| 1 | /*- |
| 2 | * SPDX-License-Identifier: BSD-3-Clause |
| 3 | * |
| 4 | * Copyright (c) 2001 Wind River Systems, Inc. |
| 5 | * All rights reserved. |
| 6 | * Written by: John Baldwin <jhb@FreeBSD.org> |
| 7 | * |
| 8 | * Redistribution and use in source and binary forms, with or without |
| 9 | * modification, are permitted provided that the following conditions |
| 10 | * are met: |
| 11 | * 1. Redistributions of source code must retain the above copyright |
| 12 | * notice, this list of conditions and the following disclaimer. |
| 13 | * 2. Redistributions in binary form must reproduce the above copyright |
| 14 | * notice, this list of conditions and the following disclaimer in the |
| 15 | * documentation and/or other materials provided with the distribution. |
| 16 | * 3. Neither the name of the author nor the names of any co-contributors |
| 17 | * may be used to endorse or promote products derived from this software |
| 18 | * without specific prior written permission. |
| 19 | * |
| 20 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND |
| 21 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 22 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 23 | * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE |
| 24 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 25 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 26 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 27 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 28 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 29 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 30 | * SUCH DAMAGE. |
| 31 | */ |
| 32 | |
| 33 | #ifndef _SYS_PCPU_H_ |
| 34 | #define	_SYS_PCPU_H_ |
| 35 | |
| 36 | #ifdef LOCORE |
| 37 | #error "no assembler-serviceable parts inside" |
| 38 | #endif |
| 39 | |
| 40 | #include <sys/param.h> |
| 41 | #include <sys/_cpuset.h> |
| 42 | #include <sys/_lock.h> |
| 43 | #include <sys/_mutex.h> |
| 44 | #include <sys/_sx.h> |
| 45 | #include <sys/queue.h> |
| 46 | #include <sys/_rmlock.h> |
| 47 | #include <sys/resource.h> |
| 48 | #include <machine/pcpu.h> |
| 49 | |
| 50 | #define	DPCPU_SETNAME		"set_pcpu" |
| 51 | #define	DPCPU_SYMPREFIX		"pcpu_entry_" |
| 52 | |
| 53 | #ifdef _KERNEL |
| 54 | |
| 55 | /* |
| 56 | * Define a set for pcpu data. |
| 57 | */ |
| 58 | extern uintptr_t *__start_set_pcpu; |
| 59 | __GLOBL(__start_set_pcpu); |
| 60 | extern uintptr_t *__stop_set_pcpu; |
| 61 | __GLOBL(__stop_set_pcpu); |
| 62 | |
| 63 | /* |
| 64 | * Array of dynamic pcpu base offsets. Indexed by id. |
| 65 | */ |
| 66 | extern uintptr_t dpcpu_off[]; |
| 67 | |
| 68 | /* |
| 69 | * Convenience defines. |
| 70 | */ |
| 71 | #define	DPCPU_START		((uintptr_t)&__start_set_pcpu) |
| 72 | #define	DPCPU_STOP		((uintptr_t)&__stop_set_pcpu) |
| 73 | #define	DPCPU_BYTES		(DPCPU_STOP - DPCPU_START) |
| 74 | #define	DPCPU_MODMIN		2048 |
| 75 | #define	DPCPU_SIZE		roundup2(DPCPU_BYTES, PAGE_SIZE) |
| 76 | #define	DPCPU_MODSIZE		(DPCPU_SIZE - (DPCPU_BYTES - DPCPU_MODMIN)) |
| 77 | |
| 78 | /* |
| 79 | * Declaration and definition. |
| 80 | */ |
| 81 | #define	DPCPU_NAME(n)		pcpu_entry_##n |
| 82 | #define	DPCPU_DECLARE(t, n)	extern t DPCPU_NAME(n) |
