linux/arch/x86/include/asm/uv/uv_hub.h

885 lines
25 KiB
C
Raw Normal View History

/*
* 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 architectural definitions
*
* Copyright (C) 2007-2014 Silicon Graphics, Inc. All rights reserved.
*/
#ifndef _ASM_X86_UV_UV_HUB_H
#define _ASM_X86_UV_UV_HUB_H
#ifdef CONFIG_X86_64
#include <linux/numa.h>
#include <linux/percpu.h>
#include <linux/timer.h>
#include <linux/io.h>
#include <linux/topology.h>
#include <asm/types.h>
#include <asm/percpu.h>
#include <asm/uv/uv_mmrs.h>
#include <asm/uv/bios.h>
#include <asm/irq_vectors.h>
#include <asm/io_apic.h>
/*
* Addressing Terminology
*
* M - The low M bits of a physical address represent the offset
* into the blade local memory. RAM memory on a blade is physically
* contiguous (although various IO spaces may punch holes in
* it)..
*
* N - Number of bits in the node portion of a socket physical
* address.
*
* NASID - network ID of a router, Mbrick or Cbrick. Nasid values of
* routers always have low bit of 1, C/MBricks have low bit
* equal to 0. Most addressing macros that target UV hub chips
* right shift the NASID by 1 to exclude the always-zero bit.
* NASIDs contain up to 15 bits.
*
* GNODE - NASID right shifted by 1 bit. Most mmrs contain gnodes instead
* of nasids.
*
* PNODE - the low N bits of the GNODE. The PNODE is the most useful variant
* of the nasid for socket usage.
*
* GPA - (global physical address) a socket physical address converted
* so that it can be used by the GRU as a global address. Socket
* physical addresses 1) need additional NASID (node) bits added
* to the high end of the address, and 2) unaliased if the
* partition does not have a physical address 0. In addition, on
* UV2 rev 1, GPAs need the gnode left shifted to bits 39 or 40.
*
*
* NumaLink Global Physical Address Format:
* +--------------------------------+---------------------+
* |00..000| GNODE | NodeOffset |
* +--------------------------------+---------------------+
* |<-------53 - M bits --->|<--------M bits ----->
*
* M - number of node offset bits (35 .. 40)
*
*
* Memory/UV-HUB Processor Socket Address Format:
* +----------------+---------------+---------------------+
* |00..000000000000| PNODE | NodeOffset |
* +----------------+---------------+---------------------+
* <--- N bits --->|<--------M bits ----->
*
* M - number of node offset bits (35 .. 40)
* N - number of PNODE bits (0 .. 10)
*
* Note: M + N cannot currently exceed 44 (x86_64) or 46 (IA64).
* The actual values are configuration dependent and are set at
* boot time. M & N values are set by the hardware/BIOS at boot.
*
*
* APICID format
* NOTE!!!!!! This is the current format of the APICID. However, code
* should assume that this will change in the future. Use functions
* in this file for all APICID bit manipulations and conversion.
*
* 1111110000000000
* 5432109876543210
* pppppppppplc0cch Nehalem-EX (12 bits in hdw reg)
* ppppppppplcc0cch Westmere-EX (12 bits in hdw reg)
* pppppppppppcccch SandyBridge (15 bits in hdw reg)
* sssssssssss
*
* p = pnode bits
* l = socket number on board
* c = core
* h = hyperthread
* s = bits that are in the SOCKET_ID CSR
*
* Note: Processor may support fewer bits in the APICID register. The ACPI
* tables hold all 16 bits. Software needs to be aware of this.
*
* Unless otherwise specified, all references to APICID refer to
* the FULL value contained in ACPI tables, not the subset in the
* processor APICID register.
*/
/*
* Maximum number of bricks in all partitions and in all coherency domains.
* This is the total number of bricks accessible in the numalink fabric. It
* includes all C & M bricks. Routers are NOT included.
*
* This value is also the value of the maximum number of non-router NASIDs
* in the numalink fabric.
*
* NOTE: a brick may contain 1 or 2 OS nodes. Don't get these confused.
*/
#define UV_MAX_NUMALINK_BLADES 16384
/*
* Maximum number of C/Mbricks within a software SSI (hardware may support
* more).
*/
#define UV_MAX_SSI_BLADES 256
/*
* The largest possible NASID of a C or M brick (+ 2)
*/
#define UV_MAX_NASID_VALUE (UV_MAX_NUMALINK_BLADES * 2)
/* System Controller Interface Reg info */
struct uv_scir_s {
struct timer_list timer;
unsigned long offset;
unsigned long last;
unsigned long idle_on;
unsigned long idle_off;
unsigned char state;
unsigned char enabled;
};
/* GAM (globally addressed memory) range table */
struct uv_gam_range_s {
u32 limit; /* PA bits 56:26 (GAM_RANGE_SHFT) */
u16 nasid; /* node's global physical address */
s8 base; /* entry index of node's base addr */
u8 reserved;
};
/*
* The following defines attributes of the HUB chip. These attributes are
* frequently referenced and are kept in a common per hub struct.
* After setup, the struct is read only, so it should be readily
* available in the L3 cache on the cpu socket for the node.
