Adding plat-eznps to ARC
Subsystems (7)
v4 cover Adding plat-eznps to ARC
From: Noam Camus <[email protected]>
v4:
1) irqchip -- use irq_domain_add_linear()
2) clocksource -- use of_clk_get()
3) New header at include/soc use by platform and drivers
4) update DTS file for above fixed drivers
v3:
1) irqchip: use MACROS instead of structures to decribe
registers.
2) clocksource: use 32bit counter and avoid 2 halfs read
of 64bit dance.
v2:
1) Remove out of tree platform include path
2) Move atomic/bitop/cmpxchg for platform to end.
Remove macro duplication.
Fix some bad implementation.
3) define cpu_relax_lowlatency() for platform.
4) rename init_irq_cpu() to init_per_cpu()
reorder call to init_per_cpu() for secondary use it instead of init_cpu_smp().
5) set res_service to call stext
6) fix build failure for CTOP_AUX_BASE at assembly code
7) Use ilog2 for mtm_init_nat()
8) Add CLKSRC_NPS option to Kconfig
change nps_clksrc_read() to be more readable.
General summay:
This set introduce new platform to ARC architecture. Platform name called "eznps" for working with EZchip NPS400 Network Proccessor.
NPS400 is targeted to service "fast path" network applications.
NPS400 got mesh of 256 extended ARC cores (AKA CTOP), each core got 16 HW threads. This is basically SMT core where at any point of time only one HW thread is active.
Each core have HW scheduler that round robin between eligible HW threads. Totaly, kernel sees 4096 CPUs which I belive is a high record. There is no cache coherency between cores so generic user applications and kernel do not use D$.
Cores got special memory mappings for huge pages (8MB). Mapping is static and should provide application enough memory without any "TLB miss". This mapping is on top of TLB mapping.
This is a basic set that will later be followed with additional set of patches with all advanced features.
Many thanks to all people helping to make this happen.
Regards,
Noam Camus
Noam Camus (18):
Documentation: Add EZchip vendor to binding list
soc: Support for EZchip SoC
ARC: [plat-eznps] define IPI_IRQ
clocksource: Add NPS400 timers driver
irqchip: add nps Internal and external irqchips
ARC: Set vmalloc size from configuration
ARC: rwlock: disable interrupts in !LLSC variant
ARC: rename smp operation init_irq_cpu() to init_per_cpu()
ARC: Mark secondary cpu online only after all HW setup is done
ARC: Add clock from device tree to time_init()
ARC: [plat-eznps] Add eznps board defconfig and dts
ARC: [plat-eznps] Add eznps platform
ARC: [plat-eznps] Use dedicated user stack top
ARC: [plat-eznps] Use dedicated atomic/bitops/cmpxchg
ARC: [plat-eznps] Use dedicated SMP barriers
ARC: [plat-eznps] Use dedicated identity auxiliary register.
ARC: [plat-eznps] Use dedicated COMMAND_LINE_SIZE
ARC: Add eznps platform to Kconfig and Makefile
Tal Zilcer (1):
ARC: [plat-eznps] Use dedicated cpu_relax()
Documentation/devicetree/bindings/arc/eznps.txt | 7 + .../interrupt-controller/ezchip,nps400-ic.txt | 17 ++ .../bindings/timer/ezchip,nps400-timer.txt | 15 ++ .../devicetree/bindings/vendor-prefixes.txt | 1 + MAINTAINERS | 6 + arch/arc/Kconfig | 9 + arch/arc/Makefile | 5 + arch/arc/boot/dts/eznps.dts | 93 +++++++++ arch/arc/configs/nps_defconfig | 85 +++++++++ arch/arc/include/asm/atomic.h | 79 ++++++++- arch/arc/include/asm/barrier.h | 8 + arch/arc/include/asm/bitops.h | 54 ++++++ arch/arc/include/asm/cmpxchg.h | 87 +++++++-- arch/arc/include/asm/entry-compact.h | 8 + arch/arc/include/asm/irq.h | 4 + arch/arc/include/asm/pgtable.h | 2 +- arch/arc/include/asm/processor.h | 36 +++- arch/arc/include/asm/setup.h | 4 + arch/arc/include/asm/smp.h | 4 +- arch/arc/include/asm/spinlock.h | 14 ++ arch/arc/kernel/ctx_sw.c | 13 ++ arch/arc/kernel/irq.c | 4 +- arch/arc/kernel/mcip.c | 2 +- arch/arc/kernel/smp.c | 14 +- arch/arc/kernel/time.c | 10 +- arch/arc/mm/tlb.c | 12 ++ arch/arc/plat-eznps/Kconfig | 34 ++++ arch/arc/plat-eznps/Makefile | 7 + arch/arc/plat-eznps/entry.S | 75 ++++++++ arch/arc/plat-eznps/include/plat/ctop.h | 198 ++++++++++++++++++++ arch/arc/plat-eznps/include/plat/mtm.h | 60 ++++++ arch/arc/plat-eznps/include/plat/smp.h | 26 +++ arch/arc/plat-eznps/mtm.c | 133 +++++++++++++ arch/arc/plat-eznps/platform.c | 102 ++++++++++ arch/arc/plat-eznps/smp.c | 156 +++++++++++++++ drivers/clocksource/Kconfig | 7 + drivers/clocksource/Makefile | 1 + drivers/clocksource/timer-nps.c | 68 +++++++ drivers/irqchip/Makefile | 1 + drivers/irqchip/irq-eznps.c | 131 +++++++++++++ include/soc/nps/common.h | 123 ++++++++++++ 41 files changed, 1678 insertions(+), 37 deletions(-) create mode 100644 Documentation/devicetree/bindings/arc/eznps.txt create mode 100644 Documentation/devicetree/bindings/interrupt-controller/ezchip,nps400-ic.txt create mode 100644 Documentation/devicetree/bindings/timer/ezchip,nps400-timer.txt create mode 100644 arch/arc/boot/dts/eznps.dts create mode 100644 arch/arc/configs/nps_defconfig create mode 100644 arch/arc/plat-eznps/Kconfig create mode 100644 arch/arc/plat-eznps/Makefile create mode 100644 arch/arc/plat-eznps/entry.S create mode 100644 arch/arc/plat-eznps/include/plat/ctop.h create mode 100644 arch/arc/plat-eznps/include/plat/mtm.h create mode 100644 arch/arc/plat-eznps/include/plat/smp.h create mode 100644 arch/arc/plat-eznps/mtm.c create mode 100644 arch/arc/plat-eznps/platform.c create mode 100644 arch/arc/plat-eznps/smp.c create mode 100644 drivers/clocksource/timer-nps.c create mode 100644 drivers/irqchip/irq-eznps.c create mode 100644 include/soc/nps/common.h
PATCH v4 1/19 Documentation: Add EZchip vendor to binding list
From: Noam Camus <[email protected]>
Add EZchip to vendor prefixes list.
EZchip introduce the NPS platform for the ARC architecture.
--- Documentation/devicetree/bindings/arc/eznps.txt | 7 +++++++ .../devicetree/bindings/vendor-prefixes.txt | 1 + 2 files changed, 8 insertions(+), 0 deletions(-) create mode 100644 Documentation/devicetree/bindings/arc/eznps.txt
Documentation/devicetree/bindings/arc/eznps.txt new file +7 -0
@@ -0,0 +1,7 @@ +EZchip NPS Network Processor Platforms Device Tree Bindings +--------------------------------------------------------------------------- + +Appliance main board with NPS400 ASIC. + +Required root node properties: + - compatible = "ezchip,arc-nps";
Documentation/devicetree/bindings/vendor-prefixes.txt +1 -0
@@ -82,6 +82,7 @@ eukrea Eukr??a Electromatique everest Everest Semiconductor Co. Ltd. everspin Everspin Technologies, Inc. excito Excito +ezchip EZchip Semiconductor fcs Fairchild Semiconductor firefly Firefly focaltech FocalTech Systems Co.,Ltd
Signature · 1 line
1.7.1
PATCH v4 2/19 soc: Support for EZchip SoC
In reply to Noam CamusFrom: Noam Camus <[email protected]>
This header file is for NPS400 SoC.
It includes macros for accessing memory mapped registers. These are functional registers that core can use to configure SoC.
--- include/soc/nps/common.h | 123 ++++++++++++++++++++++++++++++++++++++++++++++ 1 files changed, 123 insertions(+), 0 deletions(-) create mode 100644 include/soc/nps/common.h
include/soc/nps/common.h new file +123 -0
@@ -0,0 +1,123 @@ +/* +* Copyright(c) 2015 EZchip Technologies. +* +* This program is free software; you can redistribute it and/or modify it +* under the terms and conditions of the GNU General Public License, +* version 2, as published by the Free Software Foundation. +* +* This program is distributed in the hope 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. +* +* The full GNU General Public License is included in this distribution in +* the file called "COPYING". +*/ + +#ifndef SOC_NPS_COMMON_H +#define SOC_NPS_COMMON_H + +#define NPS_HOST_REG_BASE 0xF6000000 +#define CTOP_INST_MOV2B_FLIP_R3_B1_B2_INST 0x5B60 +#define CTOP_INST_MOV2B_FLIP_R3_B1_B2_LIMM 0x00010422 + +#ifndef __ASSEMBLY__ +/* CPU global ID */ +struct global_id { +union { +struct { +#ifdef CONFIG_EZNPS_MTM_EXT +u32 __reserved:20 , cluster:4 , core:4 , thread:4 ; +#else +u32 __reserved:24 , cluster:4 , core:4 ; +#endif + }; +u32 value; + }; +}; + +/* +* Convert logical to physical CPU IDs +* +* The conversion swap bits 1 and 2 of cluster id (out of 4 bits) +* Now quad of logical clusters id's are adjacent physically, +* and not like the id's physically came with each cluster. +* Below table is 4x4 mesh of core clusters as it layout on chip. +* Cluster ids are in format: logical (physical) +* +* ----------------- ------------------ +* 3 | 5 (3) 7 (7) | | 13 (11) 15 (15)| +* +* 2 | 4 (2) 6 (6) | | 12 (10) 14 (14)| +* ----------------- ------------------ +* 1 | 1 (1) 3 (5) | | 9 (9) 11 (13)| +* +* 0 | 0 (0) 2 (4) | | 8 (8) 10 (12)| +* ----------------- ------------------ +* 0 1 2 3 +*/ +static inline int nps_cluster_logic_to_phys (int cluster) +{ + __asm__ __volatile__( +" mov r3,%0\n" +" .short %1\n" +" .word %2\n" +" mov %0,r3\n" + :"+r" (cluster) + :"i" (CTOP_INST_MOV2B_FLIP_R3_B1_B2_INST ), +"i" (CTOP_INST_MOV2B_FLIP_R3_B1_B2_LIMM ) + :"r3" ); + +return cluster; +} + +#define NPS_CPU_TO_CLUSTER_NUM (cpu) \ + ({ struct global_id gid; gid.value = cpu; \ + nps_cluster_logic_to_phys(gid.cluster); }) + +struct nps_host_reg_address { +union { +struct { +u32 base:8 , cl_x:4 , cl_y:4 , + blkid:6 , reg:8 , __reserved:2 ; + }; +u32 value; + }; +}; + +struct nps_host_reg_address_non_cl { +union { +struct { +u32 base:7 , blkid:11 , reg:12 , __reserved:2 ; + }; +u32 value; + }; +}; + +static inline void *nps_host_reg_non_cl (u32 blkid,u32 reg) +{ +struct nps_host_reg_address_non_cl reg_address; + + reg_address.value =NPS_HOST_REG_BASE ; + reg_address.blkid = blkid; + reg_address.reg = reg; + +return (void *)reg_address.value; +} + +static inline void *nps_host_reg (u32 cpu,u32 blkid,u32 reg) +{ +struct nps_host_reg_address reg_address; +u32 cl =NPS_CPU_TO_CLUSTER_NUM (cpu); + + reg_address.value =NPS_HOST_REG_BASE ; + reg_address.cl_x = (cl >>2 ) &0x3 ; + reg_address.cl_y = cl &0x3 ; + reg_address.blkid = blkid; + reg_address.reg = reg; + +return (void *)reg_address.value; +} +#endif /* __ASSEMBLY__ */ + +#endif /* SOC_NPS_COMMON_H */
Signature · 1 line
1.7.1
PATCH v4 3/19 ARC: [plat-eznps] define IPI_IRQ
In reply to Noam CamusFrom: Noam Camus <[email protected]>
We add IPI irq definition to be used later by any
irqchip such NPS400 IC.
--- arch/arc/include/asm/irq.h | 4 ++++ 1 files changed, 4 insertions(+), 0 deletions(-)
arch/arc/include/asm/irq.h +4 -0
@@ -16,7 +16,11 @@#ifdef CONFIG_ISA_ARCOMPACT #define TIMER0_IRQ 3#define TIMER1_IRQ 4 +#if defined(CONFIG_SMP ) && defined(CONFIG_ARC_PLAT_EZNPS ) +#define IPI_IRQ 5 +#else #define IPI_IRQ (NR_CPU_IRQS-1)/* dummy to enable SMP build for up hardware */ +#endif #else #define TIMER0_IRQ 16#define TIMER1_IRQ 17
Signature · 1 line
1.7.1
PATCH v4 4/19 clocksource: Add NPS400 timers driver
In reply to Noam CamusFrom: Noam Camus <[email protected]>
Add internal tick generator which is shared by all cores. Each cluster of cores view it through dedicated address. This is used for SMP system where all CPUs synced by same clock source.
