diff options
Diffstat (limited to 'arch/x86/platform/uv')
| -rw-r--r-- | arch/x86/platform/uv/Makefile | 2 | ||||
| -rw-r--r-- | arch/x86/platform/uv/bios_uv.c | 269 | ||||
| -rw-r--r-- | arch/x86/platform/uv/uv_irq.c | 217 | ||||
| -rw-r--r-- | arch/x86/platform/uv/uv_nmi.c | 1096 | ||||
| -rw-r--r-- | arch/x86/platform/uv/uv_time.c | 393 |
5 files changed, 1977 insertions, 0 deletions
diff --git a/arch/x86/platform/uv/Makefile b/arch/x86/platform/uv/Makefile new file mode 100644 index 000000000..1441dda8e --- /dev/null +++ b/arch/x86/platform/uv/Makefile @@ -0,0 +1,2 @@ +# SPDX-License-Identifier: GPL-2.0-only +obj-$(CONFIG_X86_UV) += bios_uv.o uv_irq.o uv_time.o uv_nmi.o diff --git a/arch/x86/platform/uv/bios_uv.c b/arch/x86/platform/uv/bios_uv.c new file mode 100644 index 000000000..bf31af3d3 --- /dev/null +++ b/arch/x86/platform/uv/bios_uv.c @@ -0,0 +1,269 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * BIOS run time interface routines. + * + * (C) Copyright 2020 Hewlett Packard Enterprise Development LP + * Copyright (C) 2007-2017 Silicon Graphics, Inc. All rights reserved. + * Copyright (c) Russ Anderson <rja@sgi.com> + */ + +#include <linux/efi.h> +#include <linux/export.h> +#include <linux/slab.h> +#include <asm/efi.h> +#include <linux/io.h> +#include <asm/pgalloc.h> +#include <asm/uv/bios.h> +#include <asm/uv/uv_hub.h> + +unsigned long uv_systab_phys __ro_after_init = EFI_INVALID_TABLE_ADDR; + +struct uv_systab *uv_systab; + +static s64 __uv_bios_call(enum uv_bios_cmd which, u64 a1, u64 a2, u64 a3, + u64 a4, u64 a5) +{ + struct uv_systab *tab = uv_systab; + s64 ret; + + if (!tab || !tab->function) + /* + * BIOS does not support UV systab + */ + return BIOS_STATUS_UNIMPLEMENTED; + + ret = efi_call_virt_pointer(tab, function, (u64)which, a1, a2, a3, a4, a5); + + return ret; +} + +static s64 uv_bios_call(enum uv_bios_cmd which, u64 a1, u64 a2, u64 a3, u64 a4, + u64 a5) +{ + s64 ret; + + if (down_interruptible(&__efi_uv_runtime_lock)) + return BIOS_STATUS_ABORT; + + ret = __uv_bios_call(which, a1, a2, a3, a4, a5); + up(&__efi_uv_runtime_lock); + + return ret; +} + +static s64 uv_bios_call_irqsave(enum uv_bios_cmd which, u64 a1, u64 a2, u64 a3, + u64 a4, u64 a5) +{ + unsigned long bios_flags; + s64 ret; + + if (down_interruptible(&__efi_uv_runtime_lock)) + return BIOS_STATUS_ABORT; + + local_irq_save(bios_flags); + ret = __uv_bios_call(which, a1, a2, a3, a4, a5); + local_irq_restore(bios_flags); + + up(&__efi_uv_runtime_lock); + + return ret; +} + +long sn_partition_id; +EXPORT_SYMBOL_GPL(sn_partition_id); +long sn_coherency_id; +EXPORT_SYMBOL_GPL(sn_coherency_id); +long sn_region_size; +EXPORT_SYMBOL_GPL(sn_region_size); +long system_serial_number; +int uv_type; + +s64 uv_bios_get_sn_info(int fc, int *uvtype, long *partid, long *coher, + long *region, long *ssn) +{ + s64 ret; + u64 v0, v1; + union partition_info_u part; + + ret = uv_bios_call_irqsave(UV_BIOS_GET_SN_INFO, fc, + (u64)(&v0), (u64)(&v1), 0, 0); + if (ret != BIOS_STATUS_SUCCESS) + return ret; + + part.val = v0; + if (uvtype) + *uvtype = part.hub_version; + if (partid) + *partid = part.partition_id; + if (coher) + *coher = part.coherence_id; + if (region) + *region = part.region_size; + if (ssn) + *ssn = v1; + return ret; +} + +int +uv_bios_mq_watchlist_alloc(unsigned long addr, unsigned int mq_size, + unsigned long *intr_mmr_offset) +{ + u64 watchlist; + s64 ret; + + /* + * bios returns watchlist number or negative error number. + */ + ret = (int)uv_bios_call_irqsave(UV_BIOS_WATCHLIST_ALLOC, addr, + mq_size, (u64)intr_mmr_offset, + (u64)&watchlist, 0); + if (ret < BIOS_STATUS_SUCCESS) + return ret; + + return watchlist; +} +EXPORT_SYMBOL_GPL(uv_bios_mq_watchlist_alloc); + +int +uv_bios_mq_watchlist_free(int blade, int watchlist_num) +{ + return (int)uv_bios_call_irqsave(UV_BIOS_WATCHLIST_FREE, + blade, watchlist_num, 0, 0, 0); +} +EXPORT_SYMBOL_GPL(uv_bios_mq_watchlist_free); + +s64 +uv_bios_change_memprotect(u64 paddr, u64 len, enum uv_memprotect perms) +{ + return uv_bios_call_irqsave(UV_BIOS_MEMPROTECT, paddr, len, + perms, 0, 0); +} +EXPORT_SYMBOL_GPL(uv_bios_change_memprotect); + +s64 +uv_bios_reserved_page_pa(u64 buf, u64 *cookie, u64 *addr, u64 *len) +{ + return uv_bios_call_irqsave(UV_BIOS_GET_PARTITION_ADDR, (u64)cookie, + (u64)addr, buf, (u64)len, 0); +} +EXPORT_SYMBOL_GPL(uv_bios_reserved_page_pa); + +s64 uv_bios_freq_base(u64 clock_type, u64 *ticks_per_second) +{ + return uv_bios_call(UV_BIOS_FREQ_BASE, clock_type, + (u64)ticks_per_second, 0, 0, 0); +} + +/* + * uv_bios_set_legacy_vga_target - Set Legacy VGA I/O Target + * @decode: true to enable target, false to disable target + * @domain: PCI domain number + * @bus: PCI bus number + * + * Returns: + * 0: Success + * -EINVAL: Invalid domain or bus number + * -ENOSYS: Capability not available + * -EBUSY: Legacy VGA I/O cannot be retargeted at this time + */ +int uv_bios_set_legacy_vga_target(bool decode, int domain, int bus) +{ + return uv_bios_call(UV_BIOS_SET_LEGACY_VGA_TARGET, + (u64)decode, (u64)domain, (u64)bus, 0, 0); +} + +extern s64 uv_bios_get_master_nasid(u64 size, u64 *master_nasid) +{ + return uv_bios_call(UV_BIOS_EXTRA, 0, UV_BIOS_EXTRA_MASTER_NASID, 0, + size, (u64)master_nasid); +} +EXPORT_SYMBOL_GPL(uv_bios_get_master_nasid); + +extern s64 uv_bios_get_heapsize(u64 nasid, u64 size, u64 *heap_size) +{ + return uv_bios_call(UV_BIOS_EXTRA, nasid, UV_BIOS_EXTRA_GET_HEAPSIZE, + 0, size, (u64)heap_size); +} +EXPORT_SYMBOL_GPL(uv_bios_get_heapsize); + +extern s64 uv_bios_install_heap(u64 nasid, u64 heap_size, u64 *bios_heap) +{ + return uv_bios_call(UV_BIOS_EXTRA, nasid, UV_BIOS_EXTRA_INSTALL_HEAP, + 0, heap_size, (u64)bios_heap); +} +EXPORT_SYMBOL_GPL(uv_bios_install_heap); + +extern s64 uv_bios_obj_count(u64 nasid, u64 size, u64 *objcnt) +{ + return uv_bios_call(UV_BIOS_EXTRA, nasid, UV_BIOS_EXTRA_OBJECT_COUNT, + 0, size, (u64)objcnt); +} +EXPORT_SYMBOL_GPL(uv_bios_obj_count); + +extern s64 uv_bios_enum_objs(u64 nasid, u64 size, u64 *objbuf) +{ + return uv_bios_call(UV_BIOS_EXTRA, nasid, UV_BIOS_EXTRA_ENUM_OBJECTS, + 0, size, (u64)objbuf); +} +EXPORT_SYMBOL_GPL(uv_bios_enum_objs); + +extern s64 uv_bios_enum_ports(u64 nasid, u64 obj_id, u64 size, u64 *portbuf) +{ + return uv_bios_call(UV_BIOS_EXTRA, nasid, UV_BIOS_EXTRA_ENUM_PORTS, + obj_id, size, (u64)portbuf); +} +EXPORT_SYMBOL_GPL(uv_bios_enum_ports); + +extern s64 uv_bios_get_geoinfo(u64 nasid, u64 size, u64 *buf) +{ + return uv_bios_call(UV_BIOS_GET_GEOINFO, nasid, (u64)buf, size, 0, 0); +} +EXPORT_SYMBOL_GPL(uv_bios_get_geoinfo); + +extern s64 uv_bios_get_pci_topology(u64 size, u64 *buf) +{ + return uv_bios_call(UV_BIOS_GET_PCI_TOPOLOGY, (u64)buf, size, 0, 0, 0); +} +EXPORT_SYMBOL_GPL(uv_bios_get_pci_topology); + +unsigned long get_uv_systab_phys(bool msg) +{ + if ((uv_systab_phys == EFI_INVALID_TABLE_ADDR) || + !uv_systab_phys || efi_runtime_disabled()) { + if (msg) + pr_crit("UV: UVsystab: missing\n"); + return 0; + } + return uv_systab_phys; +} + +int uv_bios_init(void) +{ + unsigned long uv_systab_phys_addr; + + uv_systab = NULL; + uv_systab_phys_addr = get_uv_systab_phys(1); + if (!uv_systab_phys_addr) + return -EEXIST; + + uv_systab = ioremap(uv_systab_phys_addr, sizeof(struct uv_systab)); + if (!uv_systab || strncmp(uv_systab->signature, UV_SYSTAB_SIG, 4)) { + pr_err("UV: UVsystab: bad signature!