define GICH_MISR_U (1 << 1) + +#define GICH_HCR_EN (1 << 0) +#define GICH_HCR_UIE (1 << 1) + +#define GICH_VMCR_CTLR_SHIFT 0 +#define GICH_VMCR_CTLR_MASK (0x21f << GICH_VMCR_CTLR_SHIFT) +#define GICH_VMCR_BPR1_SHIFT 18 +#define GICH_VMCR_BPR1_MASK (7 << GICH_VMCR_BPR1_SHIFT) +#define GICH_VMCR_BPR0_SHIFT 21 +#define GICH_VMCR_BPR0_MASK (7 << GICH_VMCR_BPR0_SHIFT) +#define GICH_VMCR_PMR_SHIFT 24 +#define GICH_VMCR_PMR_MASK (0xffUL << GICH_VMCR_PMR_SHIFT) + +#define ICC_EOIR1_EL1 S3_0_C12_C12_1 +#define ICC_IAR1_EL1 S3_0_C12_C12_0 +#define ICC_SGI1R_EL1 S3_0_C12_C11_5 +#define ICC_PMR_EL1 S3_0_C4_C6_0 +#define ICC_CTLR_EL1 S3_0_C12_C12_4 +#define ICC_SRE_EL1 S3_0_C12_C12_5 +#define ICC_GRPEN1_EL1 S3_0_C12_C12_7 + +#define ICC_SRE_EL2 S3_4_C12_C9_5 + +#define ICH_VSEIR_EL2 S3_4_C12_C9_4 +#define ICH_HCR_EL2 S3_4_C12_C11_0 +#define ICH_VTR_EL2 S3_4_C12_C11_1 +#define ICH_MISR_EL2 S3_4_C12_C11_2 +#define ICH_EISR_EL2 S3_4_C12_C11_3 +#define ICH_ELSR_EL2 S3_4_C12_C11_5 +#define ICH_VMCR_EL2 S3_4_C12_C11_7 + +#define __LR0_EL2 (x) S3_4_C12_C12_ ## x +#define __LR8_EL2 (x) S3_4_C12_C13_ ## x + +#define ICH_LR0_EL2 __LR0_EL2(0) +#define ICH_LR1_EL2 __LR0_EL2(1) +#define ICH_LR2_EL2 __LR0_EL2(2) +#define ICH_LR3_EL2 __LR0_EL2(3) +#define ICH_LR4_EL2 __LR0_EL2(4) +#define ICH_LR5_EL2 __LR0_EL2(5) +#define ICH_LR6_EL2 __LR0_EL2(6) +#define ICH_LR7_EL2 __LR0_EL2(7) +#define ICH_LR8_EL2 __LR8_EL2(0) +#define ICH_LR9_EL2 __LR8_EL2(1) +#define ICH_LR10_EL2 __LR8_EL2(2) +#define ICH_LR11_EL2 __LR8_EL2(3) +#define ICH_LR12_EL2 __LR8_EL2(4) +#define ICH_LR13_EL2 __LR8_EL2(5) +#define ICH_LR14_EL2 __LR8_EL2(6) +#define ICH_LR15_EL2 __LR8_EL2(7) + +#define __AP0Rx_EL2 (x) S3_4_C12_C8_ ## x +#define ICH_AP0R0_EL2 __AP0Rx_EL2(0) +#define ICH_AP0R1_EL2 __AP0Rx_EL2(1) +#define ICH_AP0R2_EL2 __AP0Rx_EL2(2) +#define ICH_AP0R3_EL2 __AP0Rx_EL2(3) +
Quoted 8 lines
include/linux/irqchip/arm-gic-v3.h line 174
Quoted 5 lines
include/linux/irqchip/arm-gic-v3.h new file +5
-0+#define __AP1Rx_EL2 (x) S3_4_C12_C9_ ## x +#define ICH_AP1R0_EL2 __AP1Rx_EL2(0) +#define ICH_AP1R1_EL2 __AP1Rx_EL2(1) +#define ICH_AP1R2_EL2 __AP1Rx_EL2(2) +#define ICH_AP1R3_EL2 __AP1Rx_EL2(3)I'm going to rely on Christoffer having looked at all these magic numbers (he at least had ice cream!).
Quite! ;-)
Quoted 5 lines
Quoted 5 lines
include/linux/irqchip/arm-gic-v3.h new file +5
-0+#ifndef __ASSEMBLY__ + +#include <linux/stringify.h> + +#define reg (x) __stringify(x)
Quoted 8 lines
include/linux/irqchip/arm-gic-v3.h line 186
Quoted 6 lines
include/linux/irqchip/arm-gic-v3.h new file +6
-0+ +static inline void gic_write_eoir (u64 irq) +{ + asm volatile("msr " reg(ICC_EOIR1_EL1) ", %0" : : "r" (irq)); +isb (); +}Why is this in the header file?
For the benefit of the ITS part of the driver, which needs to EOI its own stuff.
Thanks for the excellent review!
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
PATCH 1/18 arm64: initial support for GICv3
Quoted 6 lines · Marc Zyngier
Quoted 2 lines · Will Deacon
Not sure if you care, but readq is a show-stopper if you want to build this with
arch/arm/.I realized that recently. Any reason why we don't have writeq/readq on LPAE-capable stuff (other than the obvious runtime detection madness)?
I'm not sure that you'd be guaranteed single-copy atomicity for such an access, which is probably assumed by readq/writeq.
Quoted 16 lines
Quoted 16 lines
Quoted 15 lines
drivers/irqchip/irq-gic-v3.c new file +15
-0+if ((typer >>32 ) == aff) { +gic_data_rdist_rd_base () = ptr; +pr_info ("CPU%d: found redistributor %llx @%p\n" , +smp_processor_id (), + (unsigned long long ) mpidr, ptr); +return 0 ; + } + +if (gic_data.redist_stride) { + ptr += gic_data.redist_stride; + }else { + ptr +=SZ_64K *2 ;/* Skip RD_base + SGI_base */ +if (typer &GICR_TYPER_VLPIS ) + ptr +=SZ_64K *2 ;/* Skip VLPI_base + reserved page */ + }
Quoted 4 lines
Quoted 1 line
VLPIS is RES0 in GICv3.
Indeed. But we still want this driver to run on GICv4 HW, even if we don't use VLPIs just yet.
Ok, just mention that in the commit then.
Quoted 17 lines
Quoted 8 lines
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c new file +8
-0+static int gic_set_affinity (struct irq_data *d,const struct cpumask *mask_val, +bool force) +{ +unsigned int cpu =cpumask_any_and (mask_val, cpu_online_mask); +void __iomem *reg; +u64 val; + +if (gic_irq (d) <32 )Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 507
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+return -EINVAL ; + + reg =gic_dist_base (d) +GICD_IROUTER + (gic_irq (d) *8 ); + val =gic_mpidr_to_affinity (cpu_logical_map (cpu)); + +writeq_relaxed (val, reg);How do you ensure that the interrupt has actually moved?
Quoted 2 lines
There is no way to ensure it actually did (other than reading back from the same register, if that's what you're aiming for?).
I guess it depends how angry Linux will get if the interrupt fires on the wrong CPU. If the interrupt is pending when the moving takes place, will it be asserted on the new target?
Will
PATCH 1/18 arm64: initial support for GICv3
Hi Marc,
I am from Cavium and we have tried your GICv3 driver on our internal emulation platform. We have booted systems with >8 cores and tested basic SGIs, PPIs and SPIs.
However we haven't exercised LPIs and ITS yet. We plan to do this soon.
If you have a new version of the patches coming out soon, we can exercise that as well on our platform.
regards,
Radha Mohan
Quoted 63 lines · Will Deacon
Quoted 6 lines · Marc Zyngier
Quoted 2 lines · Will Deacon
Not sure if you care, but readq is a show-stopper if you want to build this with
arch/arm/.I realized that recently. Any reason why we don't have writeq/readq on LPAE-capable stuff (other than the obvious runtime detection madness)?
I'm not sure that you'd be guaranteed single-copy atomicity for such an access, which is probably assumed by readq/writeq.
Quoted 16 lines
Quoted 16 lines
Quoted 15 lines
drivers/irqchip/irq-gic-v3.c new file +15
-0+if ((typer >>32 ) == aff) { +gic_data_rdist_rd_base () = ptr; +pr_info ("CPU%d: found redistributor %llx @%p\n" , +smp_processor_id (), + (unsigned long long ) mpidr, ptr); +return 0 ; + } + +if (gic_data.redist_stride) { + ptr += gic_data.redist_stride; + }else { + ptr +=SZ_64K *2 ;/* Skip RD_base + SGI_base */ +if (typer &GICR_TYPER_VLPIS ) + ptr +=SZ_64K *2 ;/* Skip VLPI_base + reserved page */ + }Quoted 4 lines
Quoted 1 line
VLPIS is RES0 in GICv3.
Indeed. But we still want this driver to run on GICv4 HW, even if we don't use VLPIs just yet.
Ok, just mention that in the commit then.
Quoted 17 lines
Quoted 8 lines
Quoted 8 lines
drivers/irqchip/irq-gic-v3.c new file +8
-0+static int gic_set_affinity (struct irq_data *d,const struct cpumask *mask_val, +bool force) +{ +unsigned int cpu =cpumask_any_and (mask_val, cpu_online_mask); +void __iomem *reg; +u64 val; + +if (gic_irq (d) <32 )Quoted 8 lines
drivers/irqchip/irq-gic-v3.c line 507
Quoted 6 lines
drivers/irqchip/irq-gic-v3.c new file +6
-0+return -EINVAL ; + + reg =gic_dist_base (d) +GICD_IROUTER + (gic_irq (d) *8 ); + val =gic_mpidr_to_affinity (cpu_logical_map (cpu)); + +writeq_relaxed (val, reg);How do you ensure that the interrupt has actually moved?
Quoted 2 lines
There is no way to ensure it actually did (other than reading back from the same register, if that's what you're aiming for?).
I guess it depends how angry Linux will get if the interrupt fires on the wrong CPU. If the interrupt is pending when the moving takes place, will it be asserted on the new target?
Will
_______________________________________________
kvmarm mailing list
kvmarm at lists.cs.columbia.edu
https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
PATCH 2/18 arm64: GICv3 device tree binding documentation
In reply to Marc ZyngierAdd the necessary documentation to support GICv3.
--- Documentation/devicetree/bindings/arm/gic-v3.txt | 81 ++++++++++++++++++++++++ 1 file changed, 81 insertions(+) create mode 100644 Documentation/devicetree/bindings/arm/gic-v3.txt
Documentation/devicetree/bindings/arm/gic-v3.txt new file +81 -0
@@ -0,0 +1,81 @@ +* ARM Generic Interrupt Controller, version 3 + +AArch64 SMP cores are often associated with a GICv3, providing private +peripheral interrupts (PPI), shared peripheral interrupts (SPI), +software generated interrupts (SGI), and locality-specific peripheral +Interrupts (LPI). + +Main node required properties: + +- compatible : should at least contain "arm,gic-v3". +- interrupt-controller : Identifies the node as an interrupt controller +- #interrupt-cells : Specifies the number of cells needed to encode an + interrupt source. Must be a single cell with a value of at least 3. + + The 1st cell is the interrupt type; 0 for SPI interrupts, 1 for PPI + interrupts. Other values are reserved for future use. + + The 2nd cell contains the interrupt number for the interrupt type. + SPI interrupts are in the range [0-987]. PPI interrupts are in the + range [0-15]. + + The 3rd cell is the flags, encoded as follows: + bits[3:0] trigger type and level flags. + 1 = edge triggered + 2 = edge triggered (deprecated, for compatibility with GICv2) + 4 = level triggered + 8 = level triggered (deprecated, for compatibility with GICv2) + + Cells 4 and beyond are reserved for future use. Where the 1st cell + has a value of 0 or 1, cells 4 and beyond act as padding, and may be + ignored. It is recommended that padding cells have a value of 0. + +- reg : Specifies base physical address(s) and size of the GIC + registers, in the following order: + - GIC Distributor interface (GICD) + - GIC Redistributors (GICR), one range per redistributor region + - GIC CPU interface (GICC) + - GIC Hypervisor interface (GICH) + - GIC Virtual CPU interface (GICV) + + GICC, GICH and GICV are optional. + +- interrupts : Interrupt source of the VGIC maintenance interrupt. + +Optional + +- redistributor-stride : If using padding pages, specifies the stride + of consecutive redistributors. Must be a multiple of 64kB. + +- #redistributor-regions: The number of independent contiguous regions + occupied by the redistributors. Required if more than one such + region is present. + +Examples: + + gic: interrupt-controller at 2cf00000 { + compatible = "arm,gic-v3"; + #interrupt-cells = <3>; + interrupt-controller; + reg = <0x0 0x2f000000 0 0x10000>, // GICD + <0x0 0x2f100000 0 0x200000>, // GICR + <0x0 0x2c000000 0 0x2000>, // GICC + <0x0 0x2c010000 0 0x2000>, // GICH + <0x0 0x2c020000 0 0x2000>; // GICV + interrupts = <1 9 4>; + }; + + gic: interrupt-controller at 2c010000 { + compatible = "arm,gic-v3"; + #interrupt-cells = <3>; + interrupt-controller; + redistributor-stride = 0x40000; // 256kB stride + #redistributor-regions = <2>; + reg = <0x0 0x2c010000 0 0x10000>, // GICD + <0x0 0x2d000000 0 0x800000>, // GICR 1: CPUs 0-31 + <0x0 0x2e000000 0 0x800000>; // GICR 2: CPUs 32-63 + <0x0 0x2c040000 0 0x2000>, // GICC + <0x0 0x2c060000 0 0x2000>, // GICH + <0x0 0x2c080000 0 0x2000>; // GICV + interrupts = <1 9 4>; + };
Signature · 1 line
1.8.3.4
PATCH 2/18 arm64: GICv3 device tree binding documentation
Hi Marc
Quoted 5 lines · Marc Zyngier <marc.zyngier <at> arm.com> writes:
Documentation/devicetree/bindings/arm/gic-v3.txt new file +5
-0+ +AArch64 SMP cores are often associated with a GICv3, providing private +peripheral interrupts (PPI), shared peripheral interrupts (SPI), +software generated interrupts (SGI), and locality-specific peripheral +Interrupts (LPI).
Documentation/devicetree/bindings/arm/gic-v3.txt line 16
Quoted 10 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +10
-0+ +Main node required properties: + +- compatible : should at least contain "arm,gic-v3". +- interrupt-controller : Identifies the node as an interrupt controller +- #interrupt-cells : Specifies the number of cells needed to encode an + interrupt source. Must be a single cell with a value of at least 3. + + The 1st cell is the interrupt type; 0 for SPI interrupts, 1 for PPI + interrupts. Other values are reserved for future use.
These values are defined in "include/dt-bindings/interrupt-controller/arm-gic.h" maybe we should start mentioning that here and encourage future device treese to use those defines to improve readability.
Documentation/devicetree/bindings/arm/gic-v3.txt line 27
Quoted 11 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +11
-0+ + The 2nd cell contains the interrupt number for the interrupt type. + SPI interrupts are in the range [0-987]. PPI interrupts are in the + range [0-15]. + + The 3rd cell is the flags, encoded as follows: + bits[3:0] trigger type and level flags. + 1 = edge triggered + 2 = edge triggered (deprecated, for compatibility with GICv2) + 4 = level triggered + 8 = level triggered (deprecated, for compatibility with GICv2)
Similar to the above comment "include/dt-bindings/interrupt-controller/irq.h" defines the trigger type and level flags. Although this file currently contains the GICv2 bindings, we could update them.
Thanks
Arnab
Quoted 17 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +17
-0+ + Cells 4 and beyond are reserved for future use. Where the 1st cell + has a value of 0 or 1, cells 4 and beyond act as padding, and may be + ignored. It is recommended that padding cells have a value of 0. + +- reg : Specifies base physical address(s) and size of the GIC + registers, in the following order: + - GIC Distributor interface (GICD) + - GIC Redistributors (GICR), one range per redistributor region + - GIC CPU interface (GICC) + - GIC Hypervisor interface (GICH) + - GIC Virtual CPU interface (GICV) + + GICC, GICH and GICV are optional. + +- interrupts : Interrupt source of the VGIC maintenance interrupt. +
PATCH 2/18 arm64: GICv3 device tree binding documentation
Hi Arnab,
Quoted 16 lines · Arnab Basu
Hi Marc
Quoted 12 lines · Marc Zyngier <marc.zyngier <at> arm.com> writes:
Documentation/devicetree/bindings/arm/gic-v3.txt new file +12
-0+ +AArch64 SMP cores are often associated with a GICv3, providing private +peripheral interrupts (PPI), shared peripheral interrupts (SPI), +software generated interrupts (SGI), and locality-specific peripheral +Interrupts (LPI). + +Main node required properties: + +- compatible : should at least contain "arm,gic-v3". +- interrupt-controller : Identifies the node as an interrupt controller +- #interrupt-cells : Specifies the number of cells needed to encode an + interrupt source. Must be a single cell with a value of at least 3.
Quoted 8 lines
Documentation/devicetree/bindings/arm/gic-v3.txt line 16
Quoted 3 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +3
-0+ + The 1st cell is the interrupt type; 0 for SPI interrupts, 1 for PPI + interrupts. Other values are reserved for future use.These values are defined in "
include/dt-bindings/interrupt-controller/arm-gic.h" maybe we should start mentioning that here and encourage future device treese to use those defines to improve readability.
It may improve readability, but it makes the definition rely on something else. Definition and usage are two different things, and I want the definition to be completely self-contained and non ambiguous.
In DTS files, people can use whatever macro they decide, and it is their problem. They will even have out of tree DTS files, for which the include file is not available.
Quoted 17 lines
Documentation/devicetree/bindings/arm/gic-v3.txt line 27
Quoted 12 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +11
-0+ + The 2nd cell contains the interrupt number for the interrupt type. + SPI interrupts are in the range [0-987]. PPI interrupts are in the + range [0-15]. + + The 3rd cell is the flags, encoded as follows: + bits[3:0] trigger type and level flags. + 1 = edge triggered + 2 = edge triggered (deprecated, for compatibility with GICv2) + 4 = level triggered + 8 = level triggered (deprecated, for compatibility with GICv2)Similar to the above comment "
include/dt-bindings/interrupt-controller/irq.h" defines the trigger type and level flags. Although this file currently contains the GICv2 bindings, we could update them.
Same as above.
Thanks for the review!
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
PATCH 2/18 arm64: GICv3 device tree binding documentation
Quoted 28 lines · Marc Zyngier
Hi Arnab,
Quoted 16 lines · Arnab Basu
Hi Marc
Quoted 12 lines · Marc Zyngier <marc.zyngier <at> arm.com> writes:
Documentation/devicetree/bindings/arm/gic-v3.txt new file +12
-0+ +AArch64 SMP cores are often associated with a GICv3, providing private +peripheral interrupts (PPI), shared peripheral interrupts (SPI), +software generated interrupts (SGI), and locality-specific peripheral +Interrupts (LPI). + +Main node required properties: + +- compatible : should at least contain "arm,gic-v3". +- interrupt-controller : Identifies the node as an interrupt controller +- #interrupt-cells : Specifies the number of cells needed to encode an + interrupt source. Must be a single cell with a value of at least 3.Quoted 8 lines
Documentation/devicetree/bindings/arm/gic-v3.txt line 16
Quoted 3 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +3
-0+ + The 1st cell is the interrupt type; 0 for SPI interrupts, 1 for PPI + interrupts. Other values are reserved for future use.These values are defined in "
include/dt-bindings/interrupt-controller/arm-gic.h" maybe we should start mentioning that here and encourage future device treese to use those defines to improve readability.