--- .../bindings/timer/ezchip,nps400-timer.txt | 15 ++++ drivers/clocksource/Kconfig | 7 ++ drivers/clocksource/Makefile | 1 + drivers/clocksource/timer-nps.c | 68 ++++++++++++++++++++ 4 files changed, 91 insertions(+), 0 deletions(-) create mode 100644 Documentation/devicetree/bindings/timer/ezchip,nps400-timer.txt create mode 100644 drivers/clocksource/timer-nps.c
Documentation/devicetree/bindings/timer/ezchip,nps400-timer.txt new file +15 -0
@@ -0,0 +1,15 @@ +NPS Network Processor + +Required properties: + +- compatible : should be "ezchip,nps400-timer" + +Clocks required for compatible = "ezchip,nps400-timer": +- clocks : Must contain a single entry describing the clock input + +Example: + +timer { + compatible = "ezchip,nps400-timer"; + clocks = <&sysclk>; +};
drivers/clocksource/Kconfig +7 -0
@@ -131,6 +131,13 @@ config CLKSRC_TI_32K This option enables support for Texas Instruments 32.768 Hz clocksource available on many OMAP-like platforms. +config CLKSRC_NPS + bool "NPS400 clocksource driver" if COMPILE_TEST + select CLKSRC_OF if OF + help + NPS400 clocksource support. + Got 64 bit counter with update rate up to 1000MHz. + config CLKSRC_STM32 bool "Clocksource for STM32 SoCs" if !ARCH_STM32 depends on OF && ARM && (ARCH_STM32 || COMPILE_TEST)
drivers/clocksource/Makefile +1 -0
@@ -46,6 +46,7 @@ obj-$(CONFIG_CLKSRC_QCOM) += qcom-timer.o obj-$(CONFIG_MTK_TIMER) += mtk_timer.o obj-$(CONFIG_CLKSRC_PISTACHIO) += time-pistachio.o obj-$(CONFIG_CLKSRC_TI_32K) += timer-ti-32k.o +obj-$(CONFIG_CLKSRC_NPS) += timer-nps.o obj-$(CONFIG_ARM_ARCH_TIMER) += arm_arch_timer.o obj-$(CONFIG_ARM_GLOBAL_TIMER) += arm_global_timer.o
drivers/clocksource/timer-nps.c new file +68 -0
68 lines not on this page · raw message
Signature · 1 line
1.7.1
Re: [PATCH v4 04/19] clocksource: Add NPS400 timers driver
Quoted 13 lines · Noam Camus
From: Noam Camus <[email protected]>
Add internal tick generator which is shared by all cores. Each cluster of cores view it through dedicated address. This is used for SMP system where all CPUs synced by same clock source.
Signed-off-by Noam Camus <[email protected]>Cc Daniel Lezcano <[email protected]>Cc Rob Herring <[email protected]>Cc Thomas Gleixner <[email protected]>Cc John Stultz <[email protected]>Acked-by Vineet Gupta <[email protected]>
[ ... ]
drivers/clocksource/timer-nps.c line 60
Quoted 2 lines
drivers/clocksource/timer-nps.c new file +2
-0+clk_prepare_enable (clk); + rate =clk_get_rate (clk);
With the return value checked:
Thanks !
-- Daniel
Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog
PATCH v4 5/19 irqchip: add nps Internal and external irqchips
In reply to Noam CamusFrom: Noam Camus <[email protected]>
Adding EZchip NPS400 support.
NPS internal interrupts are internally handled at Multi Thread Manager (MTM) that is signaled for deactivating an interrupt.
External interrupts is handled also at Global Interrupt Controller (GIC) e.g. serial and network devices.
--- .../interrupt-controller/ezchip,nps400-ic.txt | 17 +++ drivers/irqchip/Makefile | 1 + drivers/irqchip/irq-eznps.c | 131 ++++++++++++++++++++ 3 files changed, 149 insertions(+), 0 deletions(-) create mode 100644 Documentation/devicetree/bindings/interrupt-controller/ezchip,nps400-ic.txt create mode 100644 drivers/irqchip/irq-eznps.c
Documentation/devicetree/bindings/interrupt-controller/ezchip,nps400-ic.txt new file +17 -0
@@ -0,0 +1,17 @@ +EZchip NPS Interrupt Controller + +Required properties: + +- compatible : should be "ezchip,nps400-ic" +- interrupt-controller : Identifies the node as an interrupt controller +- #interrupt-cells : Specifies the number of cells needed to encode an + interrupt source. The value shall be 1. + + +Example: + +intc: interrupt-controller { + compatible = "ezchip,nps400-ic"; + interrupt-controller; + #interrupt-cells = <1>; +};
drivers/irqchip/Makefile +1 -0
@@ -55,3 +55,4 @@ obj-$(CONFIG_RENESAS_H8S_INTC) += irq-renesas-h8s.o obj-$(CONFIG_ARCH_SA1100) += irq-sa11x0.o obj-$(CONFIG_INGENIC_IRQ) += irq-ingenic.o obj-$(CONFIG_IMX_GPCV2) += irq-imx-gpcv2.o +obj-$(CONFIG_ARC_PLAT_EZNPS) += irq-eznps.o
drivers/irqchip/irq-eznps.c new file +131 -0
131 lines not on this page · raw message
Signature · 1 line
1.7.1
Re: [PATCH v4 05/19] irqchip: add nps Internal and external irqchips
Quoted 176 lines · Noam Camus
From: Noam Camus <noamc at ezchip.com>
Adding EZchip NPS400 support.
NPS internal interrupts are internally handled at Multi Thread Manager (MTM) that is signaled for deactivating an interrupt.
External interrupts is handled also at Global Interrupt Controller (GIC) e.g. serial and network devices.Signed-off-by Noam Camus <noamc at ezchip.com>Cc Thomas Gleixner <tglx at linutronix.de>Cc Jason Cooper <jason at lakedaemon.net>Cc Marc Zyngier <marc.zyngier at arm.com>--- .../interrupt-controller/ezchip,nps400-ic.txt | 17 +++ drivers/irqchip/Makefile | 1 + drivers/irqchip/irq-eznps.c | 131 ++++++++++++++++++++ 3 files changed, 149 insertions(+), 0 deletions(-) create mode 100644 Documentation/devicetree/bindings/interrupt-controller/ezchip,nps400-ic.txt create mode 100644 drivers/irqchip/irq-eznps.cDocumentation/devicetree/bindings/interrupt-controller/ezchip,nps400-ic.txt new file +17
-0@@ -0,0 +1,17 @@ +EZchip NPS Interrupt Controller + +Required properties: + +- compatible : should be "ezchip,nps400-ic" +- interrupt-controller : Identifies the node as an interrupt controller +- #interrupt-cells : Specifies the number of cells needed to encode an + interrupt source. The value shall be 1. + + +Example: + +intc: interrupt-controller { + compatible = "ezchip,nps400-ic"; + interrupt-controller; + #interrupt-cells = <1>; +};drivers/irqchip/Makefile +1
-0@@ -55,3 +55,4 @@ obj-$(CONFIG_RENESAS_H8S_INTC) += irq-renesas-h8s.o obj-$(CONFIG_ARCH_SA1100) += irq-sa11x0.o obj-$(CONFIG_INGENIC_IRQ) += irq-ingenic.o obj-$(CONFIG_IMX_GPCV2) += irq-imx-gpcv2.o +obj-$(CONFIG_ARC_PLAT_EZNPS) += irq-eznps.odrivers/irqchip/irq-eznps.c new file +117
-0@@ -0,0 +1,131 @@ +/* +* Copyright(c) 2015 EZchip Technologies. +* +* This program is free software; you can redistribute it and/or modify it +* under the terms and conditions of the GNU General Public License, +* version 2, as published by the Free Software Foundation. +* +* This program is distributed in the hope 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. +* +* The full GNU General Public License is included in this distribution in +* the file called "COPYING". +*/ + +#include <linux/interrupt.h> +#include <linux/module.h> +#include <linux/of.h> +#include <linux/irqdomain.h> +#include <linux/irqchip.h> +#include <asm/irq.h> + +/* +* NPS400 core includes a Interrupt Controller (IC) support. +* All cores can deactivate level irqs at first level control +* at cores mesh layer called MTM. +* For devices out side chip e.g. uart, network there is another +* level called Global Interrupt Manager (GIM). +* This second level can control level and edge interrupt. +* +* NOTE: AUX_IENABLE and CTOP_AUX_IACK are auxiliary registers +* with private HW copy per CPU. +*/ + +static void nps400_irq_mask (struct irq_data *data) +{ +unsigned int ienb; + + ienb =read_aux_reg (AUX_IENABLE ); + ienb &= ~(1 << data->hwirq); +write_aux_reg (AUX_IENABLE , ienb); +} + +static void nps400_irq_unmask (struct irq_data *data) +{ +unsigned int ienb; + + ienb =read_aux_reg (AUX_IENABLE ); + ienb |= (1 << data->hwirq); +write_aux_reg (AUX_IENABLE , ienb); +} + +static void nps400_irq_eoi_global (struct irq_data *data) +{ +write_aux_reg (CTOP_AUX_IACK ,1 << data->hwirq); + +/* Don't ack before all device access is done */ +mb (); + + __asm__ __volatile__ ( +" .word %0\n" + : + :"i" (CTOP_INST_RSPI_GIC_0_R12 ) + :"memory" ); +} + +static void nps400_irq_eoi (struct irq_data *data) +{ +write_aux_reg (CTOP_AUX_IACK ,1 << data->hwirq); +} + +static struct irq_chip nps400_irq_chip_fasteoi = { + .name ="NPS400 IC Global" , + .irq_mask = nps400_irq_mask, + .irq_unmask = nps400_irq_unmask, + .irq_eoi = nps400_irq_eoi_global, +}; + +static struct irq_chip nps400_irq_chip_percpu = { + .name ="NPS400 IC" , + .irq_mask = nps400_irq_mask, + .irq_unmask = nps400_irq_unmask, + .irq_eoi = nps400_irq_eoi, +}; + +static int nps400_irq_map (struct irq_domain *d,unsigned int virq, +irq_hw_number_t hw) +{ +switch (hw) { +case TIMER0_IRQ : +#if defined(CONFIG_SMP ) +case IPI_IRQ : +#endif +irq_set_chip_and_handler (virq, &nps400_irq_chip_percpu, + handle_percpu_irq); +break ; +default : +irq_set_chip_and_handler (virq, &nps400_irq_chip_fasteoi, + handle_fasteoi_irq); +break ; + } + +return 0 ; +} + +static const struct irq_domain_ops nps400_irq_ops = { + .xlate = irq_domain_xlate_onecell, + .map = nps400_irq_map, +}; + +static struct irq_domain *nps400_root_domain; + +static int __init nps400_of_init (struct device_node *node, +struct device_node *parent) +{ +if (parent)
drivers/irqchip/irq-eznps.c line 127
Quoted 10 lines
drivers/irqchip/irq-eznps.c new file +10
-0+panic ("DeviceTree incore ic not a root irq controller\n" ); + + nps400_root_domain =irq_domain_add_linear (node,NR_CPU_IRQS , + &nps400_irq_ops,NULL ); + +if (!nps400_root_domain) +panic ("nps400 root irq domain not avail\n" ); + +/* with this we don't need to export nps400_root_domain */ +irq_set_default_host (nps400_root_domain);
Why do you need this? Devices should have their interrupt-parent pointing to this node, and irq_find_host should sort it out. I must be missing some information (only being CC'd on this single patch).
drivers/irqchip/irq-eznps.c line 131
Quoted 5 lines
drivers/irqchip/irq-eznps.c new file +4
-0+ +return 0 ; +} +IRQCHIP_DECLARE (ezchip_nps400_ic,"ezchip,nps400-ic" , nps400_of_init);
Another thing I'm not seeing here is where is the interrupt actually taken. This code only contains the EOI part, but not the ACK side, as well as the reverse lookup hwirq -> irq). Where is that code?
Thanks,
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
RE: [PATCH v4 05/19] irqchip: add nps Internal and external irqchips
From: Marc Zyngier [mailto:[email protected]]
Sent: Wednesday, December 16, 2015 11:31 AM
Quoted 8 lines
Quoted 8 lines
drivers/irqchip/irq-eznps.c new file +8
-0+static int __init nps400_of_init (struct device_node *node, +struct device_node *parent) +{ +if (parent) +panic ("DeviceTree incore ic not a root irq controller\n" ); + + nps400_root_domain =irq_domain_add_linear (node,NR_CPU_IRQS , + &nps400_irq_ops,NULL );
Quoted 8 lines
drivers/irqchip/irq-eznps.c line 127
Quoted 6 lines
drivers/irqchip/irq-eznps.c new file +6
-0+ +if (!nps400_root_domain) +panic ("nps400 root irq domain not avail\n" ); + +/* with this we don't need to export nps400_root_domain */ +irq_set_default_host (nps400_root_domain);Why do you need this? Devices should have their interrupt-parent pointing to this node, and
irq_find_hostshould sort it >out. I must be missing some information (only being CC'd on this single patch).
Sorry, I will CC you by my next version, in the meantime please refer to:
https://lkml.org/lkml/2015/12/15/864
I need this for my per CPU irqs such timer and IPI which do not come from some external device but from CPUs.
For these IRQs I am calling to irq_create_mapping() from my platform at arch/arc and at that point I got no irqdomain and using irq_find_host() is not good since I got no device_node (at most I can have DT root). Is there device_node for DT root?
Please advise what to do?
Quoted 8 lines
drivers/irqchip/irq-eznps.c
Quoted 6 lines
drivers/irqchip/irq-eznps.c new file +5
-0+ +return 0 ; +} +IRQCHIP_DECLARE (ezchip_nps400_ic,"ezchip,nps400-ic" , +nps400_of_init);Another thing I'm not seeing here is where is the interrupt actually taken. This code only contains the EOI part, but not the ACK side, as well as the reverse lookup hwirq -> irq). Where is that code?
ACK is an optional handler and is not needed by my platform. I will add comment that since my IRQs are EOI based I do not need an ACK.
I do not understand reverse lookup remark, where is it missing? Could you point me to an example for such reverse lookup?
Regards,
Noam
Re: [PATCH v4 05/19] irqchip: add nps Internal and external irqchips
Quoted 19 lines · Noam Camus
From: Marc Zyngier [mailto:marc.zyngier at arm.com]
Sent: Wednesday, December 16, 2015 11:31 AMQuoted 8 lines
Quoted 8 lines
drivers/irqchip/irq-eznps.c new file +8
-0+static int __init nps400_of_init (struct device_node *node, +struct device_node *parent) +{ +if (parent) +panic ("DeviceTree incore ic not a root irq controller\n" ); + + nps400_root_domain =irq_domain_add_linear (node,NR_CPU_IRQS , + &nps400_irq_ops,NULL );Quoted 8 lines
drivers/irqchip/irq-eznps.c line 127
Quoted 6 lines
drivers/irqchip/irq-eznps.c new file +6
-0+ +if (!nps400_root_domain) +panic ("nps400 root irq domain not avail\n" ); + +/* with this we don't need to export nps400_root_domain */ +irq_set_default_host (nps400_root_domain);Why do you need this? Devices should have their interrupt-parent pointing to this node, and
irq_find_hostshould sort it >out. I must be missing some information (only being CC'd on this single patch).