\n"); + iounmap(uv_systab); + return -EINVAL; + } + + /* Starting with UV4 the UV systab size is variable */ + if (uv_systab->revision >= UV_SYSTAB_VERSION_UV4) { + int size = uv_systab->size; + + iounmap(uv_systab); + uv_systab = ioremap(uv_systab_phys_addr, size); + if (!uv_systab) { + pr_err("UV: UVsystab: ioremap(%d) failed!\n", size); + return -EFAULT; + } + } + pr_info("UV: UVsystab: Revision:%x\n", uv_systab->revision); + return 0; +} diff --git a/arch/x86/platform/uv/uv_irq.c b/arch/x86/platform/uv/uv_irq.c new file mode 100644 index 000000000..1a536a187 --- /dev/null +++ b/arch/x86/platform/uv/uv_irq.c @@ -0,0 +1,217 @@ +/* + * This file is subject to the terms and conditions of the GNU General Public + * License. See the file "COPYING" in the main directory of this archive + * for more details. + * + * SGI UV IRQ functions + * + * Copyright (C) 2008 Silicon Graphics, Inc. All rights reserved. + */ + +#include <linux/export.h> +#include <linux/rbtree.h> +#include <linux/slab.h> +#include <linux/irq.h> + +#include <asm/irqdomain.h> +#include <asm/apic.h> +#include <asm/uv/uv_irq.h> +#include <asm/uv/uv_hub.h> + +/* MMR offset and pnode of hub sourcing interrupts for a given irq */ +struct uv_irq_2_mmr_pnode { + unsigned long offset; + int pnode; +}; + +static void uv_program_mmr(struct irq_cfg *cfg, struct uv_irq_2_mmr_pnode *info) +{ + unsigned long mmr_value; + struct uv_IO_APIC_route_entry *entry; + + BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != + sizeof(unsigned long)); + + mmr_value = 0; + entry = (struct uv_IO_APIC_route_entry *)&mmr_value; + entry->vector = cfg->vector; + entry->delivery_mode = apic->delivery_mode; + entry->dest_mode = apic->dest_mode_logical; + entry->polarity = 0; + entry->trigger = 0; + entry->mask = 0; + entry->dest = cfg->dest_apicid; + + uv_write_global_mmr64(info->pnode, info->offset, mmr_value); +} + +static void uv_noop(struct irq_data *data) { } + +static int +uv_set_irq_affinity(struct irq_data *data, const struct cpumask *mask, + bool force) +{ + struct irq_data *parent = data->parent_data; + struct irq_cfg *cfg = irqd_cfg(data); + int ret; + + ret = parent->chip->irq_set_affinity(parent, mask, force); + if (ret >= 0) { + uv_program_mmr(cfg, data->chip_data); + send_cleanup_vector(cfg); + } + + return ret; +} + +static struct irq_chip uv_irq_chip = { + .name = "UV-CORE", + .irq_mask = uv_noop, + .irq_unmask = uv_noop, + .irq_eoi = apic_ack_irq, + .irq_set_affinity = uv_set_irq_affinity, +}; + +static int uv_domain_alloc(struct irq_domain *domain, unsigned int virq, + unsigned int nr_irqs, void *arg) +{ + struct uv_irq_2_mmr_pnode *chip_data; + struct irq_alloc_info *info = arg; + struct irq_data *irq_data = irq_domain_get_irq_data(domain, virq); + int ret; + + if (nr_irqs > 1 || !info || info->type != X86_IRQ_ALLOC_TYPE_UV) + return -EINVAL; + + chip_data = kmalloc_node(sizeof(*chip_data), GFP_KERNEL, + irq_data_get_node(irq_data)); + if (!chip_data) + return -ENOMEM; + + ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg); + if (ret >= 0) { + if (info->uv.limit == UV_AFFINITY_CPU) + irq_set_status_flags(virq, IRQ_NO_BALANCING); + else + irq_set_status_flags(virq, IRQ_MOVE_PCNTXT); + + chip_data->pnode = uv_blade_to_pnode(info->uv.blade); + chip_data->offset = info->uv.offset; + irq_domain_set_info(domain, virq, virq, &uv_irq_chip, chip_data, + handle_percpu_irq, NULL, info->uv.name); + } else { + kfree(chip_data); + } + + return ret; +} + +static void uv_domain_free(struct irq_domain *domain, unsigned int virq, + unsigned int nr_irqs) +{ + struct irq_data *irq_data = irq_domain_get_irq_data(domain, virq); + + BUG_ON(nr_irqs != 1); + kfree(irq_data->chip_data); + irq_clear_status_flags(virq, IRQ_MOVE_PCNTXT); + irq_clear_status_flags(virq, IRQ_NO_BALANCING); + irq_domain_free_irqs_top(domain, virq, nr_irqs); +} + +/* + * Re-target the irq to the specified CPU and enable the specified MMR located + * on the specified blade to allow the sending of MSIs to the specified CPU. + */ +static int uv_domain_activate(struct irq_domain *domain, + struct irq_data *irq_data, bool reserve) +{ + uv_program_mmr(irqd_cfg(irq_data), irq_data->chip_data); + return 0; +} + +/* + * Disable the specified MMR located on the specified blade so that MSIs are + * longer allowed to be sent. + */ +static void uv_domain_deactivate(struct irq_domain *domain, + struct irq_data *irq_data) +{ + unsigned long mmr_value; + struct uv_IO_APIC_route_entry *entry; + + mmr_value = 0; + entry = (struct uv_IO_APIC_route_entry *)&mmr_value; + entry->mask = 1; + uv_program_mmr(irqd_cfg(irq_data), irq_data->chip_data); +} + +static const struct irq_domain_ops uv_domain_ops = { + .alloc = uv_domain_alloc, + .free = uv_domain_free, + .activate = uv_domain_activate, + .deactivate = uv_domain_deactivate, +}; + +static struct irq_domain *uv_get_irq_domain(void) +{ + static struct irq_domain *uv_domain; + static DEFINE_MUTEX(uv_lock); + struct fwnode_handle *fn; + + mutex_lock(&uv_lock); + if (uv_domain) + goto out; + + fn = irq_domain_alloc_named_fwnode("UV-CORE"); + if (!fn) + goto out; + + uv_domain = irq_domain_create_tree(fn, &uv_domain_ops, NULL); + if (uv_domain) + uv_domain->parent = x86_vector_domain; + else + irq_domain_free_fwnode(fn); +out: + mutex_unlock(&uv_lock); + + return uv_domain; +} + +/* + * Set up a mapping of an available irq and vector, and enable the specified + * MMR that defines the MSI that is to be sent to the specified CPU when an + * interrupt is raised. + */ +int uv_setup_irq(char *irq_name, int cpu, int mmr_blade, + unsigned long mmr_offset, int limit) +{ + struct irq_alloc_info info; + struct irq_domain *domain = uv_get_irq_domain(); + + if (!domain) + return -ENOMEM; + + init_irq_alloc_info(&info, cpumask_of(cpu)); + info.type = X86_IRQ_ALLOC_TYPE_UV; + info.uv.limit = limit; + info.uv.blade = mmr_blade; + info.uv.offset = mmr_offset; + info.uv.name = irq_name; + + return irq_domain_alloc_irqs(domain, 