| 83 | /* struct _hack is to stop this from being used with the static keyword. */ |
| 84 | #define	DPCPU_DEFINE(t, n)	\ |
| 85 | struct _hack; t DPCPU_NAME(n) __section(DPCPU_SETNAME) __used |
| 86 | #if defined(KLD_MODULE) && (defined(__aarch64__) || defined(__riscv) \ |
| 87 | 		|| defined(__powerpc64__) || defined(__i386__)) |
| 88 | /* |
| 89 | * On some architectures the compiler will use PC-relative load to |
| 90 | * find the address of DPCPU data with the static keyword. We then |
| 91 | * use this to find the offset of the data in a per-CPU region. |
| 92 | * This works for in the kernel as we can allocate the space ahead |
| 93 | * of time, however modules need to allocate a sepatate space and |
| 94 | * then use relocations to fix the address of the data. As |
| 95 | * PC-relative data doesn't have a relocation there is nothing for |
| 96 | * the kernel module linker to fix so data is accessed from the |
| 97 | * wrong location. |
| 98 | * |
| 99 | * This is a workaround until a better solution can be found. |
| 100 | * |
| 101 | * VNET_DEFINE_STATIC also has the same workaround. |
| 102 | */ |
| 103 | #define	DPCPU_DEFINE_STATIC(t, n)	\ |
| 104 | t DPCPU_NAME(n) __section(DPCPU_SETNAME) __used |
| 105 | #else |
| 106 | #define	DPCPU_DEFINE_STATIC(t, n)	\ |
| 107 | static t DPCPU_NAME(n) __section(DPCPU_SETNAME) __used |
| 108 | #endif |
| 109 | |
| 110 | /* |
| 111 | * Accessors with a given base. |
| 112 | */ |
| 113 | #define	_DPCPU_PTR(b, n)						\ |
| 114 | (__typeof(DPCPU_NAME(n))*)((b) + (uintptr_t)&DPCPU_NAME(n)) |
| 115 | #define	_DPCPU_GET(b, n)	(*_DPCPU_PTR(b, n)) |
| 116 | #define	_DPCPU_SET(b, n, v)	(*_DPCPU_PTR(b, n) = v) |
| 117 | |
| 118 | /* |
| 119 | * Accessors for the current cpu. |
| 120 | */ |
| 121 | #define	DPCPU_PTR(n)		_DPCPU_PTR(PCPU_GET(dynamic), n) |
| 122 | #define	DPCPU_GET(n)		(*DPCPU_PTR(n)) |
| 123 | #define	DPCPU_SET(n, v)		(*DPCPU_PTR(n) = v) |
| 124 | |
| 125 | /* |
| 126 | * Accessors for remote cpus. |
| 127 | */ |
| 128 | #define	DPCPU_ID_PTR(i, n)	_DPCPU_PTR(dpcpu_off[(i)], n) |
| 129 | #define	DPCPU_ID_GET(i, n)	(*DPCPU_ID_PTR(i, n)) |
| 130 | #define	DPCPU_ID_SET(i, n, v)	(*DPCPU_ID_PTR(i, n) = v) |
| 131 | |
| 132 | /* |
| 133 | * Utility macros. |
| 134 | */ |
| 135 | #define	DPCPU_SUM(n) __extension__					\ |
| 136 | ({									\ |
| 137 | 	u_int _i;							\ |
| 138 | 	__typeof(*DPCPU_PTR(n)) sum;					\ |
| 139 | 									\ |
| 140 | 	sum = 0;							\ |
| 141 | 	CPU_FOREACH(_i) {						\ |
| 142 | 		sum += *DPCPU_ID_PTR(_i, n);				\ |
| 143 | 	}								\ |
| 144 | 	sum;								\ |
| 145 | }) |
| 146 | |
| 147 | #define	DPCPU_VARSUM(n, var) __extension__				\ |
| 148 | ({									\ |
| 149 | 	u_int _i;							\ |
| 150 | 	__typeof((DPCPU_PTR(n))->var) sum;				\ |
| 151 | 									\ |
| 152 | 	sum = 0;							\ |
| 153 | 	CPU_FOREACH(_i) {						\ |
| 154 | 		sum += (DPCPU_ID_PTR(_i, n))->var;			\ |
| 155 | 	}								\ |
| 156 | 	sum;								\ |
| 157 | }) |
| 158 | |
| 159 | #define	DPCPU_ZERO(n) do {						\ |
| 160 | 	u_int _i;							\ |
| 161 | 									\ |
| 162 | 	CPU_FOREACH(_i) {						\ |
| 163 | 		bzero(DPCPU_ID_PTR(_i, n), sizeof(*DPCPU_PTR(n)));	\ |
| 164 | 	}								\ |
| 165 | } while (0) |
| 166 | |
| 167 | #endif /* _KERNEL */ |