*/
struct uv_hub_info_s {
unsigned long global_mmr_base;
unsigned long global_mmr_shift;
unsigned long gpa_mask;
unsigned short *socket_to_node;
unsigned short *socket_to_pnode;
unsigned short *pnode_to_socket;
struct uv_gam_range_s *gr_table;
unsigned short min_socket;
unsigned short min_pnode;
unsigned char m_val;
unsigned char n_val;
unsigned char gr_table_len;
unsigned char hub_revision;
unsigned char apic_pnode_shift;
unsigned char gpa_shift;
unsigned char m_shift;
unsigned char n_lshift;
unsigned int gnode_extra;
unsigned long gnode_upper;
unsigned long lowmem_remap_top;
unsigned long lowmem_remap_base;
unsigned long global_gru_base;
unsigned long global_gru_shift;
unsigned short pnode;
unsigned short pnode_mask;
unsigned short coherency_domain_number;
unsigned short numa_blade_id;
unsigned short nr_possible_cpus;
unsigned short nr_online_cpus;
short memory_nid;
};
/* CPU specific info with a pointer to the hub common info struct */
struct uv_cpu_info_s {
void *p_uv_hub_info;
unsigned char blade_cpu_id;
struct uv_scir_s scir;
};
DECLARE_PER_CPU(struct uv_cpu_info_s, __uv_cpu_info);
#define uv_cpu_info this_cpu_ptr(&__uv_cpu_info)
#define uv_cpu_info_per(cpu) (&per_cpu(__uv_cpu_info, cpu))
#define uv_scir_info (&uv_cpu_info->scir)
#define uv_cpu_scir_info(cpu) (&uv_cpu_info_per(cpu)->scir)
x86/platform/UV: Allocate common per node hub info structs on local node Allocate and setup per node hub info structs. CPU 0/Node 0 hub info is statically allocated to be accessible early in system startup. The remaining hub info structs are allocated on the node's local memory, and shared among the CPU's on that node. This leaves the small amount of info unique to each CPU in the per CPU info struct. Memory is saved by combining the common per node info fields to common node local structs. In addtion, since the info is read only only after setup, it should stay in the L3 cache of the local processor socket. This should therefore improve the cache hit rate when a group of cpus on a node are all interrupted for a common task. Tested-by: John Estabrook <estabrook@sgi.com> Tested-by: Gary Kroening <gfk@sgi.com> Tested-by: Nathan Zimmer <nzimmer@sgi.com> Signed-off-by: Mike Travis <travis@sgi.com> Reviewed-by: Dimitri Sivanich <sivanich@sgi.com> Reviewed-by: Andrew Banman <abanman@sgi.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Len Brown <len.brown@intel.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Russ Anderson <rja@sgi.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/20160429215404.813051625@asylum.americas.sgi.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-29 21:54:15 +00:00
/* Node specific hub common info struct */
extern void **__uv_hub_info_list;
static inline struct uv_hub_info_s *uv_hub_info_list(int node)
{
return (struct uv_hub_info_s *)__uv_hub_info_list[node];
}
static inline struct uv_hub_info_s *_uv_hub_info(void)
{
return (struct uv_hub_info_s *)uv_cpu_info->p_uv_hub_info;
}
#define uv_hub_info _uv_hub_info()
static inline struct uv_hub_info_s *uv_cpu_hub_info(int cpu)
{
return (struct uv_hub_info_s *)uv_cpu_info_per(cpu)->p_uv_hub_info;
}
#define UV_HUB_INFO_VERSION 0x7150
extern int uv_hub_info_version(void);
static inline int uv_hub_info_check(int version)
{
if (uv_hub_info_version() == version)
return 0;
pr_crit("UV: uv_hub_info version(%x) mismatch, expecting(%x)\n",
uv_hub_info_version(), version);
BUG(); /* Catastrophic - cannot continue on unknown UV system */
}
#define _uv_hub_info_check() uv_hub_info_check(UV_HUB_INFO_VERSION)
/*
* HUB revision ranges for each UV HUB architecture.
* This is a software convention - NOT the hardware revision numbers in
* the hub chip.
*/
#define UV1_HUB_REVISION_BASE 1
#define UV2_HUB_REVISION_BASE 3
#define UV3_HUB_REVISION_BASE 5
#define UV4_HUB_REVISION_BASE 7
#ifdef UV1_HUB_IS_SUPPORTED
static inline int is_uv1_hub(void)
{
return uv_hub_info->hub_revision < UV2_HUB_REVISION_BASE;
}
#else
static inline int is_uv1_hub(void)
{
return 0;
}
#endif
#ifdef UV2_HUB_IS_SUPPORTED
static inline int is_uv2_hub(void)
{
return ((uv_hub_info->hub_revision >= UV2_HUB_REVISION_BASE) &&
(uv_hub_info->hub_revision < UV3_HUB_REVISION_BASE));
}
#else
static inline int is_uv2_hub(void)
{
return 0;
}
#endif
#ifdef UV3_HUB_IS_SUPPORTED
static inline int is_uv3_hub(void)
{
return ((uv_hub_info->hub_revision >= UV3_HUB_REVISION_BASE) &&
(uv_hub_info->hub_revision < UV4_HUB_REVISION_BASE));
}
#else
static inline int is_uv3_hub(void)
{
return 0;
}
#endif
#ifdef UV4_HUB_IS_SUPPORTED
static inline int is_uv4_hub(void)
{
return uv_hub_info->hub_revision >= UV4_HUB_REVISION_BASE;
}
#else
static inline int is_uv4_hub(void)
{
return 0;
}
#endif
static inline int is_uvx_hub(void)
{
if (uv_hub_info->hub_revision >= UV2_HUB_REVISION_BASE)
return uv_hub_info->hub_revision;
return 0;
}
static inline int is_uv_hub(void)
{
#ifdef UV1_HUB_IS_SUPPORTED
return uv_hub_info->hub_revision;
#endif
return is_uvx_hub();
}
union uvh_apicid {
unsigned long v;
struct uvh_apicid_s {
unsigned long local_apic_mask : 24;
unsigned long local_apic_shift : 5;
unsigned long unused1 : 3;
unsigned long pnode_mask : 24;
unsigned long pnode_shift : 5;
unsigned long unused2 : 3;
} s;
};
/*
* Local & Global MMR space macros.
* Note: macros are intended to be used ONLY by inline functions
* in this file - not by other kernel code.