Quoted 7 lines
It may improve readability, but it makes the definition rely on something else. Definition and usage are two different things, and I want the definition to be completely self-contained and non ambiguous.
In DTS files, people can use whatever macro they decide, and it is their problem. They will even have out of tree DTS files, for which the include file is not available.
Agreed.
Rob
PATCH 2/18 arm64: GICv3 device tree binding documentation
In reply to Marc ZyngierQuoted 32 lines · Marc Zyngier
Add the necessary documentation to support GICv3.
Acked-by Catalin Marinas <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- Documentation/devicetree/bindings/arm/gic-v3.txt | 81 ++++++++++++++++++++++++ 1 file changed, 81 insertions(+) create mode 100644 Documentation/devicetree/bindings/arm/gic-v3.txtDocumentation/devicetree/bindings/arm/gic-v3.txt new file +17
-0@@ -0,0 +1,81 @@ +* ARM Generic Interrupt Controller, version 3 + +AArch64 SMP cores are often associated with a GICv3, providing private +peripheral interrupts (PPI), shared peripheral interrupts (SPI), +software generated interrupts (SGI), and locality-specific peripheral +Interrupts (LPI). + +Main node required properties: + +- compatible : should at least contain "arm,gic-v3". +- interrupt-controller : Identifies the node as an interrupt controller +- #interrupt-cells : Specifies the number of cells needed to encode an + interrupt source. Must be a single cell with a value of at least 3. + + The 1st cell is the interrupt type; 0 for SPI interrupts, 1 for PPI + interrupts. Other values are reserved for future use. +
Documentation/devicetree/bindings/arm/gic-v3.txt line 27
Quoted 10 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +10
-0+ The 2nd cell contains the interrupt number for the interrupt type. + SPI interrupts are in the range [0-987]. PPI interrupts are in the + range [0-15]. + + The 3rd cell is the flags, encoded as follows: + bits[3:0] trigger type and level flags. + 1 = edge triggered + 2 = edge triggered (deprecated, for compatibility with GICv2) + 4 = level triggered + 8 = level triggered (deprecated, for compatibility with GICv2)
I don't really think compatibility is needed here. I'd just list 1 and 4.
Documentation/devicetree/bindings/arm/gic-v3.txt line 31
Quoted 4 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +4
-0+ + Cells 4 and beyond are reserved for future use. Where the 1st cell + has a value of 0 or 1, cells 4 and beyond act as padding, and may be + ignored. It is recommended that padding cells have a value of 0.
What future use? LPIs I suppose.
Otherwise, looks fine to me:
Rob
PATCH 2/18 arm64: GICv3 device tree binding documentation
Quoted 37 lines · Rob Herring
Quoted 36 lines · Marc Zyngier
Add the necessary documentation to support GICv3.
Acked-by Catalin Marinas <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- Documentation/devicetree/bindings/arm/gic-v3.txt | 81 ++++++++++++++++++++++++ 1 file changed, 81 insertions(+) create mode 100644 Documentation/devicetree/bindings/arm/gic-v3.txtDocumentation/devicetree/bindings/arm/gic-v3.txt new file +21
-0@@ -0,0 +1,81 @@ +* ARM Generic Interrupt Controller, version 3 + +AArch64 SMP cores are often associated with a GICv3, providing private +peripheral interrupts (PPI), shared peripheral interrupts (SPI), +software generated interrupts (SGI), and locality-specific peripheral +Interrupts (LPI). + +Main node required properties: + +- compatible : should at least contain "arm,gic-v3". +- interrupt-controller : Identifies the node as an interrupt controller +- #interrupt-cells : Specifies the number of cells needed to encode an + interrupt source. Must be a single cell with a value of at least 3. + + The 1st cell is the interrupt type; 0 for SPI interrupts, 1 for PPI + interrupts. Other values are reserved for future use. + + The 2nd cell contains the interrupt number for the interrupt type. + SPI interrupts are in the range [0-987]. PPI interrupts are in the + range [0-15]. +
Quoted 8 lines
Documentation/devicetree/bindings/arm/gic-v3.txt line 27
Quoted 6 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +6
-0+ The 3rd cell is the flags, encoded as follows: + bits[3:0] trigger type and level flags. + 1 = edge triggered + 2 = edge triggered (deprecated, for compatibility with GICv2) + 4 = level triggered + 8 = level triggered (deprecated, for compatibility with GICv2)I don't really think compatibility is needed here. I'd just list 1 and 4.
Fair enough.
Quoted 6 lines
Documentation/devicetree/bindings/arm/gic-v3.txt line 31
Quoted 4 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +4
-0+ + Cells 4 and beyond are reserved for future use. Where the 1st cell + has a value of 0 or 1, cells 4 and beyond act as padding, and may be + ignored. It is recommended that padding cells have a value of 0.What future use? LPIs I suppose.
That's a possibility, for non-PCI devices. I'd rather not do that though, but I thought that leaving some room for (wacky) expansion.
Quoted 3 lines
Otherwise, looks fine to me:
Acked-by Rob Herring <[email protected]>
Thanks!
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
PATCH 2/18 arm64: GICv3 device tree binding documentation
In reply to Marc ZyngierQuoted 15 lines · Marc Zyngier
Add the necessary documentation to support GICv3.
Acked-by Catalin Marinas <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- Documentation/devicetree/bindings/arm/gic-v3.txt | 81 ++++++++++++++++++++++++ 1 file changed, 81 insertions(+) create mode 100644 Documentation/devicetree/bindings/arm/gic-v3.txtDocumentation/devicetree/bindings/arm/gic-v3.txt new file +0
-0
Documentation/devicetree/bindings/arm/gic-v3.txt line 6
Quoted 6 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +6
-0@@ -0,0 +1,81 @@ +* ARM Generic Interrupt Controller, version 3 + +AArch64 SMP cores are often associated with a GICv3, providing private +peripheral interrupts (PPI), shared peripheral interrupts (SPI), +software generated interrupts (SGI), and locality-specific peripheral +Interrupts (LPI).
Super-over-ridiculous-nit: capitalize the definitions, e.g. Private Peripheral Interrupts (PPI)
Quoted 15 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +15
-0+ +Main node required properties: + +- compatible : should at least contain "arm,gic-v3". +- interrupt-controller : Identifies the node as an interrupt controller +- #interrupt-cells : Specifies the number of cells needed to encode an + interrupt source. Must be a single cell with a value of at least 3. + + The 1st cell is the interrupt type; 0 for SPI interrupts, 1 for PPI + interrupts. Other values are reserved for future use. + + The 2nd cell contains the interrupt number for the interrupt type. + SPI interrupts are in the range [0-987]. PPI interrupts are in the + range [0-15]. +
Documentation/devicetree/bindings/arm/gic-v3.txt line 31
Quoted 10 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +10
-0+ The 3rd cell is the flags, encoded as follows: + bits[3:0] trigger type and level flags. + 1 = edge triggered + 2 = edge triggered (deprecated, for compatibility with GICv2) + 4 = level triggered + 8 = level triggered (deprecated, for compatibility with GICv2) + + Cells 4 and beyond are reserved for future use. Where the 1st cell + has a value of 0 or 1, cells 4 and beyond act as padding, and may be + ignored. It is recommended that padding cells have a value of 0.
another-super-nit: Saying "If the 1st cell" or "When the 1st cell" sounds more clear to me, but I may be wrong.
Quoted 7 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +7
-0+ +- reg : Specifies base physical address(s) and size of the GIC + registers, in the following order: + - GIC Distributor interface (GICD) + - GIC Redistributors (GICR), one range per redistributor region + - GIC CPU interface (GICC) + - GIC Hypervisor interface (GICH)
Documentation/devicetree/bindings/arm/gic-v3.txt line 48
Quoted 10 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +10
-0+ - GIC Virtual CPU interface (GICV) + + GICC, GICH and GICV are optional. + +- interrupts : Interrupt source of the VGIC maintenance interrupt. + +Optional + +- redistributor-stride : If using padding pages, specifies the stride + of consecutive redistributors. Must be a multiple of 64kB.
How is this going to be used? What would the size of the padding then be? I couldn't find anything about this in the specs.
Or does this mean the total size of the set of contiguous pages for each of the redistributors? That seems to make the math in the example below work.
Quoted 14 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +14
-0+ +- #redistributor-regions: The number of independent contiguous regions + occupied by the redistributors. Required if more than one such + region is present. + +Examples: + + gic: interrupt-controller at 2cf00000 { + compatible = "arm,gic-v3"; + #interrupt-cells = <3>; + interrupt-controller; + reg = <0x0 0x2f000000 0 0x10000>, // GICD + <0x0 0x2f100000 0 0x200000>, // GICR + <0x0 0x2c000000 0 0x2000>, // GICC
Documentation/devicetree/bindings/arm/gic-v3.txt line 72
Quoted 10 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +10
-0+ <0x0 0x2c010000 0 0x2000>, // GICH + <0x0 0x2c020000 0 0x2000>; // GICV + interrupts = <1 9 4>; + }; + + gic: interrupt-controller at 2c010000 { + compatible = "arm,gic-v3"; + #interrupt-cells = <3>; + interrupt-controller; + redistributor-stride = 0x40000; // 256kB stride
I raised the point in patch 1 about the stide value being read as a 64-bit value, which would imply that the example should be the following, I think:
+ redistributor-stride = <0x0 0x40000>; // 256kB stride
Quoted 17 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +9
-0+ #redistributor-regions = <2>; + reg = <0x0 0x2c010000 0 0x10000>, // GICD + <0x0 0x2d000000 0 0x800000>, // GICR 1: CPUs 0-31 + <0x0 0x2e000000 0 0x800000>; // GICR 2: CPUs 32-63 + <0x0 0x2c040000 0 0x2000>, // GICC + <0x0 0x2c060000 0 0x2000>, // GICH + <0x0 0x2c080000 0 0x2000>; // GICV + interrupts = <1 9 4>; + };Signature · 7 lines
1.8.3.4 _______________________________________________ linux-arm-kernel mailing list linux-arm-kernel at lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Otherwise it looks good.
PATCH 2/18 arm64: GICv3 device tree binding documentation
Quoted 10 lines · Christoffer Dall
Quoted 9 lines · Marc Zyngier
Add the necessary documentation to support GICv3.
Acked-by Catalin Marinas <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- Documentation/devicetree/bindings/arm/gic-v3.txt | 81 ++++++++++++++++++++++++ 1 file changed, 81 insertions(+) create mode 100644 Documentation/devicetree/bindings/arm/gic-v3.txt
Quoted 15 lines
Documentation/devicetree/bindings/arm/gic-v3.txt line 6
Quoted 12 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +6
-0@@ -0,0 +1,81 @@ +* ARM Generic Interrupt Controller, version 3 + +AArch64 SMP cores are often associated with a GICv3, providing private +peripheral interrupts (PPI), shared peripheral interrupts (SPI), +software generated interrupts (SGI), and locality-specific peripheral +Interrupts (LPI).Super-over-ridiculous-nit: capitalize the definitions, e.g. Private Peripheral Interrupts (PPI)
Sure.
Quoted 20 lines
Quoted 20 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +20
-0+ +Main node required properties: + +- compatible : should at least contain "arm,gic-v3". +- interrupt-controller : Identifies the node as an interrupt controller +- #interrupt-cells : Specifies the number of cells needed to encode an + interrupt source. Must be a single cell with a value of at least 3. + + The 1st cell is the interrupt type; 0 for SPI interrupts, 1 for PPI + interrupts. Other values are reserved for future use. + + The 2nd cell contains the interrupt number for the interrupt type. + SPI interrupts are in the range [0-987]. PPI interrupts are in the + range [0-15]. + + The 3rd cell is the flags, encoded as follows: + bits[3:0] trigger type and level flags. + 1 = edge triggered + 2 = edge triggered (deprecated, for compatibility with GICv2) + 4 = level triggered
Quoted 8 lines
Documentation/devicetree/bindings/arm/gic-v3.txt line 31
Quoted 5 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +5
-0+ 8 = level triggered (deprecated, for compatibility with GICv2) + + Cells 4 and beyond are reserved for future use. Where the 1st cell + has a value of 0 or 1, cells 4 and beyond act as padding, and may be + ignored. It is recommended that padding cells have a value of 0.another-super-nit: Saying "If the 1st cell" or "When the 1st cell" sounds more clear to me, but I may be wrong.
Works for me.
Quoted 21 lines
Quoted 7 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +7
-0+ +- reg : Specifies base physical address(s) and size of the GIC + registers, in the following order: + - GIC Distributor interface (GICD) + - GIC Redistributors (GICR), one range per redistributor region + - GIC CPU interface (GICC) + - GIC Hypervisor interface (GICH)Documentation/devicetree/bindings/arm/gic-v3.txt line 48
Quoted 10 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +10
-0+ - GIC Virtual CPU interface (GICV) + + GICC, GICH and GICV are optional. + +- interrupts : Interrupt source of the VGIC maintenance interrupt. + +Optional + +- redistributor-stride : If using padding pages, specifies the stride + of consecutive redistributors. Must be a multiple of 64kB.How is this going to be used? What would the size of the padding then be? I couldn't find anything about this in the specs.
Quoted 3 lines
Or does this mean the total size of the set of contiguous pages for each of the redistributors? That seems to make the math in the example below work.
Indeed. This is to allow overriding of the size of a single redistributor. The driver will follow the spec to the letter (two or four 64k pages, depending on GICR_TYPER.VLPI), except when this parameter is present.
Quoted 29 lines
Quoted 14 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +14
-0+ +- #redistributor-regions: The number of independent contiguous regions + occupied by the redistributors. Required if more than one such + region is present. + +Examples: + + gic: interrupt-controller at 2cf00000 { + compatible = "arm,gic-v3"; + #interrupt-cells = <3>; + interrupt-controller; + reg = <0x0 0x2f000000 0 0x10000>, // GICD + <0x0 0x2f100000 0 0x200000>, // GICR + <0x0 0x2c000000 0 0x2000>, // GICCDocumentation/devicetree/bindings/arm/gic-v3.txt line 72
Quoted 10 lines
Documentation/devicetree/bindings/arm/gic-v3.txt new file +10
-0+ <0x0 0x2c010000 0 0x2000>, // GICH + <0x0 0x2c020000 0 0x2000>; // GICV + interrupts = <1 9 4>; + }; + + gic: interrupt-controller at 2c010000 { + compatible = "arm,gic-v3"; + #interrupt-cells = <3>; + interrupt-controller; + redistributor-stride = 0x40000; // 256kB strideI raised the point in patch 1 about the stide value being read as a 64-bit value, which would imply that the example should be the following, I think:
Review
Quoted 1 line
(quoted diff) +1
-0+ redistributor-stride = <0x0 0x40000>; // 256kB stride
Indeed. I'll fix this.
Thanks for the review,
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
PATCH 3/18 arm64: boot protocol documentation update for GICv3
In reply to Marc ZyngierLinux has some requirements that must be satisfied in order to boot on a system built with a GICv3.
--- Documentation/arm64/booting.txt | 7 +++++++ 1 file changed, 7 insertions(+)
Documentation/arm64/booting.txt +7 -0
@@ -131,6 +131,13 @@ Before jumping into the kernel, the following conditions must be met: the kernel image will be entered must be initialised by software at a higher exception level to prevent execution in an UNKNOWN state. + For systems with a GICv3 interrupt controller, it is expected that: + - ID_AA64PFR0_EL1.GIC (bits [27:24]) must have the value 0b0001 + - If EL3 is present, it must program ICC_SRE_EL3.Enable (bit 3) to + 0b1 and ICC_SRE_EL3.SRE (bit 0) to 0b1. + - If the kernel is entered at EL1, EL2 must set ICC_SRE_EL2.Enable + (bit 3) to 0b1 and ICC_SRE_EL2.SRE (bit 0) to 0b1. + The requirements described above for CPU mode, caches, MMUs, architected timers, coherency and system registers apply to all CPUs. All CPUs must enter the kernel in the same exception level.
Signature · 1 line
1.8.3.4
Catalin Marinas
maintainerPATCH 3/18 arm64: boot protocol documentation update for GICv3
Quoted 4 lines · Marc Zyngier
Linux has some requirements that must be satisfied in order to boot on a system built with a GICv3.
Signed-off-by Marc Zyngier <[email protected]>
There are some patches I haven't acked in this series, so here it goes:
PATCH 3/18 arm64: boot protocol documentation update for GICv3
In reply to Marc ZyngierQuoted 13 lines · Marc Zyngier
Linux has some requirements that must be satisfied in order to boot on a system built with a GICv3.
Signed-off-by Marc Zyngier <[email protected]>--- Documentation/arm64/booting.txt | 7 +++++++ 1 file changed, 7 insertions(+)Documentation/arm64/booting.txt +0
-0
Documentation/arm64/booting.txt Before jumping into the kernel, the following conditions must be met: · line 135
Quoted 5 lines
Documentation/arm64/booting.txt +2
-0@@ -131,6 +131,13 @@ Before jumping into the kernel, the following conditions must be met: the kernel image will be entered must be initialised by software at a higher exception level to prevent execution in an UNKNOWN state. + For systems with a GICv3 interrupt controller, it is expected that: + - ID_AA64PFR0_EL1.GIC (bits [27:24]) must have the value 0b0001
Since ID_AA64PFR0_EL1 is read-only at all exception levels, I don't see the value of this statement.
Documentation/arm64/booting.txt Before jumping into the kernel, the following conditions must be met: · line 139
Quoted 4 lines
Documentation/arm64/booting.txt +4
-0+ - If EL3 is present, it must program ICC_SRE_EL3.Enable (bit 3) to + 0b1 and ICC_SRE_EL3.SRE (bit 0) to 0b1. + - If the kernel is entered at EL1, EL2 must set ICC_SRE_EL2.Enable + (bit 3) to 0b1 and ICC_SRE_EL2.SRE (bit 0) to 0b1.
Does this force the kernel to use SRE?
Will
PATCH 3/18 arm64: boot protocol documentation update for GICv3
Quoted 17 lines · Will Deacon
Quoted 16 lines · Marc Zyngier
Linux has some requirements that must be satisfied in order to boot on a system built with a GICv3.
Signed-off-by Marc Zyngier <[email protected]>--- Documentation/arm64/booting.txt | 7 +++++++ 1 file changed, 7 insertions(+)Documentation/arm64/booting.txt +0
-0@@ -131,6 +131,13 @@ Before jumping into the kernel, the following conditions must be met: the kernel image will be entered must be initialised by software at a higher exception level to prevent execution in an UNKNOWN state.
Quoted 5 lines
Documentation/arm64/booting.txt Before jumping into the kernel, the following conditions must be met: · line 135
Quoted 2 lines
Documentation/arm64/booting.txt +2
-0+ For systems with a GICv3 interrupt controller, it is expected that: + - ID_AA64PFR0_EL1.GIC (bits [27:24]) must have the value 0b0001Since
ID_AA64PFR0_EL1is read-only at all exception levels, I don't see the value of this statement.