| 168 | |
| 169 | /* |
| 170 | * This structure maps out the global data that needs to be kept on a |
| 171 | * per-cpu basis. The members are accessed via the PCPU_GET/SET/PTR |
| 172 | * macros defined in <machine/pcpu.h>. Machine dependent fields are |
| 173 | * defined in the PCPU_MD_FIELDS macro defined in <machine/pcpu.h>. |
| 174 | */ |
| 175 | struct pcpu { |
| 176 | 	struct thread	*pc_curthread;		/* Current thread */ |
| 177 | 	struct thread	*pc_idlethread;		/* Idle thread */ |
| 178 | 	struct thread	*pc_fpcurthread;	/* Fp state owner */ |
| 179 | 	struct thread	*pc_deadthread;		/* Zombie thread or NULL */ |
| 180 | 	struct pcb	*pc_curpcb;		/* Current pcb */ |
| 181 | 	void		*pc_sched;		/* Scheduler state */ |
| 182 | 	uint64_t	pc_switchtime;		/* cpu_ticks() at last csw */ |
| 183 | 	int		pc_switchticks;		/* `ticks' at last csw */ |
| 184 | 	u_int		pc_cpuid;		/* This cpu number */ |
| 185 | 	STAILQ_ENTRY(pcpu) pc_allcpu; |
| 186 | 	struct lock_list_entry *pc_spinlocks; |
| 187 | 	long		pc_cp_time[CPUSTATES];	/* statclock ticks */ |
| 188 | 	struct _device	*pc_device;		/* CPU device handle */ |
| 189 | 	void		*pc_netisr;		/* netisr SWI cookie */ |
| 190 | 	int8_t		pc_vfs_freevnodes;	/* freevnodes counter */ |
| 191 | 	char		pc_unused1[3];		/* unused pad */ |
| 192 | 	int		pc_domain;		/* Memory domain. */ |
| 193 | 	struct rm_queue	pc_rm_queue;		/* rmlock list of trackers */ |
| 194 | 	uintptr_t	pc_dynamic;		/* Dynamic per-cpu data area */ |
| 195 | 	uint64_t	pc_early_dummy_counter;	/* Startup time counter(9) */ |
| 196 | 	uintptr_t	pc_zpcpu_offset;	/* Offset into zpcpu allocs */ |
| 197 | |
| 198 | 	/* |
| 199 | 	 * Keep MD fields last, so that CPU-specific variations on a |
| 200 | 	 * single architecture don't result in offset variations of |
| 201 | 	 * the machine-independent fields of the pcpu. Even though |
| 202 | 	 * the pcpu structure is private to the kernel, some ports |
| 203 | 	 * (e.g., lsof, part of gtop) define _KERNEL and include this |
| 204 | 	 * header. While strictly speaking this is wrong, there's no |
| 205 | 	 * reason not to keep the offsets of the MI fields constant |
| 206 | 	 * if only to make kernel debugging easier. |
| 207 | 	 */ |
| 208 | 	PCPU_MD_FIELDS; |
| 209 | } __aligned(CACHE_LINE_SIZE); |
| 210 | |
| 211 | #ifdef _KERNEL |
| 212 | |
| 213 | STAILQ_HEAD(cpuhead, pcpu); |
| 214 | |
| 215 | extern struct cpuhead cpuhead; |
| 216 | extern struct pcpu *cpuid_to_pcpu[]; |
| 217 | |
| 218 | #define	curcpu		PCPU_GET(cpuid) |
| 219 | #define	curvidata	PCPU_GET(vidata) |
| 220 | |
| 221 | #define UMA_PCPU_ALLOC_SIZE		PAGE_SIZE |
| 222 | |
| 223 | #include <machine/pcpu_aux.h> |
| 224 | |
| 225 | #ifndef curthread |
| 226 | #define	curthread	PCPU_GET(curthread) |
| 227 | #endif |
| 228 | #define	curproc		(curthread->td_proc) |
| 229 | |
| 230 | #ifndef ZPCPU_ASSERT_PROTECTED |
| 231 | #define ZPCPU_ASSERT_PROTECTED() MPASS(curthread->td_critnest > 0) |
| 232 | #endif |
| 233 | |
| 234 | #ifndef zpcpu_offset_cpu |
| 235 | #define zpcpu_offset_cpu(cpu)	(UMA_PCPU_ALLOC_SIZE * cpu) |
| 236 | #endif |
| 237 | #ifndef zpcpu_offset |
| 238 | #define zpcpu_offset()		(PCPU_GET(zpcpu_offset)) |
| 239 | #endif |
| 240 | |