* n - NASID (full 15-bit global nasid)
* g - GNODE (full 15-bit global nasid, right shifted 1)
* p - PNODE (local part of nsids, right shifted 1)
*/
#define UV_NASID_TO_PNODE(n) (((n) >> 1) & uv_hub_info->pnode_mask)
#define UV_PNODE_TO_GNODE(p) ((p) |uv_hub_info->gnode_extra)
#define UV_PNODE_TO_NASID(p) (UV_PNODE_TO_GNODE(p) << 1)
#define UV1_LOCAL_MMR_BASE 0xf4000000UL
#define UV1_GLOBAL_MMR32_BASE 0xf8000000UL
#define UV1_LOCAL_MMR_SIZE (64UL * 1024 * 1024)
#define UV1_GLOBAL_MMR32_SIZE (64UL * 1024 * 1024)
#define UV2_LOCAL_MMR_BASE 0xfa000000UL
#define UV2_GLOBAL_MMR32_BASE 0xfc000000UL
#define UV2_LOCAL_MMR_SIZE (32UL * 1024 * 1024)
#define UV2_GLOBAL_MMR32_SIZE (32UL * 1024 * 1024)
#define UV3_LOCAL_MMR_BASE 0xfa000000UL
#define UV3_GLOBAL_MMR32_BASE 0xfc000000UL
#define UV3_LOCAL_MMR_SIZE (32UL * 1024 * 1024)
#define UV3_GLOBAL_MMR32_SIZE (32UL * 1024 * 1024)
#define UV4_LOCAL_MMR_BASE 0xfa000000UL
#define UV4_GLOBAL_MMR32_BASE 0xfc000000UL
#define UV4_LOCAL_MMR_SIZE (32UL * 1024 * 1024)
#define UV4_GLOBAL_MMR32_SIZE (16UL * 1024 * 1024)
#define UV_LOCAL_MMR_BASE ( \
is_uv1_hub() ? UV1_LOCAL_MMR_BASE : \
is_uv2_hub() ? UV2_LOCAL_MMR_BASE : \
is_uv3_hub() ? UV3_LOCAL_MMR_BASE : \
/*is_uv4_hub*/ UV4_LOCAL_MMR_BASE)
#define UV_GLOBAL_MMR32_BASE ( \
is_uv1_hub() ? UV1_GLOBAL_MMR32_BASE : \
is_uv2_hub() ? UV2_GLOBAL_MMR32_BASE : \
is_uv3_hub() ? UV3_GLOBAL_MMR32_BASE : \
/*is_uv4_hub*/ UV4_GLOBAL_MMR32_BASE)
#define UV_LOCAL_MMR_SIZE ( \
is_uv1_hub() ? UV1_LOCAL_MMR_SIZE : \
is_uv2_hub() ? UV2_LOCAL_MMR_SIZE : \
is_uv3_hub() ? UV3_LOCAL_MMR_SIZE : \
/*is_uv4_hub*/ UV4_LOCAL_MMR_SIZE)
#define UV_GLOBAL_MMR32_SIZE ( \
is_uv1_hub() ? UV1_GLOBAL_MMR32_SIZE : \
is_uv2_hub() ? UV2_GLOBAL_MMR32_SIZE : \
is_uv3_hub() ? UV3_GLOBAL_MMR32_SIZE : \
/*is_uv4_hub*/ UV4_GLOBAL_MMR32_SIZE)
#define UV_GLOBAL_MMR64_BASE (uv_hub_info->global_mmr_base)
#define UV_GLOBAL_GRU_MMR_BASE 0x4000000
#define UV_GLOBAL_MMR32_PNODE_SHIFT 15
#define _UV_GLOBAL_MMR64_PNODE_SHIFT 26
#define UV_GLOBAL_MMR64_PNODE_SHIFT (uv_hub_info->global_mmr_shift)
#define UV_GLOBAL_MMR32_PNODE_BITS(p) ((p) << (UV_GLOBAL_MMR32_PNODE_SHIFT))
#define UV_GLOBAL_MMR64_PNODE_BITS(p) \
(((unsigned long)(p)) << UV_GLOBAL_MMR64_PNODE_SHIFT)
#define UVH_APICID 0x002D0E00L
#define UV_APIC_PNODE_SHIFT 6
#define UV_APICID_HIBIT_MASK 0xffff0000
/* Local Bus from cpu's perspective */
#define LOCAL_BUS_BASE 0x1c00000
#define LOCAL_BUS_SIZE (4 * 1024 * 1024)
/*
* System Controller Interface Reg
*
* Note there are NO leds on a UV system. This register is only
* used by the system controller to monitor system-wide operation.
* There are 64 regs per node. With Nahelem cpus (2 cores per node,
* 8 cpus per core, 2 threads per cpu) there are 32 cpu threads on
* a node.
*
* The window is located at top of ACPI MMR space
*/
#define SCIR_WINDOW_COUNT 64
#define SCIR_LOCAL_MMR_BASE (LOCAL_BUS_BASE + \
LOCAL_BUS_SIZE - \
SCIR_WINDOW_COUNT)
#define SCIR_CPU_HEARTBEAT 0x01 /* timer interrupt */
#define SCIR_CPU_ACTIVITY 0x02 /* not idle */
#define SCIR_CPU_HB_INTERVAL (HZ) /* once per second */
/* Loop through all installed blades */
#define for_each_possible_blade(bid) \
for ((bid) = 0; (bid) < uv_num_possible_blades(); (bid)++)
/*
* Macros for converting between kernel virtual addresses, socket local physical
* addresses, and UV global physical addresses.
* Note: use the standard __pa() & __va() macros for converting
* between socket virtual and socket physical addresses.