Think virtualization. A hypervisor can control reads of ID_AA64PFR0_EL1 by setting HCR_EL2.TID3, and report whatever it wants.
Quoted 6 lines
Documentation/arm64/booting.txt Before jumping into the kernel, the following conditions must be met: · line 139
Quoted 4 lines
Documentation/arm64/booting.txt +4
-0+ - If EL3 is present, it must program ICC_SRE_EL3.Enable (bit 3) to + 0b1 and ICC_SRE_EL3.SRE (bit 0) to 0b1. + - If the kernel is entered at EL1, EL2 must set ICC_SRE_EL2.Enable + (bit 3) to 0b1 and ICC_SRE_EL2.SRE (bit 0) to 0b1.Does this force the kernel to use SRE?
No. The kernel still have to explicitly set ICC_SRE_EL1.SRE to 1.
Cheers,
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
PATCH 3/18 arm64: boot protocol documentation update for GICv3
Quoted 18 lines · Marc Zyngier
Quoted 17 lines · Will Deacon
Quoted 16 lines · Marc Zyngier
Linux has some requirements that must be satisfied in order to boot on a system built with a GICv3.
Signed-off-by Marc Zyngier <[email protected]>--- Documentation/arm64/booting.txt | 7 +++++++ 1 file changed, 7 insertions(+)Documentation/arm64/booting.txt +0
-0@@ -131,6 +131,13 @@ Before jumping into the kernel, the following conditions must be met: the kernel image will be entered must be initialised by software at a higher exception level to prevent execution in an UNKNOWN state.
Quoted 8 lines
Quoted 5 lines
Documentation/arm64/booting.txt Before jumping into the kernel, the following conditions must be met: · line 135
Quoted 2 lines
Documentation/arm64/booting.txt +2
-0+ For systems with a GICv3 interrupt controller, it is expected that: + - ID_AA64PFR0_EL1.GIC (bits [27:24]) must have the value 0b0001Since
ID_AA64PFR0_EL1is read-only at all exception levels, I don't see the value of this statement.Think virtualization. A hypervisor can control reads of
ID_AA64PFR0_EL1by setting HCR_EL2.TID3, and report whatever it wants.
Sure, but it seems unreasonable to me that we require a hypervisor to tell a guest about GICv3 if the system happens to have one. What if it wants to emulate a GICv2? In other words, requiring this in booting.txt seems superflous.
Will
PATCH 3/18 arm64: boot protocol documentation update for GICv3
Quoted 25 lines · Will Deacon
Quoted 24 lines · Marc Zyngier
Quoted 22 lines · Will Deacon
Quoted 13 lines · Marc Zyngier
Linux has some requirements that must be satisfied in order to boot on a system built with a GICv3.
Signed-off-by Marc Zyngier <[email protected]>--- Documentation/arm64/booting.txt | 7 +++++++ 1 file changed, 7 insertions(+)Documentation/arm64/booting.txt +0
-0Documentation/arm64/booting.txt Before jumping into the kernel, the following conditions must be met: · line 135
Quoted 5 lines
Documentation/arm64/booting.txt +2
-0@@ -131,6 +131,13 @@ Before jumping into the kernel, the following conditions must be met: the kernel image will be entered must be initialised by software at a higher exception level to prevent execution in an UNKNOWN state. + For systems with a GICv3 interrupt controller, it is expected that: + - ID_AA64PFR0_EL1.GIC (bits [27:24]) must have the value 0b0001Since
ID_AA64PFR0_EL1is read-only at all exception levels, I don't see the value of this statement.
Quoted 7 lines
Quoted 2 lines
Think virtualization. A hypervisor can control reads of
ID_AA64PFR0_EL1by setting HCR_EL2.TID3, and report whatever it wants.Sure, but it seems unreasonable to me that we require a hypervisor to tell a guest about GICv3 if the system happens to have one. What if it wants to emulate a GICv2? In other words, requiring this in booting.txt seems superflous.
The hypervisor can perfectly mask out the GICv3 feature if it wants to. The point I'm trying to make here is that if a guest is expected to be able to use the GICv3 system registers, then ID_AA64PFR0_EL1 should report so.
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
PATCH 3/18 arm64: boot protocol documentation update for GICv3
Quoted 34 lines · Marc Zyngier
Quoted 25 lines · Will Deacon
Quoted 24 lines · Marc Zyngier
Quoted 22 lines · Will Deacon
Quoted 13 lines · Marc Zyngier
Linux has some requirements that must be satisfied in order to boot on a system built with a GICv3.
Signed-off-by Marc Zyngier <[email protected]>--- Documentation/arm64/booting.txt | 7 +++++++ 1 file changed, 7 insertions(+)Documentation/arm64/booting.txt +0
-0Documentation/arm64/booting.txt Before jumping into the kernel, the following conditions must be met: · line 135
Quoted 5 lines
Documentation/arm64/booting.txt +2
-0@@ -131,6 +131,13 @@ Before jumping into the kernel, the following conditions must be met: the kernel image will be entered must be initialised by software at a higher exception level to prevent execution in an UNKNOWN state. + For systems with a GICv3 interrupt controller, it is expected that: + - ID_AA64PFR0_EL1.GIC (bits [27:24]) must have the value 0b0001Since
ID_AA64PFR0_EL1is read-only at all exception levels, I don't see the value of this statement.Quoted 7 lines
Quoted 2 lines
Think virtualization. A hypervisor can control reads of
ID_AA64PFR0_EL1by setting HCR_EL2.TID3, and report whatever it wants.Sure, but it seems unreasonable to me that we require a hypervisor to tell a guest about GICv3 if the system happens to have one. What if it wants to emulate a GICv2? In other words, requiring this in booting.txt seems superflous.
Quoted 4 lines
The hypervisor can perfectly mask out the GICv3 feature if it wants to. The point I'm trying to make here is that if a guest is expected to be able to use the GICv3 system registers, then
ID_AA64PFR0_EL1should report so.
Fair enough, but that's not what the text says :)
Will
PATCH 4/18 KVM: arm/arm64: vgic: move GICv2 registers to their own structure
In reply to Marc ZyngierIn order to make way for the GICv3 registers, move the v2-specific registers to their own structure.
--- arch/arm/kernel/asm-offsets.c | 14 +++++------ arch/arm/kvm/interrupts_head.S | 26 ++++++++++----------- arch/arm64/kernel/asm-offsets.c | 14 +++++------ arch/arm64/kvm/hyp.S | 26 ++++++++++----------- include/kvm/arm_vgic.h | 20 ++++++++++------ virt/kvm/arm/vgic.c | 52 ++++++++++++++++++++--------------------- 6 files changed, 79 insertions(+), 73 deletions(-)
arch/arm/kernel/asm-offsets.c +7 -7
@@ -181,13 +181,13 @@ int main(void) DEFINE(VCPU_HYP_PC, offsetof(struct kvm_vcpu, arch.fault.hyp_pc));#ifdef CONFIG_KVM_ARM_VGIC DEFINE(VCPU_VGIC_CPU, offsetof(struct kvm_vcpu, arch.vgic_cpu)); -DEFINE (VGIC_CPU_HCR , offsetof(struct vgic_cpu , vgic_hcr)); -DEFINE (VGIC_CPU_VMCR , offsetof(struct vgic_cpu , vgic_vmcr)); -DEFINE (VGIC_CPU_MISR , offsetof(struct vgic_cpu , vgic_misr)); -DEFINE (VGIC_CPU_EISR , offsetof(struct vgic_cpu , vgic_eisr)); -DEFINE (VGIC_CPU_ELRSR , offsetof(struct vgic_cpu , vgic_elrsr)); -DEFINE (VGIC_CPU_APR , offsetof(struct vgic_cpu , vgic_apr)); -DEFINE (VGIC_CPU_LR , offsetof(struct vgic_cpu , vgic_lr)); + DEFINE(VGIC_V2_CPU_HCR, offsetof(struct vgic_cpu, vgic_v2.vgic_hcr)); + DEFINE(VGIC_V2_CPU_VMCR, offsetof(struct vgic_cpu, vgic_v2.vgic_vmcr)); + DEFINE(VGIC_V2_CPU_MISR, offsetof(struct vgic_cpu, vgic_v2.vgic_misr)); + DEFINE(VGIC_V2_CPU_EISR, offsetof(struct vgic_cpu, vgic_v2.vgic_eisr)); + DEFINE(VGIC_V2_CPU_ELRSR, offsetof(struct vgic_cpu, vgic_v2.vgic_elrsr)); + DEFINE(VGIC_V2_CPU_APR, offsetof(struct vgic_cpu, vgic_v2.vgic_apr)); + DEFINE(VGIC_V2_CPU_LR, offsetof(struct vgic_cpu, vgic_v2.vgic_lr));DEFINE (VGIC_CPU_NR_LR , offsetof(struct vgic_cpu , nr_lr));#ifdef CONFIG_KVM_ARM_TIMER DEFINE(VCPU_TIMER_CNTV_CTL, offsetof(struct kvm_vcpu, arch.timer_cpu.cntv_ctl));
arch/arm/kvm/interrupts_head.S +13 -13
@@ -413,14 +413,14 @@ vcpu .req r0 @ vcpu pointer always in r0 ldr r9, [r2, #GICH_ELRSR1] ldr r10, [r2, #GICH_APR] - str r3, [r11, #VGIC_CPU_HCR] - str r4, [r11, #VGIC_CPU_VMCR] - str r5, [r11, #VGIC_CPU_MISR] - str r6, [r11, #VGIC_CPU_EISR] - str r7, [r11, #(VGIC_CPU_EISR + 4)] - str r8, [r11, #VGIC_CPU_ELRSR] - str r9, [r11, #(VGIC_CPU_ELRSR + 4)] - str r10, [r11, #VGIC_CPU_APR] + str r3, [r11, #VGIC_V2_CPU_HCR] + str r4, [r11, #VGIC_V2_CPU_VMCR] + str r5, [r11, #VGIC_V2_CPU_MISR] + str r6, [r11, #VGIC_V2_CPU_EISR] + str r7, [r11, #(VGIC_V2_CPU_EISR + 4)] + str r8, [r11, #VGIC_V2_CPU_ELRSR] + str r9, [r11, #(VGIC_V2_CPU_ELRSR + 4)] + str r10, [r11, #VGIC_V2_CPU_APR] /* Clear GICH_HCR */ mov r5, #0
@@ -428,7 +428,7 @@ vcpu .req r0 @ vcpu pointer always in r0 /* Save list registers */ add r2, r2, #GICH_LR0 - add r3, r11, #VGIC_CPU_LR + add r3, r11, #VGIC_V2_CPU_LR ldr r4, [r11, #VGIC_CPU_NR_LR] 1: ldr r6, [r2], #4 str r6, [r3], #4
@@ -455,9 +455,9 @@ vcpu .req r0 @ vcpu pointer always in r0 add r11, vcpu, #VCPU_VGIC_CPU /* We only restore a minimal set of registers */ - ldr r3, [r11, #VGIC_CPU_HCR] - ldr r4, [r11, #VGIC_CPU_VMCR] - ldr r8, [r11, #VGIC_CPU_APR] + ldr r3, [r11, #VGIC_V2_CPU_HCR] + ldr r4, [r11, #VGIC_V2_CPU_VMCR] + ldr r8, [r11, #VGIC_V2_CPU_APR] str r3, [r2, #GICH_HCR] str r4, [r2, #GICH_VMCR]
@@ -465,7 +465,7 @@ vcpu .req r0 @ vcpu pointer always in r0 /* Restore list registers */ add r2, r2, #GICH_LR0 - add r3, r11, #VGIC_CPU_LR + add r3, r11, #VGIC_V2_CPU_LR ldr r4, [r11, #VGIC_CPU_NR_LR] 1: ldr r6, [r3], #4 str r6, [r2], #4
arch/arm64/kernel/asm-offsets.c +7 -7
@@ -129,13 +129,13 @@ int main(void) DEFINE(KVM_TIMER_ENABLED, offsetof(struct kvm, arch.timer.enabled));DEFINE (VCPU_KVM , offsetof(struct kvm_vcpu , kvm)); DEFINE(VCPU_VGIC_CPU, offsetof(struct kvm_vcpu, arch.vgic_cpu)); -DEFINE (VGIC_CPU_HCR , offsetof(struct vgic_cpu , vgic_hcr)); -DEFINE (VGIC_CPU_VMCR , offsetof(struct vgic_cpu , vgic_vmcr)); -DEFINE (VGIC_CPU_MISR , offsetof(struct vgic_cpu , vgic_misr)); -DEFINE (VGIC_CPU_EISR , offsetof(struct vgic_cpu , vgic_eisr)); -DEFINE (VGIC_CPU_ELRSR , offsetof(struct vgic_cpu , vgic_elrsr)); -DEFINE (VGIC_CPU_APR , offsetof(struct vgic_cpu , vgic_apr)); -DEFINE (VGIC_CPU_LR , offsetof(struct vgic_cpu , vgic_lr)); + DEFINE(VGIC_V2_CPU_HCR, offsetof(struct vgic_cpu, vgic_v2.vgic_hcr)); + DEFINE(VGIC_V2_CPU_VMCR, offsetof(struct vgic_cpu, vgic_v2.vgic_vmcr)); + DEFINE(VGIC_V2_CPU_MISR, offsetof(struct vgic_cpu, vgic_v2.vgic_misr)); + DEFINE(VGIC_V2_CPU_EISR, offsetof(struct vgic_cpu, vgic_v2.vgic_eisr)); + DEFINE(VGIC_V2_CPU_ELRSR, offsetof(struct vgic_cpu, vgic_v2.vgic_elrsr)); + DEFINE(VGIC_V2_CPU_APR, offsetof(struct vgic_cpu, vgic_v2.vgic_apr)); + DEFINE(VGIC_V2_CPU_LR, offsetof(struct vgic_cpu, vgic_v2.vgic_lr));DEFINE (VGIC_CPU_NR_LR , offsetof(struct vgic_cpu , nr_lr)); DEFINE(KVM_VTTBR, offsetof(struct kvm, arch.vttbr)); DEFINE(KVM_VGIC_VCTRL, offsetof(struct kvm, arch.vgic.vctrl_base));
arch/arm64/kvm/hyp.S +13 -13
@@ -412,14 +412,14 @@ CPU_BE( rev w9, w9 ) CPU_BE( rev w10, w10 ) CPU_BE( rev w11, w11 ) - str w4, [x3, #VGIC_CPU_HCR] - str w5, [x3, #VGIC_CPU_VMCR] - str w6, [x3, #VGIC_CPU_MISR] - str w7, [x3, #VGIC_CPU_EISR] - str w8, [x3, #(VGIC_CPU_EISR + 4)] - str w9, [x3, #VGIC_CPU_ELRSR] - str w10, [x3, #(VGIC_CPU_ELRSR + 4)] - str w11, [x3, #VGIC_CPU_APR] + str w4, [x3, #VGIC_V2_CPU_HCR] + str w5, [x3, #VGIC_V2_CPU_VMCR] + str w6, [x3, #VGIC_V2_CPU_MISR] + str w7, [x3, #VGIC_V2_CPU_EISR] + str w8, [x3, #(VGIC_V2_CPU_EISR + 4)] + str w9, [x3, #VGIC_V2_CPU_ELRSR] + str w10, [x3, #(VGIC_V2_CPU_ELRSR + 4)] + str w11, [x3, #VGIC_V2_CPU_APR] /* Clear GICH_HCR */ str wzr, [x2, #GICH_HCR]
@@ -427,7 +427,7 @@ CPU_BE( rev w11, w11 ) /* Save list registers */ add x2, x2, #GICH_LR0 ldr w4, [x3, #VGIC_CPU_NR_LR] - add x3, x3, #VGIC_CPU_LR + add x3, x3, #VGIC_V2_CPU_LR 1: ldr w5, [x2], #4 CPU_BE( rev w5, w5 ) str w5, [x3], #4
@@ -452,9 +452,9 @@ CPU_BE( rev w5, w5 ) add x3, x0, #VCPU_VGIC_CPU /* We only restore a minimal set of registers */ - ldr w4, [x3, #VGIC_CPU_HCR] - ldr w5, [x3, #VGIC_CPU_VMCR] - ldr w6, [x3, #VGIC_CPU_APR] + ldr w4, [x3, #VGIC_V2_CPU_HCR] + ldr w5, [x3, #VGIC_V2_CPU_VMCR] + ldr w6, [x3, #VGIC_V2_CPU_APR] CPU_BE( rev w4, w4 ) CPU_BE( rev w5, w5 ) CPU_BE( rev w6, w6 )
@@ -466,7 +466,7 @@ CPU_BE( rev w6, w6 ) /* Restore list registers */ add x2, x2, #GICH_LR0 ldr w4, [x3, #VGIC_CPU_NR_LR] - add x3, x3, #VGIC_CPU_LR + add x3, x3, #VGIC_V2_CPU_LR 1: ldr w5, [x3], #4 CPU_BE( rev w5, w5 ) str w5, [x2], #4
include/kvm/arm_vgic.h +13 -7
@@ -110,6 +110,16 @@ struct vgic_dist {#endif }; +struct vgic_v2_cpu_if { +u32 vgic_hcr; +u32 vgic_vmcr; +u32 vgic_misr;/* Saved only */ +u32 vgic_eisr[2 ];/* Saved only */ +u32 vgic_elrsr[2 ];/* Saved only */ +u32 vgic_apr; +u32 vgic_lr[VGIC_MAX_LRS ]; +}; +struct vgic_cpu {#ifdef CONFIG_KVM_ARM_VGIC /* per IRQ to LR mapping */
@@ -126,13 +136,9 @@ struct vgic_cpu {int nr_lr;/* CPU vif control registers for world switch */ -u32 vgic_hcr; -u32 vgic_vmcr; -u32 vgic_misr;/* Saved only */ -u32 vgic_eisr[2 ];/* Saved only */ -u32 vgic_elrsr[2 ];/* Saved only */ -u32 vgic_apr; -u32 vgic_lr[VGIC_MAX_LRS ]; +union { +struct vgic_v2_cpu_if vgic_v2; + };#endif };
virt/kvm/arm/vgic.c +26 -26
152 lines not on this page · raw message
Signature · 1 line
1.8.3.4
PATCH 4/18 KVM: arm/arm64: vgic: move GICv2 registers to their own structure
Quoted 223 lines · Marc Zyngier
In order to make way for the GICv3 registers, move the v2-specific registers to their own structure.