Quoted 8 lines
Sorry, I will CC you by my next version, in the meantime please refer to:
https://lkml.org/lkml/2015/12/15/864I need this for my per CPU irqs such timer and IPI which do not come from some external device but from CPUs. For these IRQs I am calling to
irq_create_mapping()from my platform atarch/arcand at that point I got no irqdomain and usingirq_find_host()is not good since I got nodevice_node(at most I can have DT root).
That's a problem. You should never do that for your timer (doing a request_irq will do the right thing, and that's what your timer driver already does).
As for initializing your IPIs, they are usually outside of the IRQ space, so you should handle them separately (and get your irqchip to initialize them).
Quoted 12 lines
Is there
device_nodefor DT root?
Please advise what to do?Quoted 8 lines
drivers/irqchip/irq-eznps.c
Quoted 6 lines
drivers/irqchip/irq-eznps.c new file +5
-0+ +return 0 ; +} +IRQCHIP_DECLARE (ezchip_nps400_ic,"ezchip,nps400-ic" , +nps400_of_init);Another thing I'm not seeing here is where is the interrupt actually taken. This code only contains the EOI part, but not the ACK side, as well as the reverse lookup hwirq -> irq). Where is that code?
Quoted 2 lines
ACK is an optional handler and is not needed by my platform. I will add comment that since my IRQs are EOI based I do not need an ACK.
What I'm talking about is not the irq_ack method that the irqchip can provide, but the action your perform on your interrupt controller to extract the hwirq number and find out what corresponding Linux interrupt has fired.
Quoted 3 lines
I do not understand reverse lookup remark, where is it missing? Could you point me to an example for such reverse lookup?
See for example:
This is the function (sun4i_handle_irq) that is executed almost immediately after the IRQ is asserted. The CPU reads the hwirq from the interrupt controller, and use handle_domain_irq() to call the corresponding handler (by doing a lookup in the domain).
I couldn't find out in your platform code where you are doing that.
Thanks,
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
RE: [PATCH v4 05/19] irqchip: add nps Internal and external irqchips
Quoted 10 lines
From: Marc Zyngier [mailto:marc.zyngier at arm.com]
Sent: Friday, December 18, 2015 1:21 PMQuoted 5 lines
I need this for my per CPU irqs such timer and IPI which do not come from some external device but from CPUs. For these IRQs I am calling to
irq_create_mapping()from my platform atarch/arcand at that point I got no irqdomain and usingirq_find_host()is not good since I got nodevice_node(at most I can have DT root).That's a problem. You should never do that for your timer (doing a
request_irqwill do the right thing, and that's what your timer driver already does).
Please be more specific, from all that I wrote what is the problem?
When I use request_irq() it fail without the mapping and mapping fail without a domain. Never do what?
What should I do then?
Quoted 1 line
As for initializing your IPIs, they are usually outside of the IRQ space, so you should handle them separately (and get your irqchip to initialize them).
I am handling all my IRQs within same irqchip, which is the only one I have. So I am not sure what you expect here. Please be more elaborate.
Quoted 12 lines
Quoted 5 lines
Another thing I'm not seeing here is where is the interrupt actually taken. This code only contains the EOI part, but not the ACK side, as well as the reverse lookup hwirq -> irq). Where is that code?
ACK is an optional handler and is not needed by my platform. I will add comment that since my IRQs are EOI based I do not need an ACK.
What I'm talking about is not the
irq_ackmethod that the irqchip can provide, but the action your perform on your interrupt controller to extract the hwirq number and find out what corresponding Linux interrupt has fired.Quoted 3 lines
I do not understand reverse lookup remark, where is it missing? Could you point me to an example for such reverse lookup?
Quoted 7 lines
See for example:
This is the function (
sun4i_handle_irq) that is executed almost immediately after the IRQ is asserted. The CPU reads the hwirq from the interrupt controller, and usehandle_domain_irq()to call the corresponding handler (by doing a lookup in the domain).I couldn't find out in your platform code where you are doing that.
OK, this is seem much like exclusively ARM stuff.
Note that I am working with ARC (seem alike) here and we do not define CONFIG_HANDLE_DOMAIN_IRQ and do not implement set_handle_irq().
So for ARC this reverse mapping is something we can leave without (maybe because we are kind of a legacy domain).
Note that this patch is part of a set I introduce to ARC as a new platform. You can view my tree and the patch set at:
https://github.com/EZchip/linux/commits/EZ-arc-for-next-basic-reviewd-DO-NOT-COMMENT
-Noam
Re: [PATCH v4 05/19] irqchip: add nps Internal and external irqchips
Quoted 9 lines · Noam Camus
Quoted 8 lines
From: Marc Zyngier [mailto:marc.zyngier at arm.com]
Sent: Friday, December 18, 2015 1:21 PMQuoted 5 lines
I need this for my per CPU irqs such timer and IPI which do not come from some external device but from CPUs. For these IRQs I am calling to
irq_create_mapping()from my platform atarch/arcand at that point I got no irqdomain and usingirq_find_host()is not good since I got nodevice_node(at most I can have DT root).
Quoted 3 lines
Quoted 1 line
That's a problem. You should never do that for your timer (doing a
request_irqwill do the right thing, and that's what your timer driver already does).Please be more specific, from all that I wrote what is the problem?
Calling irq_create_mapping out of the blue like you do it here:
+static void eznps_init_per_cpu(int cpu)
+{
+ /* Create mapping for all per cpu IRQs */
+ if (cpu == 0) {
+ irq_create_mapping(NULL, TIMER0_IRQ);
+ irq_create_mapping(NULL, IPI_IRQ);
+ }
is simply not acceptable.
Quoted 1 line
When I use
request_irq()it fail without the mapping and mapping fail without a domain.
Grmbl...
That's not completely surprising:
+ timer {
+ compatible = "ezchip,nps400-timer";
+ clocks = <&sysclk>;
+ clock-names="sysclk";
+ };
Where is the interrupt?
Quoted 2 lines
Never do what?
What should I do then?
Maybe you should start by looking how the other architectures have solved that exact problem at least half a dozen time.
Quoted 7 lines
Quoted 4 lines
As for initializing your IPIs, they are usually outside of the IRQ space, so you should handle them separately (and get your irqchip to initialize them).
I am handling all my IRQs within same irqchip, which is the only one I have. So I am not sure what you expect here. Please be more elaborate.
Do not create a mapping for IPIs. Full stop. Handle them independently from your normal IRQs.
Quoted 14 lines
Quoted 13 lines
Quoted 5 lines
Another thing I'm not seeing here is where is the interrupt actually taken. This code only contains the EOI part, but not the ACK side, as well as the reverse lookup hwirq -> irq). Where is that code?
ACK is an optional handler and is not needed by my platform. I will add comment that since my IRQs are EOI based I do not need an ACK.
What I'm talking about is not the
irq_ackmethod that the irqchip can provide, but the action your perform on your interrupt controller to extract the hwirq number and find out what corresponding Linux interrupt has fired.Quoted 3 lines
I do not understand reverse lookup remark, where is it missing? Could you point me to an example for such reverse lookup?
See for example:
Quoted 7 lines
Quoted 5 lines
This is the function (
sun4i_handle_irq) that is executed almost immediately after the IRQ is asserted. The CPU reads the hwirq from the interrupt controller, and usehandle_domain_irq()to call the corresponding handler (by doing a lookup in the domain).I couldn't find out in your platform code where you are doing that.
OK, this is seem much like exclusively ARM stuff.
No, this is not. Can you please stop looking at the surface of things and start taking an interest in how things actually *work*? Almost *nothing* in the interrupt handling code is architecture specific.
Quoted 6 lines
Note that I am working with ARC (seem alike) here and we do not define
CONFIG_HANDLE_DOMAIN_IRQand do not implementset_handle_irq().So for ARC this reverse mapping is something we can leave without (maybe because we are kind of a legacy domain).
Yeah, I just located the crap: arch_do_IRQ() happily takes a hwirq (the vector number), and uses that as a Linux IRQ. This looks a lot like ARM pre-DT, about 10 years ago.
Well, time to meet the 21st century. If you intend to use DT, please fix your arch port. Otherwise, just hardcode everything in your platform and don't pretend to support device tree.
Thanks,
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
RE: [PATCH v4 05/19] irqchip: add nps Internal and external irqchips
Quoted 6 lines
From: Marc Zyngier [mailto:marc.zyngier at arm.com]
Sent: Friday, December 18, 2015 6:31 PMQuoted 3 lines
Note that I am working with ARC (seem alike) here and we do not define
CONFIG_HANDLE_DOMAIN_IRQand do not implementset_handle_irq().
Quoted 6 lines
Quoted 2 lines
So for ARC this reverse mapping is something we can leave without (maybe because we are kind of a legacy domain).
Yeah, I just located the crap:
arch_do_IRQ()happily takes a hwirq (the vector number), and uses that as a Linux IRQ. This looks a lot like ARM pre-DT, about 10 years ago.Well, time to meet the 21st century. If you intend to use DT, please fix your arch port. Otherwise, just hardcode everything in your platform and don't pretend to support device tree.
Thank you very much.
I have the complete picture now.
Will discuss with ARC maintainer and see what best way to do this change.
-Noam
Re: [PATCH v4 05/19] irqchip: add nps Internal and external irqchips
In reply to Marc ZyngierQuoted 90 lines · Marc Zyngier
Quoted 9 lines · Noam Camus
Quoted 8 lines
From: Marc Zyngier [mailto:marc.zyngier at arm.com]
Sent: Friday, December 18, 2015 1:21 PMQuoted 5 lines
I need this for my per CPU irqs such timer and IPI which do not come from some external device but from CPUs. For these IRQs I am calling to
irq_create_mapping()from my platform atarch/arcand at that point I got no irqdomain and usingirq_find_host()is not good since I got nodevice_node(at most I can have DT root).Quoted 3 lines
Quoted 1 line
That's a problem. You should never do that for your timer (doing a
request_irqwill do the right thing, and that's what your timer driver already does).Please be more specific, from all that I wrote what is the problem?
Calling
irq_create_mappingout of the blue like you do it here:(quoted diff) +7
-0+static void eznps_init_per_cpu(int cpu) +{ + /* Create mapping for all per cpu IRQs */ + if (cpu == 0) { + irq_create_mapping(NULL, TIMER0_IRQ); + irq_create_mapping(NULL, IPI_IRQ); + }is simply not acceptable.
Quoted 1 line
When I use
request_irq()it fail without the mapping and mapping fail without a domain.Grmbl...
That's not completely surprising:
(quoted diff) +5
-0+ timer { + compatible = "ezchip,nps400-timer"; + clocks = <&sysclk>; + clock-names="sysclk"; + };Where is the interrupt?
Quoted 2 lines
Never do what?
What should I do then?Maybe you should start by looking how the other architectures have solved that exact problem at least half a dozen time.
Quoted 7 lines
Quoted 4 lines
As for initializing your IPIs, they are usually outside of the IRQ space, so you should handle them separately (and get your irqchip to initialize them).
I am handling all my IRQs within same irqchip, which is the only one I have. So I am not sure what you expect here. Please be more elaborate.
Do not create a mapping for IPIs. Full stop. Handle them independently from your normal IRQs.
Quoted 14 lines
Quoted 13 lines
Quoted 5 lines
Another thing I'm not seeing here is where is the interrupt actually taken. This code only contains the EOI part, but not the ACK side, as well as the reverse lookup hwirq -> irq). Where is that code?
ACK is an optional handler and is not needed by my platform. I will add comment that since my IRQs are EOI based I do not need an ACK.
What I'm talking about is not the
irq_ackmethod that the irqchip can provide, but the action your perform on your interrupt controller to extract the hwirq number and find out what corresponding Linux interrupt has fired.Quoted 3 lines
I do not understand reverse lookup remark, where is it missing? Could you point me to an example for such reverse lookup?
See for example:
Quoted 7 lines
Quoted 5 lines
This is the function (
sun4i_handle_irq) that is executed almost immediately after the IRQ is asserted. The CPU reads the hwirq from the interrupt controller, and usehandle_domain_irq()to call the corresponding handler (by doing a lookup in the domain).I couldn't find out in your platform code where you are doing that.
OK, this is seem much like exclusively ARM stuff.
No, this is not. Can you please stop looking at the surface of things and start taking an interest in how things actually *work*? Almost *nothing* in the interrupt handling code is architecture specific.
Quoted 6 lines
Note that I am working with ARC (seem alike) here and we do not define
CONFIG_HANDLE_DOMAIN_IRQand do not implementset_handle_irq().So for ARC this reverse mapping is something we can leave without (maybe because we are kind of a legacy domain).
Quoted 7 lines
Yeah, I just located the crap:
arch_do_IRQ()happily takes a hwirq (the vector number), and uses that as a Linux IRQ. This looks a lot like ARM pre-DT, about 10 years ago.Well, time to meet the 21st century. If you intend to use DT, please fix your arch port. Otherwise, just hardcode everything in your platform and don't pretend to support device tree.
Hi Marc,
Taking your rant in a positive stride and I'm all up for making this as nice/modern as possible. I don't have issues with enabling CONFIG_HANDLE_DOMAIN_IRQ for ARC (although it might add a few cycles o/h to each ISR)
However currently (4.4-rcX) it is only enabled for arm/arm64/openrisc and from the looks of it in drivers/irqchip, only ARM based SoCs use the handle_domain_irq() calls by plugging into ARM top level handler.
Why is that not a problem for other arches like PPC/MIPS which also use DT heavily. Or perhaps they are also subtly broken as ARC is !