*/
/* global bits offset - number of local address bits in gpa for this UV arch */
static inline unsigned int uv_gpa_shift(void)
{
return uv_hub_info->gpa_shift;
}
#define _uv_gpa_shift
/* Find node that has the address range that contains global address */
static inline struct uv_gam_range_s *uv_gam_range(unsigned long pa)
{
struct uv_gam_range_s *gr = uv_hub_info->gr_table;
unsigned long pal = (pa & uv_hub_info->gpa_mask) >> UV_GAM_RANGE_SHFT;
int i, num = uv_hub_info->gr_table_len;
if (gr) {
for (i = 0; i < num; i++, gr++) {
if (pal < gr->limit)
return gr;
}
}
pr_crit("UV: GAM Range for 0x%lx not found at %p!\n", pa, gr);
BUG();
}
/* Return base address of node that contains global address */
static inline unsigned long uv_gam_range_base(unsigned long pa)
{
struct uv_gam_range_s *gr = uv_gam_range(pa);
int base = gr->base;
if (base < 0)
return 0UL;
return uv_hub_info->gr_table[base].limit;
}
/* socket phys RAM --> UV global NASID (UV4+) */
static inline unsigned long uv_soc_phys_ram_to_nasid(unsigned long paddr)
{
return uv_gam_range(paddr)->nasid;
}
#define _uv_soc_phys_ram_to_nasid
/* socket virtual --> UV global NASID (UV4+) */
static inline unsigned long uv_gpa_nasid(void *v)
{
return uv_soc_phys_ram_to_nasid(__pa(v));
}
/* socket phys RAM --> UV global physical address */
static inline unsigned long uv_soc_phys_ram_to_gpa(unsigned long paddr)
{
unsigned int m_val = uv_hub_info->m_val;
if (paddr < uv_hub_info->lowmem_remap_top)
paddr |= uv_hub_info->lowmem_remap_base;
if (m_val) {
paddr |= uv_hub_info->gnode_upper;
paddr = ((paddr << uv_hub_info->m_shift)
>> uv_hub_info->m_shift) |
((paddr >> uv_hub_info->m_val)
<< uv_hub_info->n_lshift);
} else {
paddr |= uv_soc_phys_ram_to_nasid(paddr)
<< uv_hub_info->gpa_shift;
}
return paddr;
}
/* socket virtual --> UV global physical address */
static inline unsigned long uv_gpa(void *v)
{
return uv_soc_phys_ram_to_gpa(__pa(v));
}
/* Top two bits indicate the requested address is in MMR space. */
static inline int
uv_gpa_in_mmr_space(unsigned long gpa)
{
return (gpa >> 62) == 0x3UL;
}
/* UV global physical address --> socket phys RAM */
static inline unsigned long uv_gpa_to_soc_phys_ram(unsigned long gpa)
{
unsigned long paddr;
unsigned long remap_base = uv_hub_info->lowmem_remap_base;
unsigned long remap_top = uv_hub_info->lowmem_remap_top;
unsigned int m_val = uv_hub_info->m_val;
if (m_val)
gpa = ((gpa << uv_hub_info->m_shift) >> uv_hub_info->m_shift) |
((gpa >> uv_hub_info->n_lshift) << uv_hub_info->m_val);
paddr = gpa & uv_hub_info->gpa_mask;
if (paddr >= remap_base && paddr < remap_base + remap_top)
paddr -= remap_base;
return paddr;
}
/* gpa -> gnode */
static inline unsigned long uv_gpa_to_gnode(unsigned long gpa)
{
unsigned int n_lshift = uv_hub_info->n_lshift;
if (n_lshift)
return gpa >> n_lshift;
return uv_gam_range(gpa)->nasid >> 1;
}
/* gpa -> pnode */
static inline int uv_gpa_to_pnode(unsigned long gpa)
{
return uv_gpa_to_gnode(gpa) & uv_hub_info->pnode_mask;
}
/* gpa -> node offset */
static inline unsigned long uv_gpa_to_offset(unsigned long gpa)
{
unsigned int m_shift = uv_hub_info->m_shift;
if (m_shift)
return (gpa << m_shift) >> m_shift;
return (gpa & uv_hub_info->gpa_mask) - uv_gam_range_base(gpa);
}
/* Convert socket to node */
static inline int _uv_socket_to_node(int socket, unsigned short *s2nid)
{
return s2nid ? s2nid[socket - uv_hub_info->min_socket] : socket;
}
static inline int uv_socket_to_node(int socket)
{
return _uv_socket_to_node(socket, uv_hub_info->socket_to_node);
}
/* pnode, offset --> socket virtual */
static inline void *uv_pnode_offset_to_vaddr(int pnode, unsigned long offset)
{
unsigned int m_val = uv_hub_info->m_val;
unsigned long base;
unsigned short sockid, node, *p2s;
if (m_val)
return __va(((unsigned long)pnode << m_val) | offset);
p2s = uv_hub_info->pnode_to_socket;
sockid = p2s ? p2s[pnode - uv_hub_info->min_pnode] : pnode;
node = uv_socket_to_node(sockid);
/* limit address of previous socket is our base, except node 0 is 0 */
if (!node)
return __va((unsigned long)offset);
base = (unsigned long)(uv_hub_info->gr_table[node - 1].limit);
return __va(base << UV_GAM_RANGE_SHFT | offset);
}
/* Extract/Convert a PNODE from an APICID (full apicid, not processor subset) */
static inline int uv_apicid_to_pnode(int apicid)
{
int pnode = apicid >> uv_hub_info->apic_pnode_shift;
unsigned short *s2pn = uv_hub_info->socket_to_pnode;
return s2pn ? s2pn[pnode - uv_hub_info->min_socket] : pnode;
}
/* Convert an apicid to the socket number on the blade */
static inline int uv_apicid_to_socket(int apicid)
{
if (is_uv1_hub())
return (apicid >> (uv_hub_info->apic_pnode_shift - 1)) & 1;
else
return 0;
}
/*
* Access global MMRs using the low memory MMR32 space. This region supports
* faster MMR access but not all MMRs are accessible in this space.