Acked-by Catalin Marinas <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- arch/arm/kernel/asm-offsets.c | 14 +++++------ arch/arm/kvm/interrupts_head.S | 26 ++++++++++----------- arch/arm64/kernel/asm-offsets.c | 14 +++++------ arch/arm64/kvm/hyp.S | 26 ++++++++++----------- include/kvm/arm_vgic.h | 20 ++++++++++------ virt/kvm/arm/vgic.c | 52 ++++++++++++++++++++--------------------- 6 files changed, 79 insertions(+), 73 deletions(-)arch/arm/kernel/asm-offsets.c +7
-7@@ -181,13 +181,13 @@ int main(void) DEFINE(VCPU_HYP_PC, offsetof(struct kvm_vcpu, arch.fault.hyp_pc));#ifdef CONFIG_KVM_ARM_VGIC DEFINE(VCPU_VGIC_CPU, offsetof(struct kvm_vcpu, arch.vgic_cpu)); -DEFINE (VGIC_CPU_HCR , offsetof(struct vgic_cpu , vgic_hcr)); -DEFINE (VGIC_CPU_VMCR , offsetof(struct vgic_cpu , vgic_vmcr)); -DEFINE (VGIC_CPU_MISR , offsetof(struct vgic_cpu , vgic_misr)); -DEFINE (VGIC_CPU_EISR , offsetof(struct vgic_cpu , vgic_eisr)); -DEFINE (VGIC_CPU_ELRSR , offsetof(struct vgic_cpu , vgic_elrsr)); -DEFINE (VGIC_CPU_APR , offsetof(struct vgic_cpu , vgic_apr)); -DEFINE (VGIC_CPU_LR , offsetof(struct vgic_cpu , vgic_lr)); + DEFINE(VGIC_V2_CPU_HCR, offsetof(struct vgic_cpu, vgic_v2.vgic_hcr)); + DEFINE(VGIC_V2_CPU_VMCR, offsetof(struct vgic_cpu, vgic_v2.vgic_vmcr)); + DEFINE(VGIC_V2_CPU_MISR, offsetof(struct vgic_cpu, vgic_v2.vgic_misr)); + DEFINE(VGIC_V2_CPU_EISR, offsetof(struct vgic_cpu, vgic_v2.vgic_eisr)); + DEFINE(VGIC_V2_CPU_ELRSR, offsetof(struct vgic_cpu, vgic_v2.vgic_elrsr)); + DEFINE(VGIC_V2_CPU_APR, offsetof(struct vgic_cpu, vgic_v2.vgic_apr)); + DEFINE(VGIC_V2_CPU_LR, offsetof(struct vgic_cpu, vgic_v2.vgic_lr));DEFINE (VGIC_CPU_NR_LR , offsetof(struct vgic_cpu , nr_lr));#ifdef CONFIG_KVM_ARM_TIMER DEFINE(VCPU_TIMER_CNTV_CTL, offsetof(struct kvm_vcpu, arch.timer_cpu.cntv_ctl));arch/arm/kvm/interrupts_head.S +13
-13@@ -413,14 +413,14 @@ vcpu .req r0 @ vcpu pointer always in r0 ldr r9, [r2, #GICH_ELRSR1] ldr r10, [r2, #GICH_APR] - str r3, [r11, #VGIC_CPU_HCR] - str r4, [r11, #VGIC_CPU_VMCR] - str r5, [r11, #VGIC_CPU_MISR] - str r6, [r11, #VGIC_CPU_EISR] - str r7, [r11, #(VGIC_CPU_EISR + 4)] - str r8, [r11, #VGIC_CPU_ELRSR] - str r9, [r11, #(VGIC_CPU_ELRSR + 4)] - str r10, [r11, #VGIC_CPU_APR] + str r3, [r11, #VGIC_V2_CPU_HCR] + str r4, [r11, #VGIC_V2_CPU_VMCR] + str r5, [r11, #VGIC_V2_CPU_MISR] + str r6, [r11, #VGIC_V2_CPU_EISR] + str r7, [r11, #(VGIC_V2_CPU_EISR + 4)] + str r8, [r11, #VGIC_V2_CPU_ELRSR] + str r9, [r11, #(VGIC_V2_CPU_ELRSR + 4)] + str r10, [r11, #VGIC_V2_CPU_APR] /* Clear GICH_HCR */ mov r5, #0@@ -428,7 +428,7 @@ vcpu .req r0 @ vcpu pointer always in r0 /* Save list registers */ add r2, r2, #GICH_LR0 - add r3, r11, #VGIC_CPU_LR + add r3, r11, #VGIC_V2_CPU_LR ldr r4, [r11, #VGIC_CPU_NR_LR] 1: ldr r6, [r2], #4 str r6, [r3], #4@@ -455,9 +455,9 @@ vcpu .req r0 @ vcpu pointer always in r0 add r11, vcpu, #VCPU_VGIC_CPU /* We only restore a minimal set of registers */ - ldr r3, [r11, #VGIC_CPU_HCR] - ldr r4, [r11, #VGIC_CPU_VMCR] - ldr r8, [r11, #VGIC_CPU_APR] + ldr r3, [r11, #VGIC_V2_CPU_HCR] + ldr r4, [r11, #VGIC_V2_CPU_VMCR] + ldr r8, [r11, #VGIC_V2_CPU_APR] str r3, [r2, #GICH_HCR] str r4, [r2, #GICH_VMCR]@@ -465,7 +465,7 @@ vcpu .req r0 @ vcpu pointer always in r0 /* Restore list registers */ add r2, r2, #GICH_LR0 - add r3, r11, #VGIC_CPU_LR + add r3, r11, #VGIC_V2_CPU_LR ldr r4, [r11, #VGIC_CPU_NR_LR] 1: ldr r6, [r3], #4 str r6, [r2], #4arch/arm64/kernel/asm-offsets.c +7
-7@@ -129,13 +129,13 @@ int main(void) DEFINE(KVM_TIMER_ENABLED, offsetof(struct kvm, arch.timer.enabled));DEFINE (VCPU_KVM , offsetof(struct kvm_vcpu , kvm)); DEFINE(VCPU_VGIC_CPU, offsetof(struct kvm_vcpu, arch.vgic_cpu)); -DEFINE (VGIC_CPU_HCR , offsetof(struct vgic_cpu , vgic_hcr)); -DEFINE (VGIC_CPU_VMCR , offsetof(struct vgic_cpu , vgic_vmcr)); -DEFINE (VGIC_CPU_MISR , offsetof(struct vgic_cpu , vgic_misr)); -DEFINE (VGIC_CPU_EISR , offsetof(struct vgic_cpu , vgic_eisr)); -DEFINE (VGIC_CPU_ELRSR , offsetof(struct vgic_cpu , vgic_elrsr)); -DEFINE (VGIC_CPU_APR , offsetof(struct vgic_cpu , vgic_apr)); -DEFINE (VGIC_CPU_LR , offsetof(struct vgic_cpu , vgic_lr)); + DEFINE(VGIC_V2_CPU_HCR, offsetof(struct vgic_cpu, vgic_v2.vgic_hcr)); + DEFINE(VGIC_V2_CPU_VMCR, offsetof(struct vgic_cpu, vgic_v2.vgic_vmcr)); + DEFINE(VGIC_V2_CPU_MISR, offsetof(struct vgic_cpu, vgic_v2.vgic_misr)); + DEFINE(VGIC_V2_CPU_EISR, offsetof(struct vgic_cpu, vgic_v2.vgic_eisr)); + DEFINE(VGIC_V2_CPU_ELRSR, offsetof(struct vgic_cpu, vgic_v2.vgic_elrsr)); + DEFINE(VGIC_V2_CPU_APR, offsetof(struct vgic_cpu, vgic_v2.vgic_apr)); + DEFINE(VGIC_V2_CPU_LR, offsetof(struct vgic_cpu, vgic_v2.vgic_lr));DEFINE (VGIC_CPU_NR_LR , offsetof(struct vgic_cpu , nr_lr)); DEFINE(KVM_VTTBR, offsetof(struct kvm, arch.vttbr)); DEFINE(KVM_VGIC_VCTRL, offsetof(struct kvm, arch.vgic.vctrl_base));arch/arm64/kvm/hyp.S +13
-13@@ -412,14 +412,14 @@ CPU_BE( rev w9, w9 ) CPU_BE( rev w10, w10 ) CPU_BE( rev w11, w11 ) - str w4, [x3, #VGIC_CPU_HCR] - str w5, [x3, #VGIC_CPU_VMCR] - str w6, [x3, #VGIC_CPU_MISR] - str w7, [x3, #VGIC_CPU_EISR] - str w8, [x3, #(VGIC_CPU_EISR + 4)] - str w9, [x3, #VGIC_CPU_ELRSR] - str w10, [x3, #(VGIC_CPU_ELRSR + 4)] - str w11, [x3, #VGIC_CPU_APR] + str w4, [x3, #VGIC_V2_CPU_HCR] + str w5, [x3, #VGIC_V2_CPU_VMCR] + str w6, [x3, #VGIC_V2_CPU_MISR] + str w7, [x3, #VGIC_V2_CPU_EISR] + str w8, [x3, #(VGIC_V2_CPU_EISR + 4)] + str w9, [x3, #VGIC_V2_CPU_ELRSR] + str w10, [x3, #(VGIC_V2_CPU_ELRSR + 4)] + str w11, [x3, #VGIC_V2_CPU_APR] /* Clear GICH_HCR */ str wzr, [x2, #GICH_HCR]@@ -427,7 +427,7 @@ CPU_BE( rev w11, w11 ) /* Save list registers */ add x2, x2, #GICH_LR0 ldr w4, [x3, #VGIC_CPU_NR_LR] - add x3, x3, #VGIC_CPU_LR + add x3, x3, #VGIC_V2_CPU_LR 1: ldr w5, [x2], #4 CPU_BE( rev w5, w5 ) str w5, [x3], #4@@ -452,9 +452,9 @@ CPU_BE( rev w5, w5 ) add x3, x0, #VCPU_VGIC_CPU /* We only restore a minimal set of registers */ - ldr w4, [x3, #VGIC_CPU_HCR] - ldr w5, [x3, #VGIC_CPU_VMCR] - ldr w6, [x3, #VGIC_CPU_APR] + ldr w4, [x3, #VGIC_V2_CPU_HCR] + ldr w5, [x3, #VGIC_V2_CPU_VMCR] + ldr w6, [x3, #VGIC_V2_CPU_APR] CPU_BE( rev w4, w4 ) CPU_BE( rev w5, w5 ) CPU_BE( rev w6, w6 )@@ -466,7 +466,7 @@ CPU_BE( rev w6, w6 ) /* Restore list registers */ add x2, x2, #GICH_LR0 ldr w4, [x3, #VGIC_CPU_NR_LR] - add x3, x3, #VGIC_CPU_LR + add x3, x3, #VGIC_V2_CPU_LR 1: ldr w5, [x3], #4 CPU_BE( rev w5, w5 ) str w5, [x2], #4include/kvm/arm_vgic.h +13
-7@@ -110,6 +110,16 @@ struct vgic_dist {#endif }; +struct vgic_v2_cpu_if { +u32 vgic_hcr; +u32 vgic_vmcr; +u32 vgic_misr;/* Saved only */ +u32 vgic_eisr[2 ];/* Saved only */ +u32 vgic_elrsr[2 ];/* Saved only */ +u32 vgic_apr; +u32 vgic_lr[VGIC_MAX_LRS ]; +}; +struct vgic_cpu {#ifdef CONFIG_KVM_ARM_VGIC /* per IRQ to LR mapping */ @@ -126,13 +136,9 @@ struct vgic_cpu {int nr_lr;/* CPU vif control registers for world switch */ -u32 vgic_hcr; -u32 vgic_vmcr; -u32 vgic_misr;/* Saved only */ -u32 vgic_eisr[2 ];/* Saved only */ -u32 vgic_elrsr[2 ];/* Saved only */ -u32 vgic_apr; -u32 vgic_lr[VGIC_MAX_LRS ]; +union { +struct vgic_v2_cpu_if vgic_v2; + };#endif };virt/kvm/arm/vgic.c +0
-0
virt/kvm/arm/vgic.c static bool handle_mmio_sgi_reg(struct kvm_vcpu *vcpu, · line 605
Quoted 5 lines
virt/kvm/arm/vgic.c +1
-1@@ -602,7 +602,7 @@ static bool handle_mmio_sgi_reg(struct kvm_vcpu *vcpu,static void vgic_retire_lr (int lr_nr,int irq,struct vgic_cpu *vgic_cpu) {clear_bit (lr_nr, vgic_cpu->lr_used); - vgic_cpu->vgic_lr[lr_nr] &= ~GICH_LR_STATE ; + vgic_cpu->vgic_v2.vgic_lr[lr] &= ~GICH_LR_STATE ;
lr_nr, this breaks bisectability
Quoted 138 lines
virt/kvm/arm/vgic.c +22
-22vgic_cpu->vgic_irq_lr_map[irq] =LR_EMPTY ; }@@ -1007,7 +1007,7 @@ static void vgic_retire_disabled_irqs(struct kvm_vcpu *vcpu)int lr;for_each_set_bit (lr, vgic_cpu->lr_used, vgic_cpu->nr_lr) { -int irq = vgic_cpu->vgic_lr[lr] &GICH_LR_VIRTUALID ; +int irq = vgic_cpu->vgic_v2.vgic_lr[lr] &GICH_LR_VIRTUALID ;if (!vgic_irq_is_enabled (vcpu, irq)) {vgic_retire_lr (lr, irq, vgic_cpu);@@ -1037,11 +1037,11 @@ static bool vgic_queue_irq(struct kvm_vcpu *vcpu, u8 sgi_source_id, int irq)/* Do we have an active interrupt for the same CPUID? */ if (lr !=LR_EMPTY && - (LR_CPUID (vgic_cpu->vgic_lr[lr]) == sgi_source_id)) { + (LR_CPUID (vgic_cpu->vgic_v2.vgic_lr[lr]) == sgi_source_id)) {kvm_debug ("LR%d piggyback for IRQ%d %x\n" , - lr, irq, vgic_cpu->vgic_lr[lr]); + lr, irq, vgic_cpu->vgic_v2.vgic_lr[lr]);BUG_ON (!test_bit (lr, vgic_cpu->lr_used)); - vgic_cpu->vgic_lr[lr] |=GICH_LR_PENDING_BIT ; + vgic_cpu->vgic_v2.vgic_lr[lr] |=GICH_LR_PENDING_BIT ;return true; }@@ -1052,12 +1052,12 @@ static bool vgic_queue_irq(struct kvm_vcpu *vcpu, u8 sgi_source_id, int irq)return false;kvm_debug ("LR%d allocated for IRQ%d %x\n" , lr, irq, sgi_source_id); - vgic_cpu->vgic_lr[lr] =MK_LR_PEND (sgi_source_id, irq); + vgic_cpu->vgic_v2.vgic_lr[lr] =MK_LR_PEND (sgi_source_id, irq); vgic_cpu->vgic_irq_lr_map[irq] = lr;set_bit (lr, vgic_cpu->lr_used);if (!vgic_irq_is_edge (vcpu, irq)) - vgic_cpu->vgic_lr[lr] |=GICH_LR_EOI ; + vgic_cpu->vgic_v2.vgic_lr[lr] |=GICH_LR_EOI ;return true; }@@ -1155,9 +1155,9 @@ static void __kvm_vgic_flush_hwstate(struct kvm_vcpu *vcpu) epilog:if (overflow) { - vgic_cpu->vgic_hcr |=GICH_HCR_UIE ; + vgic_cpu->vgic_v2.vgic_hcr |=GICH_HCR_UIE ; }else { - vgic_cpu->vgic_hcr &= ~GICH_HCR_UIE ; + vgic_cpu->vgic_v2.vgic_hcr &= ~GICH_HCR_UIE ;/* * We're about to run this VCPU, and we've consumed * everything the distributor had in store for @@ -1173,21 +1173,21 @@ static bool vgic_process_maintenance(struct kvm_vcpu *vcpu)struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;bool level_pending = false; -kvm_debug ("MISR = %08x\n" , vgic_cpu->vgic_misr); +kvm_debug ("MISR = %08x\n" , vgic_cpu->vgic_v2.vgic_misr); -if (vgic_cpu->vgic_misr &GICH_MISR_EOI ) { +if (vgic_cpu->vgic_v2.vgic_misr &GICH_MISR_EOI ) {/* * Some level interrupts have been EOIed. Clear their * active bit. */ int lr, irq; -for_each_set_bit (lr, (unsigned long *)vgic_cpu->vgic_eisr, +for_each_set_bit (lr, (unsigned long *)vgic_cpu->vgic_v2.vgic_eisr, vgic_cpu->nr_lr) { - irq = vgic_cpu->vgic_lr[lr] &GICH_LR_VIRTUALID ; + irq = vgic_cpu->vgic_v2.vgic_lr[lr] &GICH_LR_VIRTUALID ;vgic_irq_clear_active (vcpu, irq); - vgic_cpu->vgic_lr[lr] &= ~GICH_LR_EOI ; + vgic_cpu->vgic_v2.vgic_lr[lr] &= ~GICH_LR_EOI ;/* Any additional pending interrupt? */ if (vgic_dist_irq_is_pending (vcpu, irq)) {@@ -1201,13 +1201,13 @@ static bool vgic_process_maintenance(struct kvm_vcpu *vcpu)* Despite being EOIed, the LR may not have * been marked as empty. */ -set_bit (lr, (unsigned long *)vgic_cpu->vgic_elrsr); - vgic_cpu->vgic_lr[lr] &= ~GICH_LR_ACTIVE_BIT ; +set_bit (lr, (unsigned long *)vgic_cpu->vgic_v2.vgic_elrsr); + vgic_cpu->vgic_v2.vgic_lr[lr] &= ~GICH_LR_ACTIVE_BIT ; } } -if (vgic_cpu->vgic_misr &GICH_MISR_U ) - vgic_cpu->vgic_hcr &= ~GICH_HCR_UIE ; +if (vgic_cpu->vgic_v2.vgic_misr &GICH_MISR_U ) + vgic_cpu->vgic_v2.vgic_hcr &= ~GICH_HCR_UIE ;return level_pending; }@@ -1226,21 +1226,21 @@ static void __kvm_vgic_sync_hwstate(struct kvm_vcpu *vcpu) level_pending =vgic_process_maintenance (vcpu);/* Clear mappings for empty LRs */ -for_each_set_bit (lr, (unsigned long *)vgic_cpu->vgic_elrsr, +for_each_set_bit (lr, (unsigned long *)vgic_cpu->vgic_v2.vgic_elrsr, vgic_cpu->nr_lr) {int irq;if (!test_and_clear_bit (lr, vgic_cpu->lr_used))continue ; - irq = vgic_cpu->vgic_lr[lr] &GICH_LR_VIRTUALID ; + irq = vgic_cpu->vgic_v2.vgic_lr[lr] &GICH_LR_VIRTUALID ;BUG_ON (irq >=VGIC_NR_IRQS ); vgic_cpu->vgic_irq_lr_map[irq] =LR_EMPTY ; }/* Check if we still have something up our sleeve... */ - pending =find_first_zero_bit ((unsigned long *)vgic_cpu->vgic_elrsr, + pending =find_first_zero_bit ((unsigned long *)vgic_cpu->vgic_v2.vgic_elrsr, vgic_cpu->nr_lr);if (level_pending || pending < vgic_cpu->nr_lr)set_bit (vcpu->vcpu_id, &dist->irq_pending_on_cpu);@@ -1436,10 +1436,10 @@ int kvm_vgic_vcpu_init(struct kvm_vcpu *vcpu)* points to their reset values. Anything else resets to zero * anyway. */ - vgic_cpu->vgic_vmcr =0 ; + vgic_cpu->vgic_v2.vgic_vmcr =0 ; vgic_cpu->nr_lr = vgic_nr_lr; - vgic_cpu->vgic_hcr =GICH_HCR_EN ;/* Get the show on the road... */ + vgic_cpu->vgic_v2.vgic_hcr =GICH_HCR_EN ;/* Get the show on the road... */ return 0 ; }
virt/kvm/arm/vgic.c static bool handle_cpu_mmio_misc(struct kvm_vcpu *vcpu, · line 1753
Quoted 10 lines
virt/kvm/arm/vgic.c +1
-1@@ -1745,15 +1745,15 @@ static bool handle_cpu_mmio_misc(struct kvm_vcpu *vcpu, }if (!mmio->is_write) { - reg = (vgic_cpu->vgic_vmcr & mask) >> shift; + reg = (vgic_cpu->vgic_v2.vgic_vmcr & mask) >> shift;mmio_data_write (mmio, ~0 , reg); }else { reg =mmio_data_read (mmio, ~0 ); reg = (reg << shift) & mask;if (reg != (vgic_cpu->vgic_vmcr & mask))
You forgot one here
Quoted 11 lines
virt/kvm/arm/vgic.c +2
-2updated = true; - vgic_cpu->vgic_vmcr &= ~mask; - vgic_cpu->vgic_vmcr |= reg; + vgic_cpu->vgic_v2.vgic_vmcr &= ~mask; + vgic_cpu->vgic_v2.vgic_vmcr |= reg; }return updated; }Signature · 1 line
1.8.3.4
I think you also left one out in vgic_unqueue_irqs.