What am I missing ?
Thx,
-Vineet
Re: [PATCH v4 05/19] irqchip: add nps Internal and external irqchips
Hi Marc,
Quoted 105 lines · Vineet Gupta
Quoted 90 lines · Marc Zyngier
Quoted 9 lines · Noam Camus
Quoted 8 lines
From: Marc Zyngier [mailto:marc.zyngier at arm.com]
Sent: Friday, December 18, 2015 1:21 PMQuoted 5 lines
I need this for my per CPU irqs such timer and IPI which do not come from some external device but from CPUs. For these IRQs I am calling to
irq_create_mapping()from my platform atarch/arcand at that point I got no irqdomain and usingirq_find_host()is not good since I got nodevice_node(at most I can have DT root).Quoted 3 lines
Quoted 1 line
That's a problem. You should never do that for your timer (doing a
request_irqwill do the right thing, and that's what your timer driver already does).Please be more specific, from all that I wrote what is the problem?
Calling
irq_create_mappingout of the blue like you do it here:(quoted diff) +7
-0+static void eznps_init_per_cpu(int cpu) +{ + /* Create mapping for all per cpu IRQs */ + if (cpu == 0) { + irq_create_mapping(NULL, TIMER0_IRQ); + irq_create_mapping(NULL, IPI_IRQ); + }is simply not acceptable.
Quoted 1 line
When I use
request_irq()it fail without the mapping and mapping fail without a domain.Grmbl...
That's not completely surprising:
(quoted diff) +5
-0+ timer { + compatible = "ezchip,nps400-timer"; + clocks = <&sysclk>; + clock-names="sysclk"; + };Where is the interrupt?
Quoted 2 lines
Never do what?
What should I do then?Maybe you should start by looking how the other architectures have solved that exact problem at least half a dozen time.
Quoted 7 lines
Quoted 4 lines
As for initializing your IPIs, they are usually outside of the IRQ space, so you should handle them separately (and get your irqchip to initialize them).
I am handling all my IRQs within same irqchip, which is the only one I have. So I am not sure what you expect here. Please be more elaborate.
Do not create a mapping for IPIs. Full stop. Handle them independently from your normal IRQs.
Quoted 14 lines
Quoted 13 lines
Quoted 5 lines
Another thing I'm not seeing here is where is the interrupt actually taken. This code only contains the EOI part, but not the ACK side, as well as the reverse lookup hwirq -> irq). Where is that code?
ACK is an optional handler and is not needed by my platform. I will add comment that since my IRQs are EOI based I do not need an ACK.
What I'm talking about is not the
irq_ackmethod that the irqchip can provide, but the action your perform on your interrupt controller to extract the hwirq number and find out what corresponding Linux interrupt has fired.Quoted 3 lines
I do not understand reverse lookup remark, where is it missing? Could you point me to an example for such reverse lookup?
See for example:
Quoted 7 lines
Quoted 5 lines
This is the function (
sun4i_handle_irq) that is executed almost immediately after the IRQ is asserted. The CPU reads the hwirq from the interrupt controller, and usehandle_domain_irq()to call the corresponding handler (by doing a lookup in the domain).I couldn't find out in your platform code where you are doing that.
OK, this is seem much like exclusively ARM stuff.
No, this is not. Can you please stop looking at the surface of things and start taking an interest in how things actually *work*? Almost *nothing* in the interrupt handling code is architecture specific.
Quoted 6 lines
Note that I am working with ARC (seem alike) here and we do not define
CONFIG_HANDLE_DOMAIN_IRQand do not implementset_handle_irq().So for ARC this reverse mapping is something we can leave without (maybe because we are kind of a legacy domain).
Quoted 7 lines
Yeah, I just located the crap:
arch_do_IRQ()happily takes a hwirq (the vector number), and uses that as a Linux IRQ. This looks a lot like ARM pre-DT, about 10 years ago.Well, time to meet the 21st century. If you intend to use DT, please fix your arch port. Otherwise, just hardcode everything in your platform and don't pretend to support device tree.
Hi Marc,
Taking your rant in a positive stride and I'm all up for making this as nice/modern as possible. I don't have issues with enabling
CONFIG_HANDLE_DOMAIN_IRQfor ARC (although it might add a few cycles o/h to each ISR)
Quoted 8 lines
However currently (4.4-rcX) it is only enabled for arm/arm64/openrisc and from the looks of it in
drivers/irqchip, only ARM based SoCs use thehandle_domain_irq()calls by plugging into ARM top level handler.Why is that not a problem for other arches like PPC/MIPS which also use DT heavily. Or perhaps they are also subtly broken as ARC is !
What am I missing ?
Marc, I hope you are back from holidays. When u get a chance could you please respond to above.
Also I tool a stab at it anyways.
1. Enabled HANDLE_DOMAIN_IRQ
2. arch_do_IRQ() calls handle_domain_irq(NULL, hwirq, regs) since that code doesn't know about domains.
It fails w/o irq_set_default_host() being called.
If we go back to sunxi example you quoted above, it relies on driver passing the domain to handle_domain_irq(). IMHO it is simpler if we had the default domain.
So long story short, ARC can be made to use handle_domain_irq() w/o the song and dance of registering another callback from irqchip code if we retained the default domain setting.
-Vineet
Re: [PATCH v4 05/19] irqchip: add nps Internal and external irqchips
Hi Vineet,
Sorry I missed that one, it must have been caught in the mother of all "mark as read" I did on coming back from holiday.
Quoted 116 lines · Vineet Gupta
Hi Marc,
Quoted 113 lines · Vineet Gupta
Quoted 90 lines · Marc Zyngier
Quoted 9 lines · Noam Camus
Quoted 8 lines
From: Marc Zyngier [mailto:marc.zyngier at arm.com]
Sent: Friday, December 18, 2015 1:21 PMQuoted 5 lines
I need this for my per CPU irqs such timer and IPI which do not come from some external device but from CPUs. For these IRQs I am calling to
irq_create_mapping()from my platform atarch/arcand at that point I got no irqdomain and usingirq_find_host()is not good since I got nodevice_node(at most I can have DT root).Quoted 3 lines
Quoted 1 line
That's a problem. You should never do that for your timer (doing a
request_irqwill do the right thing, and that's what your timer driver already does).Please be more specific, from all that I wrote what is the problem?
Calling
irq_create_mappingout of the blue like you do it here:(quoted diff) +7
-0+static void eznps_init_per_cpu(int cpu) +{ + /* Create mapping for all per cpu IRQs */ + if (cpu == 0) { + irq_create_mapping(NULL, TIMER0_IRQ); + irq_create_mapping(NULL, IPI_IRQ); + }is simply not acceptable.
Quoted 1 line
When I use
request_irq()it fail without the mapping and mapping fail without a domain.Grmbl...
That's not completely surprising:
(quoted diff) +5
-0+ timer { + compatible = "ezchip,nps400-timer"; + clocks = <&sysclk>; + clock-names="sysclk"; + };Where is the interrupt?
Quoted 2 lines
Never do what?
What should I do then?Maybe you should start by looking how the other architectures have solved that exact problem at least half a dozen time.
Quoted 7 lines
Quoted 4 lines
As for initializing your IPIs, they are usually outside of the IRQ space, so you should handle them separately (and get your irqchip to initialize them).
I am handling all my IRQs within same irqchip, which is the only one I have. So I am not sure what you expect here. Please be more elaborate.
Do not create a mapping for IPIs. Full stop. Handle them independently from your normal IRQs.
Quoted 14 lines
Quoted 13 lines
Quoted 5 lines
Another thing I'm not seeing here is where is the interrupt actually taken. This code only contains the EOI part, but not the ACK side, as well as the reverse lookup hwirq -> irq). Where is that code?
ACK is an optional handler and is not needed by my platform. I will add comment that since my IRQs are EOI based I do not need an ACK.
What I'm talking about is not the
irq_ackmethod that the irqchip can provide, but the action your perform on your interrupt controller to extract the hwirq number and find out what corresponding Linux interrupt has fired.Quoted 3 lines
I do not understand reverse lookup remark, where is it missing? Could you point me to an example for such reverse lookup?
See for example:
Quoted 7 lines
Quoted 5 lines
This is the function (
sun4i_handle_irq) that is executed almost immediately after the IRQ is asserted. The CPU reads the hwirq from the interrupt controller, and usehandle_domain_irq()to call the corresponding handler (by doing a lookup in the domain).I couldn't find out in your platform code where you are doing that.
OK, this is seem much like exclusively ARM stuff.
No, this is not. Can you please stop looking at the surface of things and start taking an interest in how things actually *work*? Almost *nothing* in the interrupt handling code is architecture specific.
Quoted 6 lines
Note that I am working with ARC (seem alike) here and we do not define
CONFIG_HANDLE_DOMAIN_IRQand do not implementset_handle_irq().So for ARC this reverse mapping is something we can leave without (maybe because we are kind of a legacy domain).
Quoted 7 lines
Yeah, I just located the crap:
arch_do_IRQ()happily takes a hwirq (the vector number), and uses that as a Linux IRQ. This looks a lot like ARM pre-DT, about 10 years ago.Well, time to meet the 21st century. If you intend to use DT, please fix your arch port. Otherwise, just hardcode everything in your platform and don't pretend to support device tree.
Hi Marc,
Taking your rant in a positive stride and I'm all up for making this as nice/modern as possible. I don't have issues with enabling
CONFIG_HANDLE_DOMAIN_IRQfor ARC (although it might add a few cycles o/h to each ISR)However currently (4.4-rcX) it is only enabled for arm/arm64/openrisc and from the looks of it in
drivers/irqchip, only ARM based SoCs use thehandle_domain_irq()calls by plugging into ARM top level handler.Why is that not a problem for other arches like PPC/MIPS which also use DT heavily. Or perhaps they are also subtly broken as ARC is !
What am I missing ?
HANDLE_DOMAIN_IRQ is not mandatory at all - a number of architectures had something open-coded in the past (with some drawbacks and/or bugs), and this config option is just one of the ways to get it right.
MIPS/PPC perform the reverse lookup directly, without using this infrastructure, and that would be a valid approach for ARC too.
Quoted 9 lines
Marc, I hope you are back from holidays. When u get a chance could you please respond to above.
Also I tool a stab at it anyways.
1. EnabledHANDLE_DOMAIN_IRQ
2.arch_do_IRQ()callshandle_domain_irq(NULL, hwirq, regs)since that code doesn't know about domains.It fails w/o
irq_set_default_host()being called.
Well, that's expected. unless you use the underlying primitive (__handle_domain_irq) and pass false as the parameter for lookup (see handle_IRQ() in arch/arm/kernel/irq.c). This is what we use for "legacy" platforms that do not support domains.
Quoted 6 lines
If we go back to sunxi example you quoted above, it relies on driver passing the domain to
handle_domain_irq(). IMHO it is simpler if we had the default domain.So long story short, ARC can be made to use
handle_domain_irq()w/o the song and dance of registering another callback from irqchip code if we retained the default domain setting.
This only works if you can guarantee that you will never have another irqchip calling irq_set_default_host()... If your system is always simple enough to guarantee that, why not.
Thanks,
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
Re: [PATCH v4 05/19] irqchip: add nps Internal and external irqchips
Quoted 4 lines · Marc Zyngier
Hi Vineet,
Sorry I missed that one, it must have been caught in the mother of all "mark as read" I did on coming back from holiday.
That was expected - NP :-)
Quoted 117 lines · Vineet Gupta
Quoted 115 lines
Quoted 115 lines
Hi Marc,
Quoted 112 lines · Vineet Gupta
Quoted 112 lines · Marc Zyngier
Quoted 90 lines
Quoted 90 lines · Noam Camus
Quoted 9 lines
Quoted 9 lines
Quoted 8 lines
Quoted 8 lines
From: Marc Zyngier [mailto:marc.zyngier at arm.com]
Sent: Friday, December 18, 2015 1:21 PMQuoted 5 lines
Quoted 5 lines
I need this for my per CPU irqs such timer and IPI which do not come
from some external device but from CPUs. For these IRQs I am calling
to irq_create_mapping() from my platform at arch/arc and at that point
I got no irqdomain and using irq_find_host() is not good since I got
no device_node (at most I can have DT root).Quoted 3 lines
Quoted 3 lines
Quoted 1 line
Quoted 1 line
That's a problem. You should never do that for your timer (doing a
request_irqwill do the right thing, and that's what your timer driver already does).Please be more specific, from all that I wrote what is the problem?
Calling
irq_create_mappingout of the blue like you do it here:(quoted diff) +7
-0+static void eznps_init_per_cpu(int cpu) +{ + /* Create mapping for all per cpu IRQs */ + if (cpu == 0) { + irq_create_mapping(NULL, TIMER0_IRQ); + irq_create_mapping(NULL, IPI_IRQ); + }is simply not acceptable.
Quoted 1 line
Quoted 1 line
When I use
request_irq()it fail without the mapping and mapping fail without a domain.Grmbl...
That's not completely surprising:
(quoted diff) +5
-0+ timer { + compatible = "ezchip,nps400-timer"; + clocks = <&sysclk>; + clock-names="sysclk"; + };Where is the interrupt?
Quoted 2 lines
Quoted 2 lines
Never do what?
What should I do then?Maybe you should start by looking how the other architectures have solved that exact problem at least half a dozen time.
Quoted 7 lines
Quoted 7 lines
Quoted 4 lines
Quoted 4 lines
As for initializing your IPIs, they are usually outside of the IRQ space, so you should handle them separately (and get your irqchip to initialize them).
I am handling all my IRQs within same irqchip, which is the only one I have. So I am not sure what you expect here. Please be more elaborate.
Do not create a mapping for IPIs. Full stop. Handle them independently from your normal IRQs.
Quoted 14 lines
Quoted 14 lines
Quoted 13 lines
Quoted 13 lines
Quoted 5 lines
Quoted 5 lines
Another thing I'm not seeing here is where is the interrupt actually taken. This code only contains the EOI part, but not the ACK side, as well as the reverse lookup hwirq -> irq). Where is that code?
ACK is an optional handler and is not needed by my platform.