1, + uv_blade_to_memory_nid(mmr_blade), &info); +} +EXPORT_SYMBOL_GPL(uv_setup_irq); + +/* + * Tear down a mapping of an irq and vector, and disable the specified MMR that + * defined the MSI that was to be sent to the specified CPU when an interrupt + * was raised. + * + * Set mmr_blade and mmr_offset to what was passed in on uv_setup_irq(). + */ +void uv_teardown_irq(unsigned int irq) +{ + irq_domain_free_irqs(irq, 1); +} +EXPORT_SYMBOL_GPL(uv_teardown_irq); diff --git a/arch/x86/platform/uv/uv_nmi.c b/arch/x86/platform/uv/uv_nmi.c new file mode 100644 index 000000000..a60af0230 --- /dev/null +++ b/arch/x86/platform/uv/uv_nmi.c @@ -0,0 +1,1096 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * SGI NMI support routines + * + * (C) Copyright 2020 Hewlett Packard Enterprise Development LP + * Copyright (C) 2007-2017 Silicon Graphics, Inc. All rights reserved. + * Copyright (c) Mike Travis + */ + +#include <linux/cpu.h> +#include <linux/delay.h> +#include <linux/kdb.h> +#include <linux/kexec.h> +#include <linux/kgdb.h> +#include <linux/moduleparam.h> +#include <linux/nmi.h> +#include <linux/sched.h> +#include <linux/sched/debug.h> +#include <linux/slab.h> +#include <linux/clocksource.h> + +#include <asm/apic.h> +#include <asm/current.h> +#include <asm/kdebug.h> +#include <asm/local64.h> +#include <asm/nmi.h> +#include <asm/reboot.h> +#include <asm/traps.h> +#include <asm/uv/uv.h> +#include <asm/uv/uv_hub.h> +#include <asm/uv/uv_mmrs.h> + +/* + * UV handler for NMI + * + * Handle system-wide NMI events generated by the global 'power nmi' command. + * + * Basic operation is to field the NMI interrupt on each CPU and wait + * until all CPU's have arrived into the nmi handler. If some CPU's do not + * make it into the handler, try and force them in with the IPI(NMI) signal. + * + * We also have to lessen UV Hub MMR accesses as much as possible as this + * disrupts the UV Hub's primary mission of directing NumaLink traffic and + * can cause system problems to occur. + * + * To do this we register our primary NMI notifier on the NMI_UNKNOWN + * chain. This reduces the number of false NMI calls when the perf + * tools are running which generate an enormous number of NMIs per + * second (~4M/s for 1024 CPU threads). Our secondary NMI handler is + * very short as it only checks that if it has been "pinged" with the + * IPI(NMI) signal as mentioned above, and does not read the UV Hub's MMR. + * + */ + +static struct uv_hub_nmi_s **uv_hub_nmi_list; + +DEFINE_PER_CPU(struct uv_cpu_nmi_s, uv_cpu_nmi); + +/* Newer SMM NMI handler, not present in all systems */ +static unsigned long uvh_nmi_mmrx; /* UVH_EVENT_OCCURRED0/1 */ +static unsigned long uvh_nmi_mmrx_clear; /* UVH_EVENT_OCCURRED0/1_ALIAS */ +static int uvh_nmi_mmrx_shift; /* UVH_EVENT_OCCURRED0/1_EXTIO_INT0_SHFT */ +static char *uvh_nmi_mmrx_type; /* "EXTIO_INT0" */ + +/* Non-zero indicates newer SMM NMI handler present */ +static unsigned long uvh_nmi_mmrx_supported; /* UVH_EXTIO_INT0_BROADCAST */ + +/* Indicates to BIOS that we want to use the newer SMM NMI handler */ +static unsigned long uvh_nmi_mmrx_req; /* UVH_BIOS_KERNEL_MMR_ALIAS_2 */ +static int uvh_nmi_mmrx_req_shift; /* 62 */ + +/* UV hubless values */ +#define NMI_CONTROL_PORT 0x70 +#define NMI_DUMMY_PORT 0x71 +#define PAD_OWN_GPP_D_0 0x2c +#define GPI_NMI_STS_GPP_D_0 0x164 +#define GPI_NMI_ENA_GPP_D_0 0x174 +#define STS_GPP_D_0_MASK 0x1 +#define PAD_CFG_DW0_GPP_D_0 0x4c0 +#define GPIROUTNMI (1ul << 17) +#define PCH_PCR_GPIO_1_BASE 0xfdae0000ul +#define PCH_PCR_GPIO_ADDRESS(offset) (int *)((u64)(pch_base) | (u64)(offset)) + +static u64 *pch_base; +static unsigned long nmi_mmr; +static unsigned long nmi_mmr_clear; +static unsigned long nmi_mmr_pending; + +static atomic_t uv_in_nmi; +static atomic_t uv_nmi_cpu = ATOMIC_INIT(-1); +static atomic_t uv_nmi_cpus_in_nmi = ATOMIC_INIT(-1); +static atomic_t uv_nmi_slave_continue; +static cpumask_var_t uv_nmi_cpu_mask; + +static atomic_t uv_nmi_kexec_failed; + +/* Values for uv_nmi_slave_continue */ +#define SLAVE_CLEAR 0 +#define SLAVE_CONTINUE 1 +#define SLAVE_EXIT 2 + +/* + * Default is all stack dumps go to the console and buffer. + * Lower level to send to log buffer only. + */ +static int uv_nmi_loglevel = CONSOLE_LOGLEVEL_DEFAULT; +module_param_named(dump_loglevel, uv_nmi_loglevel, int, 0644); + +/* + * The following values show statistics on how perf events are affecting + * this system. + */ +static int param_get_local64(char *buffer, const struct kernel_param *kp) +{ + return sprintf(buffer, "%lu\n", local64_read((local64_t *)kp->arg)); +} + +static int param_set_local64(const char *val, const struct kernel_param *kp) +{ + /* Clear on any write */ + local64_set((local64_t *)kp->arg, 0); + return 0; +} + +static const struct kernel_param_ops param_ops_local64 = { + .get = param_get_local64, + .set = param_set_local64, +}; +#define param_check_local64(name, p) __param_check(name, p, local64_t) + +static local64_t uv_nmi_count; +module_param_named(nmi_count, uv_nmi_count, local64, 0644); + +static local64_t uv_nmi_misses; +module_param_named(nmi_misses, uv_nmi_misses, local64, 0644); + +static local64_t uv_nmi_ping_count; +module_param_named(ping_count, uv_nmi_ping_count, local64, 0644); + +static local64_t uv_nmi_ping_misses; +module_param_named(ping_misses, uv_nmi_ping_misses, local64, 0644); + +/* + * Following values allow tuning for large systems under heavy loading + */ +static int uv_nmi_initial_delay = 100; +module_param_named(initial_delay, uv_nmi_initial_delay, int, 0644); + +static int uv_nmi_slave_delay = 100; +module_param_named(slave_delay, uv_nmi_slave_delay, int, 0644); + +static int uv_nmi_loop_delay = 100; +module_param_named(loop_delay, uv_nmi_loop_delay, int, 0644); + +static int uv_nmi_trigger_delay = 10000; +module_param_named(trigger_delay, uv_nmi_trigger_delay, int, 0644); + +static int uv_nmi_wait_count = 100; +module_param_named(wait_count, uv_nmi_wait_count, int, 0644); + +static int uv_nmi_retry_count = 500; +module_param_named(retry_count, uv_nmi_retry_count, int, 0644); + +static bool uv_pch_intr_enable = true; +static bool uv_pch_intr_now_enabled; +module_param_named(pch_intr_enable, uv_pch_intr_enable, bool, 0644); + +static bool uv_pch_init_enable = true; +module_param_named(pch_init_enable, uv_pch_init_enable, bool, 0644); + +static int uv_nmi_debug; +module_param_named(debug, uv_nmi_debug, int, 0644); + +#define nmi_debug(fmt, ...) \ + do { \ + if (uv_nmi_debug) \ + pr_info(fmt, ##__VA_ARGS__); \ + } while (0) + +/* Valid NMI Actions */ +#define ACTION_LEN 16 +static struct nmi_action { + char *action; + char *desc; +} valid_acts[] = { + { "kdump", "do kernel