PATCH 5/18 KVM: ARM: vgic: introduce vgic_ops and LR manipulation primitives
In reply to Marc ZyngierIn order to split the various register manipulation from the main vgic code, introduce a vgic_ops structure, and start by abstracting the LR manipulation code with a couple of accessors.
--- include/kvm/arm_vgic.h | 18 ++++++ virt/kvm/arm/vgic.c | 170 +++++++++++++++++++++++++++++++++---------------- 2 files changed, 132 insertions(+), 56 deletions(-)
include/kvm/arm_vgic.h +18 -0
@@ -68,6 +68,24 @@ struct vgic_bytemap {u32 shared[VGIC_NR_SHARED_IRQS /4 ]; }; +struct kvm_vcpu ; + +#define LR_STATE_PENDING (1 << 0) +#define LR_STATE_ACTIVE (1 << 1) +#define LR_STATE_MASK (3 << 0) +#define LR_EOI_INT (1 << 2) + +struct vgic_lr { +u16 irq; +u8 source; +u8 state; +}; + +struct vgic_ops { +struct vgic_lr (*get_lr)(const struct kvm_vcpu *,int ); +void (*set_lr)(struct kvm_vcpu *,int ,struct vgic_lr ); +}; +struct vgic_dist {#ifdef CONFIG_KVM_ARM_VGIC spinlock_t lock;
virt/kvm/arm/vgic.c +114 -56
275 lines not on this page · raw message
Signature · 1 line
1.8.3.4
PATCH 6/18 KVM: ARM: vgic: abstract access to the ELRSR bitmap
In reply to Marc ZyngierMove the GICH_ELRSR access to its own function, and add it to the vgic_ops structure.
--- include/kvm/arm_vgic.h | 1 + virt/kvm/arm/vgic.c | 20 ++++++++++++++++---- 2 files changed, 17 insertions(+), 4 deletions(-)
include/kvm/arm_vgic.h +1 -0
@@ -84,6 +84,7 @@ struct vgic_lr {struct vgic_ops {struct vgic_lr (*get_lr)(const struct kvm_vcpu *,int );void (*set_lr)(struct kvm_vcpu *,int ,struct vgic_lr ); +u64 (*get_elrsr)(const struct kvm_vcpu *vcpu); };struct vgic_dist {
virt/kvm/arm/vgic.c +16 -4
@@ -1022,9 +1022,16 @@ static void vgic_v2_set_lr(struct kvm_vcpu *vcpu, int lr,set_bit (lr, (unsigned long *)vcpu->arch.vgic_cpu.vgic_v2.vgic_elrsr); } +static u64 vgic_v2_get_elrsr (const struct kvm_vcpu *vcpu) +{ +const u32 *elrsr = vcpu->arch.vgic_cpu.vgic_v2.vgic_elrsr; +return *(u64 *)elrsr; +} +static const struct vgic_ops vgic_ops = { .get_lr = vgic_v2_get_lr, .set_lr = vgic_v2_set_lr, + .get_elrsr = vgic_v2_get_elrsr, };static inline struct vgic_lr vgic_get_lr (const struct kvm_vcpu *vcpu,int lr)
@@ -1038,6 +1045,11 @@ static inline void vgic_set_lr(struct kvm_vcpu *vcpu, int lr, vgic_ops.set_lr (vcpu, lr, vlr); } +static inline u64 vgic_get_elrsr (struct kvm_vcpu *vcpu) +{ +return vgic_ops.get_elrsr (vcpu); +} +static void vgic_retire_lr (int lr_nr,int irq,struct kvm_vcpu *vcpu) {struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
@@ -1278,14 +1290,15 @@ static void __kvm_vgic_sync_hwstate(struct kvm_vcpu *vcpu) {struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;struct vgic_dist *dist = &vcpu->kvm->arch.vgic; +u64 elrsr =vgic_get_elrsr (vcpu); +unsigned long *elrsr_ptr = (unsigned long *)&elrsr;int lr, pending;bool level_pending; level_pending =vgic_process_maintenance (vcpu);/* Clear mappings for empty LRs */ -for_each_set_bit (lr, (unsigned long *)vgic_cpu->vgic_v2.vgic_elrsr, - vgic_cpu->nr_lr) { +for_each_set_bit (lr, elrsr_ptr, vgic_cpu->nr_lr) {struct vgic_lr vlr;if (!test_and_clear_bit (lr, vgic_cpu->lr_used))
@@ -1298,8 +1311,7 @@ static void __kvm_vgic_sync_hwstate(struct kvm_vcpu *vcpu) }/* Check if we still have something up our sleeve... */ - pending =find_first_zero_bit ((unsigned long *)vgic_cpu->vgic_v2.vgic_elrsr, - vgic_cpu->nr_lr); + pending =find_first_zero_bit (elrsr_ptr, vgic_cpu->nr_lr);if (level_pending || pending < vgic_cpu->nr_lr)set_bit (vcpu->vcpu_id, &dist->irq_pending_on_cpu); }
Signature · 1 line
1.8.3.4
PATCH 7/18 KVM: ARM: vgic: abstract EISR bitmap access
In reply to Marc ZyngierMove the GICH_EISR access to its own function.
--- include/kvm/arm_vgic.h | 1 + virt/kvm/arm/vgic.c | 16 +++++++++++++++- 2 files changed, 16 insertions(+), 1 deletion(-)
include/kvm/arm_vgic.h +1 -0
@@ -85,6 +85,7 @@ struct vgic_ops {struct vgic_lr (*get_lr)(const struct kvm_vcpu *,int );void (*set_lr)(struct kvm_vcpu *,int ,struct vgic_lr );u64 (*get_elrsr)(const struct kvm_vcpu *vcpu); +u64 (*get_eisr)(const struct kvm_vcpu *vcpu); };struct vgic_dist {
virt/kvm/arm/vgic.c +15 -1
@@ -1028,10 +1028,17 @@ static u64 vgic_v2_get_elrsr(const struct kvm_vcpu *vcpu)return *(u64 *)elrsr; } +static u64 vgic_v2_get_eisr (const struct kvm_vcpu *vcpu) +{ +const u32 *eisr = vcpu->arch.vgic_cpu.vgic_v2.vgic_eisr; +return *(u64 *)eisr; +} +static const struct vgic_ops vgic_ops = { .get_lr = vgic_v2_get_lr, .set_lr = vgic_v2_set_lr, .get_elrsr = vgic_v2_get_elrsr, + .get_eisr = vgic_v2_get_eisr, };static inline struct vgic_lr vgic_get_lr (const struct kvm_vcpu *vcpu,int lr)
@@ -1050,6 +1057,11 @@ static inline u64 vgic_get_elrsr(struct kvm_vcpu *vcpu)return vgic_ops.get_elrsr (vcpu); } +static inline u64 vgic_get_eisr (struct kvm_vcpu *vcpu) +{ +return vgic_ops.get_eisr (vcpu); +} +static void vgic_retire_lr (int lr_nr,int irq,struct kvm_vcpu *vcpu) {struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
@@ -1256,9 +1268,11 @@ static bool vgic_process_maintenance(struct kvm_vcpu *vcpu)* Some level interrupts have been EOIed. Clear their * active bit. */ +u64 eisr =vgic_get_eisr (vcpu); +unsigned long *eisr_ptr = (unsigned long *)&eisr;int lr; -for_each_set_bit (lr, (unsigned long *)vgic_cpu->vgic_v2.vgic_eisr, +for_each_set_bit (lr, eisr_ptr, vgic_cpu->nr_lr) { vgic_cpu->nr_lr) {struct vgic_lr vlr =vgic_get_lr (vcpu, lr);
Signature · 1 line
1.8.3.4
PATCH 8/18 KVM: ARM: vgic: abstract MISR decoding
In reply to Marc ZyngierInstead of directly dealing with the GICH_MISR bits, move the code to its own function and use a couple of public flags to represent the actual state.
--- include/kvm/arm_vgic.h | 4 ++++ virt/kvm/arm/vgic.c | 26 +++++++++++++++++++++++--- 2 files changed, 27 insertions(+), 3 deletions(-)
include/kvm/arm_vgic.h +4 -0
@@ -86,6 +86,7 @@ struct vgic_ops {void (*set_lr)(struct kvm_vcpu *,int ,struct vgic_lr );u64 (*get_elrsr)(const struct kvm_vcpu *vcpu);u64 (*get_eisr)(const struct kvm_vcpu *vcpu); +u32 (*get_interrupt_status)(const struct kvm_vcpu *vcpu); };struct vgic_dist {
@@ -164,6 +165,9 @@ struct vgic_cpu {#define LR_EMPTY 0xff +#define INT_STATUS_EOI (1 << 0) +#define INT_STATUS_UNDERFLOW (1 << 1) +struct kvm ;struct kvm_vcpu ;struct kvm_run ;
virt/kvm/arm/vgic.c +23 -3
@@ -1034,11 +1034,25 @@ static u64 vgic_v2_get_eisr(const struct kvm_vcpu *vcpu)return *(u64 *)eisr; } +static u32 vgic_v2_get_interrupt_status (const struct kvm_vcpu *vcpu) +{ +u32 misr = vcpu->arch.vgic_cpu.vgic_v2.vgic_misr; +u32 ret =0 ; + +if (misr &GICH_MISR_EOI ) + ret |=INT_STATUS_EOI ; +if (misr &GICH_MISR_U ) + ret |=INT_STATUS_UNDERFLOW ; + +return ret; +} +static const struct vgic_ops vgic_ops = { .get_lr = vgic_v2_get_lr, .set_lr = vgic_v2_set_lr, .get_elrsr = vgic_v2_get_elrsr, .get_eisr = vgic_v2_get_eisr, + .get_interrupt_status = vgic_v2_get_interrupt_status, };static inline struct vgic_lr vgic_get_lr (const struct kvm_vcpu *vcpu,int lr)
@@ -1062,6 +1076,11 @@ static inline u64 vgic_get_eisr(struct kvm_vcpu *vcpu)return vgic_ops.get_eisr (vcpu); } +static inline u32 vgic_get_interrupt_status (struct kvm_vcpu *vcpu) +{ +return vgic_ops.get_interrupt_status (vcpu); +} +static void vgic_retire_lr (int lr_nr,int irq,struct kvm_vcpu *vcpu) {struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
@@ -1259,11 +1278,12 @@ epilog:static bool vgic_process_maintenance (struct kvm_vcpu *vcpu) {struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu; +u32 status =vgic_get_interrupt_status (vcpu);bool level_pending = false; -kvm_debug ("MISR = %08x\n" , vgic_cpu->vgic_v2.vgic_misr); +kvm_debug ("STATUS = %08x\n" , status); -if (vgic_cpu->vgic_v2.vgic_misr &GICH_MISR_EOI ) { +if (status &INT_STATUS_EOI ) {/* * Some level interrupts have been EOIed. Clear their * active bit.
@@ -1290,7 +1310,7 @@ static bool vgic_process_maintenance(struct kvm_vcpu *vcpu) } } -if (vgic_cpu->vgic_v2.vgic_misr &GICH_MISR_U ) +if (status &INT_STATUS_UNDERFLOW ) vgic_cpu->vgic_v2.vgic_hcr &= ~GICH_HCR_UIE ;return level_pending;
Signature · 1 line
1.8.3.4
PATCH 9/18 KVM: ARM: vgic: move underflow handling to vgic_ops
In reply to Marc ZyngierMove the code dealing with LR underflow handling to its own functions, and make them accessible through vgic_ops.
--- include/kvm/arm_vgic.h | 2 ++ virt/kvm/arm/vgic.c | 28 +++++++++++++++++++++++++--- 2 files changed, 27 insertions(+), 3 deletions(-)
include/kvm/arm_vgic.h +2 -0
@@ -87,6 +87,8 @@ struct vgic_ops {u64 (*get_elrsr)(const struct kvm_vcpu *vcpu);u64 (*get_eisr)(const struct kvm_vcpu *vcpu);u32 (*get_interrupt_status)(const struct kvm_vcpu *vcpu); +void (*set_underflow)(struct kvm_vcpu *vcpu); +void (*clear_underflow)(struct kvm_vcpu *vcpu); };struct vgic_dist {
virt/kvm/arm/vgic.c +25 -3
@@ -1047,12 +1047,24 @@ static u32 vgic_v2_get_interrupt_status(const struct kvm_vcpu *vcpu)return ret; } +static void vgic_v2_set_underflow (struct kvm_vcpu *vcpu) +{ + vcpu->arch.vgic_cpu.vgic_v2.vgic_hcr |=GICH_HCR_UIE ; +} + +static void vgic_v2_clear_underflow (struct kvm_vcpu *vcpu) +{ + vcpu->arch.vgic_cpu.vgic_v2.vgic_hcr &= ~GICH_HCR_UIE ; +} +static const struct vgic_ops vgic_ops = { .get_lr = vgic_v2_get_lr, .set_lr = vgic_v2_set_lr, .get_elrsr = vgic_v2_get_elrsr, .get_eisr = vgic_v2_get_eisr, .get_interrupt_status = vgic_v2_get_interrupt_status, + .set_underflow = vgic_v2_set_underflow, + .clear_underflow = vgic_v2_clear_underflow, };static inline struct vgic_lr vgic_get_lr (const struct kvm_vcpu *vcpu,int lr)
@@ -1081,6 +1093,16 @@ static inline u32 vgic_get_interrupt_status(struct kvm_vcpu *vcpu)return vgic_ops.get_interrupt_status (vcpu); } +static inline void vgic_set_underflow (struct kvm_vcpu *vcpu) +{ + vgic_ops.set_underflow (vcpu); +} + +static inline void vgic_clear_underflow (struct kvm_vcpu *vcpu) +{ + vgic_ops.clear_underflow (vcpu); +} +static void vgic_retire_lr (int lr_nr,int irq,struct kvm_vcpu *vcpu) {struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
@@ -1262,9 +1284,9 @@ static void __kvm_vgic_flush_hwstate(struct kvm_vcpu *vcpu) epilog:if (overflow) { - vgic_cpu->vgic_v2.vgic_hcr |=GICH_HCR_UIE ; +vgic_set_underflow (vcpu); }else { - vgic_cpu->vgic_v2.vgic_hcr &= ~GICH_HCR_UIE ; +vgic_clear_underflow (vcpu);/* * We're about to run this VCPU, and we've consumed * everything the distributor had in store for
@@ -1311,7 +1333,7 @@ static bool vgic_process_maintenance(struct kvm_vcpu *vcpu) }if (status &INT_STATUS_UNDERFLOW ) - vgic_cpu->vgic_v2.vgic_hcr &= ~GICH_HCR_UIE ; +vgic_clear_underflow (vcpu);return level_pending; }
Signature · 1 line
1.8.3.4
PATCH 10/18 KVM: ARM: vgic: abstract VMCR access
In reply to Marc ZyngierInstead of directly messing with with the GICH_VMCR bits for the CPU interface save/restore code, add accessors that encode/decode the entire set of registers exposed by VMCR.
Not the most efficient thing, but given that this code is only used by the save/restore code, performance is far from being critical.
--- include/kvm/arm_vgic.h | 9 +++++++ virt/kvm/arm/vgic.c | 69 ++++++++++++++++++++++++++++++++++++++------------ 2 files changed, 62 insertions(+), 16 deletions(-)
include/kvm/arm_vgic.h +9 -0
@@ -81,6 +81,13 @@ struct vgic_lr {u8 state; }; +struct vgic_vmcr { +u32 ctlr; +u32 abpr; +u32 bpr; +u32 pmr; +}; +struct vgic_ops {struct vgic_lr (*get_lr)(const struct kvm_vcpu *,int );void (*set_lr)(struct kvm_vcpu *,int ,struct vgic_lr );
@@ -89,6 +96,8 @@ struct vgic_ops {u32 (*get_interrupt_status)(const struct kvm_vcpu *vcpu);void (*set_underflow)(struct kvm_vcpu *vcpu);void (*clear_underflow)(struct kvm_vcpu *vcpu); +void (*get_vmcr)(struct kvm_vcpu *vcpu,struct vgic_vmcr *vmcr); +void (*set_vmcr)(struct kvm_vcpu *vcpu,struct vgic_vmcr *vmcr); };struct vgic_dist {
virt/kvm/arm/vgic.c +53 -16
118 lines not on this page · raw message
Signature · 1 line
1.8.3.4
PATCH 11/18 KVM: ARM: vgic: introduce vgic_enable
In reply to Marc ZyngierMove the code dealing with kicking the VGIC on to vgic_ops.
--- include/kvm/arm_vgic.h | 1 + virt/kvm/arm/vgic.c | 29 +++++++++++++++++++++-------- 2 files changed, 22 insertions(+), 8 deletions(-)
include/kvm/arm_vgic.h +1 -0
@@ -98,6 +98,7 @@ struct vgic_ops {void (*clear_underflow)(struct kvm_vcpu *vcpu);void (*get_vmcr)(struct kvm_vcpu *vcpu,struct vgic_vmcr *vmcr);void (*set_vmcr)(struct kvm_vcpu *vcpu,struct vgic_vmcr *vmcr); +void (*enable)(struct kvm_vcpu *vcpu); };struct vgic_dist {
virt/kvm/arm/vgic.c +21 -8
@@ -1081,6 +1081,19 @@ static void vgic_v2_set_vmcr(struct kvm_vcpu *vcpu, struct vgic_vmcr *vmcrp) vcpu->arch.vgic_cpu.vgic_v2.vgic_vmcr = vmcr; } +static void vgic_v2_enable (struct kvm_vcpu *vcpu) +{ +/* +* By forcing VMCR to zero, the GIC will restore the binary +* points to their reset values. Anything else resets to zero +* anyway. +*/ + vcpu->arch.vgic_cpu.vgic_v2.vgic_vmcr =0 ; + +/* Get the show on the road... */ + vcpu->arch.vgic_cpu.vgic_v2.vgic_hcr =GICH_HCR_EN ; +} +static const struct vgic_ops vgic_ops = { .get_lr = vgic_v2_get_lr, .set_lr = vgic_v2_set_lr,
@@ -1091,6 +1104,7 @@ static const struct vgic_ops vgic_ops = { .clear_underflow = vgic_v2_clear_underflow, .get_vmcr = vgic_v2_get_vmcr, .set_vmcr = vgic_v2_set_vmcr, + .enable = vgic_v2_enable, };static inline struct vgic_lr vgic_get_lr (const struct kvm_vcpu *vcpu,int lr)
@@ -1139,6 +1153,11 @@ static void vgic_set_vmcr(struct kvm_vcpu *vcpu, struct vgic_vmcr *vmcr) vgic_ops.set_vmcr (vcpu, vmcr); } +static inline void vgic_enable (struct kvm_vcpu *vcpu) +{ + vgic_ops.enable (vcpu); +} +static void vgic_retire_lr (int lr_nr,int irq,struct kvm_vcpu *vcpu) {struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
@@ -1593,15 +1612,9 @@ int kvm_vgic_vcpu_init(struct kvm_vcpu *vcpu) vgic_cpu->vgic_irq_lr_map[i] =LR_EMPTY ; } -/* -* By forcing VMCR to zero, the GIC will restore the binary -* points to their reset values. Anything else resets to zero -* anyway. -*/ - vgic_cpu->vgic_v2.vgic_vmcr =0 ; - vgic_cpu->nr_lr = vgic_nr_lr; - vgic_cpu->vgic_v2.vgic_hcr =GICH_HCR_EN ;/* Get the show on the road... */ + +vgic_enable (vcpu);return 0 ; }
Signature · 1 line
1.8.3.4
PATCH 12/18 KVM: ARM: introduce vgic_params structure
In reply to Marc ZyngierMove all the data specific to a given GIC implementation into its own little structure.