I will add comment that since my IRQs are EOI based I do not need an ACK.What I'm talking about is not the
irq_ackmethod that the irqchip can provide, but the action your perform on your interrupt controller to extract the hwirq number and find out what corresponding Linux interrupt has fired.Quoted 3 lines
Quoted 3 lines
I do not understand reverse lookup remark, where is it missing?
Could you point me to an example for such reverse lookup?See for example:
Quoted 7 lines
Quoted 7 lines
Quoted 5 lines
Quoted 5 lines
This is the function (
sun4i_handle_irq) that is executed almost immediately after the IRQ is asserted. The CPU reads the hwirq from the interrupt controller, and usehandle_domain_irq()to call the corresponding handler (by doing a lookup in the domain).I couldn't find out in your platform code where you are doing that.
OK, this is seem much like exclusively ARM stuff.
No, this is not. Can you please stop looking at the surface of things and start taking an interest in how things actually *work*? Almost *nothing* in the interrupt handling code is architecture specific.
Quoted 6 lines
Quoted 6 lines
Note that I am working with ARC (seem alike) here and we do not define
CONFIG_HANDLE_DOMAIN_IRQand do not implementset_handle_irq().So for ARC this reverse mapping is something we can leave without (maybe because we are kind of a legacy domain).
Quoted 7 lines
Quoted 7 lines
Yeah, I just located the crap:
arch_do_IRQ()happily takes a hwirq (the vector number), and uses that as a Linux IRQ. This looks a lot like ARM pre-DT, about 10 years ago.Well, time to meet the 21st century. If you intend to use DT, please fix your arch port. Otherwise, just hardcode everything in your platform and don't pretend to support device tree.
Hi Marc,
Taking your rant in a positive stride and I'm all up for making this as nice/modern as possible. I don't have issues with enabling
CONFIG_HANDLE_DOMAIN_IRQfor ARC (although it might add a few cycles o/h to each ISR)However currently (4.4-rcX) it is only enabled for arm/arm64/openrisc and from the looks of it in
drivers/irqchip, only ARM based SoCs use thehandle_domain_irq()calls by plugging into ARM top level handler.Why is that not a problem for other arches like PPC/MIPS which also use DT heavily. Or perhaps they are also subtly broken as ARC is !
Quoted 7 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
What am I missing ?
HANDLE_DOMAIN_IRQis not mandatory at all - a number of architectures had something open-coded in the past (with some drawbacks and/or bugs), and this config option is just one of the ways to get it right.MIPS/PPC perform the reverse lookup directly, without using this infrastructure, and that would be a valid approach for ARC too.
I'd rather use the generic infrastructure and improve it if needed !
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Marc, I hope you are back from holidays. When u get a chance could you please respond to above.
Also I tool a stab at it anyways.
1. EnabledHANDLE_DOMAIN_IRQ
2.arch_do_IRQ()callshandle_domain_irq(NULL, hwirq, regs)since that code doesn't know about domains.
Quoted 5 lines
Quoted 1 line
Quoted 1 line
It fails w/o
irq_set_default_host()being called.Well, that's expected. unless you use the underlying primitive (
__handle_domain_irq) and pass false as the parameter for lookup (seehandle_IRQ()inarch/arm/kernel/irq.c). This is what we use for "legacy" platforms that do not support domains.
Right I saw that and that causes virq = hwirq - kind of defeats the purpose if we were doing this on ARC.
Quoted 10 lines
Quoted 7 lines
Quoted 7 lines
If we go back to sunxi example you quoted above, it relies on driver passing the domain to
handle_domain_irq(). IMHO it is simpler if we had the default domain.So long story short, ARC can be made to use
handle_domain_irq()w/o the song and dance of registering another callback from irqchip code if we retained the default domain setting.This only works if you can guarantee that you will never have another irqchip calling
irq_set_default_host()... If your system is always simple enough to guarantee that, why not.
ATM we can certainly assume that.
However with ARM approach two irqchips can still call set_handle_irq() and only the first one succeeds (and others return silently). That seems wrong to me - irq_xxx.c will still use the handler registered by say irq_sun41.c ?
Thx,
-Vineet
Re: [PATCH v4 05/19] irqchip: add nps Internal and external irqchips
On 12/01/16 09:12, Vineet Gupta wrote:
[...]
Quoted 17 lines
Quoted 6 lines
HANDLE_DOMAIN_IRQis not mandatory at all - a number of architectures had something open-coded in the past (with some drawbacks and/or bugs), and this config option is just one of the ways to get it right.MIPS/PPC perform the reverse lookup directly, without using this infrastructure, and that would be a valid approach for ARC too.
I'd rather use the generic infrastructure and improve it if needed !
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Marc, I hope you are back from holidays. When u get a chance could you please respond to above.
Also I tool a stab at it anyways.
1. EnabledHANDLE_DOMAIN_IRQ
2.arch_do_IRQ()callshandle_domain_irq(NULL, hwirq, regs)since that code doesn't know about domains.
Quoted 8 lines
Quoted 5 lines
Quoted 1 line
Quoted 1 line
It fails w/o
irq_set_default_host()being called.Well, that's expected. unless you use the underlying primitive (
__handle_domain_irq) and pass false as the parameter for lookup (seehandle_IRQ()inarch/arm/kernel/irq.c). This is what we use for "legacy" platforms that do not support domains.Right I saw that and that causes virq = hwirq - kind of defeats the purpose if we were doing this on ARC.
You can forget about this if you convert all your platforms to using domains. But if you have to deal with a transition period (which, in the ARM case, will last exactly forever because some platforms will never be converted to DT), this comes in handy.
Quoted 11 lines
Quoted 10 lines
Quoted 7 lines
Quoted 7 lines
If we go back to sunxi example you quoted above, it relies on driver passing the domain to
handle_domain_irq(). IMHO it is simpler if we had the default domain.So long story short, ARC can be made to use
handle_domain_irq()w/o the song and dance of registering another callback from irqchip code if we retained the default domain setting.This only works if you can guarantee that you will never have another irqchip calling
irq_set_default_host()... If your system is always simple enough to guarantee that, why not.
Quoted 5 lines
ATM we can certainly assume that.
However with ARM approach two irqchips can still call
set_handle_irq()and only the first one succeeds (and others return silently). That seems wrong to me -irq_xxx.cwill still use the handler registered by sayirq_sun41.c?
Oh, this is by no mean foolproof. We do rely on irqchips installing their primary handler only if they are truly the primary irqchip (they do not have a parent). Anything else will fail badly.
So if you're confident that you can ensure that noone will set de default domain twice, you'll be fine.
Thanks,
M.
Signature · 1 line
Jazz is not dead. It just smells funny...
Re: [PATCH v4 05/19] irqchip: add nps Internal and external irqchips
In reply to Marc ZyngierHi Marc,
Quoted 24 lines · Marc Zyngier
Quoted 9 lines · Noam Camus
Quoted 8 lines
From: Marc Zyngier [mailto:marc.zyngier at arm.com]
Sent: Friday, December 18, 2015 1:21 PMQuoted 5 lines
I need this for my per CPU irqs such timer and IPI which do not come from some external device but from CPUs. For these IRQs I am calling to
irq_create_mapping()from my platform atarch/arcand at that point I got no irqdomain and usingirq_find_host()is not good since I got nodevice_node(at most I can have DT root).Quoted 3 lines
Quoted 1 line
That's a problem. You should never do that for your timer (doing a
request_irqwill do the right thing, and that's what your timer driver already does).Please be more specific, from all that I wrote what is the problem?
Calling
irq_create_mappingout of the blue like you do it here:(quoted diff) +7
-0+static void eznps_init_per_cpu(int cpu) +{ + /* Create mapping for all per cpu IRQs */ + if (cpu == 0) { + irq_create_mapping(NULL, TIMER0_IRQ); + irq_create_mapping(NULL, IPI_IRQ); + }
Quoted 1 line
is simply not acceptable.
I understand but... see below
Quoted 9 lines
Quoted 6 lines
Quoted 3 lines
As for initializing your IPIs, they are usually outside of the IRQ space, so you should handle them separately (and get your irqchip to initialize them).
I am handling all my IRQs within same irqchip, which is the only one I have. So I am not sure what you expect here. Please be more elaborate.
Do not create a mapping for IPIs. Full stop. Handle them independently from your normal IRQs.
why not ? IPI is a hardware construct afterall.
Anyhow I looked in arch/arm and do_IPI/handle_IPI are the handlers. do_IPI is called from asm, irq-gic.c calls handle_IPI. I can't seem to find an explicit request_irq / request_percpu_irq for IPI irq ?
...
Quoted 7 lines
Quoted 6 lines
Note that I am working with ARC (seem alike) here and we do not define
CONFIG_HANDLE_DOMAIN_IRQand do not implementset_handle_irq().So for ARC this reverse mapping is something we can leave without (maybe because we are kind of a legacy domain).
Quoted 7 lines
Yeah, I just located the crap:
arch_do_IRQ()happily takes a hwirq (the vector number), and uses that as a Linux IRQ. This looks a lot like ARM pre-DT, about 10 years ago.Well, time to meet the 21st century. If you intend to use DT, please fix your arch port. Otherwise, just hardcode everything in your platform and don't pretend to support device tree.
Following works for me, hopefully it is closer to 21st century code :-)
----------->
Quoted 1 line
From
619eb5179d865140a723dd524d0e42fbf234b53bMon Sep 17 00:00:00 2001
From: Vineet Gupta <[email protected]>
Date: Fri, 1 Jan 2016 15:12:54 +0530
Subject: [PATCH] ARC: [intc-*] Do a domain lookup in primary handler for hwirq
-> linux virq
The primary interrupt handler arch_do_IRQ() was passing hwirq as linux virq to core code. This was fragile and worked so far as we only had legacy/linear domains.
This came out of a rant by Marc Zyngier.
http://lists.infradead.org/pipermail/linux-snps-arc/2015-December/000298.html
--- arch/arc/Kconfig | 1 + arch/arc/kernel/intc-arcv2.c | 9 ++++++--- arch/arc/kernel/intc-compact.c | 10 ++++++---- arch/arc/kernel/irq.c | 9 ++------- 4 files changed, 15 insertions(+), 14 deletions(-)
arch/arc/Kconfig +1 -0
@@ -31,6 +31,7 @@ config ARC select HAVE_MOD_ARCH_SPECIFIC if ARC_DW2_UNWIND select HAVE_OPROFILE select HAVE_PERF_EVENTS + select HANDLE_DOMAIN_IRQ select IRQ_DOMAIN select MODULES_USE_ELF_RELA select NO_BOOTMEM
arch/arc/kernel/intc-arcv2.c +6 -3
@@ -130,21 +130,24 @@ static const struct irq_domain_ops arcv2_irq_ops = { .map = arcv2_irq_map, }; -static struct irq_domain *root_domain;static int __init init_onchip_IRQ (struct device_node *intc,struct device_node *parent) { +struct irq_domain *root_domain; +if (parent)panic ("DeviceTree incore intc not a root irq controller\n" ); root_domain =irq_domain_add_legacy (intc,NR_CPU_IRQS ,0 ,0 , &arcv2_irq_ops,NULL ); -if (!root_domain)panic ("root irq domain not avail\n" ); -/* with this we don't need to export root_domain */ +/* +* Needed for primary domain lookup to succeed +* This is a primary irqchip, and can never have a parent +*/ irq_set_default_host (root_domain);return 0 ;
arch/arc/kernel/intc-compact.c +6 -4
@@ -97,21 +97,23 @@ static const struct irq_domain_ops arc_intc_domain_ops = { .map = arc_intc_domain_map, }; -static struct irq_domain *root_domain; -static int __init init_onchip_IRQ (struct device_node *intc,struct device_node *parent) { +struct irq_domain *root_domain; +if (parent)panic ("DeviceTree incore intc not a root irq controller\n" ); root_domain =irq_domain_add_legacy (intc,NR_CPU_IRQS ,0 ,0 , &arc_intc_domain_ops,NULL ); -if (!root_domain)panic ("root irq domain not avail\n" ); -/* with this we don't need to export root_domain */ +/* +* Needed for primary domain lookup to succeed +* This is a primary irqchip, and can never have a parent +*/ irq_set_default_host (root_domain);return 0 ;
arch/arc/kernel/irq.c +2 -7
@@ -41,14 +41,9 @@ void __init init_IRQ(void)* "C" Entry point for any ARC ISR, called from low level vector handler * @irq is the vector number read from ICAUSE reg of on-chip intc */ -void arch_do_IRQ (unsigned int irq,struct pt_regs *regs) +void arch_do_IRQ (unsigned int hwirq,struct pt_regs *regs) { -struct pt_regs *old_regs =set_irq_regs (regs); - -irq_enter (); -generic_handle_irq (irq); -irq_exit (); -set_irq_regs (old_regs); +handle_domain_irq (NULL , hwirq, regs); }/*
Signature · 1 line
2.5.0
Re: [PATCH v4 05/19] irqchip: add nps Internal and external irqchips
Hi Marc,
Please respond to Vineet last email.
I wish to close the IPI handling within my patch set.
Regards,
Noam
________________________________________
From: Vineet Gupta <[email protected]>
Sent: Monday, January 25, 2016 3:08:34 PM
To: Marc Zyngier; Noam Camus; linux-snps-arc at lists.infradead.org
Cc: linux-kernel at vger.kernel.org; Chris Metcalf; daniel.lezcano at linaro.org; Thomas Gleixner; Jason Cooper
Subject: Re: [PATCH v4 05/19] irqchip: add nps Internal and external irqchips
Hi Marc,
Quoted 24 lines · Marc Zyngier
Quoted 9 lines · Noam Camus
Quoted 8 lines
From: Marc Zyngier [mailto:marc.zyngier at arm.com]
Sent: Friday, December 18, 2015 1:21 PMQuoted 5 lines
I need this for my per CPU irqs such timer and IPI which do not come from some external device but from CPUs. For these IRQs I am calling to
irq_create_mapping()from my platform atarch/arcand at that point I got no irqdomain and usingirq_find_host()is not good since I got nodevice_node(at most I can have DT root).Quoted 3 lines
Quoted 1 line
That's a problem. You should never do that for your timer (doing a
request_irqwill do the right thing, and that's what your timer driver already does).Please be more specific, from all that I wrote what is the problem?