*/
static inline unsigned long *uv_global_mmr32_address(int pnode, unsigned long offset)
{
return __va(UV_GLOBAL_MMR32_BASE |
UV_GLOBAL_MMR32_PNODE_BITS(pnode) | offset);
}
static inline void uv_write_global_mmr32(int pnode, unsigned long offset, unsigned long val)
{
writeq(val, uv_global_mmr32_address(pnode, offset));
}
static inline unsigned long uv_read_global_mmr32(int pnode, unsigned long offset)
{
return readq(uv_global_mmr32_address(pnode, offset));
}
/*
* Access Global MMR space using the MMR space located at the top of physical
* memory.
*/
static inline volatile void __iomem *uv_global_mmr64_address(int pnode, unsigned long offset)
{
return __va(UV_GLOBAL_MMR64_BASE |
UV_GLOBAL_MMR64_PNODE_BITS(pnode) | offset);
}
static inline void uv_write_global_mmr64(int pnode, unsigned long offset, unsigned long val)
{
writeq(val, uv_global_mmr64_address(pnode, offset));
}
static inline unsigned long uv_read_global_mmr64(int pnode, unsigned long offset)
{
return readq(uv_global_mmr64_address(pnode, offset));
}
static inline void uv_write_global_mmr8(int pnode, unsigned long offset, unsigned char val)
{
writeb(val, uv_global_mmr64_address(pnode, offset));
}
static inline unsigned char uv_read_global_mmr8(int pnode, unsigned long offset)
{
return readb(uv_global_mmr64_address(pnode, offset));
}
/*
* Access hub local MMRs. Faster than using global space but only local MMRs
* are accessible.
*/
static inline unsigned long *uv_local_mmr_address(unsigned long offset)
{
return __va(UV_LOCAL_MMR_BASE | offset);
}
static inline unsigned long uv_read_local_mmr(unsigned long offset)
{
return readq(uv_local_mmr_address(offset));
}
static inline void uv_write_local_mmr(unsigned long offset, unsigned long val)
{
writeq(val, uv_local_mmr_address(offset));
}
static inline unsigned char uv_read_local_mmr8(unsigned long offset)
{
return readb(uv_local_mmr_address(offset));
}
static inline void uv_write_local_mmr8(unsigned long offset, unsigned char val)
{
writeb(val, uv_local_mmr_address(offset));
}
/* Blade-local cpu number of current cpu. Numbered 0 .. <# cpus on the blade> */
static inline int uv_blade_processor_id(void)
{
return uv_cpu_info->blade_cpu_id;
}
/* Blade-local cpu number of cpu N. Numbered 0 .. <# cpus on the blade> */
static inline int uv_cpu_blade_processor_id(int cpu)
{
return uv_cpu_info_per(cpu)->blade_cpu_id;
}
#define _uv_cpu_blade_processor_id 1 /* indicate function available */
/* Blade number to Node number (UV1..UV4 is 1:1) */
static inline int uv_blade_to_node(int blade)
{
return blade;
}
/* Blade number of current cpu. Numnbered 0 .. <#blades -1> */
static inline int uv_numa_blade_id(void)
{
return uv_hub_info->numa_blade_id;
}
/*
* Convert linux node number to the UV blade number.
* .. Currently for UV1 thru UV4 the node and the blade are identical.
* .. If this changes then you MUST check references to this function!
*/
static inline int uv_node_to_blade_id(int nid)
{
return nid;
}
/* Convert a cpu number to the the UV blade number */
static inline int uv_cpu_to_blade_id(int cpu)
{
return uv_node_to_blade_id(cpu_to_node(cpu));
}
/* Convert a blade id to the PNODE of the blade */
static inline int uv_blade_to_pnode(int bid)
{
return uv_hub_info_list(uv_blade_to_node(bid))->pnode;
}
/* Nid of memory node on blade. -1 if no blade-local memory */
static inline int uv_blade_to_memory_nid(int bid)
{
return uv_hub_info_list(uv_blade_to_node(bid))->memory_nid;
}
/* Determine the number of possible cpus on a blade */
static inline int uv_blade_nr_possible_cpus(int bid)
{
return uv_hub_info_list(uv_blade_to_node(bid))->nr_possible_cpus;
}
/* Determine the number of online cpus on a blade */
static inline int uv_blade_nr_online_cpus(int bid)
{
return uv_hub_info_list(uv_blade_to_node(bid))->nr_online_cpus;
}
/* Convert a cpu id to the PNODE of the blade containing the cpu */
static inline int uv_cpu_to_pnode(int cpu)
{
return uv_cpu_hub_info(cpu)->pnode;
}
/* Convert a linux node number to the PNODE of the blade */
static inline int uv_node_to_pnode(int nid)
{
return uv_hub_info_list(nid)->pnode;
}
/* Maximum possible number of blades */
extern short uv_possible_blades;
static inline int uv_num_possible_blades(void)
{
return uv_possible_blades;
}
x86/UV: Update UV support for external NMI signals The current UV NMI handler has not been updated for the changes in the system NMI handler and the perf operations. The UV NMI handler reads an MMR in the UV Hub to check to see if the NMI event was caused by the external 'system NMI' that the operator can initiate on the System Mgmt Controller. The problem arises when the perf tools are running, causing millions of perf events per second on very large CPU count systems. Previously this was okay because the perf NMI handler ran at a higher priority on the NMI call chain and if the NMI was a perf event, it would stop calling other NMI handlers remaining on the NMI call chain. Now the system NMI handler calls all the handlers on the NMI call chain including the UV NMI handler. This causes the UV NMI handler to read the MMRs at the same millions per second rate. This can lead to significant performance loss and possible system failures. It also can cause thousands of 'Dazed and Confused' messages being sent to the system console. This effectively makes perf tools unusable on UV systems. To avoid this excessive overhead when perf tools are running, this code has been optimized to minimize reading of the MMRs as much as possible, by moving to the NMI_UNKNOWN notifier chain. This chain is called only when all the users on the standard NMI_LOCAL call chain have been called and none of them have claimed this NMI. There is an exception where the NMI_LOCAL notifier chain is used. When the perf tools are in use, it's possible that the UV NMI was captured by some other NMI handler and then either ignored or mistakenly processed as a perf event. We set a per_cpu ('ping') flag for those CPUs that ignored the initial NMI, and then send them an IPI NMI signal. The NMI_LOCAL handler on each cpu does not need to read the MMR, but instead checks the in memory flag indicating it was pinged. There are two module variables, 'ping_count' indicating how many requested NMI events occurred, and 'ping_misses' indicating how many stray NMI events. These most likely are perf events so it shows the overhead of the perf NMI interrupts and how many MMR reads were avoided. This patch also minimizes the reads of the MMRs by having the first cpu entering the NMI handler on each node set a per HUB in-memory atomic value. (Having a per HUB value avoids sending lock traffic over NumaLink.) Both types of UV NMIs from the SMI layer are supported. Signed-off-by: Mike Travis <travis@sgi.com> Reviewed-by: Dimitri Sivanich <sivanich@sgi.com> Reviewed-by: Hedi Berriche <hedi@sgi.