--- include/kvm/arm_vgic.h | 11 +++++++++ virt/kvm/arm/vgic.c | 66 +++++++++++++++++++++----------------------------- 2 files changed, 39 insertions(+), 38 deletions(-)
include/kvm/arm_vgic.h +11 -0
@@ -101,6 +101,17 @@ struct vgic_ops {void (*enable)(struct kvm_vcpu *vcpu); }; +struct vgic_params { +/* Physical address of vgic virtual cpu interface */ +phys_addr_t vcpu_base; +/* Number of list registers */ +u32 nr_lr; +/* Interrupt number */ +unsigned int maint_irq; +/* Virtual control interface base address */ +void __iomem *vctrl_base; +}; +struct vgic_dist {#ifdef CONFIG_KVM_ARM_VGIC spinlock_t lock;
virt/kvm/arm/vgic.c +28 -38
192 lines not on this page · raw message
Signature · 1 line
1.8.3.4
PATCH 13/18 KVM: ARM: vgic: split GICv2 backend from the main vgic code
In reply to Marc ZyngierBrutally hack the innocent vgic code, and move the GICv2 specific code to its own file, using vgic_ops and vgic_params as a way to pass information between the two blocks.
--- arch/arm/kvm/Makefile | 1 + arch/arm64/kvm/Makefile | 2 +- include/kvm/arm_vgic.h | 3 + virt/kvm/arm/vgic-v2.c | 229 ++++++++++++++++++++++++++++++++++++++++++++++ virt/kvm/arm/vgic.c | 239 +++++++----------------------------------------- 5 files changed, 266 insertions(+), 208 deletions(-) create mode 100644 virt/kvm/arm/vgic-v2.c
arch/arm/kvm/Makefile +1 -0
@@ -21,4 +21,5 @@ obj-y += kvm-arm.o init.o interrupts.o obj-y += arm.o handle_exit.o guest.o mmu.o emulate.o reset.o obj-y += coproc.o coproc_a15.o coproc_a7.o mmio.o psci.o perf.o obj-$(CONFIG_KVM_ARM_VGIC) += $(KVM)/arm/vgic.o +obj-$(CONFIG_KVM_ARM_VGIC) += $(KVM)/arm/vgic-v2.o obj-$(CONFIG_KVM_ARM_TIMER) += $(KVM)/arm/arch_timer.o
arch/arm64/kvm/Makefile +1 -1
@@ -19,5 +19,5 @@ kvm-$(CONFIG_KVM_ARM_HOST) += emulate.o inject_fault.o regmap.o kvm-$(CONFIG_KVM_ARM_HOST) += hyp.o hyp-init.o handle_exit.o kvm-$(CONFIG_KVM_ARM_HOST) += guest.o reset.o sys_regs.o sys_regs_generic_v8.o -kvm-$(CONFIG_KVM_ARM_VGIC) += $(KVM)/arm/vgic.o +kvm-$(CONFIG_KVM_ARM_VGIC) += $(KVM)/arm/vgic.o $(KVM)/arm/vgic-v2.o kvm-$(CONFIG_KVM_ARM_TIMER) += $(KVM)/arm/arch_timer.o
include/kvm/arm_vgic.h +3 -0
@@ -213,6 +213,9 @@ bool vgic_handle_mmio(struct kvm_vcpu *vcpu, struct kvm_run *run,#define irqchip_in_kernel (k) (!!((k)->arch.vgic.vctrl_base))#define vgic_initialized (k) ((k)->arch.vgic.ready) +int vgic_v2_probe (const struct vgic_ops **ops, +const struct vgic_params **params); +#else static inline int kvm_vgic_hyp_init (void ) {
virt/kvm/arm/vgic-v2.c new file +229 -0
229 lines not on this page · raw message
virt/kvm/arm/vgic.c +32 -207
370 lines not on this page · raw message
Signature · 1 line
1.8.3.4
PATCH 14/18 arm64: KVM: remove __kvm_hyp_code_{start, end} from hyp.S
In reply to Marc ZyngierWe already have __hyp_text_{start,end} to express the boundaries of the HYP text section, and __kvm_hyp_code_{start,end} are getting in the way of a more modular world switch code.
Just turn __kvm_hyp_code_{start,end} into #defines mapping the linker-emited symbols.
--- arch/arm64/include/asm/kvm_asm.h | 6 ++++-- arch/arm64/include/asm/virt.h | 4 ++++ arch/arm64/kvm/hyp.S | 6 ------ 3 files changed, 8 insertions(+), 8 deletions(-)
arch/arm64/include/asm/kvm_asm.h +4 -2
@@ -18,6 +18,8 @@#ifndef __ARM_KVM_ASM_H__#define __ARM_KVM_ASM_H__ +#include <asm/virt.h> +/* * 0 is reserved as an invalid value. * Order *must* be kept in sync with the hyp switch code.
@@ -95,8 +97,8 @@ extern char __kvm_hyp_init_end[];extern char __kvm_hyp_vector[]; -extern char __kvm_hyp_code_start[]; -extern char __kvm_hyp_code_end[]; +#define __kvm_hyp_code_start __hyp_text_start +#define __kvm_hyp_code_end __hyp_text_endextern void __kvm_flush_vm_context (void );extern void __kvm_tlb_flush_vmid_ipa (struct kvm *kvm,phys_addr_t ipa);
arch/arm64/include/asm/virt.h +4 -0
@@ -63,6 +63,10 @@ static inline bool is_hyp_mode_mismatched(void)return __boot_cpu_mode[0 ] != __boot_cpu_mode[1 ]; } +/* The section containing the hypervisor text */ +extern char __hyp_text_start[]; +extern char __hyp_text_end[]; +#endif /* __ASSEMBLY__ */ #endif /* ! __ASM__VIRT_H */
arch/arm64/kvm/hyp.S +0 -6
@@ -36,9 +36,6 @@ .pushsection .hyp.text, "ax" .align PAGE_SHIFT -__kvm_hyp_code_start: - .globl __kvm_hyp_code_start - .macro save_common_regs // x2: base address for cpu context // x3: tmp register
@@ -851,7 +848,4 @@ ENTRY(__kvm_hyp_vector) ventry el1_error_invalid // Error 32-bit EL1 ENDPROC(__kvm_hyp_vector) -__kvm_hyp_code_end: - .globl __kvm_hyp_code_end - .popsection
Signature · 1 line
1.8.3.4
PATCH 15/18 arm64: KVM: split GICv2 world switch from hyp code
In reply to Marc ZyngierMove the GICv2 world switch code into its own file, and add the necessary indirection to the arm64 switch code.
Also introduce a new type field to the vgic_params structure.
--- arch/arm/include/asm/kvm_host.h | 5 ++ arch/arm64/include/asm/kvm_asm.h | 8 +++ arch/arm64/include/asm/kvm_host.h | 16 +++++ arch/arm64/kvm/Makefile | 4 +- arch/arm64/kvm/hyp.S | 119 +++++--------------------------- arch/arm64/kvm/vgic-v2-switch.S | 141 ++++++++++++++++++++++++++++++++++++++ include/kvm/arm_vgic.h | 7 +- virt/kvm/arm/vgic-v2.c | 15 ++-- virt/kvm/arm/vgic.c | 3 + 9 files changed, 208 insertions(+), 110 deletions(-) create mode 100644 arch/arm64/kvm/vgic-v2-switch.S
arch/arm/include/asm/kvm_host.h +5 -0
@@ -222,6 +222,11 @@ static inline int kvm_arch_dev_ioctl_check_extension(long ext)return 0 ; } +static inline void vgic_arch_setup (const struct vgic_params *vgic) +{ +BUG_ON (vgic->type !=VGIC_V2 ); +} +int kvm_perf_init (void );int kvm_perf_teardown (void );
arch/arm64/include/asm/kvm_asm.h +8 -0
@@ -104,6 +104,14 @@ extern void __kvm_flush_vm_context(void);extern void __kvm_tlb_flush_vmid_ipa (struct kvm *kvm,phys_addr_t ipa);extern int __kvm_vcpu_run (struct kvm_vcpu *vcpu); + +/* +* These are really code, but as they do not use the PCS we don't +* describe them as functions... +*/ +extern char __save_vgic_v2_state[]; +extern char __restore_vgic_v2_state[]; +#endif #endif /* __ARM_KVM_ASM_H__ */
arch/arm64/include/asm/kvm_host.h +16 -0
@@ -200,4 +200,20 @@ static inline void __cpu_init_hyp_mode(phys_addr_t boot_pgd_ptr, hyp_stack_ptr, vector_ptr); } +static inline void vgic_arch_setup (const struct vgic_params *vgic) +{ +extern char *__vgic_sr_vectors[2 ]; + +switch (vgic->type) + { +case VGIC_V2 : + __vgic_sr_vectors[0 ] = __save_vgic_v2_state; + __vgic_sr_vectors[1 ] = __restore_vgic_v2_state; +break ; + +default : +BUG (); + } +} +#endif /* __ARM64_KVM_HOST_H__ */
arch/arm64/kvm/Makefile +3 -1
@@ -19,5 +19,7 @@ kvm-$(CONFIG_KVM_ARM_HOST) += emulate.o inject_fault.o regmap.o kvm-$(CONFIG_KVM_ARM_HOST) += hyp.o hyp-init.o handle_exit.o kvm-$(CONFIG_KVM_ARM_HOST) += guest.o reset.o sys_regs.o sys_regs_generic_v8.o -kvm-$(CONFIG_KVM_ARM_VGIC) += $(KVM)/arm/vgic.o $(KVM)/arm/vgic-v2.o +kvm-$(CONFIG_KVM_ARM_VGIC) += $(KVM)/arm/vgic.o +kvm-$(CONFIG_KVM_ARM_VGIC) += $(KVM)/arm/vgic-v2.o +kvm-$(CONFIG_KVM_ARM_VGIC) += vgic-v2-switch.o kvm-$(CONFIG_KVM_ARM_TIMER) += $(KVM)/arm/arch_timer.o
arch/arm64/kvm/hyp.S +19 -100
155 lines not on this page · raw message
arch/arm64/kvm/vgic-v2-switch.S new file +141 -0
141 lines not on this page · raw message
include/kvm/arm_vgic.h +6 -1
@@ -24,7 +24,6 @@#include <linux/irqreturn.h> #include <linux/spinlock.h> #include <linux/types.h> -#include <linux/irqchip/arm-gic.h> #define VGIC_NR_IRQS 256#define VGIC_NR_SGIS 16
@@ -70,6 +69,10 @@ struct vgic_bytemap {struct kvm_vcpu ; +enum vgic_type { +VGIC_V2 ,/* Good ol' GICv2 */ +}; +#define LR_STATE_PENDING (1 << 0)#define LR_STATE_ACTIVE (1 << 1)#define LR_STATE_MASK (3 << 0)
@@ -102,6 +105,8 @@ struct vgic_ops { };struct vgic_params { +/* vgic type */ +enum vgic_type type;/* Physical address of vgic virtual cpu interface */ phys_addr_t vcpu_base;/* Number of list registers */
virt/kvm/arm/vgic-v2.c +7 -8
@@ -170,7 +170,7 @@ int vgic_v2_probe(const struct vgic_ops **ops, vgic_node =of_find_compatible_node (NULL ,NULL ,"arm,cortex-a15-gic" );if (!vgic_node) { -kvm_err ("error: no compatible vgic node in DT\n" ); +kvm_err ("error: no compatible GICv2 node in DT\n" );return -ENODEV ; }
@@ -183,15 +183,15 @@ int vgic_v2_probe(const struct vgic_ops **ops, ret =of_address_to_resource (vgic_node,2 , &vctrl_res);if (ret) { -kvm_err ("Cannot obtain VCTRL resource\n" ); - goto out_free_irq; +kvm_err ("Cannot obtain GICH resource\n" ); + goto out; } vgic->vctrl_base =of_iomap (vgic_node,2 );if (!vgic->vctrl_base) { -kvm_err ("Cannot ioremap VCTRL\n" ); +kvm_err ("Cannot ioremap GICH\n" ); ret = -ENOMEM ; - goto out_free_irq; + goto out; } vgic->nr_lr =readl_relaxed (vgic->vctrl_base +GICH_VTR );
@@ -206,7 +206,7 @@ int vgic_v2_probe(const struct vgic_ops **ops, }if (of_address_to_resource (vgic_node,3 , &vcpu_res)) { -kvm_err ("Cannot obtain VCPU resource\n" ); +kvm_err ("Cannot obtain GICV resource\n" ); ret = -ENXIO ; goto out_unmap; }
@@ -215,14 +215,13 @@ int vgic_v2_probe(const struct vgic_ops **ops,kvm_info ("%s@%llx IRQ%d\n" , vgic_node->name, vctrl_res.start, vgic->maint_irq); + vgic->type =VGIC_V2 ; *ops = &vgic_v2_ops; *params = vgic; goto out; out_unmap:iounmap (vgic->vctrl_base); -out_free_irq: -free_percpu_irq (vgic->maint_irq,kvm_get_running_vcpus ()); out:of_node_put (vgic_node);return ret;
virt/kvm/arm/vgic.c +3 -0
@@ -1531,6 +1531,9 @@ int kvm_vgic_hyp_init(void)on_each_cpu (vgic_init_maintenance_interrupt,NULL ,1 ); +/* Callback into for arch code for setup */ +vgic_arch_setup (vgic); +return 0 ; out_free_irq:
Signature · 1 line
1.8.3.4
PATCH 15/18 arm64: KVM: split GICv2 world switch from hyp code
Quoted 41 lines · Marc Zyngier
Move the GICv2 world switch code into its own file, and add the necessary indirection to the arm64 switch code.
Also introduce a new type field to the
vgic_paramsstructure.Acked-by Catalin Marinas <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- arch/arm/include/asm/kvm_host.h | 5 ++ arch/arm64/include/asm/kvm_asm.h | 8 +++ arch/arm64/include/asm/kvm_host.h | 16 +++++ arch/arm64/kvm/Makefile | 4 +- arch/arm64/kvm/hyp.S | 119 +++++--------------------------- arch/arm64/kvm/vgic-v2-switch.S | 141 ++++++++++++++++++++++++++++++++++++++ include/kvm/arm_vgic.h | 7 +- virt/kvm/arm/vgic-v2.c | 15 ++-- virt/kvm/arm/vgic.c | 3 + 9 files changed, 208 insertions(+), 110 deletions(-) create mode 100644 arch/arm64/kvm/vgic-v2-switch.Sarch/arm/include/asm/kvm_host.h +5
-0@@ -222,6 +222,11 @@ static inline int kvm_arch_dev_ioctl_check_extension(long ext)return 0 ; } +static inline void vgic_arch_setup (const struct vgic_params *vgic) +{ +BUG_ON (vgic->type !=VGIC_V2 ); +} +int kvm_perf_init (void );int kvm_perf_teardown (void );arch/arm64/include/asm/kvm_asm.h +0
-0
arch/arm64/include/asm/kvm_asm.h extern void __kvm_flush_vm_context(void); · line 111
Quoted 8 lines
arch/arm64/include/asm/kvm_asm.h +5
-0@@ -104,6 +104,14 @@ extern void __kvm_flush_vm_context(void);extern void __kvm_tlb_flush_vmid_ipa (struct kvm *kvm,phys_addr_t ipa);extern int __kvm_vcpu_run (struct kvm_vcpu *vcpu); + +/* +* These are really code, but as they do not use the PCS we don't +* describe them as functions... +*/
I don't understand what the PCS has to do with anything.
Quoted 13 lines
arch/arm64/include/asm/kvm_asm.h +3
-0+extern char __save_vgic_v2_state[]; +extern char __restore_vgic_v2_state[]; +#endif #endif /* __ARM_KVM_ASM_H__ */ arch/arm64/include/asm/kvm_host.h +0
-0@@ -200,4 +200,20 @@ static inline void __cpu_init_hyp_mode(phys_addr_t boot_pgd_ptr, hyp_stack_ptr, vector_ptr); }
arch/arm64/include/asm/kvm_host.h static inline void __cpu_init_hyp_mode(phys_addr_t boot_pgd_ptr, · line 211
Quoted 10 lines
arch/arm64/include/asm/kvm_host.h +9
-0+static inline void vgic_arch_setup (const struct vgic_params *vgic) +{ +extern char *__vgic_sr_vectors[2 ]; + +switch (vgic->type) + { +case VGIC_V2 : + __vgic_sr_vectors[0 ] = __save_vgic_v2_state; + __vgic_sr_vectors[1 ] = __restore_vgic_v2_state;
If you make this array a struct then you can use asm_offsets to generate indices for your asm.
Will
PATCH 16/18 arm64: KVM: move hcr_el2 setting into vgic-v2-switch.S
In reply to Marc ZyngierGICv3 requires the hcr_el2 switch to be tightly coupled with some of the interrupt controller's register switch.
In order to have similar code paths, start moving the hcr_el2 manipulation code to the GICv2 switch code.
--- arch/arm64/kvm/hyp.S | 7 ------- arch/arm64/kvm/vgic-v2-switch.S | 8 ++++++++ 2 files changed, 8 insertions(+), 7 deletions(-)
arch/arm64/kvm/hyp.S +0 -7
@@ -335,11 +335,6 @@ .endm .macro activate_traps - ldr x2, [x0, #VCPU_IRQ_LINES] - ldr x1, [x0, #VCPU_HCR_EL2] - orr x2, x2, x1 - msr hcr_el2, x2 - ldr x2, =(CPTR_EL2_TTA) msr cptr_el2, x2
@@ -353,8 +348,6 @@ .endm .macro deactivate_traps - mov x2, #HCR_RW - msr hcr_el2, x2 msr cptr_el2, xzr msr hstr_el2, xzr
arch/arm64/kvm/vgic-v2-switch.S +8 -0
@@ -85,6 +85,9 @@ CPU_BE( rev w5, w5 ) sub w4, w4, #1 cbnz w4, 1b 2: + mov x2, #HCR_RW + msr hcr_el2, x2 + isb .endm /*
@@ -92,6 +95,11 @@ CPU_BE( rev w5, w5 ) * x0: Register pointing to VCPU struct */ .macro restore_vgic_v2_state + ldr x2, [x0, #VCPU_IRQ_LINES] + ldr x1, [x0, #VCPU_HCR_EL2] + orr x2, x2, x1 + msr hcr_el2, x2 + /* Get VGIC VCTRL base into x2 */ ldr x2, [x0, #VCPU_KVM] kern_hyp_va x2
Signature · 1 line
1.8.3.4
PATCH 17/18 KVM: ARM: vgic: add the GICv3 backend
In reply to Marc ZyngierIntroduce the support code for emulating a GICv2 on top of GICv3 hardware.