Calling
irq_create_mappingout of the blue like you do it here:(quoted diff) +7
-0+static void eznps_init_per_cpu(int cpu) +{ + /* Create mapping for all per cpu IRQs */ + if (cpu == 0) { + irq_create_mapping(NULL, TIMER0_IRQ); + irq_create_mapping(NULL, IPI_IRQ); + }
Quoted 1 line
is simply not acceptable.
I understand but... see below
Quoted 9 lines
Quoted 6 lines
Quoted 3 lines
As for initializing your IPIs, they are usually outside of the IRQ space, so you should handle them separately (and get your irqchip to initialize them).
I am handling all my IRQs within same irqchip, which is the only one I have. So I am not sure what you expect here. Please be more elaborate.
Do not create a mapping for IPIs. Full stop. Handle them independently from your normal IRQs.
why not ? IPI is a hardware construct afterall.
Anyhow I looked in arch/arm and do_IPI/handle_IPI are the handlers. do_IPI is called from asm, irq-gic.c calls handle_IPI. I can't seem to find an explicit request_irq / request_percpu_irq for IPI irq ?
...
Quoted 7 lines
Quoted 6 lines
Note that I am working with ARC (seem alike) here and we do not define
CONFIG_HANDLE_DOMAIN_IRQand do not implementset_handle_irq().So for ARC this reverse mapping is something we can leave without (maybe because we are kind of a legacy domain).
Quoted 7 lines
Yeah, I just located the crap:
arch_do_IRQ()happily takes a hwirq (the vector number), and uses that as a Linux IRQ. This looks a lot like ARM pre-DT, about 10 years ago.Well, time to meet the 21st century. If you intend to use DT, please fix your arch port. Otherwise, just hardcode everything in your platform and don't pretend to support device tree.
Following works for me, hopefully it is closer to 21st century code :-)
----------->
Quoted 1 line
From
619eb5179d865140a723dd524d0e42fbf234b53bMon Sep 17 00:00:00 2001
From: Vineet Gupta <[email protected]>
Date: Fri, 1 Jan 2016 15:12:54 +0530
Subject: [PATCH] ARC: [intc-*] Do a domain lookup in primary handler for hwirq
-> linux virq
The primary interrupt handler arch_do_IRQ() was passing hwirq as linux virq to core code. This was fragile and worked so far as we only had legacy/linear domains.
This came out of a rant by Marc Zyngier.
http://lists.infradead.org/pipermail/linux-snps-arc/2015-December/000298.html
--- arch/arc/Kconfig | 1 + arch/arc/kernel/intc-arcv2.c | 9 ++++++--- arch/arc/kernel/intc-compact.c | 10 ++++++---- arch/arc/kernel/irq.c | 9 ++------- 4 files changed, 15 insertions(+), 14 deletions(-)
arch/arc/Kconfig +1 -0
@@ -31,6 +31,7 @@ config ARC select HAVE_MOD_ARCH_SPECIFIC if ARC_DW2_UNWIND select HAVE_OPROFILE select HAVE_PERF_EVENTS + select HANDLE_DOMAIN_IRQ select IRQ_DOMAIN select MODULES_USE_ELF_RELA select NO_BOOTMEM
arch/arc/kernel/intc-arcv2.c +6 -3
@@ -130,21 +130,24 @@ static const struct irq_domain_ops arcv2_irq_ops = { .map = arcv2_irq_map, }; -static struct irq_domain *root_domain;static int __init init_onchip_IRQ (struct device_node *intc,struct device_node *parent) { +struct irq_domain *root_domain; +if (parent)panic ("DeviceTree incore intc not a root irq controller\n" ); root_domain =irq_domain_add_legacy (intc,NR_CPU_IRQS ,0 ,0 , &arcv2_irq_ops,NULL ); -if (!root_domain)panic ("root irq domain not avail\n" ); -/* with this we don't need to export root_domain */ +/* +* Needed for primary domain lookup to succeed +* This is a primary irqchip, and can never have a parent +*/ irq_set_default_host (root_domain);return 0 ;
arch/arc/kernel/intc-compact.c +6 -4
@@ -97,21 +97,23 @@ static const struct irq_domain_ops arc_intc_domain_ops = { .map = arc_intc_domain_map, }; -static struct irq_domain *root_domain; -static int __init init_onchip_IRQ (struct device_node *intc,struct device_node *parent) { +struct irq_domain *root_domain; +if (parent)panic ("DeviceTree incore intc not a root irq controller\n" ); root_domain =irq_domain_add_legacy (intc,NR_CPU_IRQS ,0 ,0 , &arc_intc_domain_ops,NULL ); -if (!root_domain)panic ("root irq domain not avail\n" ); -/* with this we don't need to export root_domain */ +/* +* Needed for primary domain lookup to succeed +* This is a primary irqchip, and can never have a parent +*/ irq_set_default_host (root_domain);return 0 ;
arch/arc/kernel/irq.c +2 -7
@@ -41,14 +41,9 @@ void __init init_IRQ(void)* "C" Entry point for any ARC ISR, called from low level vector handler * @irq is the vector number read from ICAUSE reg of on-chip intc */ -void arch_do_IRQ (unsigned int irq,struct pt_regs *regs) +void arch_do_IRQ (unsigned int hwirq,struct pt_regs *regs) { -struct pt_regs *old_regs =set_irq_regs (regs); - -irq_enter (); -generic_handle_irq (irq); -irq_exit (); -set_irq_regs (old_regs); +handle_domain_irq (NULL , hwirq, regs); }/*
--
2.5.0
PATCH v4 6/19 ARC: Set vmalloc size from configuration
In reply to Noam CamusFrom: Noam Camus <[email protected]>
User space use lower 2G of the virtual address space. However kernel steals upper 512M of this space.
This stolen space is used partially for vmalloc and the rest serves as gutter between kernel and user space.
The vmalloc size is depend on NR_CPUS since "per cpu" mechanism use vmalloc to allocate chunks for its use.
Historically vmalloc size was 256M however it is not enough in case of many CPUs e.g. 4K CPUs. For such case an extra 192M is allocated.
We are setting default vmalloc size to be 256M, any addtional space will be taken from kernel/user gutter.
This introduce build error since PGDIR_SIZE() cannot use "1UL" ibecause it is called from mm/tlbex.S by VMALLOC_START. Using _BITUL() solves this build error.
--- arch/arc/Kconfig | 8 ++++++++ arch/arc/include/asm/pgtable.h | 2 +- arch/arc/include/asm/processor.h | 10 +++++----- arch/arc/mm/tlb.c | 6 ++++++ 4 files changed, 20 insertions(+), 6 deletions(-)
arch/arc/Kconfig +8 -0
@@ -468,6 +468,14 @@ config ARCH_PHYS_ADDR_T_64BIT config ARCH_DMA_ADDR_T_64BIT bool +config ARC_VMALLOC_SIZE + hex "Vmalloc size (MB)" + range 0 512 + default "256" + help + By default equals to 256MB and the rest of 512MB is + left for gutter between kernel and user space. + config ARC_CURR_IN_REG bool "Dedicate Register r25 for current_task pointer" default y
arch/arc/include/asm/pgtable.h +1 -1
@@ -210,7 +210,7 @@#define BITS_FOR_PGD (32 - BITS_FOR_PTE - BITS_IN_PAGE)#define PGDIR_SHIFT (32 - BITS_FOR_PGD) -#define PGDIR_SIZE (1UL << PGDIR_SHIFT)/* vaddr span, not PDG sz */ +#define PGDIR_SIZE _BITUL(PGDIR_SHIFT)/* vaddr span, not PDG sz */ #define PGDIR_MASK (~(PGDIR_SIZE-1))#define PTRS_PER_PTE _BITUL(BITS_FOR_PTE)
arch/arc/include/asm/processor.h +5 -5
@@ -109,18 +109,18 @@ extern unsigned int get_wchan(struct task_struct *p);* 0xC000_0000 0xFFFF_FFFF (peripheral uncached space) * ----------------------------------------------------------------------------- */ -#define VMALLOC_START 0x70000000/* * 1 PGDIR_SIZE each for fixmap/pkmap, 2 PGDIR_SIZE gutter * See asm/highmem.h for details */ -#define VMALLOC_SIZE (PAGE_OFFSET - VMALLOC_START - PGDIR_SIZE * 4) -#define VMALLOC_END (VMALLOC_START + VMALLOC_SIZE) +#define VMALLOC_END (PAGE_OFFSET - PGDIR_SIZE * 4) +#define VMALLOC_SIZE (CONFIG_ARC_VMALLOC_SIZE << 20) +#define VMALLOC_START (VMALLOC_END - VMALLOC_SIZE) -#define USER_KERNEL_GUTTER 0x10000000 +#define TASK_SIZE 0x60000000 -#define TASK_SIZE (VMALLOC_START - USER_KERNEL_GUTTER) +#define USER_KERNEL_GUTTER (VMALLOC_START - TASK_SIZE)#define STACK_TOP TASK_SIZE#define STACK_TOP_MAX STACK_TOP
arch/arc/mm/tlb.c +6 -0
@@ -814,6 +814,12 @@ void arc_mmu_init(void)printk (arc_mmu_mumbojumbo (0 , str,sizeof (str))); +/* +* vmalloc size (in MB) sanity check, +* Can't be done in processor.h due to header include depenedencies +*/ +BUILD_BUG_ON (!IS_ALIGNED ((CONFIG_ARC_VMALLOC_SIZE <<20 ),PMD_SIZE )); +/* For efficiency sake, kernel is compile time built for a MMU ver * This must match the hardware it is running on. * Linux built for MMU V2, if run on MMU V1 will break down because V1
Signature · 1 line
1.7.1
PATCH v4 7/19 ARC: rwlock: disable interrupts in !LLSC variant
In reply to Noam CamusFrom: Noam Camus <[email protected]>
If we hold rwlock and interrupt occures we may end up spinning on it for ever during softirq. Note that this lock is an internal lock
and since the lock is free to be used from any context, the lock needs to be IRQ-safe.
Below you may see an example for interrupt we get while nl_table_lock is holding its rw->lock_mutex and we spinned on it for ever.
The concept for the fix was taken from SPARC.
[2015-05-12 19:16:12] Stack Trace:
[2015-05-12 19:16:12] arc_unwind_core+0xb8/0x11c [2015-05-12 19:16:12] dump_stack+0x68/0xac [2015-05-12 19:16:12] _raw_read_lock+0xa8/0xac [2015-05-12 19:16:12] netlink_broadcast_filtered+0x56/0x35c [2015-05-12 19:16:12] nlmsg_notify+0x42/0xa4 [2015-05-12 19:16:13] neigh_update+0x1fe/0x44c [2015-05-12 19:16:13] neigh_event_ns+0x40/0xa4 [2015-05-12 19:16:13] arp_process+0x46e/0x5a8 [2015-05-12 19:16:13] __netif_receive_skb_core+0x358/0x500 [2015-05-12 19:16:13] process_backlog+0x92/0x154 [2015-05-12 19:16:13] net_rx_action+0xb8/0x188 [2015-05-12 19:16:13] __do_softirq+0xda/0x1d8 [2015-05-12 19:16:14] irq_exit+0x8a/0x8c [2015-05-12 19:16:14] arch_do_IRQ+0x6c/0xa8 [2015-05-12 19:16:14] handle_interrupt_level1+0xe4/0xf0
--- arch/arc/include/asm/spinlock.h | 14 ++++++++++++++ 1 files changed, 14 insertions(+), 0 deletions(-)
arch/arc/include/asm/spinlock.h +14 -0
@@ -610,7 +610,9 @@ static inline void arch_spin_unlock(arch_spinlock_t *lock)static inline int arch_read_trylock (arch_rwlock_t *rw) {int ret =0 ; +unsigned long flags; +local_irq_save (flags);arch_spin_lock (&(rw->lock_mutex));/*
@@ -623,6 +625,7 @@ static inline int arch_read_trylock(arch_rwlock_t *rw) }arch_spin_unlock (&(rw->lock_mutex)); +local_irq_restore (flags);smp_mb ();return ret;
@@ -632,7 +635,9 @@ static inline int arch_read_trylock(arch_rwlock_t *rw)static inline int arch_write_trylock (arch_rwlock_t *rw) {int ret =0 ; +unsigned long flags; +local_irq_save (flags);arch_spin_lock (&(rw->lock_mutex));/*
@@ -646,6 +651,7 @@ static inline int arch_write_trylock(arch_rwlock_t *rw) ret =1 ; }arch_spin_unlock (&(rw->lock_mutex)); +local_irq_restore (flags);return ret; }
@@ -664,16 +670,24 @@ static inline void arch_write_lock(arch_rwlock_t *rw)static inline void arch_read_unlock (arch_rwlock_t *rw) { +unsigned long flags; + +local_irq_save (flags);arch_spin_lock (&(rw->lock_mutex)); rw->counter++;arch_spin_unlock (&(rw->lock_mutex)); +local_irq_restore (flags); }static inline void arch_write_unlock (arch_rwlock_t *rw) { +unsigned long flags; + +local_irq_save (flags);arch_spin_lock (&(rw->lock_mutex)); rw->counter = __ARCH_RW_LOCK_UNLOCKED__;arch_spin_unlock (&(rw->lock_mutex)); +local_irq_restore (flags); }#endif
Signature · 1 line
1.7.1
PATCH v4 8/19 ARC: rename smp operation init_irq_cpu() to init_per_cpu()
In reply to Noam CamusFrom: Noam Camus <[email protected]>
This will better reflect its description i.e. "any needed setup..." and not just do an "IPI request".