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Cc: Jason Wessel <jason.wessel@windriver.com> Link: http://lkml.kernel.org/r/20130923212500.353547733@asylum.americas.sgi.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2013-09-23 21:25:01 +00:00
/* Per Hub NMI support */
extern void uv_nmi_setup(void);
extern void uv_nmi_setup_hubless(void);
x86/UV: Update UV support for external NMI signals The current UV NMI handler has not been updated for the changes in the system NMI handler and the perf operations. The UV NMI handler reads an MMR in the UV Hub to check to see if the NMI event was caused by the external 'system NMI' that the operator can initiate on the System Mgmt Controller. The problem arises when the perf tools are running, causing millions of perf events per second on very large CPU count systems. Previously this was okay because the perf NMI handler ran at a higher priority on the NMI call chain and if the NMI was a perf event, it would stop calling other NMI handlers remaining on the NMI call chain. Now the system NMI handler calls all the handlers on the NMI call chain including the UV NMI handler. This causes the UV NMI handler to read the MMRs at the same millions per second rate. This can lead to significant performance loss and possible system failures. It also can cause thousands of 'Dazed and Confused' messages being sent to the system console. This effectively makes perf tools unusable on UV systems. To avoid this excessive overhead when perf tools are running, this code has been optimized to minimize reading of the MMRs as much as possible, by moving to the NMI_UNKNOWN notifier chain. This chain is called only when all the users on the standard NMI_LOCAL call chain have been called and none of them have claimed this NMI. There is an exception where the NMI_LOCAL notifier chain is used. When the perf tools are in use, it's possible that the UV NMI was captured by some other NMI handler and then either ignored or mistakenly processed as a perf event. We set a per_cpu ('ping') flag for those CPUs that ignored the initial NMI, and then send them an IPI NMI signal. The NMI_LOCAL handler on each cpu does not need to read the MMR, but instead checks the in memory flag indicating it was pinged. There are two module variables, 'ping_count' indicating how many requested NMI events occurred, and 'ping_misses' indicating how many stray NMI events. These most likely are perf events so it shows the overhead of the perf NMI interrupts and how many MMR reads were avoided. This patch also minimizes the reads of the MMRs by having the first cpu entering the NMI handler on each node set a per HUB in-memory atomic value. (Having a per HUB value avoids sending lock traffic over NumaLink.) Both types of UV NMIs from the SMI layer are supported. Signed-off-by: Mike Travis <travis@sgi.com> Reviewed-by: Dimitri Sivanich <sivanich@sgi.com> Reviewed-by: Hedi Berriche <hedi@sgi.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Cc: Jason Wessel <jason.wessel@windriver.com> Link: http://lkml.kernel.org/r/20130923212500.353547733@asylum.americas.sgi.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2013-09-23 21:25:01 +00:00
/* BMC sets a bit this MMR non-zero before sending an NMI */
#define UVH_NMI_MMR UVH_SCRATCH5
#define UVH_NMI_MMR_CLEAR UVH_SCRATCH5_ALIAS
#define UVH_NMI_MMR_SHIFT 63
#define UVH_NMI_MMR_TYPE "SCRATCH5"
/* Newer SMM NMI handler, not present in all systems */
#define UVH_NMI_MMRX UVH_EVENT_OCCURRED0
#define UVH_NMI_MMRX_CLEAR UVH_EVENT_OCCURRED0_ALIAS
#define UVH_NMI_MMRX_SHIFT UVH_EVENT_OCCURRED0_EXTIO_INT0_SHFT
x86/UV: Update UV support for external NMI signals The current UV NMI handler has not been updated for the changes in the system NMI handler and the perf operations. The UV NMI handler reads an MMR in the UV Hub to check to see if the NMI event was caused by the external 'system NMI' that the operator can initiate on the System Mgmt Controller. The problem arises when the perf tools are running, causing millions of perf events per second on very large CPU count systems. Previously this was okay because the perf NMI handler ran at a higher priority on the NMI call chain and if the NMI was a perf event, it would stop calling other NMI handlers remaining on the NMI call chain. Now the system NMI handler calls all the handlers on the NMI call chain including the UV NMI handler. This causes the UV NMI handler to read the MMRs at the same millions per second rate. This can lead to significant performance loss and possible system failures. It also can cause thousands of 'Dazed and Confused' messages being sent to the system console. This effectively makes perf tools unusable on UV systems. To avoid this excessive overhead when perf tools are running, this code has been optimized to minimize reading of the MMRs as much as possible, by moving to the NMI_UNKNOWN notifier chain. This chain is called only when all the users on the standard NMI_LOCAL call chain have been called and none of them have claimed this NMI. There is an exception where the NMI_LOCAL notifier chain is used. When the perf tools are in use, it's possible that the UV NMI was captured by some other NMI handler and then either ignored or mistakenly processed as a perf event. We set a per_cpu ('ping') flag for those CPUs that ignored the initial NMI, and then send them an IPI NMI signal. The NMI_LOCAL handler on each cpu does not need to read the MMR, but instead checks the in memory flag indicating it was pinged. There are two module variables, 'ping_count' indicating how many requested NMI events occurred, and 'ping_misses' indicating how many stray NMI events. These most likely are perf events so it shows the overhead of the perf NMI interrupts and how many MMR reads were avoided. This patch also minimizes the reads of the MMRs by having the first cpu entering the NMI handler on each node set a per HUB in-memory atomic value. (Having a per HUB value avoids sending lock traffic over NumaLink.) Both types of UV NMIs from the SMI layer are supported. Signed-off-by: Mike Travis <travis@sgi.com> Reviewed-by: Dimitri Sivanich <sivanich@sgi.com> Reviewed-by: Hedi Berriche <hedi@sgi.