crash dump" }, + { "dump", "dump process stack for each cpu" }, + { "ips", "dump Inst Ptr info for each cpu" }, + { "kdb", "enter KDB (needs kgdboc= assignment)" }, + { "kgdb", "enter KGDB (needs gdb target remote)" }, + { "health", "check if CPUs respond to NMI" }, +}; +typedef char action_t[ACTION_LEN]; +static action_t uv_nmi_action = { "dump" }; + +static int param_get_action(char *buffer, const struct kernel_param *kp) +{ + return sprintf(buffer, "%s\n", uv_nmi_action); +} + +static int param_set_action(const char *val, const struct kernel_param *kp) +{ + int i; + int n = ARRAY_SIZE(valid_acts); + char arg[ACTION_LEN], *p; + + /* (remove possible '\n') */ + strncpy(arg, val, ACTION_LEN - 1); + arg[ACTION_LEN - 1] = '\0'; + p = strchr(arg, '\n'); + if (p) + *p = '\0'; + + for (i = 0; i < n; i++) + if (!strcmp(arg, valid_acts[i].action)) + break; + + if (i < n) { + strcpy(uv_nmi_action, arg); + pr_info("UV: New NMI action:%s\n", uv_nmi_action); + return 0; + } + + pr_err("UV: Invalid NMI action:%s, valid actions are:\n", arg); + for (i = 0; i < n; i++) + pr_err("UV: %-8s - %s\n", + valid_acts[i].action, valid_acts[i].desc); + return -EINVAL; +} + +static const struct kernel_param_ops param_ops_action = { + .get = param_get_action, + .set = param_set_action, +}; +#define param_check_action(name, p) __param_check(name, p, action_t) + +module_param_named(action, uv_nmi_action, action, 0644); + +static inline bool uv_nmi_action_is(const char *action) +{ + return (strncmp(uv_nmi_action, action, strlen(action)) == 0); +} + +/* Setup which NMI support is present in system */ +static void uv_nmi_setup_mmrs(void) +{ + bool new_nmi_method_only = false; + + /* First determine arch specific MMRs to handshake with BIOS */ + if (UVH_EVENT_OCCURRED0_EXTIO_INT0_MASK) { /* UV2,3,4 setup */ + uvh_nmi_mmrx = UVH_EVENT_OCCURRED0; + uvh_nmi_mmrx_clear = UVH_EVENT_OCCURRED0_ALIAS; + uvh_nmi_mmrx_shift = UVH_EVENT_OCCURRED0_EXTIO_INT0_SHFT; + uvh_nmi_mmrx_type = "OCRD0-EXTIO_INT0"; + + uvh_nmi_mmrx_supported = UVH_EXTIO_INT0_BROADCAST; + uvh_nmi_mmrx_req = UVH_BIOS_KERNEL_MMR_ALIAS_2; + uvh_nmi_mmrx_req_shift = 62; + + } else if (UVH_EVENT_OCCURRED1_EXTIO_INT0_MASK) { /* UV5+ setup */ + uvh_nmi_mmrx = UVH_EVENT_OCCURRED1; + uvh_nmi_mmrx_clear = UVH_EVENT_OCCURRED1_ALIAS; + uvh_nmi_mmrx_shift = UVH_EVENT_OCCURRED1_EXTIO_INT0_SHFT; + uvh_nmi_mmrx_type = "OCRD1-EXTIO_INT0"; + + new_nmi_method_only = true; /* Newer nmi always valid on UV5+ */ + uvh_nmi_mmrx_req = 0; /* no request bit to clear */ + + } else { + pr_err("UV:%s:NMI support not available on this system\n", __func__); + return; + } + + /* Then find out if new NMI is supported */ + if (new_nmi_method_only || uv_read_local_mmr(uvh_nmi_mmrx_supported)) { + if (uvh_nmi_mmrx_req) + uv_write_local_mmr(uvh_nmi_mmrx_req, + 1UL << uvh_nmi_mmrx_req_shift); + nmi_mmr = uvh_nmi_mmrx; + nmi_mmr_clear = uvh_nmi_mmrx_clear; + nmi_mmr_pending = 1UL << uvh_nmi_mmrx_shift; + pr_info("UV: SMI NMI support: %s\n", uvh_nmi_mmrx_type); + } else { + nmi_mmr = UVH_NMI_MMR; + nmi_mmr_clear = UVH_NMI_MMR_CLEAR; + nmi_mmr_pending = 1UL << UVH_NMI_MMR_SHIFT; + pr_info("UV: SMI NMI support: %s\n", UVH_NMI_MMR_TYPE); + } +} + +/* Read NMI MMR and check if NMI flag was set by BMC. */ +static inline int uv_nmi_test_mmr(struct uv_hub_nmi_s *hub_nmi) +{ + hub_nmi->nmi_value = uv_read_local_mmr(nmi_mmr); + atomic_inc(&hub_nmi->read_mmr_count); + return !!(hub_nmi->nmi_value & nmi_mmr_pending); +} + +static inline void uv_local_mmr_clear_nmi(void) +{ + uv_write_local_mmr(nmi_mmr_clear, nmi_mmr_pending); +} + +/* + * UV hubless NMI handler functions + */ +static inline void uv_reassert_nmi(void) +{ + /* (from arch/x86/include/asm/mach_traps.h) */ + outb(0x8f, NMI_CONTROL_PORT); + inb(NMI_DUMMY_PORT); /* dummy read */ + outb(0x0f, NMI_CONTROL_PORT); + inb(NMI_DUMMY_PORT); /* dummy read */ +} + +static void uv_init_hubless_pch_io(int offset, int mask, int data) +{ + int *addr = PCH_PCR_GPIO_ADDRESS(offset); + int readd = readl(addr); + + if (mask) { /* OR in new data */ + int writed = (readd & ~mask) | data; + + nmi_debug("UV:PCH: %p = %x & %x | %x (%x)\n", + addr, readd, ~mask, data, writed); + writel(writed, addr); + } else if (readd & data) { /* clear status bit */ + nmi_debug("UV:PCH: %p = %x\n", addr, data); + writel(data, addr); + } + + (void)readl(addr); /* flush write data */ +} + +static void uv_nmi_setup_hubless_intr(void) +{ + uv_pch_intr_now_enabled = uv_pch_intr_enable; + + uv_init_hubless_pch_io( + PAD_CFG_DW0_GPP_D_0, GPIROUTNMI, + uv_pch_intr_now_enabled ? GPIROUTNMI : 0); + + nmi_debug("UV:NMI: GPP_D_0 interrupt %s\n", + uv_pch_intr_now_enabled ? "enabled" : "disabled"); +} + +static struct init_nmi { + unsigned int offset; + unsigned int mask; + unsigned int data; +} init_nmi[] = { + { /* HOSTSW_OWN_GPP_D_0 */ + .offset = 0x84, + .mask = 0x1, + .data = 0x0, /* ACPI Mode */ + }, + +/* Clear status: */ + { /* GPI_INT_STS_GPP_D_0 */ + .offset = 0x104, + .mask = 0x0, + .data = 0x1, /* Clear Status */ + }, + { /* GPI_GPE_STS_GPP_D_0 */ + .offset = 0x124, + .mask = 0x0, + .data = 0x1, /* Clear Status */ + }, + { /* GPI_SMI_STS_GPP_D_0 */ + .offset = 0x144, + .mask = 0x0, + .data = 0x1, /* Clear Status */ + }, + { /* GPI_NMI_STS_GPP_D_0 */ + .offset = 0x164, + .mask = 0x0, + .data = 0x1, /* Clear Status */ + }, + +/* Disable interrupts: */ + { /* GPI_INT_EN_GPP_D_0 */ + .offset = 0x114, + .mask = 0x1, + .data = 0x0, /* Disable interrupt generation */ + }, + { /* GPI_GPE_EN_GPP_D_0 */ + .offset = 0x134, + .mask = 0x1, + .data = 0x0, /* Disable interrupt generation */ + }, + { /* GPI_SMI_EN_GPP_D_0 */ + .offset = 0x154, + .mask = 0x1, + .data = 0x0, /* Disable interrupt generation */ + }, + { /* GPI_NMI_EN_GPP_D_0 */ + .offset = 0x174, + .mask = 0x1, + .data = 0x0, /* Disable interrupt generation */ + }, + +/* Setup GPP_D_0 Pad Config: */ + { /* PAD_CFG_DW0_GPP_D_0 */ + .offset = 0x4c0, + .mask = 0xffffffff, + .data = 0x82020100, +/* + * 31:30 Pad Reset Config (PADRSTCFG): = 2h # PLTRST# (default) + * + * 29 RX Pad State Select (RXPADSTSEL): = 0 # Raw RX pad state directly + * from RX buffer (default) + * + * 28 RX Raw Override to '1' (RXRAW1): = 0 # No Override + * + * 26:25 RX Level/Edge Configuration (RXEVCFG): + * = 0h # Level + * = 1h # Edge + * + * 23 RX Invert (RXINV): = 0 # No Inversion (signal active high) + * + * 20 GPIO Input Route IOxAPIC (GPIROUTIOXAPIC): + * = 0 # Routing does not cause peripheral IRQ... + * # (we want an NMI not an IRQ) + * + * 19 GPIO Input Route SCI (GPIROUTSCI): = 0 # Routing does not cause SCI. + * 18 GPIO Input Route SMI (GPIROUTSMI): = 0 # Routing does not cause SMI. + * 17 GPIO Input