--- arch/arc/include/asm/smp.h | 4 ++-- arch/arc/kernel/irq.c | 4 ++-- arch/arc/kernel/mcip.c | 2 +- arch/arc/kernel/smp.c | 4 ++-- 4 files changed, 7 insertions(+), 7 deletions(-)
arch/arc/include/asm/smp.h +2 -2
@@ -48,7 +48,7 @@ extern int smp_ipi_irq_setup(int cpu, int irq);* @init_early_smp: A SMP specific h/w block can init itself * Could be common across platforms so not covered by * mach_desc->init_early() -* @init_irq_cpu: Called for each core so SMP h/w block driver can do +* @init_per_cpu: Called for each core so SMP h/w block driver can do * any needed setup per cpu (e.g. IPI request) * @cpu_kick: For Master to kickstart a cpu (optionally at a PC) * @ipi_send: To send IPI to a @cpu
@@ -57,7 +57,7 @@ extern int smp_ipi_irq_setup(int cpu, int irq);struct plat_smp_ops {const char *info;void (*init_early_smp)(void ); -void (*init_irq_cpu)(int cpu); +void (*init_per_cpu)(int cpu);void (*cpu_kick)(int cpu,unsigned long pc);void (*ipi_send)(int cpu);void (*ipi_clear)(int irq);
arch/arc/kernel/irq.c +2 -2
@@ -29,8 +29,8 @@ void __init init_IRQ(void)#ifdef CONFIG_SMP /* a SMP H/w block could do IPI IRQ request here */ -if (plat_smp_ops.init_irq_cpu) - plat_smp_ops.init_irq_cpu (smp_processor_id ()); +if (plat_smp_ops.init_per_cpu) + plat_smp_ops.init_per_cpu (smp_processor_id ());if (machine_desc->init_cpu_smp) machine_desc->init_cpu_smp (smp_processor_id ());
arch/arc/kernel/mcip.c +1 -1
@@ -132,7 +132,7 @@ static void mcip_probe_n_setup(void)struct plat_smp_ops plat_smp_ops = { .info = smp_cpuinfo_buf, .init_early_smp = mcip_probe_n_setup, - .init_irq_cpu = mcip_setup_per_cpu, + .init_per_cpu = mcip_setup_per_cpu, .ipi_send = mcip_ipi_send, .ipi_clear = mcip_ipi_clear, };
arch/arc/kernel/smp.c +2 -2
@@ -132,8 +132,8 @@ void start_kernel_secondary(void)pr_info ("## CPU%u LIVE ##: Executing Code...\n" , cpu);/* Some SMP H/w setup - for each cpu */ -if (plat_smp_ops.init_irq_cpu) - plat_smp_ops.init_irq_cpu (cpu); +if (plat_smp_ops.init_per_cpu) + plat_smp_ops.init_per_cpu (cpu);if (machine_desc->init_cpu_smp) machine_desc->init_cpu_smp (cpu);
Signature · 1 line
1.7.1
PATCH v4 9/19 ARC: Mark secondary cpu online only after all HW setup is done
In reply to Noam CamusFrom: Noam Camus <[email protected]>
In SMP setup, master loops for each_present_cpu calling cpu_up(). For ARC it returns as soon as new cpu's status becomes online, However secondary may still do HW initializing,
machine or platform hook level.
So turn secondary online only after all HW setup is done.
Signed-off-by: Noam Camus <noamc at ezchip.com>
Acked-by: Vineet Gupta <vgupta at synopsys.com>
--- arch/arc/kernel/smp.c | 10 +++++----- 1 files changed, 5 insertions(+), 5 deletions(-)
arch/arc/kernel/smp.c +5 -5
@@ -126,11 +126,6 @@ void start_kernel_secondary(void) current->active_mm = mm;cpumask_set_cpu (cpu,mm_cpumask (mm)); -notify_cpu_starting (cpu); -set_cpu_online (cpu, true); - -pr_info ("## CPU%u LIVE ##: Executing Code...\n" , cpu); -/* Some SMP H/w setup - for each cpu */ if (plat_smp_ops.init_per_cpu) plat_smp_ops.init_per_cpu (cpu);
@@ -138,6 +133,11 @@ void start_kernel_secondary(void)if (machine_desc->init_cpu_smp) machine_desc->init_cpu_smp (cpu); +notify_cpu_starting (cpu); +set_cpu_online (cpu, true); + +pr_info ("## CPU%u LIVE ##: Executing Code...\n" , cpu); +arc_local_timer_setup ();local_irq_enable ();
Signature · 1 line
1.7.1
PATCH v4 10/19 ARC: Add clock from device tree to time_init()
In reply to Noam CamusFrom: Noam Camus <[email protected]>
External clock source can be used if included by one of DTS file of a chosen platform.
Clocksource may be depend on some clk from DTS as the real source of this clock.
--- arch/arc/kernel/time.c | 10 +++++++++- 1 files changed, 9 insertions(+), 1 deletions(-)
arch/arc/kernel/time.c +9 -1
@@ -29,6 +29,8 @@* which however is currently broken */ +#include <linux/clk-provider.h> +#include <linux/clocksource.h> #include <linux/spinlock.h> #include <linux/interrupt.h> #include <linux/module.h>
@@ -38,7 +40,6 @@#include <linux/init.h> #include <linux/timex.h> #include <linux/profile.h> -#include <linux/clocksource.h> #include <linux/clockchips.h> #include <asm/irq.h> #include <asm/arcregs.h>
@@ -283,6 +284,13 @@ void __init time_init(void)*/ clocksource_register_hz (&arc_counter,arc_get_core_freq ()); +#ifdef CONFIG_COMMON_CLK +of_clk_init (NULL ); +#endif +#ifdef CONFIG_CLKSRC_OF +clocksource_probe (); +#endif +/* sets up the periodic event timer */ arc_local_timer_setup (); }
Signature · 1 line
1.7.1
PATCH v4 11/19 ARC: [plat-eznps] Add eznps board defconfig and dts
In reply to Noam CamusFrom: Noam Camus <[email protected]>
Adding default configuration file and DTS file
--- arch/arc/boot/dts/eznps.dts | 93 ++++++++++++++++++++++++++++++++++++++++ arch/arc/configs/nps_defconfig | 85 ++++++++++++++++++++++++++++++++++++ 2 files changed, 178 insertions(+), 0 deletions(-) create mode 100644 arch/arc/boot/dts/eznps.dts create mode 100644 arch/arc/configs/nps_defconfig
arch/arc/boot/dts/eznps.dts new file +93 -0
@@ -0,0 +1,93 @@ +/* + * Copyright(c) 2015 EZchip Technologies. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope 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. + * + * The full GNU General Public License is included in this distribution in + * the file called "COPYING". + */ + +/dts-v1/; + +/include/ "skeleton.dtsi" + +/ { + compatible = "ezchip,arc-nps"; + clock-frequency = <83333333>; /* 83.333333 MHZ */ + #address-cells = <1>; + #size-cells = <1>; + interrupt-parent = <&intc>; + present-cpus = "0-1,16-17"; + possible-cpus = "0-4095"; + + aliases { + ethernet0 = &gmac0; + pll = &sysclk; + }; + + chosen { + bootargs = "earlycon=uart8250,mmio32be,0xf7209000,115200n8 console=ttyS0,115200n8"; + }; + + memory { + device_type = "memory"; + reg = <0x00000000 0x20000000>; /* 512M */ + }; + + clocks { + sysclk: sysclk { + compatible = "fixed-clock"; + #clock-cells = <0>; + clock-frequency = <83333333>; + }; + }; + + soc { + compatible = "simple-bus"; + #address-cells = <1>; + #size-cells = <1>; + + /* child and parent address space 1:1 mapped */ + ranges; + + intc: interrupt-controller { + compatible = "ezchip,nps400-ic"; + interrupt-controller; + #interrupt-cells = <1>; + }; + + timer { + compatible = "ezchip,nps400-timer"; + clocks = <&sysclk>; + clock-names="sysclk"; + }; + + uart at f7209000 { + compatible = "snps,dw-apb-uart"; + device_type = "serial"; + reg = <0xf7209000 0x100>; + interrupts = <6>; + clocks = <&sysclk>; + clock-names="baudclk"; + baud = <115200>; + reg-shift = <2>; + reg-io-width = <4>; + native-endian; + }; + + gmac0: ethernet at f7470000 { + compatible = "ezchip,nps-mgt-enet"; + reg = <0xf7470000 0x1940>; + interrupts = <7>; + /* Filled in by U-Boot */ + mac-address = [ 00 C0 00 F0 04 03 ]; + }; + }; +};
arch/arc/configs/nps_defconfig new file +85 -0
85 lines not on this page · raw message
Signature · 1 line
1.7.1
PATCH v4 12/19 ARC: [plat-eznps] Add eznps platform
In reply to Noam CamusFrom: Noam Camus <[email protected]>
This platform include boards:
Hardware Emulator (HE)
Simulator based upon nSIM.
--- MAINTAINERS | 6 + arch/arc/plat-eznps/Kconfig | 34 ++++++ arch/arc/plat-eznps/Makefile | 7 + arch/arc/plat-eznps/entry.S | 75 ++++++++++++ arch/arc/plat-eznps/include/plat/ctop.h | 198 +++++++++++++++++++++++++++++++ arch/arc/plat-eznps/include/plat/mtm.h | 60 +++++++++ arch/arc/plat-eznps/include/plat/smp.h | 26 ++++ arch/arc/plat-eznps/mtm.c | 133 +++++++++++++++++++++ arch/arc/plat-eznps/platform.c | 102 ++++++++++++++++ arch/arc/plat-eznps/smp.c | 156 ++++++++++++++++++++++++ 10 files changed, 797 insertions(+), 0 deletions(-) create mode 100644 arch/arc/plat-eznps/Kconfig create mode 100644 arch/arc/plat-eznps/Makefile create mode 100644 arch/arc/plat-eznps/entry.S create mode 100644 arch/arc/plat-eznps/include/plat/ctop.h create mode 100644 arch/arc/plat-eznps/include/plat/mtm.h create mode 100644 arch/arc/plat-eznps/include/plat/smp.h create mode 100644 arch/arc/plat-eznps/mtm.c create mode 100644 arch/arc/plat-eznps/platform.c create mode 100644 arch/arc/plat-eznps/smp.c
MAINTAINERS +6 -0
@@ -4257,6 +4257,12 @@ S: Maintained F: drivers/video/fbdev/exynos/exynos_mipi* F: include/video/exynos_mipi* +EZchip NPS platform support +M: Noam Camus <noamc at ezchip.com> +S: Supported +F: arch/arc/plat-eznps +F: arch/arc/boot/dts/eznps.dts + F71805F HARDWARE MONITORING DRIVER M: Jean Delvare <jdelvare at suse.com> L: lm-sensors at lm-sensors.org
arch/arc/plat-eznps/Kconfig new file +34 -0
@@ -0,0 +1,34 @@ +# +# For a description of the syntax of this configuration file, +# see Documentation/kbuild/kconfig-language.txt. +# + +menuconfig ARC_PLAT_EZNPS + bool "\"EZchip\" ARC dev platform" + select ARC_HAS_COH_CACHES if SMP + select CPU_BIG_ENDIAN + select CLKSRC_NPS + select EZCHIP_NPS_MANAGEMENT_ENET if ETHERNET + help + Support for EZchip development platforms, + based on ARC700 cores. + We handle few flavours: + - Hardware Emulator AKA HE which is FPGA based chasis + - Simulator based on MetaWare nSIM + - NPS400 chip based on ASIC + +config EZNPS_MTM_EXT + bool "ARC-EZchip MTM Extensions" + select CPUMASK_OFFSTACK + depends on ARC_PLAT_EZNPS && SMP + default y + help + Here we add new hierarchy for CPUs topology. + We got: + Core + Thread + At the new thread level each CPU represent one HW thread. + At highest hierarchy each core contain 16 threads, + any of them seem like CPU from Linux point of view. + All threads within same core share the execution unit of the + core and HW scheduler round robin between them.
arch/arc/plat-eznps/Makefile new file +7 -0
@@ -0,0 +1,7 @@ +# +# Makefile for the linux kernel. +# + +obj-y := entry.o platform.o +obj-$(CONFIG_SMP) += smp.o +obj-$(CONFIG_EZNPS_MTM_EXT) += mtm.o
arch/arc/plat-eznps/entry.S new file +75 -0
75 lines not on this page · raw message
arch/arc/plat-eznps/include/plat/ctop.h new file +198 -0
198 lines not on this page · raw message
arch/arc/plat-eznps/include/plat/mtm.h new file +60 -0
60 lines not on this page · raw message
arch/arc/plat-eznps/include/plat/smp.h new file +26 -0
@@ -0,0 +1,26 @@ +/* +* Copyright(c) 2015 EZchip Technologies. +* +* This program is free software; you can redistribute it and/or modify it +* under the terms and conditions of the GNU General Public License, +* version 2, as published by the Free Software Foundation. +* +* This program is distributed in the hope 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. +* +* The full GNU General Public License is included in this distribution in +* the file called "COPYING". +*/ + +#ifndef __PLAT_EZNPS_SMP_H +#define __PLAT_EZNPS_SMP_H + +#ifdef CONFIG_SMP + +extern void res_service (void ); + +#endif /* CONFIG_SMP */ + +#endif
arch/arc/plat-eznps/mtm.c new file +133 -0
133 lines not on this page · raw message
arch/arc/plat-eznps/platform.c new file +102 -0
102 lines not on this page · raw message
arch/arc/plat-eznps/smp.c new file +156 -0
156 lines not on this page · raw message
Signature · 1 line
1.7.1
PATCH v4 13/19 ARC: [plat-eznps] Use dedicated user stack top
In reply to Noam CamusFrom: Noam Camus <[email protected]>
NPS use special mapping right below TASK_SIZE. Hence we need to lower STACK_TOP so that user stack won't overlap NPS special mapping.