--- include/kvm/arm_vgic.h | 26 ++++++ virt/kvm/arm/vgic-v3.c | 220 +++++++++++++++++++++++++++++++++++++++++++++++++ virt/kvm/arm/vgic.c | 2 + 3 files changed, 248 insertions(+) create mode 100644 virt/kvm/arm/vgic-v3.c
include/kvm/arm_vgic.h +26 -0
@@ -32,6 +32,7 @@#define VGIC_NR_SHARED_IRQS (VGIC_NR_IRQS - VGIC_NR_PRIVATE_IRQS)#define VGIC_MAX_CPUS KVM_MAX_VCPUS#define VGIC_MAX_LRS (1 << 6) +#define VGIC_V3_MAX_LRS 16/* Sanity checks... */ #if (VGIC_MAX_CPUS >8 )
@@ -71,6 +72,7 @@ struct kvm_vcpu;enum vgic_type {VGIC_V2 ,/* Good ol' GICv2 */ +VGIC_V3 ,/* v2 on v3, really */ };#define LR_STATE_PENDING (1 << 0)
@@ -169,6 +171,19 @@ struct vgic_v2_cpu_if {u32 vgic_lr[VGIC_MAX_LRS ]; }; +struct vgic_v3_cpu_if { +#ifdef CONFIG_ARM_GIC_V3 +u32 vgic_hcr; +u32 vgic_vmcr; +u32 vgic_misr;/* Saved only */ +u32 vgic_eisr;/* Saved only */ +u32 vgic_elrsr;/* Saved only */ +u32 vgic_ap0r[4 ]; +u32 vgic_ap1r[4 ]; +u64 vgic_lr[VGIC_V3_MAX_LRS ]; +#endif +}; +struct vgic_cpu {#ifdef CONFIG_KVM_ARM_VGIC /* per IRQ to LR mapping */
@@ -187,6 +202,7 @@ struct vgic_cpu {/* CPU vif control registers for world switch */ union {struct vgic_v2_cpu_if vgic_v2; +struct vgic_v3_cpu_if vgic_v3; };#endif };
@@ -220,6 +236,16 @@ bool vgic_handle_mmio(struct kvm_vcpu *vcpu, struct kvm_run *run,int vgic_v2_probe (const struct vgic_ops **ops,const struct vgic_params **params); +#ifdef CONFIG_ARM_GIC_V3 +int vgic_v3_probe (const struct vgic_ops **ops, +const struct vgic_params **params); +#else +static inline int vgic_v3_probe (const struct vgic_ops **ops, +const struct vgic_params **params) +{ +return -ENODEV ; +} +#endif #else static inline int kvm_vgic_hyp_init (void )
virt/kvm/arm/vgic-v3.c new file +220 -0
220 lines not on this page · raw message
virt/kvm/arm/vgic.c +2 -0
@@ -1514,6 +1514,8 @@ int kvm_vgic_hyp_init(void) ret =vgic_v2_probe (&vgic_ops, &vgic);if (ret) + ret =vgic_v3_probe (&vgic_ops, &vgic); +if (ret)return ret; ret = request_percpu_irq(vgic->maint_irq, vgic_maintenance_handler,
Signature · 1 line
1.8.3.4
PATCH 17/18 KVM: ARM: vgic: add the GICv3 backend
Quoted 17 lines · Marc Zyngier
Introduce the support code for emulating a GICv2 on top of GICv3 hardware.
Acked-by Catalin Marinas <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- include/kvm/arm_vgic.h | 26 ++++++ virt/kvm/arm/vgic-v3.c | 220 +++++++++++++++++++++++++++++++++++++++++++++++++ virt/kvm/arm/vgic.c | 2 + 3 files changed, 248 insertions(+) create mode 100644 virt/kvm/arm/vgic-v3.cinclude/kvm/arm_vgic.h +0
-0
include/kvm/arm_vgic.h line 35
Quoted 4 lines
include/kvm/arm_vgic.h +1
-0@@ -32,6 +32,7 @@#define VGIC_NR_SHARED_IRQS (VGIC_NR_IRQS - VGIC_NR_PRIVATE_IRQS)#define VGIC_MAX_CPUS KVM_MAX_VCPUS#define VGIC_MAX_LRS (1 << 6) +#define VGIC_V3_MAX_LRS 16
Since we have fewer list registers, doesn't the code in vgic.c need updating to honour the relevant bounds? (e.g. the use of find_first_zero_bit in vgic_queue_irq).
Will
PATCH 17/18 KVM: ARM: vgic: add the GICv3 backend
Quoted 13 lines · Will Deacon
Quoted 12 lines · Marc Zyngier
Introduce the support code for emulating a GICv2 on top of GICv3 hardware.
Acked-by Catalin Marinas <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- include/kvm/arm_vgic.h | 26 ++++++ virt/kvm/arm/vgic-v3.c | 220 +++++++++++++++++++++++++++++++++++++++++++++++++ virt/kvm/arm/vgic.c | 2 + 3 files changed, 248 insertions(+) create mode 100644 virt/kvm/arm/vgic-v3.c
Quoted 13 lines
include/kvm/arm_vgic.h line 35
Quoted 9 lines
include/kvm/arm_vgic.h +1
-0@@ -32,6 +32,7 @@#define VGIC_NR_SHARED_IRQS (VGIC_NR_IRQS - VGIC_NR_PRIVATE_IRQS)#define VGIC_MAX_CPUS KVM_MAX_VCPUS#define VGIC_MAX_LRS (1 << 6) +#define VGIC_V3_MAX_LRS 16Since we have fewer list registers, doesn't the code in
vgic.cneed updating to honour the relevant bounds? (e.g. the use offind_first_zero_bitinvgic_queue_irq).
I'm confused. We've always used a variable, as even with GICv2, we usually have far less list registers than the maximum (4 vs 64). Looking at the code you mention, I see this:
/* Try to use another LR for this interrupt */ lr = find_first_zero_bit((unsigned long *)vgic_cpu->lr_used, vgic->nr_lr);
Am I looking at the wrong place?
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
PATCH 17/18 KVM: ARM: vgic: add the GICv3 backend
Quoted 32 lines · Marc Zyngier
Quoted 13 lines · Will Deacon
Quoted 12 lines · Marc Zyngier
Introduce the support code for emulating a GICv2 on top of GICv3 hardware.
Acked-by Catalin Marinas <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- include/kvm/arm_vgic.h | 26 ++++++ virt/kvm/arm/vgic-v3.c | 220 +++++++++++++++++++++++++++++++++++++++++++++++++ virt/kvm/arm/vgic.c | 2 + 3 files changed, 248 insertions(+) create mode 100644 virt/kvm/arm/vgic-v3.cQuoted 13 lines
include/kvm/arm_vgic.h line 35
Quoted 9 lines
include/kvm/arm_vgic.h +1
-0@@ -32,6 +32,7 @@#define VGIC_NR_SHARED_IRQS (VGIC_NR_IRQS - VGIC_NR_PRIVATE_IRQS)#define VGIC_MAX_CPUS KVM_MAX_VCPUS#define VGIC_MAX_LRS (1 << 6) +#define VGIC_V3_MAX_LRS 16Since we have fewer list registers, doesn't the code in
vgic.cneed updating to honour the relevant bounds? (e.g. the use offind_first_zero_bitinvgic_queue_irq).I'm confused. We've always used a variable, as even with GICv2, we usually have far less list registers than the maximum (4 vs 64). Looking at the code you mention, I see this:
Quoted 5 lines
/* Try to use another LR for this interrupt */ lr = find_first_zero_bit((unsigned long *)vgic_cpu->lr_used, vgic->nr_lr);Am I looking at the wrong place?
No, I'm being stupid. I was looking at mainline rather than your GICv3 branch and thought we were still using VGIC_MAX_LRS.
Will
PATCH 18/18 arm64: KVM: vgic: add GICv3 world switch
In reply to Marc ZyngierIntroduce the GICv3 world switch code and helper functions, enabling GICv2 emulation on GICv3 hardware.
--- arch/arm64/include/asm/kvm_asm.h | 4 + arch/arm64/include/asm/kvm_host.h | 5 + arch/arm64/kernel/asm-offsets.c | 8 ++ arch/arm64/kvm/Makefile | 2 + arch/arm64/kvm/vgic-v3-switch.S | 275 ++++++++++++++++++++++++++++++++++++++ 5 files changed, 294 insertions(+) create mode 100644 arch/arm64/kvm/vgic-v3-switch.S
arch/arm64/include/asm/kvm_asm.h +4 -0
@@ -105,12 +105,16 @@ extern void __kvm_tlb_flush_vmid_ipa(struct kvm *kvm, phys_addr_t ipa);extern int __kvm_vcpu_run (struct kvm_vcpu *vcpu); +extern u64 __vgic_v3_get_ich_vtr_el2 (void ); +/* * These are really code, but as they do not use the PCS we don't * describe them as functions... */ extern char __save_vgic_v2_state[];extern char __restore_vgic_v2_state[]; +extern char __save_vgic_v3_state[]; +extern char __restore_vgic_v3_state[];#endif
arch/arm64/include/asm/kvm_host.h +5 -0
@@ -211,6 +211,11 @@ static inline void vgic_arch_setup(const struct vgic_params *vgic) __vgic_sr_vectors[1 ] = __restore_vgic_v2_state;break ; +case VGIC_V3 : + __vgic_sr_vectors[0 ] = __save_vgic_v3_state; + __vgic_sr_vectors[1 ] = __restore_vgic_v3_state; +break ; +default :BUG (); }
arch/arm64/kernel/asm-offsets.c +8 -0
@@ -136,6 +136,14 @@ int main(void) DEFINE(VGIC_V2_CPU_ELRSR, offsetof(struct vgic_cpu, vgic_v2.vgic_elrsr)); DEFINE(VGIC_V2_CPU_APR, offsetof(struct vgic_cpu, vgic_v2.vgic_apr)); DEFINE(VGIC_V2_CPU_LR, offsetof(struct vgic_cpu, vgic_v2.vgic_lr)); + DEFINE(VGIC_V3_CPU_HCR, offsetof(struct vgic_cpu, vgic_v3.vgic_hcr)); + DEFINE(VGIC_V3_CPU_VMCR, offsetof(struct vgic_cpu, vgic_v3.vgic_vmcr)); + DEFINE(VGIC_V3_CPU_MISR, offsetof(struct vgic_cpu, vgic_v3.vgic_misr)); + DEFINE(VGIC_V3_CPU_EISR, offsetof(struct vgic_cpu, vgic_v3.vgic_eisr)); + DEFINE(VGIC_V3_CPU_ELRSR, offsetof(struct vgic_cpu, vgic_v3.vgic_elrsr)); + DEFINE(VGIC_V3_CPU_AP0R, offsetof(struct vgic_cpu, vgic_v3.vgic_ap0r)); + DEFINE(VGIC_V3_CPU_AP1R, offsetof(struct vgic_cpu, vgic_v3.vgic_ap1r)); + DEFINE(VGIC_V3_CPU_LR, offsetof(struct vgic_cpu, vgic_v3.vgic_lr));DEFINE (VGIC_CPU_NR_LR , offsetof(struct vgic_cpu , nr_lr)); DEFINE(KVM_VTTBR, offsetof(struct kvm, arch.vttbr)); DEFINE(KVM_VGIC_VCTRL, offsetof(struct kvm, arch.vgic.vctrl_base));
arch/arm64/kvm/Makefile +2 -0
@@ -22,4 +22,6 @@ kvm-$(CONFIG_KVM_ARM_HOST) += guest.o reset.o sys_regs.o sys_regs_generic_v8.o kvm-$(CONFIG_KVM_ARM_VGIC) += $(KVM)/arm/vgic.o kvm-$(CONFIG_KVM_ARM_VGIC) += $(KVM)/arm/vgic-v2.o kvm-$(CONFIG_KVM_ARM_VGIC) += vgic-v2-switch.o +kvm-$(CONFIG_KVM_ARM_VGIC) += $(KVM)/arm/vgic-v3.o +kvm-$(CONFIG_KVM_ARM_VGIC) += vgic-v3-switch.o kvm-$(CONFIG_KVM_ARM_TIMER) += $(KVM)/arm/arch_timer.o
arch/arm64/kvm/vgic-v3-switch.S new file +275 -0
275 lines not on this page · raw message
Signature · 1 line
1.8.3.4
PATCH 18/18 arm64: KVM: vgic: add GICv3 world switch
Quoted 13 lines · Marc Zyngier
Introduce the GICv3 world switch code and helper functions, enabling GICv2 emulation on GICv3 hardware.
Acked-by Catalin Marinas <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- arch/arm64/include/asm/kvm_asm.h | 4 + arch/arm64/include/asm/kvm_host.h | 5 + arch/arm64/kernel/asm-offsets.c | 8 ++ arch/arm64/kvm/Makefile | 2 + arch/arm64/kvm/vgic-v3-switch.S | 275 ++++++++++++++++++++++++++++++++++++++ 5 files changed, 294 insertions(+) create mode 100644 arch/arm64/kvm/vgic-v3-switch.S
[...]
arch/arm64/kvm/vgic-v3-switch.S line 45
Quoted 10 lines
+.macro
save_vgic_v3_statearch/arm64/kvm/vgic-v3-switch.S new file +9
-0+ /* Compute the address of struct vgic_cpu */ + add x3, x0, #VCPU_VGIC_CPU + + mrs x5, ICC_SRE_EL2 + orr x5, x5, #(1 << 3) + msr ICC_SRE_EL2, x5 + isb + + dsb st
Aha, so you *do* use a dsb here (see my comments to patch 1/18). Does it really need to be full-system, and does -st apply to an msr?
Saying that, why would ICC_SRE_EL2 not already be initialised by the hyp setup code?
Quoted 6 lines
arch/arm64/kvm/vgic-v3-switch.S new file +6
-0+ + /* Save all interesting registers */ + mrs x4, ICH_HCR_EL2 + mrs x5, ICH_VMCR_EL2 + mrs x6, ICH_MISR_EL2 + mrs x7, ICH_EISR_EL2
arch/arm64/kvm/vgic-v3-switch.S line 61
Quoted 10 lines
arch/arm64/kvm/vgic-v3-switch.S new file +10
-0+ mrs x8, ICH_ELSR_EL2 + + str w4, [x3, #VGIC_V3_CPU_HCR] + str w5, [x3, #VGIC_V3_CPU_VMCR] + str w6, [x3, #VGIC_V3_CPU_MISR] + str w7, [x3, #VGIC_V3_CPU_EISR] + str w8, [x3, #VGIC_V3_CPU_ELRSR] + + msr ICH_HCR_EL2, xzr + isb
Why do you need an isb here?
Quoted 72 lines
arch/arm64/kvm/vgic-v3-switch.S new file +72
-0+ + mrs x21, ICH_VTR_EL2 + and w22, w21, #0xf + mov w23, #0xf + sub w23, w23, w22 // How many regs we have to skip + + adr x24, 1f + add x24, x24, x23, lsl #2 + br x24 + +1: + mrs x20, ICH_LR15_EL2 + mrs x19, ICH_LR14_EL2 + mrs x18, ICH_LR13_EL2 + mrs x17, ICH_LR12_EL2 + mrs x16, ICH_LR11_EL2 + mrs x15, ICH_LR10_EL2 + mrs x14, ICH_LR9_EL2 + mrs x13, ICH_LR8_EL2 + mrs x12, ICH_LR7_EL2 + mrs x11, ICH_LR6_EL2 + mrs x10, ICH_LR5_EL2 + mrs x9, ICH_LR4_EL2 + mrs x8, ICH_LR3_EL2 + mrs x7, ICH_LR2_EL2 + mrs x6, ICH_LR1_EL2 + mrs x5, ICH_LR0_EL2 + + adr x24, 1f + add x24, x24, x23, lsl #2 // adr(1f) + unimp_nr*4 + br x24 + +1: + str x20, [x3, #(VGIC_V3_CPU_LR + 15*8)] + str x19, [x3, #(VGIC_V3_CPU_LR + 14*8)] + str x18, [x3, #(VGIC_V3_CPU_LR + 13*8)] + str x17, [x3, #(VGIC_V3_CPU_LR + 12*8)] + str x16, [x3, #(VGIC_V3_CPU_LR + 11*8)] + str x15, [x3, #(VGIC_V3_CPU_LR + 10*8)] + str x14, [x3, #(VGIC_V3_CPU_LR + 9*8)] + str x13, [x3, #(VGIC_V3_CPU_LR + 8*8)] + str x12, [x3, #(VGIC_V3_CPU_LR + 7*8)] + str x11, [x3, #(VGIC_V3_CPU_LR + 6*8)] + str x10, [x3, #(VGIC_V3_CPU_LR + 5*8)] + str x9, [x3, #(VGIC_V3_CPU_LR + 4*8)] + str x8, [x3, #(VGIC_V3_CPU_LR + 3*8)] + str x7, [x3, #(VGIC_V3_CPU_LR + 2*8)] + str x6, [x3, #(VGIC_V3_CPU_LR + 1*8)] + str x5, [x3, #VGIC_V3_CPU_LR] + + lsr w22, w21, #29 // Get PRIbits + cmp w22, #4 // 5 bits + b.eq 5f + cmp w22, #5 // 6 bits + b.eq 6f + // 7 bits + mrs x20, ICH_AP0R3_EL2 + str w20, [x3, #(VGIC_V3_CPU_AP0R + 3*4)] + mrs x19, ICH_AP0R2_EL2 + str w19, [x3, #(VGIC_V3_CPU_AP0R + 2*4)] +6: mrs x18, ICH_AP0R1_EL2 + str w18, [x3, #(VGIC_V3_CPU_AP0R + 1*4)] +5: mrs x17, ICH_AP0R0_EL2 + str w17, [x3, #VGIC_V3_CPU_AP0R] + + cmp w22, #4 // 5 bits + b.eq 5f + cmp w22, #5 // 6 bits + b.eq 6f + // 7 bits + mrs x20, ICH_AP1R3_EL2 + str w20, [x3, #(VGIC_V3_CPU_AP1R + 3*4)]
arch/arm64/kvm/vgic-v3-switch.S line 143
Quoted 10 lines
arch/arm64/kvm/vgic-v3-switch.S new file +10
-0+ mrs x19, ICH_AP1R2_EL2 + str w19, [x3, #(VGIC_V3_CPU_AP1R + 2*4)] +6: mrs x18, ICH_AP1R1_EL2 + str w18, [x3, #(VGIC_V3_CPU_AP1R + 1*4)] +5: mrs x17, ICH_AP1R0_EL2 + str w17, [x3, #VGIC_V3_CPU_AP1R] + + mov x2, #HCR_RW + msr hcr_el2, x2 + isb
Same here -- HCR_EL2.RW is for lower exception levels.