Route NMI (GPIROUTNMI): = 1 # Routing can cause NMI. + * + * 11:10 Pad Mode (PMODE1/0): = 0h = GPIO control the Pad. + * 9 GPIO RX Disable (GPIORXDIS): + * = 0 # Enable the input buffer (active low enable) + * + * 8 GPIO TX Disable (GPIOTXDIS): + * = 1 # Disable the output buffer; i.e. Hi-Z + * + * 1 GPIO RX State (GPIORXSTATE): This is the current internal RX pad state.. + * 0 GPIO TX State (GPIOTXSTATE): + * = 0 # (Leave at default) + */ + }, + +/* Pad Config DW1 */ + { /* PAD_CFG_DW1_GPP_D_0 */ + .offset = 0x4c4, + .mask = 0x3c00, + .data = 0, /* Termination = none (default) */ + }, +}; + +static void uv_init_hubless_pch_d0(void) +{ + int i, read; + + read = *PCH_PCR_GPIO_ADDRESS(PAD_OWN_GPP_D_0); + if (read != 0) { + pr_info("UV: Hubless NMI already configured\n"); + return; + } + + nmi_debug("UV: Initializing UV Hubless NMI on PCH\n"); + for (i = 0; i < ARRAY_SIZE(init_nmi); i++) { + uv_init_hubless_pch_io(init_nmi[i].offset, + init_nmi[i].mask, + init_nmi[i].data); + } +} + +static int uv_nmi_test_hubless(struct uv_hub_nmi_s *hub_nmi) +{ + int *pstat = PCH_PCR_GPIO_ADDRESS(GPI_NMI_STS_GPP_D_0); + int status = *pstat; + + hub_nmi->nmi_value = status; + atomic_inc(&hub_nmi->read_mmr_count); + + if (!(status & STS_GPP_D_0_MASK)) /* Not a UV external NMI */ + return 0; + + *pstat = STS_GPP_D_0_MASK; /* Is a UV NMI: clear GPP_D_0 status */ + (void)*pstat; /* Flush write */ + + return 1; +} + +static int uv_test_nmi(struct uv_hub_nmi_s *hub_nmi) +{ + if (hub_nmi->hub_present) + return uv_nmi_test_mmr(hub_nmi); + + if (hub_nmi->pch_owner) /* Only PCH owner can check status */ + return uv_nmi_test_hubless(hub_nmi); + + return -1; +} + +/* + * If first CPU in on this hub, set hub_nmi "in_nmi" and "owner" values and + * return true. If first CPU in on the system, set global "in_nmi" flag. + */ +static int uv_set_in_nmi(int cpu, struct uv_hub_nmi_s *hub_nmi) +{ + int first = atomic_add_unless(&hub_nmi->in_nmi, 1, 1); + + if (first) { + atomic_set(&hub_nmi->cpu_owner, cpu); + if (atomic_add_unless(&uv_in_nmi, 1, 1)) + atomic_set(&uv_nmi_cpu, cpu); + + atomic_inc(&hub_nmi->nmi_count); + } + return first; +} + +/* Check if this is a system NMI event */ +static int uv_check_nmi(struct uv_hub_nmi_s *hub_nmi) +{ + int cpu = smp_processor_id(); + int nmi = 0; + int nmi_detected = 0; + + local64_inc(&uv_nmi_count); + this_cpu_inc(uv_cpu_nmi.queries); + + do { + nmi = atomic_read(&hub_nmi->in_nmi); + if (nmi) + break; + + if (raw_spin_trylock(&hub_nmi->nmi_lock)) { + nmi_detected = uv_test_nmi(hub_nmi); + + /* Check flag for UV external NMI */ + if (nmi_detected > 0) { + uv_set_in_nmi(cpu, hub_nmi); + nmi = 1; + break; + } + + /* A non-PCH node in a hubless system waits for NMI */ + else if (nmi_detected < 0) + goto slave_wait; + + /* MMR/PCH NMI flag is clear */ + raw_spin_unlock(&hub_nmi->nmi_lock); + + } else { + + /* Wait a moment for the HUB NMI locker to set flag */ +slave_wait: cpu_relax(); + udelay(uv_nmi_slave_delay); + + /* Re-check hub in_nmi flag */ + nmi = atomic_read(&hub_nmi->in_nmi); + if (nmi) + break; + } + + /* + * Check if this BMC missed setting the MMR NMI flag (or) + * UV hubless system where only PCH owner can check flag + */ + if (!nmi) { + nmi = atomic_read(&uv_in_nmi); + if (nmi) + uv_set_in_nmi(cpu, hub_nmi); + } + + /* If we're holding the hub lock, release it now */ + if (nmi_detected < 0) + raw_spin_unlock(&hub_nmi->nmi_lock); + + } while (0); + + if (!nmi) + local64_inc(&uv_nmi_misses); + + return nmi; +} + +/* Need to reset the NMI MMR register, but only once per hub. */ +static inline void uv_clear_nmi(int cpu) +{ + struct uv_hub_nmi_s *hub_nmi = uv_hub_nmi; + + if (cpu == atomic_read(&hub_nmi->cpu_owner)) { + atomic_set(&hub_nmi->cpu_owner, -1); + atomic_set(&hub_nmi->in_nmi, 0); + if (hub_nmi->hub_present) + uv_local_mmr_clear_nmi(); + else + uv_reassert_nmi(); + raw_spin_unlock(&hub_nmi->nmi_lock); + } +} + +/* Ping non-responding CPU's attempting to force them into the NMI handler */ +static void uv_nmi_nr_cpus_ping(void) +{ + int cpu; + + for_each_cpu(cpu, uv_nmi_cpu_mask) + uv_cpu_nmi_per(cpu).pinging = 1; + + apic->send_IPI_mask(uv_nmi_cpu_mask, APIC_DM_NMI); +} + +/* Clean up flags for CPU's that ignored both NMI and ping */ +static void uv_nmi_cleanup_mask(void) +{ + int cpu; + + for_each_cpu(cpu, uv_nmi_cpu_mask) { + uv_cpu_nmi_per(cpu).pinging = 0; + uv_cpu_nmi_per(cpu).state = UV_NMI_STATE_OUT; + cpumask_clear_cpu(cpu, uv_nmi_cpu_mask); + } +} + +/* Loop waiting as CPU's enter NMI handler */ +static int uv_nmi_wait_cpus(int first) +{ + int i, j, k, n = num_online_cpus(); + int last_k = 0, waiting = 0; + int cpu = smp_processor_id(); + + if (first) { + cpumask_copy(uv_nmi_cpu_mask, cpu_online_mask); + k = 0; + } else { + k = n - cpumask_weight(uv_nmi_cpu_mask); + } + + /* PCH NMI causes only one CPU to respond */ + if (first && uv_pch_intr_now_enabled) { + cpumask_clear_cpu(cpu, uv_nmi_cpu_mask); + return n - k - 1; + } + + udelay(uv_nmi_initial_delay); + for (i = 0; i < uv_nmi_retry_count; i++) { + int loop_delay = uv_nmi_loop_delay; + + for_each_cpu(j, uv_nmi_cpu_mask) { + if (uv_cpu_nmi_per(j).state) { + cpumask_clear_cpu(j, uv_nmi_cpu_mask); + if (++k >= n) + break; + } + } + if (k >= n) { /* all in? */ + k = n; + break; + } + if (last_k != k) { /* abort if no new CPU's coming in */ + last_k = k; + waiting = 0; + } else if (++waiting > uv_nmi_wait_count) + break; + + /* Extend delay if waiting only for CPU 0: */ + if (waiting && (n - k) == 1 && + cpumask_test_cpu(0, uv_nmi_cpu_mask)) + loop_delay *= 100; + + udelay(loop_delay); + } + atomic_set(&uv_nmi_cpus_in_nmi, k); + return n - k; +} + +/* Wait until all slave CPU's have entered UV NMI handler */ +static void uv_nmi_wait(int master) +{ + /* Indicate this CPU is in: */ + this_cpu_write(uv_cpu_nmi.state, UV_NMI_STATE_IN); + + /* If not the first CPU in (the master), then we are a slave CPU */ + if (!master) + return; + + do { + /* Wait for all other CPU's to gather here */ + if (!uv_nmi_wait_cpus(1)) + break; + + /* If not all made it in, send IPI NMI to them */ + pr_alert("UV: Sending NMI IPI to %d CPUs: %*pbl\n", + cpumask_weight(uv_nmi_cpu_mask), + cpumask_pr_args(uv_nmi_cpu_mask)); + + uv_nmi_nr_cpus_ping(); + + /* If all CPU's are in, then done */ + if (!uv_nmi_wait_cpus(0)) + break; + + pr_alert("UV: %d CPUs not in NMI loop: %*pbl\n", + cpumask_weight(uv_nmi_cpu_mask), + cpumask_pr_args(uv_nmi_cpu_mask)); + } while (0); + + pr_alert("UV: %d of %d CPUs in NMI\n", + atomic_read(&uv_nmi_cpus_in_nmi), num_online_cpus()); +} + +/* Dump Instruction Pointer header */ +static void uv_nmi_dump_cpu_ip_hdr(void) +{ + pr_info("\nUV: %4s %6s %-32s %s (Note: PID 0 not listed)\n", + "CPU", "PID", "COMMAND", "IP"); +} + +/* Dump Instruction Pointer info */ +static void uv_nmi_dump_cpu_ip(int cpu, struct pt_regs *regs) +{ + pr_info("UV: %4d %6d %-32.32s %pS", + cpu, current->pid, current->comm, (void *)regs->ip); +} + +/* + * Dump this CPU's state. If action was set to "kdump" and the crash_kexec + * failed, then we provide "dump" as an alternate action. Action "dump" now + * also includes the show "ips" (instruction pointers) action whereas the + * action "ips" only displays instruction pointers for the non-idle CPU's. + * This is an abbreviated form of the "ps" command. + */ +static void uv_nmi_dump_state_cpu(int cpu, struct pt_regs *regs) +{ + const char *dots = " ................................. "; + + if (cpu == 0) + uv_nmi_dump_cpu_ip_hdr(); + + if (current->pid != 0 || !uv_nmi_action_is("ips")) + uv_nmi_dump_cpu_ip(cpu, regs); + + if (uv_nmi_action_is("dump")) { + pr_info("UV:%sNMI process trace for CPU %d\n", dots, cpu); + show_regs(regs); + } + + this_cpu_write(uv_cpu_nmi.state, UV_NMI_STATE_DUMP_DONE); +} + +/* Trigger a slave CPU to dump it's state */ +static void uv_nmi_trigger_dump(int cpu) +{ + int retry = uv_nmi_trigger_delay; + + if (uv_cpu_nmi_per(cpu).state != UV_NMI_STATE_IN) + return; + + uv_cpu_nmi_per(cpu).state = UV_NMI_STATE_DUMP; + do { + cpu_relax(); + udelay(10); + if (uv_cpu_nmi_per(cpu).state + != UV_NMI_STATE_DUMP) + return; + } while (--retry > 0); + + pr_crit("UV: CPU %d stuck in process dump function\n", cpu); + uv_cpu_nmi_per(cpu).state = UV_NMI_STATE_DUMP_DONE; +} + +/* Wait until all CPU's ready to exit */ +static void uv_nmi_sync_exit(int master) +{ + atomic_dec(&uv_nmi_cpus_in_nmi); + if (master) { + while (atomic_read(&uv_nmi_cpus_in_nmi) > 0) + cpu_relax(); + atomic_set(&uv_nmi_slave_continue, SLAVE_CLEAR); + } else { + while (atomic_read(&uv_nmi_slave_continue)) + cpu_relax(); + } +} + +/* Current "health" check is to check which CPU's are responsive */ +static void uv_nmi_action_health(int cpu, struct pt_regs *regs, int master) +{ + if (master) { + int in = atomic_read(&uv_nmi_cpus_in_nmi); + int out = num_online_cpus() - in; + + pr_alert("UV: NMI CPU health check (non-responding:%d)\n", out); + atomic_set(&uv_nmi_slave_continue, SLAVE_EXIT); + } else { + while (!atomic_read(&uv_nmi_slave_continue)) + cpu_relax(); + } + uv_nmi_sync_exit(master); +} + +/* Walk through CPU list and dump state of each */ +static void uv_nmi_dump_state(int cpu, struct pt_regs *regs, int master) +{ + if (master) { + int tcpu; + int ignored = 0; + int saved_console_loglevel = console_loglevel; + + pr_alert("UV: tracing %s for %d CPUs from CPU %d\n", + uv_nmi_action_is("ips") ? "IPs" : "processes", + atomic_read(&uv_nmi_cpus_in_nmi), cpu); + + console_loglevel = uv_nmi_loglevel; + atomic_set(&uv_nmi_slave_continue, SLAVE_EXIT); + for_each_online_cpu(tcpu) { + if (cpumask_test_cpu(tcpu, uv_nmi_cpu_mask)) + ignored++; + else if (tcpu == cpu) + uv_nmi_dump_state_cpu(tcpu, regs); + else + uv_nmi_trigger_dump(tcpu); + } + if (ignored) + pr_alert("UV: %d CPUs ignored NMI\n", ignored); + + console_loglevel = saved_console_loglevel; + pr_alert("UV: process trace complete\n"); + } else { + while (!atomic_read(&uv_nmi_slave_continue)) + cpu_relax(); + while (this_cpu_read(uv_cpu_nmi.state) != UV_NMI_STATE_DUMP) + cpu_relax(); + uv_nmi_dump_state_cpu(cpu, regs); + } + uv_nmi_sync_exit(master); +} + +static void uv_nmi_touch_watchdogs(void) +{ + touch_softlockup_watchdog_sync(); + clocksource_touch_watchdog(); + rcu_cpu_stall_reset(); + touch_nmi_watchdog(); +} + +static void uv_nmi_kdump(int cpu, int main, struct pt_regs *regs) +{ + /* Check if kdump kernel loaded for both main and secondary CPUs */ + if (!kexec_crash_image) { + if (main) + pr_err("UV: NMI error: kdump kernel not loaded\n"); + return; + } + + /* Call crash to dump system state */ + if (main) { + pr_emerg("UV: NMI executing crash_kexec on CPU%d\n", cpu); + crash_kexec(regs); + + pr_emerg("UV: crash_kexec unexpectedly returned\n"); + atomic_set(&uv_nmi_kexec_failed, 1); + + } else { /* secondary */ + + /* If kdump kernel fails, secondaries will exit this loop */ + while (atomic_read(&uv_nmi_kexec_failed) == 0) { + + /* Once shootdown cpus starts, they do not return */ + run_crash_ipi_callback(regs); + + mdelay(10); + } + } +} + +#ifdef CONFIG_KGDB +#ifdef CONFIG_KGDB_KDB +static inline int uv_nmi_kdb_reason(void) +{ + return KDB_REASON_SYSTEM_NMI; +} +#else /* !CONFIG_KGDB_KDB */ +static inline int uv_nmi_kdb_reason(void) +{ + /* Ensure user is expecting to attach gdb remote */ + if (uv_nmi_action_is("kgdb")) + return 0; + + pr_err("UV: NMI error: KDB is not enabled in this kernel\n"); + return -1; +} +#endif /* CONFIG_KGDB_KDB */ + +/* + * Call KGDB/KDB from NMI handler + * + * Note that if both KGDB and KDB are configured, then the action of 'kgdb' or + * 'kdb' has no affect on which is used. See the KGDB documentation for further + * information. + */ +static void uv_call_kgdb_kdb(int cpu, struct pt_regs *regs, int master) +{ + if (master) { + int reason = uv_nmi_kdb_reason(); + int ret; + + if (reason < 0) + return; + + /* Call KGDB NMI handler as MASTER */ + ret = kgdb_nmicallin(cpu, X86_TRAP_NMI, regs, reason, + &uv_nmi_slave_continue); + if (ret) { + pr_alert("KGDB returned error, is kgdboc set?\n"); + atomic_set(&uv_nmi_slave_continue, SLAVE_EXIT); + } + } else { + /* Wait for KGDB signal that it's ready for slaves to enter */ + int sig; + + do { + cpu_relax(); + sig = atomic_read(&uv_nmi_slave_continue); + } while (!sig); + + /* Call KGDB as slave */ + if (sig == SLAVE_CONTINUE) + kgdb_nmicallback(cpu, regs); + } + uv_nmi_sync_exit(master); +} + +#else /* !CONFIG_KGDB */ +static inline void uv_call_kgdb_kdb(int cpu, struct pt_regs *regs, int master) +{ + pr_err("UV: NMI error: KGDB is not enabled in this kernel\n"); +} +#endif /* !CONFIG_KGDB */ + +/* + * UV NMI handler + */ +static int uv_handle_nmi(unsigned int reason, struct pt_regs *regs) +{ + struct uv_hub_nmi_s *hub_nmi = uv_hub_nmi; + int cpu = smp_processor_id(); + int master = 0; + unsigned long flags; + + local_irq_save(flags); + + /* If not a UV System NMI, ignore */ + if (!this_cpu_read(uv_cpu_nmi.pinging) && !uv_check_nmi(hub_nmi)) { + local_irq_restore(flags); + return NMI_DONE; + } + + /* Indicate we are the first CPU into the NMI handler */ + master = (atomic_read(&uv_nmi_cpu) == cpu); + + /* If NMI action is "kdump", then attempt to do it */ + if (uv_nmi_action_is("kdump")) { + uv_nmi_kdump(cpu, master, regs); + + /* Unexpected return, revert action to "dump" */ + if (master) + strncpy(uv_nmi_action, "dump", strlen(uv_nmi_action)); + } + + /* Pause