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Cc: Jason Wessel <jason.wessel@windriver.com> Link: http://lkml.kernel.org/r/20130923212500.353547733@asylum.americas.sgi.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2013-09-23 21:25:01 +00:00
#define UVH_NMI_MMRX_TYPE "EXTIO_INT0"
/* Non-zero indicates newer SMM NMI handler present */
#define UVH_NMI_MMRX_SUPPORTED UVH_EXTIO_INT0_BROADCAST
/* Indicates to BIOS that we want to use the newer SMM NMI handler */
#define UVH_NMI_MMRX_REQ UVH_SCRATCH5_ALIAS_2
#define UVH_NMI_MMRX_REQ_SHIFT 62
struct uv_hub_nmi_s {
raw_spinlock_t nmi_lock;
atomic_t in_nmi; /* flag this node in UV NMI IRQ */
atomic_t cpu_owner; /* last locker of this struct */
atomic_t read_mmr_count; /* count of MMR reads */
atomic_t nmi_count; /* count of true UV NMIs */
unsigned long nmi_value; /* last value read from NMI MMR */
bool hub_present; /* false means UV hubless system */
bool pch_owner; /* indicates this hub owns PCH */
x86/UV: Update UV support for external NMI signals The current UV NMI handler has not been updated for the changes in the system NMI handler and the perf operations. The UV NMI handler reads an MMR in the UV Hub to check to see if the NMI event was caused by the external 'system NMI' that the operator can initiate on the System Mgmt Controller. The problem arises when the perf tools are running, causing millions of perf events per second on very large CPU count systems. Previously this was okay because the perf NMI handler ran at a higher priority on the NMI call chain and if the NMI was a perf event, it would stop calling other NMI handlers remaining on the NMI call chain. Now the system NMI handler calls all the handlers on the NMI call chain including the UV NMI handler. This causes the UV NMI handler to read the MMRs at the same millions per second rate. This can lead to significant performance loss and possible system failures. It also can cause thousands of 'Dazed and Confused' messages being sent to the system console. This effectively makes perf tools unusable on UV systems. To avoid this excessive overhead when perf tools are running, this code has been optimized to minimize reading of the MMRs as much as possible, by moving to the NMI_UNKNOWN notifier chain. This chain is called only when all the users on the standard NMI_LOCAL call chain have been called and none of them have claimed this NMI. There is an exception where the NMI_LOCAL notifier chain is used. When the perf tools are in use, it's possible that the UV NMI was captured by some other NMI handler and then either ignored or mistakenly processed as a perf event. We set a per_cpu ('ping') flag for those CPUs that ignored the initial NMI, and then send them an IPI NMI signal. The NMI_LOCAL handler on each cpu does not need to read the MMR, but instead checks the in memory flag indicating it was pinged. There are two module variables, 'ping_count' indicating how many requested NMI events occurred, and 'ping_misses' indicating how many stray NMI events. These most likely are perf events so it shows the overhead of the perf NMI interrupts and how many MMR reads were avoided. This patch also minimizes the reads of the MMRs by having the first cpu entering the NMI handler on each node set a per HUB in-memory atomic value. (Having a per HUB value avoids sending lock traffic over NumaLink.) Both types of UV NMIs from the SMI layer are supported. Signed-off-by: Mike Travis <travis@sgi.com> Reviewed-by: Dimitri Sivanich <sivanich@sgi.com> Reviewed-by: Hedi Berriche <hedi@sgi.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Cc: Jason Wessel <jason.wessel@windriver.com> Link: http://lkml.kernel.org/r/20130923212500.353547733@asylum.americas.sgi.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2013-09-23 21:25:01 +00:00
};
struct uv_cpu_nmi_s {
struct uv_hub_nmi_s *hub;
int state;
int pinging;
x86/UV: Update UV support for external NMI signals The current UV NMI handler has not been updated for the changes in the system NMI handler and the perf operations. The UV NMI handler reads an MMR in the UV Hub to check to see if the NMI event was caused by the external 'system NMI' that the operator can initiate on the System Mgmt Controller. The problem arises when the perf tools are running, causing millions of perf events per second on very large CPU count systems. Previously this was okay because the perf NMI handler ran at a higher priority on the NMI call chain and if the NMI was a perf event, it would stop calling other NMI handlers remaining on the NMI call chain. Now the system NMI handler calls all the handlers on the NMI call chain including the UV NMI handler. This causes the UV NMI handler to read the MMRs at the same millions per second rate. This can lead to significant performance loss and possible system failures. It also can cause thousands of 'Dazed and Confused' messages being sent to the system console. This effectively makes perf tools unusable on UV systems. To avoid this excessive overhead when perf tools are running, this code has been optimized to minimize reading of the MMRs as much as possible, by moving to the NMI_UNKNOWN notifier chain. This chain is called only when all the users on the standard NMI_LOCAL call chain