Quoted 9 lines
arch/arm64/kvm/vgic-v3-switch.S new file +9
-0+ + // Restore SRE_EL1 access + mov x5, #1 + msr ICC_SRE_EL1, x5 +.endm + +/* + * Restore the VGIC CPU state from memory + * x0: Register pointing to VCPU struct
arch/arm64/kvm/vgic-v3-switch.S line 162
Quoted 10 lines
arch/arm64/kvm/vgic-v3-switch.S new file +10
-0+ */ +.macro restore_vgic_v3_state + ldr x2, [x0, #VCPU_IRQ_LINES] + ldr x1, [x0, #VCPU_HCR_EL2] + orr x2, x2, x1 + msr hcr_el2, x2 + + // Disable SRE_EL1 access + msr ICC_SRE_EL1, xzr + isb
We're executing at EL2, why the isb?
Quoted 81 lines
arch/arm64/kvm/vgic-v3-switch.S new file +81
-0+ + /* Compute the address of struct vgic_cpu */ + add x3, x0, #VCPU_VGIC_CPU + + /* Restore all interesting registers */ + ldr w4, [x3, #VGIC_V3_CPU_HCR] + ldr w5, [x3, #VGIC_V3_CPU_VMCR] + + msr ICH_HCR_EL2, x4 + msr ICH_VMCR_EL2, x5 + + mrs x21, ICH_VTR_EL2 + + lsr w22, w21, #29 // Get PRIbits + cmp w22, #4 // 5 bits + b.eq 5f + cmp w22, #5 // 6 bits + b.eq 6f + // 7 bits + ldr w20, [x3, #(VGIC_V3_CPU_AP1R + 3*4)] + msr ICH_AP1R3_EL2, x20 + ldr w19, [x3, #(VGIC_V3_CPU_AP1R + 2*4)] + msr ICH_AP1R2_EL2, x19 +6: ldr w18, [x3, #(VGIC_V3_CPU_AP1R + 1*4)] + msr ICH_AP1R1_EL2, x18 +5: ldr w17, [x3, #VGIC_V3_CPU_AP1R] + msr ICH_AP1R0_EL2, x17 + + cmp w22, #4 // 5 bits + b.eq 5f + cmp w22, #5 // 6 bits + b.eq 6f + // 7 bits + ldr w20, [x3, #(VGIC_V3_CPU_AP0R + 3*4)] + msr ICH_AP0R3_EL2, x20 + ldr w19, [x3, #(VGIC_V3_CPU_AP0R + 2*4)] + msr ICH_AP0R2_EL2, x19 +6: ldr w18, [x3, #(VGIC_V3_CPU_AP0R + 1*4)] + msr ICH_AP0R1_EL2, x18 +5: ldr w17, [x3, #VGIC_V3_CPU_AP0R] + msr ICH_AP0R0_EL2, x17 + + and w22, w21, #0xf + mov w23, #0xf + sub w23, w23, w22 // How many regs we have to skip + + adr x24, 1f + add x24, x24, x23, lsl #2 // adr(1f) + unimp_nr*4 + br x24 + +1: + ldr x20, [x3, #(VGIC_V3_CPU_LR + 15*8)] + ldr x19, [x3, #(VGIC_V3_CPU_LR + 14*8)] + ldr x18, [x3, #(VGIC_V3_CPU_LR + 13*8)] + ldr x17, [x3, #(VGIC_V3_CPU_LR + 12*8)] + ldr x16, [x3, #(VGIC_V3_CPU_LR + 11*8)] + ldr x15, [x3, #(VGIC_V3_CPU_LR + 10*8)] + ldr x14, [x3, #(VGIC_V3_CPU_LR + 9*8)] + ldr x13, [x3, #(VGIC_V3_CPU_LR + 8*8)] + ldr x12, [x3, #(VGIC_V3_CPU_LR + 7*8)] + ldr x11, [x3, #(VGIC_V3_CPU_LR + 6*8)] + ldr x10, [x3, #(VGIC_V3_CPU_LR + 5*8)] + ldr x9, [x3, #(VGIC_V3_CPU_LR + 4*8)] + ldr x8, [x3, #(VGIC_V3_CPU_LR + 3*8)] + ldr x7, [x3, #(VGIC_V3_CPU_LR + 2*8)] + ldr x6, [x3, #(VGIC_V3_CPU_LR + 1*8)] + ldr x5, [x3, #VGIC_V3_CPU_LR] + + adr x24, 1f + add x24, x24, x23, lsl #2 + br x24 + +1: + msr ICH_LR15_EL2, x20 + msr ICH_LR14_EL2, x19 + msr ICH_LR13_EL2, x18 + msr ICH_LR12_EL2, x17 + msr ICH_LR11_EL2, x16 + msr ICH_LR10_EL2, x15 + msr ICH_LR9_EL2, x14 + msr ICH_LR8_EL2, x13
arch/arm64/kvm/vgic-v3-switch.S line 253
Quoted 10 lines
arch/arm64/kvm/vgic-v3-switch.S new file +10
-0+ msr ICH_LR7_EL2, x12 + msr ICH_LR6_EL2, x11 + msr ICH_LR5_EL2, x10 + msr ICH_LR4_EL2, x9 + msr ICH_LR3_EL2, x8 + msr ICH_LR2_EL2, x7 + msr ICH_LR1_EL2, x6 + msr ICH_LR0_EL2, x5 + + dsb st
Now you have a dsb without an isb! There's no consistency here and the docs don't tell me much either.
arch/arm64/kvm/vgic-v3-switch.S line 257
Quoted 4 lines
arch/arm64/kvm/vgic-v3-switch.S new file +4
-0+ + mrs x5, ICC_SRE_EL2 + and x5, x5, #~(1 << 3) + msr ICC_SRE_EL2, x5
Why are you doing this? Should't we leave bits 0 and 3 of SRE_EL2 always set?
Will
PATCH 18/18 arm64: KVM: vgic: add GICv3 world switch
Quoted 17 lines · Will Deacon
Quoted 13 lines · Marc Zyngier
Introduce the GICv3 world switch code and helper functions, enabling GICv2 emulation on GICv3 hardware.
Acked-by Catalin Marinas <[email protected]>Signed-off-by Marc Zyngier <[email protected]>--- arch/arm64/include/asm/kvm_asm.h | 4 + arch/arm64/include/asm/kvm_host.h | 5 + arch/arm64/kernel/asm-offsets.c | 8 ++ arch/arm64/kvm/Makefile | 2 + arch/arm64/kvm/vgic-v3-switch.S | 275 ++++++++++++++++++++++++++++++++++++++ 5 files changed, 294 insertions(+) create mode 100644 arch/arm64/kvm/vgic-v3-switch.S[...]
Quoted 13 lines
arch/arm64/kvm/vgic-v3-switch.S line 45
Quoted 10 lines
+.macro
save_vgic_v3_statearch/arm64/kvm/vgic-v3-switch.S new file +9
-0+ /* Compute the address of struct vgic_cpu */ + add x3, x0, #VCPU_VGIC_CPU + + mrs x5, ICC_SRE_EL2 + orr x5, x5, #(1 << 3) + msr ICC_SRE_EL2, x5 + isb + + dsb stAha, so you *do* use a dsb here (see my comments to patch 1/18). Does it really need to be full-system, and does -st apply to an msr?
It seems to apply indeed, according to the architects. As for the locality, this is unclear at best...
Quoted 2 lines
Saying that, why would
ICC_SRE_EL2not already be initialised by the hyp setup code?
We switch the SRE bit at EL2 when we enter the guest in order to prevent the guest from using SRE itself. Here, we come back to the host, hence restoring the SRE setting.
Quoted 10 lines
Quoted 10 lines
arch/arm64/kvm/vgic-v3-switch.S new file +10
-0+ + /* Save all interesting registers */ + mrs x4, ICH_HCR_EL2 + mrs x5, ICH_VMCR_EL2 + mrs x6, ICH_MISR_EL2 + mrs x7, ICH_EISR_EL2 + mrs x8, ICH_ELSR_EL2 + + str w4, [x3, #VGIC_V3_CPU_HCR] + str w5, [x3, #VGIC_V3_CPU_VMCR]
Quoted 8 lines
arch/arm64/kvm/vgic-v3-switch.S line 61
Quoted 6 lines
arch/arm64/kvm/vgic-v3-switch.S new file +6
-0+ str w6, [x3, #VGIC_V3_CPU_MISR] + str w7, [x3, #VGIC_V3_CPU_EISR] + str w8, [x3, #VGIC_V3_CPU_ELRSR] + + msr ICH_HCR_EL2, xzr + isbWhy do you need an isb here?
Hmmm. I can't really recall why I did put that one in, so I guess I should drop it.
Quoted 76 lines
Quoted 76 lines
arch/arm64/kvm/vgic-v3-switch.S new file +76
-0+ + mrs x21, ICH_VTR_EL2 + and w22, w21, #0xf + mov w23, #0xf + sub w23, w23, w22 // How many regs we have to skip + + adr x24, 1f + add x24, x24, x23, lsl #2 + br x24 + +1: + mrs x20, ICH_LR15_EL2 + mrs x19, ICH_LR14_EL2 + mrs x18, ICH_LR13_EL2 + mrs x17, ICH_LR12_EL2 + mrs x16, ICH_LR11_EL2 + mrs x15, ICH_LR10_EL2 + mrs x14, ICH_LR9_EL2 + mrs x13, ICH_LR8_EL2 + mrs x12, ICH_LR7_EL2 + mrs x11, ICH_LR6_EL2 + mrs x10, ICH_LR5_EL2 + mrs x9, ICH_LR4_EL2 + mrs x8, ICH_LR3_EL2 + mrs x7, ICH_LR2_EL2 + mrs x6, ICH_LR1_EL2 + mrs x5, ICH_LR0_EL2 + + adr x24, 1f + add x24, x24, x23, lsl #2 // adr(1f) + unimp_nr*4 + br x24 + +1: + str x20, [x3, #(VGIC_V3_CPU_LR + 15*8)] + str x19, [x3, #(VGIC_V3_CPU_LR + 14*8)] + str x18, [x3, #(VGIC_V3_CPU_LR + 13*8)] + str x17, [x3, #(VGIC_V3_CPU_LR + 12*8)] + str x16, [x3, #(VGIC_V3_CPU_LR + 11*8)] + str x15, [x3, #(VGIC_V3_CPU_LR + 10*8)] + str x14, [x3, #(VGIC_V3_CPU_LR + 9*8)] + str x13, [x3, #(VGIC_V3_CPU_LR + 8*8)] + str x12, [x3, #(VGIC_V3_CPU_LR + 7*8)] + str x11, [x3, #(VGIC_V3_CPU_LR + 6*8)] + str x10, [x3, #(VGIC_V3_CPU_LR + 5*8)] + str x9, [x3, #(VGIC_V3_CPU_LR + 4*8)] + str x8, [x3, #(VGIC_V3_CPU_LR + 3*8)] + str x7, [x3, #(VGIC_V3_CPU_LR + 2*8)] + str x6, [x3, #(VGIC_V3_CPU_LR + 1*8)] + str x5, [x3, #VGIC_V3_CPU_LR] + + lsr w22, w21, #29 // Get PRIbits + cmp w22, #4 // 5 bits + b.eq 5f + cmp w22, #5 // 6 bits + b.eq 6f + // 7 bits + mrs x20, ICH_AP0R3_EL2 + str w20, [x3, #(VGIC_V3_CPU_AP0R + 3*4)] + mrs x19, ICH_AP0R2_EL2 + str w19, [x3, #(VGIC_V3_CPU_AP0R + 2*4)] +6: mrs x18, ICH_AP0R1_EL2 + str w18, [x3, #(VGIC_V3_CPU_AP0R + 1*4)] +5: mrs x17, ICH_AP0R0_EL2 + str w17, [x3, #VGIC_V3_CPU_AP0R] + + cmp w22, #4 // 5 bits + b.eq 5f + cmp w22, #5 // 6 bits + b.eq 6f + // 7 bits + mrs x20, ICH_AP1R3_EL2 + str w20, [x3, #(VGIC_V3_CPU_AP1R + 3*4)] + mrs x19, ICH_AP1R2_EL2 + str w19, [x3, #(VGIC_V3_CPU_AP1R + 2*4)] +6: mrs x18, ICH_AP1R1_EL2 + str w18, [x3, #(VGIC_V3_CPU_AP1R + 1*4)]
Quoted 8 lines
arch/arm64/kvm/vgic-v3-switch.S line 143
Quoted 6 lines
+5: mrs x17,
ICH_AP1R0_EL2arch/arm64/kvm/vgic-v3-switch.S new file +5
-0+ str w17, [x3, #VGIC_V3_CPU_AP1R] + + mov x2, #HCR_RW + msr hcr_el2, x2 + isbSame here -- HCR_EL2.RW is for lower exception levels.
Yes, and it has an influence on the following setting of SRE just below. We have to absolutely sure that we've exited guest context before touching ICC_SRE_EL1.
Quoted 13 lines
Quoted 13 lines
arch/arm64/kvm/vgic-v3-switch.S new file +13
-0+ + // Restore SRE_EL1 access + mov x5, #1 + msr ICC_SRE_EL1, x5 +.endm + +/* + * Restore the VGIC CPU state from memory + * x0: Register pointing to VCPU struct + */ +.macro restore_vgic_v3_state + ldr x2, [x0, #VCPU_IRQ_LINES] + ldr x1, [x0, #VCPU_HCR_EL2]
Quoted 8 lines
arch/arm64/kvm/vgic-v3-switch.S line 162
Quoted 6 lines
arch/arm64/kvm/vgic-v3-switch.S new file +6
-0+ orr x2, x2, x1 + msr hcr_el2, x2 + + // Disable SRE_EL1 access + msr ICC_SRE_EL1, xzr + isbWe're executing at EL2, why the isb?
If we don't make sure that SRE is effectively disabled at EL1, interrupts injected as group 0 will appear as FIQs on the guest side. Obviously, we don't want that.
Quoted 86 lines
Quoted 86 lines
arch/arm64/kvm/vgic-v3-switch.S new file +86
-0+ + /* Compute the address of struct vgic_cpu */ + add x3, x0, #VCPU_VGIC_CPU + + /* Restore all interesting registers */ + ldr w4, [x3, #VGIC_V3_CPU_HCR] + ldr w5, [x3, #VGIC_V3_CPU_VMCR] + + msr ICH_HCR_EL2, x4 + msr ICH_VMCR_EL2, x5 + + mrs x21, ICH_VTR_EL2 + + lsr w22, w21, #29 // Get PRIbits + cmp w22, #4 // 5 bits + b.eq 5f + cmp w22, #5 // 6 bits + b.eq 6f + // 7 bits + ldr w20, [x3, #(VGIC_V3_CPU_AP1R + 3*4)] + msr ICH_AP1R3_EL2, x20 + ldr w19, [x3, #(VGIC_V3_CPU_AP1R + 2*4)] + msr ICH_AP1R2_EL2, x19 +6: ldr w18, [x3, #(VGIC_V3_CPU_AP1R + 1*4)] + msr ICH_AP1R1_EL2, x18 +5: ldr w17, [x3, #VGIC_V3_CPU_AP1R] + msr ICH_AP1R0_EL2, x17 + + cmp w22, #4 // 5 bits + b.eq 5f + cmp w22, #5 // 6 bits + b.eq 6f + // 7 bits + ldr w20, [x3, #(VGIC_V3_CPU_AP0R + 3*4)] + msr ICH_AP0R3_EL2, x20 + ldr w19, [x3, #(VGIC_V3_CPU_AP0R + 2*4)] + msr ICH_AP0R2_EL2, x19 +6: ldr w18, [x3, #(VGIC_V3_CPU_AP0R + 1*4)] + msr ICH_AP0R1_EL2, x18 +5: ldr w17, [x3, #VGIC_V3_CPU_AP0R] + msr ICH_AP0R0_EL2, x17 + + and w22, w21, #0xf + mov w23, #0xf + sub w23, w23, w22 // How many regs we have to skip + + adr x24, 1f + add x24, x24, x23, lsl #2 // adr(1f) + unimp_nr*4 + br x24 + +1: + ldr x20, [x3, #(VGIC_V3_CPU_LR + 15*8)] + ldr x19, [x3, #(VGIC_V3_CPU_LR + 14*8)] + ldr x18, [x3, #(VGIC_V3_CPU_LR + 13*8)] + ldr x17, [x3, #(VGIC_V3_CPU_LR + 12*8)] + ldr x16, [x3, #(VGIC_V3_CPU_LR + 11*8)] + ldr x15, [x3, #(VGIC_V3_CPU_LR + 10*8)] + ldr x14, [x3, #(VGIC_V3_CPU_LR + 9*8)] + ldr x13, [x3, #(VGIC_V3_CPU_LR + 8*8)] + ldr x12, [x3, #(VGIC_V3_CPU_LR + 7*8)] + ldr x11, [x3, #(VGIC_V3_CPU_LR + 6*8)] + ldr x10, [x3, #(VGIC_V3_CPU_LR + 5*8)] + ldr x9, [x3, #(VGIC_V3_CPU_LR + 4*8)] + ldr x8, [x3, #(VGIC_V3_CPU_LR + 3*8)] + ldr x7, [x3, #(VGIC_V3_CPU_LR + 2*8)] + ldr x6, [x3, #(VGIC_V3_CPU_LR + 1*8)] + ldr x5, [x3, #VGIC_V3_CPU_LR] + + adr x24, 1f + add x24, x24, x23, lsl #2 + br x24 + +1: + msr ICH_LR15_EL2, x20 + msr ICH_LR14_EL2, x19 + msr ICH_LR13_EL2, x18 + msr ICH_LR12_EL2, x17 + msr ICH_LR11_EL2, x16 + msr ICH_LR10_EL2, x15 + msr ICH_LR9_EL2, x14 + msr ICH_LR8_EL2, x13 + msr ICH_LR7_EL2, x12 + msr ICH_LR6_EL2, x11 + msr ICH_LR5_EL2, x10 + msr ICH_LR4_EL2, x9 + msr ICH_LR3_EL2, x8
Quoted 8 lines
arch/arm64/kvm/vgic-v3-switch.S line 253
Quoted 5 lines
arch/arm64/kvm/vgic-v3-switch.S new file +5
-0+ msr ICH_LR2_EL2, x7 + msr ICH_LR1_EL2, x6 + msr ICH_LR0_EL2, x5 + + dsb stNow you have a dsb without an isb! There's no consistency here and the docs don't tell me much either.
Hmmm. isb may be missing indeed. I'll grab "You Know Who" to have a chat... ;-)
Quoted 7 lines
arch/arm64/kvm/vgic-v3-switch.S line 257
Quoted 4 lines
arch/arm64/kvm/vgic-v3-switch.S new file +4
-0+ + mrs x5, ICC_SRE_EL2 + and x5, x5, #~(1 << 3) + msr ICC_SRE_EL2, x5Why are you doing this? Should't we leave bits 0 and 3 of
SRE_EL2always set?
No. We want to prevent the guest from doing SRE, as we don't handle that at all.
M.
Signature · 1 line
Jazz is not dead. It just smells funny...