| 241 | #ifndef zpcpu_base_to_offset |
| 242 | #define zpcpu_base_to_offset(base) (base) |
| 243 | #endif |
| 244 | #ifndef zpcpu_offset_to_base |
| 245 | #define zpcpu_offset_to_base(base) (base) |
| 246 | #endif |
| 247 | |
| 248 | /* Accessor to elements allocated via UMA_ZONE_PCPU zone. */ |
| 249 | #define zpcpu_get(base) ({								\ |
| 250 | 	__typeof(base) _ptr = (void *)((char *)(base) + zpcpu_offset());		\ |
| 251 | 	_ptr;										\ |
| 252 | }) |
| 253 | |
| 254 | #define zpcpu_get_cpu(base, cpu) ({							\ |
| 255 | 	__typeof(base) _ptr = (void *)((char *)(base) +	zpcpu_offset_cpu(cpu));		\ |
| 256 | 	_ptr;										\ |
| 257 | }) |
| 258 | |
| 259 | /* |
| 260 | * This operation is NOT atomic and does not post any barriers. |
| 261 | * If you use this the assumption is that the target CPU will not |
| 262 | * be modifying this variable. |
| 263 | * If you need atomicity use xchg. |
| 264 | * */ |
| 265 | #define zpcpu_replace(base, val) ({					\ |
| 266 | 	__typeof(val) *_ptr = zpcpu_get(base);				\ |
| 267 | 	__typeof(val) _old;						\ |
| 268 | 									\ |
| 269 | 	_old = *_ptr;							\ |
| 270 | 	*_ptr = val;							\ |
| 271 | 	_old;								\ |
| 272 | }) |
| 273 | |
| 274 | #define zpcpu_replace_cpu(base, val, cpu) ({				\ |
| 275 | 	__typeof(val) *_ptr = zpcpu_get_cpu(base, cpu);			\ |
| 276 | 	__typeof(val) _old;						\ |
| 277 | 									\ |
| 278 | 	_old = *_ptr;							\ |
| 279 | 	*_ptr = val;							\ |
| 280 | 	_old;								\ |
| 281 | }) |
| 282 | |
| 283 | #ifndef zpcpu_set_protected |
| 284 | #define zpcpu_set_protected(base, val) ({				\ |
| 285 | 	ZPCPU_ASSERT_PROTECTED();					\ |
| 286 | 	__typeof(val) *_ptr = zpcpu_get(base);				\ |
| 287 | 									\ |
| 288 | 	*_ptr = (val);							\ |
| 289 | }) |
| 290 | #endif |
| 291 | |
| 292 | #ifndef zpcpu_add_protected |
| 293 | #define zpcpu_add_protected(base, val) ({				\ |
| 294 | 	ZPCPU_ASSERT_PROTECTED();					\ |
| 295 | 	__typeof(val) *_ptr = zpcpu_get(base);				\ |
| 296 | 									\ |
| 297 | 	*_ptr += (val);							\ |
| 298 | }) |
| 299 | #endif |
| 300 | |
| 301 | #ifndef zpcpu_sub_protected |
| 302 | #define zpcpu_sub_protected(base, val) ({				\ |
| 303 | 	ZPCPU_ASSERT_PROTECTED();					\ |
| 304 | 	__typeof(val) *_ptr = zpcpu_get(base);				\ |
| 305 | 									\ |
| 306 | 	*_ptr -= (val);							\ |
| 307 | }) |
| 308 | #endif |
| 309 | |
| 310 | /* |
| 311 | * Machine dependent callouts. cpu_pcpu_init() is responsible for |
| 312 | * initializing machine dependent fields of struct pcpu, and |
| 313 | * db_show_mdpcpu() is responsible for handling machine dependent |
| 314 | * fields for the DDB 'show pcpu' command. |
| 315 | */ |
| 316 | void	cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size); |
| 317 | void	db_show_mdpcpu(struct pcpu *pcpu); |
| 318 | |
| 319 | void	*dpcpu_alloc(int size); |
| 320 | void	dpcpu_copy(void *s, int size); |
| 321 | void	dpcpu_free(void *s, int size); |
| 322 | void	dpcpu_init(void *dpcpu, int cpuid); |
| 323 | void	pcpu_destroy(struct pcpu *pcpu); |
| 324 | struct	pcpu *pcpu_find(u_int cpuid); |
| 325 | void	pcpu_init(struct pcpu *pcpu, int cpuid, size_t size); |
| 326 | |
| 327 | #endif /* _KERNEL */ |
| 328 | |
| 329 | #endif /* !_SYS_PCPU_H_ */ |