as all CPU's enter the NMI handler */ + uv_nmi_wait(master); + + /* Process actions other than "kdump": */ + if (uv_nmi_action_is("health")) { + uv_nmi_action_health(cpu, regs, master); + } else if (uv_nmi_action_is("ips") || uv_nmi_action_is("dump")) { + uv_nmi_dump_state(cpu, regs, master); + } else if (uv_nmi_action_is("kdb") || uv_nmi_action_is("kgdb")) { + uv_call_kgdb_kdb(cpu, regs, master); + } else { + if (master) + pr_alert("UV: unknown NMI action: %s\n", uv_nmi_action); + uv_nmi_sync_exit(master); + } + + /* Clear per_cpu "in_nmi" flag */ + this_cpu_write(uv_cpu_nmi.state, UV_NMI_STATE_OUT); + + /* Clear MMR NMI flag on each hub */ + uv_clear_nmi(cpu); + + /* Clear global flags */ + if (master) { + if (!cpumask_empty(uv_nmi_cpu_mask)) + uv_nmi_cleanup_mask(); + atomic_set(&uv_nmi_cpus_in_nmi, -1); + atomic_set(&uv_nmi_cpu, -1); + atomic_set(&uv_in_nmi, 0); + atomic_set(&uv_nmi_kexec_failed, 0); + atomic_set(&uv_nmi_slave_continue, SLAVE_CLEAR); + } + + uv_nmi_touch_watchdogs(); + local_irq_restore(flags); + + return NMI_HANDLED; +} + +/* + * NMI handler for pulling in CPU's when perf events are grabbing our NMI + */ +static int uv_handle_nmi_ping(unsigned int reason, struct pt_regs *regs) +{ + int ret; + + this_cpu_inc(uv_cpu_nmi.queries); + if (!this_cpu_read(uv_cpu_nmi.pinging)) { + local64_inc(&uv_nmi_ping_misses); + return NMI_DONE; + } + + this_cpu_inc(uv_cpu_nmi.pings); + local64_inc(&uv_nmi_ping_count); + ret = uv_handle_nmi(reason, regs); + this_cpu_write(uv_cpu_nmi.pinging, 0); + return ret; +} + +static void uv_register_nmi_notifier(void) +{ + if (register_nmi_handler(NMI_UNKNOWN, uv_handle_nmi, 0, "uv")) + pr_warn("UV: NMI handler failed to register\n"); + + if (register_nmi_handler(NMI_LOCAL, uv_handle_nmi_ping, 0, "uvping")) + pr_warn("UV: PING NMI handler failed to register\n"); +} + +void uv_nmi_init(void) +{ + unsigned int value; + + /* + * Unmask NMI on all CPU's + */ + value = apic_read(APIC_LVT1) | APIC_DM_NMI; + value &= ~APIC_LVT_MASKED; + apic_write(APIC_LVT1, value); +} + +/* Setup HUB NMI info */ +static void __init uv_nmi_setup_common(bool hubbed) +{ + int size = sizeof(void *) * (1 << NODES_SHIFT); + int cpu; + + uv_hub_nmi_list = kzalloc(size, GFP_KERNEL); + nmi_debug("UV: NMI hub list @ 0x%p (%d)\n", uv_hub_nmi_list, size); + BUG_ON(!uv_hub_nmi_list); + size = sizeof(struct uv_hub_nmi_s); + for_each_present_cpu(cpu) { + int nid = cpu_to_node(cpu); + if (uv_hub_nmi_list[nid] == NULL) { + uv_hub_nmi_list[nid] = kzalloc_node(size, + GFP_KERNEL, nid); + BUG_ON(!uv_hub_nmi_list[nid]); + raw_spin_lock_init(&(uv_hub_nmi_list[nid]->nmi_lock)); + atomic_set(&uv_hub_nmi_list[nid]->cpu_owner, -1); + uv_hub_nmi_list[nid]->hub_present = hubbed; + uv_hub_nmi_list[nid]->pch_owner = (nid == 0); + } + uv_hub_nmi_per(cpu) = uv_hub_nmi_list[nid]; + } + BUG_ON(!alloc_cpumask_var(&uv_nmi_cpu_mask, GFP_KERNEL)); +} + +/* Setup for UV Hub systems */ +void __init uv_nmi_setup(void) +{ + uv_nmi_setup_mmrs(); + uv_nmi_setup_common(true); + uv_register_nmi_notifier(); + pr_info("UV: Hub NMI enabled\n"); +} + +/* Setup for UV Hubless systems */ +void __init uv_nmi_setup_hubless(void) +{ + uv_nmi_setup_common(false); + pch_base = xlate_dev_mem_ptr(PCH_PCR_GPIO_1_BASE); + nmi_debug("UV: PCH base:%p from 0x%lx, GPP_D_0\n", + pch_base, PCH_PCR_GPIO_1_BASE); + if (uv_pch_init_enable) + uv_init_hubless_pch_d0(); + uv_init_hubless_pch_io(GPI_NMI_ENA_GPP_D_0, + STS_GPP_D_0_MASK, STS_GPP_D_0_MASK); + uv_nmi_setup_hubless_intr(); + /* Ensure NMI enabled in Processor Interface Reg: */ + uv_reassert_nmi(); + uv_register_nmi_notifier(); + pr_info("UV: PCH NMI enabled\n"); +} diff --git a/arch/x86/platform/uv/uv_time.c b/arch/x86/platform/uv/uv_time.c new file mode 100644 index 000000000..54663f3e0 --- /dev/null +++ b/arch/x86/platform/uv/uv_time.c @@ -0,0 +1,393 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * SGI RTC clock/timer routines. + * + * (C) Copyright 2020 Hewlett Packard Enterprise Development LP + * Copyright (c) 2009-2013 Silicon Graphics, Inc. All Rights Reserved. + * Copyright (c) Dimitri Sivanich + */ +#include <linux/clockchips.h> +#include <linux/slab.h> + +#include <asm/uv/uv_mmrs.h> +#include <asm/uv/uv_hub.h> +#include <asm/uv/bios.h> +#include <asm/uv/uv.h> +#include <asm/apic.h> +#include <asm/cpu.h> + +#define RTC_NAME "sgi_rtc" + +static u64 uv_read_rtc(struct clocksource *cs); +static int uv_rtc_next_event(unsigned long, struct clock_event_device *); +static int uv_rtc_shutdown(struct clock_event_device *evt); + +static struct clocksource clocksource_uv = { + .name = RTC_NAME, + .rating = 299, + .read = uv_read_rtc, + .mask = (u64)UVH_RTC_REAL_TIME_CLOCK_MASK, + .flags = CLOCK_SOURCE_IS_CONTINUOUS, +}; + +static struct clock_event_device clock_event_device_uv = { + .name = RTC_NAME, + .features = CLOCK_EVT_FEAT_ONESHOT, + .shift = 20, + .rating = 400, + .irq = -1, + .set_next_event = uv_rtc_next_event, + .set_state_shutdown = uv_rtc_shutdown, + .event_handler = NULL, +}; + +static DEFINE_PER_CPU(struct clock_event_device, cpu_ced); + +/* There is one of these allocated per node */ +struct uv_rtc_timer_head { + spinlock_t lock; + /* next cpu waiting for timer, local node relative: */ + int next_cpu; + /* number of cpus on this node: */ + int ncpus; + struct { + int lcpu; /* systemwide logical cpu number */ + u64 expires; /* next timer expiration for this cpu */ + } cpu[]; +}; + +/* + * Access to uv_rtc_timer_head via blade id. + */ +static struct uv_rtc_timer_head **blade_info __read_mostly; + +static int uv_rtc_evt_enable; + +/* + * Hardware interface routines + */ + +/* Send IPIs to another node */ +static void uv_rtc_send_IPI(int cpu) +{ + unsigned long apicid, val; + int pnode; + + apicid = cpu_physical_id(cpu); + pnode = uv_apicid_to_pnode(apicid); + val = (1UL << UVH_IPI_INT_SEND_SHFT) | + (apicid << UVH_IPI_INT_APIC_ID_SHFT) | + (X86_PLATFORM_IPI_VECTOR << UVH_IPI_INT_VECTOR_SHFT); + + uv_write_global_mmr64(pnode, UVH_IPI_INT, val); +} + +/* Check for an RTC interrupt pending */ +static int uv_intr_pending(int pnode) +{ + return uv_read_global_mmr64(pnode, UVH_EVENT_OCCURRED2) & + UVH_EVENT_OCCURRED2_RTC_1_MASK; +} + +/* Setup interrupt and return non-zero if early expiration occurred. */ +static int uv_setup_intr(int cpu, u64 expires) +{ + u64 val; + unsigned long apicid = cpu_physical_id(cpu); + int pnode = uv_cpu_to_pnode(cpu); + + uv_write_global_mmr64(pnode, UVH_RTC1_INT_CONFIG, + UVH_RTC1_INT_CONFIG_M_MASK); + uv_write_global_mmr64(pnode, UVH_INT_CMPB, -1L); + + uv_write_global_mmr64(pnode, UVH_EVENT_OCCURRED2_ALIAS, + UVH_EVENT_OCCURRED2_RTC_1_MASK); + + val = (X86_PLATFORM_IPI_VECTOR << UVH_RTC1_INT_CONFIG_VECTOR_SHFT) | + ((u64)apicid << UVH_RTC1_INT_CONFIG_APIC_ID_SHFT); + + /* Set configuration */ + uv_write_global_mmr64(pnode, UVH_RTC1_INT_CONFIG, val); + /* Initialize comparator value */ + uv_write_global_mmr64(pnode, UVH_INT_CMPB, expires); + + if (uv_read_rtc(NULL) <= expires) + return 0; + + return !uv_intr_pending(pnode); +} + +/* + * Per-cpu timer tracking routines + */ + +static __init void uv_rtc_deallocate_timers(void) +{ + int bid; + + for_each_possible_blade(bid) { + kfree(blade_info[bid]); + } + kfree(blade_info); +} + +/* Allocate per-node list of cpu timer expiration times. */ +static __init int uv_rtc_allocate_timers(void) +{ + int cpu; + + blade_info = kcalloc(uv_possible_blades, sizeof(void *), GFP_KERNEL); + if (!blade_info) + return -ENOMEM; + + for_each_present_cpu(cpu) { + int nid = cpu_to_node(cpu); + int bid = uv_cpu_to_blade_id(cpu); + int bcpu = uv_cpu_blade_processor_id(cpu); + struct uv_rtc_timer_head *head = blade_info[bid]; + + if (!head) { + head = kmalloc_node(struct_size(head, cpu, + uv_blade_nr_possible_cpus(bid)), + GFP_KERNEL, nid); + if (!head) { + uv_rtc_deallocate_timers(); + return -ENOMEM; + } + spin_lock_init(&head->lock); + head->ncpus = uv_blade_nr_possible_cpus(bid); + head->next_cpu = -1; + blade_info[bid] = head; + } + + head->cpu[bcpu].lcpu = cpu; + head->cpu[bcpu].expires = ULLONG_MAX; + } + + return 0; +} + +/* Find and set the next expiring timer. */ +static void uv_rtc_find_next_timer(struct uv_rtc_timer_head *head, int pnode) +{ + u64 lowest = ULLONG_MAX; + int c, bcpu = -1; + + head->next_cpu = -1; + for (c = 0; c < head->ncpus; c++) { + u64 exp = head->cpu[c].expires; + if (exp < lowest) { + bcpu = c; + lowest = exp; + } + } + if (bcpu >= 0) { + head->next_cpu = bcpu; + c = head->cpu[bcpu].lcpu; + if (uv_setup_intr(c, lowest)) + /* If we didn't set it up in time, trigger */ + uv_rtc_send_IPI(c); + } else { + uv_write_global_mmr64(pnode, UVH_RTC1_INT_CONFIG, + UVH_RTC1_INT_CONFIG_M_MASK); + } +} + +/* + * Set expiration time for current cpu. + * + * Returns 1 if we missed the expiration time. + */ +static int uv_rtc_set_timer(int cpu, u64 expires) +{ + int pnode = uv_cpu_to_pnode(cpu); + int bid = uv_cpu_to_blade_id(cpu); + struct uv_rtc_timer_head *head = blade_info[bid]; + int bcpu = uv_cpu_blade_processor_id(cpu); + u64 *t = &head->cpu[bcpu].expires; + unsigned long flags; + int next_cpu; + + spin_lock_irqsave(&head->lock, flags); + + next_cpu = head->next_cpu; + *t = expires; + + /* Will this one be next to go off? */ + if (next_cpu < 0 || bcpu == next_cpu || + expires < head->cpu[next_cpu].expires) { + head->next_cpu = bcpu; + if (uv_setup_intr(cpu, expires)) { + *t = ULLONG_MAX; + uv_rtc_find_next_timer(head, pnode); + spin_unlock_irqrestore(&head->lock, flags); + return -ETIME; + } + } + + spin_unlock_irqrestore(&head->lock, flags); + return 0; +} + +/* + * Unset expiration time for current cpu. + * + * Returns 1 if this timer was pending. + */ +static int uv_rtc_unset_timer(int cpu, int force) +{ + int pnode = uv_cpu_to_pnode(cpu); + int bid = uv_cpu_to_blade_id(cpu); + struct uv_rtc_timer_head *head = blade_info[bid]; + int bcpu = uv_cpu_blade_processor_id(cpu); + u64 *t = &head->cpu[bcpu].expires; + unsigned long flags; + int rc = 0; + + spin_lock_irqsave(&head->lock, flags); + + if ((head->next_cpu == bcpu && uv_read_rtc(NULL) >= *t) || force) + rc = 1; + + if (rc) { + *t = ULLONG_MAX; + /* Was the hardware setup for this timer? */ + if (head->next_cpu == bcpu) + uv_rtc_find_next_timer(head, pnode); + } + + spin_unlock_irqrestore(&head->lock, flags); + + return rc; +} + + +/* + * Kernel interface routines. + */ + +/* + * Read the RTC. + * + * Starting with HUB rev 2.0, the UV RTC register is replicated across all + * cachelines of it's own page. This allows faster simultaneous reads + * from a given socket. + */ +static u64 uv_read_rtc(struct clocksource *cs) +{ + unsigned long offset; + + if (uv_get_min_hub_revision_id() == 1) + offset = 0; + else + offset = (uv_blade_processor_id() * L1_CACHE_BYTES) % PAGE_SIZE; + + return (u64)uv_read_local_mmr(UVH_RTC | offset); +} + +/* + * Program the next event, relative to now + */ +static int uv_rtc_next_event(unsigned long delta, + struct clock_event_device *ced) +{ + int ced_cpu = cpumask_first(ced->cpumask); + + return uv_rtc_set_timer(ced_cpu, delta + uv_read_rtc(NULL)); +} + +/* + * Shutdown the RTC timer + */ +static int uv_rtc_shutdown(struct clock_event_device *evt) +{ + int ced_cpu = cpumask_first(evt->cpumask); + + uv_rtc_unset_timer(ced_cpu, 1); + return 0; +} + +static void uv_rtc_interrupt(void) +{ + int cpu = smp_processor_id(); + struct clock_event_device *ced = &per_cpu(cpu_ced, cpu); + + if (!ced || !ced->event_handler) + return; + + if (uv_rtc_unset_timer(cpu, 0) != 1) + return; + + ced->event_handler(ced); +} + +static int __init uv_enable_evt_rtc(char *str) +{ + uv_rtc_evt_enable = 1; + + return 1; +} +__setup("uvrtcevt", uv_enable_evt_rtc); + +static __init void uv_rtc_register_clockevents(struct work_struct *dummy) +{ + struct clock_event_device *ced = this_cpu_ptr(&cpu_ced); + + *ced = clock_event_device_uv; + ced->cpumask = cpumask_of(smp_processor_id()); + clockevents_register_device(ced); +} + +static __init int uv_rtc_setup_clock(void) +{ + int rc; + + if (!is_uv_system()) + return -ENODEV; + + rc = clocksource_register_hz(&clocksource_uv, sn_rtc_cycles_per_second); + if (rc) + printk(KERN_INFO "UV RTC clocksource failed rc %d\n", rc); + else + printk(KERN_INFO "UV RTC clocksource registered freq %lu MHz\n", + sn_rtc_cycles_per_second/(unsigned long)1E6); + + if (rc || !uv_rtc_evt_enable || x86_platform_ipi_callback) + return rc; + + /* Setup and register clockevents */ + rc = uv_rtc_allocate_timers(); + if (rc) + goto error; + + x86_platform_ipi_callback = uv_rtc_interrupt; + + clock_event_device_uv.mult = div_sc(sn_rtc_cycles_per_second, + NSEC_PER_SEC, clock_event_device_uv.shift); + + clock_event_device_uv.min_delta_ns = NSEC_PER_SEC / + sn_rtc_cycles_per_second; + clock_event_device_uv.min_delta_ticks = 1; + + clock_event_device_uv.max_delta_ns = clocksource_uv.mask * + (NSEC_PER_SEC / sn_rtc_cycles_per_second); + clock_event_device_uv.max_delta_ticks = clocksource_uv.mask; + + rc = schedule_on_each_cpu(uv_rtc_register_clockevents); + if (rc) { + x86_platform_ipi_callback = NULL; + uv_rtc_deallocate_timers(); + goto error; + } + + printk(KERN_INFO "UV RTC clockevents registered\n"); + + return 0; + +error: + clocksource_unregister(&clocksource_uv); + printk(KERN_INFO "UV RTC clockevents failed rc %d\n", rc); + + return rc; +} +arch_initcall(uv_rtc_setup_clock); |