have been called and none of them have claimed this NMI. There is an exception where the NMI_LOCAL notifier chain is used. When the perf tools are in use, it's possible that the UV NMI was captured by some other NMI handler and then either ignored or mistakenly processed as a perf event. We set a per_cpu ('ping') flag for those CPUs that ignored the initial NMI, and then send them an IPI NMI signal. The NMI_LOCAL handler on each cpu does not need to read the MMR, but instead checks the in memory flag indicating it was pinged. There are two module variables, 'ping_count' indicating how many requested NMI events occurred, and 'ping_misses' indicating how many stray NMI events. These most likely are perf events so it shows the overhead of the perf NMI interrupts and how many MMR reads were avoided. This patch also minimizes the reads of the MMRs by having the first cpu entering the NMI handler on each node set a per HUB in-memory atomic value. (Having a per HUB value avoids sending lock traffic over NumaLink.) Both types of UV NMIs from the SMI layer are supported. Signed-off-by: Mike Travis <travis@sgi.com> Reviewed-by: Dimitri Sivanich <sivanich@sgi.com> Reviewed-by: Hedi Berriche <hedi@sgi.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Cc: Jason Wessel <jason.wessel@windriver.com> Link: http://lkml.kernel.org/r/20130923212500.353547733@asylum.americas.sgi.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2013-09-23 21:25:01 +00:00
int queries;
int pings;
};
DECLARE_PER_CPU(struct uv_cpu_nmi_s, uv_cpu_nmi);
#define uv_hub_nmi this_cpu_read(uv_cpu_nmi.hub)
#define uv_cpu_nmi_per(cpu) (per_cpu(uv_cpu_nmi, cpu))
x86/UV: Update UV support for external NMI signals The current UV NMI handler has not been updated for the changes in the system NMI handler and the perf operations. The UV NMI handler reads an MMR in the UV Hub to check to see if the NMI event was caused by the external 'system NMI' that the operator can initiate on the System Mgmt Controller. The problem arises when the perf tools are running, causing millions of perf events per second on very large CPU count systems. Previously this was okay because the perf NMI handler ran at a higher priority on the NMI call chain and if the NMI was a perf event, it would stop calling other NMI handlers remaining on the NMI call chain. Now the system NMI handler calls all the handlers on the NMI call chain including the UV NMI handler. This causes the UV NMI handler to read the MMRs at the same millions per second rate. This can lead to significant performance loss and possible system failures. It also can cause thousands of 'Dazed and Confused' messages being sent to the system console. This effectively makes perf tools unusable on UV systems. To avoid this excessive overhead when perf tools are running, this code has been optimized to minimize reading of the MMRs as much as possible, by moving to the NMI_UNKNOWN notifier chain. This chain is called only when all the users on the standard NMI_LOCAL call chain have been called and none of them have claimed this NMI. There is an exception where the NMI_LOCAL notifier chain is used. When the perf tools are in use, it's possible that the UV NMI was captured by some other NMI handler and then either ignored or mistakenly processed as a perf event. We set a per_cpu ('ping') flag for those CPUs that ignored the initial NMI, and then send them an IPI NMI signal. The NMI_LOCAL handler on each cpu does not need to read the MMR, but instead checks the in memory flag indicating it was pinged. There are two module variables, 'ping_count' indicating how many requested NMI events occurred, and 'ping_misses' indicating how many stray NMI events. These most likely are perf events so it shows the overhead of the perf NMI interrupts and how many MMR reads were avoided. This patch also minimizes the reads of the MMRs by having the first cpu entering the NMI handler on each node set a per HUB in-memory atomic value. (Having a per HUB value avoids sending lock traffic over NumaLink.) Both types of UV NMIs from the SMI layer are supported. Signed-off-by: Mike Travis <travis@sgi.com> Reviewed-by: Dimitri Sivanich <sivanich@sgi.com> Reviewed-by: Hedi Berriche <hedi@sgi.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Cc: Jason Wessel <jason.wessel@windriver.com> Link: http://lkml.kernel.org/r/20130923212500.353547733@asylum.americas.sgi.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2013-09-23 21:25:01 +00:00
#define uv_hub_nmi_per(cpu) (uv_cpu_nmi_per(cpu).hub)
/* uv_cpu_nmi_states */
#define UV_NMI_STATE_OUT 0
#define UV_NMI_STATE_IN 1
#define UV_NMI_STATE_DUMP 2
#define UV_NMI_STATE_DUMP_DONE 3
/* Update SCIR state */
static inline void uv_set_scir_bits(unsigned char value)
{
if (uv_scir_info->state != value) {
uv_scir_info->state = value;
uv_write_local_mmr8(uv_scir_info->offset, value);
}
}
static inline unsigned long uv_scir_offset(int apicid)
{
return SCIR_LOCAL_MMR_BASE | (apicid & 0x3f);
}
static inline void uv_set_cpu_scir_bits(int cpu, unsigned char value)
{
if (uv_cpu_scir_info(cpu)->state != value) {
uv_write_global_mmr8(uv_cpu_to_pnode(cpu),
uv_cpu_scir_info(cpu)->offset, value);
uv_cpu_scir_info(cpu)->state = value;
}
}
extern unsigned int uv_apicid_hibits;
static unsigned long uv_hub_ipi_value(int apicid, int vector, int mode)
{
apicid |= uv_apicid_hibits;
return (1UL << UVH_IPI_INT_SEND_SHFT) |
((apicid) << UVH_IPI_INT_APIC_ID_SHFT) |
(mode << UVH_IPI_INT_DELIVERY_MODE_SHFT) |
(vector << UVH_IPI_INT_VECTOR_SHFT);
}
static inline void uv_hub_send_ipi(int pnode, int apicid, int vector)
{
unsigned long val;
unsigned long dmode = dest_Fixed;
if (vector == NMI_VECTOR)
dmode = dest_NMI;
val = uv_hub_ipi_value(apicid, vector, dmode);
uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
}
/*
* Get the minimum revision number of the hub chips within the partition.
* (See UVx_HUB_REVISION_BASE above for specific values.)
*/
static inline int uv_get_min_hub_revision_id(void)
{
return uv_hub_info->hub_revision;
}
#endif /* CONFIG_X86_64 */
#endif /* _ASM_X86_UV_UV_HUB_H */