--- arch/arc/include/asm/processor.h | 17 +++++++++++++++++ arch/arc/mm/tlb.c | 6 ++++++ 2 files changed, 23 insertions(+), 0 deletions(-)
arch/arc/include/asm/processor.h +17 -0
@@ -122,7 +122,24 @@ extern unsigned int get_wchan(struct task_struct *p);#define USER_KERNEL_GUTTER (VMALLOC_START - TASK_SIZE) +#ifdef CONFIG_ARC_PLAT_EZNPS +/* NPS architecture defines special window of 129M in user address space for +* special memory areas, when accessing this window the MMU do not use TLB. +* Instead MMU direct the access to: +* 0x57f00000:0x57ffffff -- 1M of closely coupled memory (aka CMEM) +* 0x58000000:0x5fffffff -- 16 huge pages, 8M each, with fixed map (aka FMTs) +* +* CMEM - is the fastest memory we got and its size is 16K. +* FMT - is used to map either to internal/external memory. +* Internal memory is the second fast memory and its size is 16M +* External memory is the biggest memory (16G) and also the slowest. +* +* STACK_TOP need to be PMD align (21bit) that is why we supply 0x57e00000. +*/ +#define STACK_TOP 0x57e00000 +#else /* CONFIG_ARC_PLAT_EZNPS */ #define STACK_TOP TASK_SIZE +#endif /* CONFIG_ARC_PLAT_EZNPS */ #define STACK_TOP_MAX STACK_TOP/* This decides where the kernel will search for a free chunk of vm
arch/arc/mm/tlb.c +6 -0
@@ -820,6 +820,12 @@ void arc_mmu_init(void)*/ BUILD_BUG_ON (!IS_ALIGNED ((CONFIG_ARC_VMALLOC_SIZE <<20 ),PMD_SIZE )); +/* +* stack top size sanity check, +* Can't be done in processor.h due to header include depenedencies +*/ +BUILD_BUG_ON (!IS_ALIGNED (STACK_TOP ,PMD_SIZE )); +/* For efficiency sake, kernel is compile time built for a MMU ver * This must match the hardware it is running on. * Linux built for MMU V2, if run on MMU V1 will break down because V1
Signature · 1 line
1.7.1
PATCH v4 14/19 ARC: [plat-eznps] Use dedicated atomic/bitops/cmpxchg
In reply to Noam CamusFrom: Noam Camus <[email protected]>
We need our own implementaions since we lack LLSC support. Our extended ISA provided with optimized solution for all 32bit operations we see in these three headers.
Signed-off-by: Noam Camus <noamc at ezchip.com>
--- arch/arc/include/asm/atomic.h | 79 +++++++++++++++++++++++++++++++++++- arch/arc/include/asm/bitops.h | 54 +++++++++++++++++++++++++ arch/arc/include/asm/cmpxchg.h | 87 +++++++++++++++++++++++++++++++++------- 3 files changed, 202 insertions(+), 18 deletions(-)
arch/arc/include/asm/atomic.h +76 -3
@@ -17,6 +17,7 @@#include <asm/barrier.h> #include <asm/smp.h> +#ifndef CONFIG_ARC_PLAT_EZNPS #define atomic_read (v) READ_ONCE((v)->counter)#ifdef CONFIG_ARC_HAS_LLSC
@@ -180,12 +181,84 @@ ATOMIC_OP(andnot, &= ~, bic) ATOMIC_OP(or, |=, or) ATOMIC_OP(xor, ^=, xor) -#undef ATOMIC_OPS -#undef ATOMIC_OP_RETURN -#undef ATOMIC_OP#undef SCOND_FAIL_RETRY_VAR_DEF#undef SCOND_FAIL_RETRY_ASM#undef SCOND_FAIL_RETRY_VARS +#else /* CONFIG_ARC_PLAT_EZNPS */ +static inline int atomic_read (const atomic_t *v) +{ +int temp; + + __asm__ __volatile__( +" ld.di %0, [%1]" + :"=r" (temp) + :"r" (&v->counter) + :"memory" ); +return temp; +} + +static inline void atomic_set (atomic_t *v,int i) +{ + __asm__ __volatile__( +" st.di %0,[%1]" + : + :"r" (i),"r" (&v->counter) + :"memory" ); +} + +#define ATOMIC_OP (op, c_op, asm_op) \ +static inline void atomic_##op(int i, atomic_t *v) \ +{ \ + __asm__ __volatile__( \ + " mov r2, %0\n" \ + " mov r3, %1\n" \ + " .word %2\n" \ + : \ + : "r"(i), "r"(&v->counter), "i"(asm_op) \ + : "r2", "r3", "memory"); \ +} \ + +#define ATOMIC_OP_RETURN (op, c_op, asm_op) \ +static inline int atomic_##op##_return(int i, atomic_t *v) \ +{ \ + unsigned int temp = i; \ + \ +/* Explicit full memory barrier needed before/after */ \ +smp_mb (); \ + \ + __asm__ __volatile__( \ +" mov r2, %0\n" \ +" mov r3, %1\n" \ +" .word %2\n" \ +" mov %0, r2" \ + :"+r" (temp) \ + :"r" (&v->counter),"i" (asm_op) \ + :"r2" ,"r3" ,"memory" ); \ + \ +smp_mb (); \ + \ + temp c_op i; \ + \ +return temp; \ +} + +#define ATOMIC_OPS (op, c_op, asm_op) \ + ATOMIC_OP(op, c_op, asm_op) \ + ATOMIC_OP_RETURN(op, c_op, asm_op) + +ATOMIC_OPS(add, +=, CTOP_INST_AADD_DI_R2_R2_R3) +#define atomic_sub (i, v) atomic_add(-(i), (v)) +#define atomic_sub_return (i, v) atomic_add_return(-(i), (v)) + +ATOMIC_OP(and, &=, CTOP_INST_AAND_DI_R2_R2_R3) +#define atomic_andnot (mask, v) atomic_and(~(mask), (v)) +ATOMIC_OP(or, |=, CTOP_INST_AOR_DI_R2_R2_R3) +ATOMIC_OP(xor, ^=, CTOP_INST_AXOR_DI_R2_R2_R3) +#endif /* CONFIG_ARC_PLAT_EZNPS */ + +#undef ATOMIC_OPS +#undef ATOMIC_OP_RETURN +#undef ATOMIC_OP/** * __atomic_add_unless - add unless the number is a given value
arch/arc/include/asm/bitops.h +54 -0
75 lines not on this page · raw message
arch/arc/include/asm/cmpxchg.h +72 -15
109 lines not on this page · raw message
Signature · 1 line
1.7.1
PATCH v4 15/19 ARC: [plat-eznps] Use dedicated SMP barriers
In reply to Noam CamusFrom: Noam Camus <[email protected]>
NPS device got 256 cores and each got 16 HW threads (SMT). We use EZchip dedicated ISA to trigger HW scheduler of the core that current HW thread belongs to.
This scheduling makes sure that data beyond barrier is available to all HW threads in core and by that to all in device (4K).
--- arch/arc/include/asm/barrier.h | 8 ++++++++ 1 files changed, 8 insertions(+), 0 deletions(-)
arch/arc/include/asm/barrier.h +8 -0
@@ -34,6 +34,7 @@#ifdef CONFIG_ISA_ARCOMPACT +#ifndef CONFIG_ARC_PLAT_EZNPS /* * ARCompact based cores (ARC700) only have SYNC instruction which is super * heavy weight as it flushes the pipeline as well.
@@ -41,6 +42,13 @@*/ #define mb () asm volatile("sync\n" : : : "memory") +#else /* CONFIG_ARC_PLAT_EZNPS */ + +#include <plat/ctop.h> +#define mb () asm volatile (".word %0" : : "i"(CTOP_INST_SCHD_RW) : "memory") +#define rmb () asm volatile (".word %0" : : "i"(CTOP_INST_SCHD_RD) : "memory") +#endif /* CONFIG_ARC_PLAT_EZNPS */ +#endif #include <asm-generic/barrier.h>
Signature · 1 line
1.7.1
PATCH v4 16/19 ARC: [plat-eznps] Use dedicated identity auxiliary register.
In reply to Noam CamusFrom: Noam Camus <[email protected]>
With generic "identity" num of CPUs is limited to 256 (8 bit). We use our alternative AUX register GLOBAL_ID (12 bit). Now we can support up to 4096 CPUs.
--- arch/arc/include/asm/entry-compact.h | 8 ++++++++ arch/arc/kernel/ctx_sw.c | 13 +++++++++++++ 2 files changed, 21 insertions(+), 0 deletions(-)
arch/arc/include/asm/entry-compact.h +8 -0
@@ -35,6 +35,7 @@#include <asm/asm-offsets.h> #include <asm/irqflags-compact.h> #include <asm/thread_info.h> /* For THREAD_SIZE */ +#include <plat/ctop.h> /*-------------------------------------------------------------- * Switch to Kernel Mode stack if SP points to User Mode stack
@@ -298,9 +299,16 @@ /* Get CPU-ID of this core */ .macro GET_CPU_ID reg +#ifdef CONFIG_ARC_PLAT_EZNPS + lr \reg, [CTOP_AUX_LOGIC_GLOBAL_ID] +#ifndef CONFIG_EZNPS_MTM_EXT + lsr \reg, \reg, 4 +#endif +#else lr \reg, [identity] lsr \reg, \reg, 8 bmsk \reg, \reg, 7 +#endif .endm #endif /* __ASM_ARC_ENTRY_COMPACT_H */
arch/arc/kernel/ctx_sw.c +13 -0
@@ -16,6 +16,9 @@#include <asm/asm-offsets.h> #include <linux/sched.h> +#ifdef CONFIG_ARC_PLAT_EZNPS +#include <plat/ctop.h> +#endif #define KSP_WORD_OFF ((TASK_THREAD + THREAD_KSP) / 4)
@@ -67,9 +70,16 @@ __switch_to(struct task_struct *prev_task, struct task_struct *next_task)#ifndef CONFIG_SMP "st %2, [@_current_task] \n\t"#else +#ifdef CONFIG_ARC_PLAT_EZNPS +"lr r24, [%4] \n\t" +#ifndef CONFIG_EZNPS_MTM_EXT + "lsr r24, r24, 4 \n\t" +#endif +#else "lr r24, [identity] \n\t" "lsr r24, r24, 8 \n\t" "bmsk r24, r24, 7 \n\t" +#endif "add2 r24, @_current_task, r24 \n\t" "st %2, [r24] \n\t" #endif
@@ -107,6 +117,9 @@ __switch_to(struct task_struct *prev_task, struct task_struct *next_task) : "=r"(tmp) : "n"(KSP_WORD_OFF), "r"(next), "r"(prev) +#ifdef CONFIG_ARC_PLAT_EZNPS + , "i"(CTOP_AUX_LOGIC_GLOBAL_ID) +#endif : "blink" );
Signature · 1 line
1.7.1
PATCH v4 17/19 ARC: [plat-eznps] Use dedicated cpu_relax()
In reply to Noam CamusFrom: Tal Zilcer <[email protected]>
Since the CTOP is SMT hardware multi-threaded, we need to hint the HW that now will be a very good time to do a hardware thread context switching. This is done by issuing the schd.rw instruction (binary coded here so as to not require specific revision of GCC to build the kernel).
sched.rw means that Thread becomes eligible for execution by the threads scheduler after all pending read/write transactions were completed.
Implementing cpu_relax_lowlatency() with barrier() Since with current semantics of cpu_relax() it may take a while till yielded CPU will get back.
--- arch/arc/include/asm/processor.h | 9 +++++++++ 1 files changed, 9 insertions(+), 0 deletions(-)
arch/arc/include/asm/processor.h +9 -0
@@ -57,9 +57,18 @@ struct task_struct;* A lot of busy-wait loops in SMP are based off of non-volatile data otherwise * get optimised away by gcc */ +#ifndef CONFIG_EZNPS_MTM_EXT #define cpu_relax () __asm__ __volatile__ ("" : : : "memory") +#else +#define cpu_relax () \ + __asm__ __volatile__ (".word %0" : : "i"(CTOP_INST_SCHD_RW) : "memory") +#endif +#ifndef CONFIG_EZNPS_MTM_EXT #define cpu_relax_lowlatency () cpu_relax() +#else +#define cpu_relax_lowlatency () barrier() +#endif #define copy_segments (tsk, mm) do { } while (0)#define release_segments (mm) do { } while (0)
Signature · 1 line
1.7.1
PATCH v4 18/19 ARC: [plat-eznps] Use dedicated COMMAND_LINE_SIZE
In reply to Noam CamusFrom: Noam Camus <[email protected]>
The default 256 bytes sometimes is just not enough. We usually provide earlycon=... and console=... and ip=... All this and more may need more room.
--- arch/arc/include/asm/setup.h | 4 ++++ 1 files changed, 4 insertions(+), 0 deletions(-)
arch/arc/include/asm/setup.h +4 -0
@@ -12,7 +12,11 @@#include <linux/types.h> #include <uapi/asm/setup.h> +#ifdef CONFIG_ARC_PLAT_EZNPS +#define COMMAND_LINE_SIZE 2048 +#else #define COMMAND_LINE_SIZE 256 +#endif /* * Data structure to map a ID to string
Signature · 1 line
1.7.1
PATCH v4 19/19 ARC: Add eznps platform to Kconfig and Makefile
In reply to Noam CamusFrom: Noam Camus <[email protected]>
This commit should be left last since only now eznps platform is in state which one can actually use.
Signed-off-by: Noam Camus <noamc at ezchip.com>
--- arch/arc/Kconfig | 1 + arch/arc/Makefile | 5 +++++ 2 files changed, 6 insertions(+), 0 deletions(-)
arch/arc/Kconfig +1 -0
@@ -91,6 +91,7 @@ source "arch/arc/plat-sim/Kconfig" source "arch/arc/plat-tb10x/Kconfig" source "arch/arc/plat-axs10x/Kconfig" #New platform adds here +source "arch/arc/plat-eznps/Kconfig" endmenu
arch/arc/Makefile +5 -0
@@ -99,6 +99,11 @@ core-y += arch/arc/boot/dts/ core-$(CONFIG_ARC_PLAT_SIM) += arch/arc/plat-sim/ core-$(CONFIG_ARC_PLAT_TB10X) += arch/arc/plat-tb10x/ core-$(CONFIG_ARC_PLAT_AXS10X) += arch/arc/plat-axs10x/ +core-$(CONFIG_ARC_PLAT_EZNPS) += arch/arc/plat-eznps/ + +ifdef CONFIG_ARC_PLAT_EZNPS +KBUILD_CPPFLAGS += -I$(srctree)/arch/arc/plat-eznps/include +endif drivers-$(CONFIG_OPROFILE) += arch/arc/oprofile/
Signature · 1 line
1.7.1