2 * Copyright (c) 2007 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * $DragonFly: src/sys/platform/vkernel/i386/mp.c,v 1.3 2007/07/01 04:02:33 dillon Exp $
38 #include <sys/interrupt.h>
39 #include <sys/kernel.h>
40 #include <sys/memrange.h>
42 #include <sys/types.h>
44 #include <vm/vm_extern.h>
45 #include <vm/vm_kern.h>
46 #include <vm/vm_object.h>
47 #include <vm/vm_page.h>
49 #include <machine/cpufunc.h>
50 #include <machine/globaldata.h>
51 #include <machine/md_var.h>
52 #include <machine/pmap.h>
53 #include <machine/smp.h>
54 #include <machine/tls.h>
61 extern pt_entry_t *KPTphys;
63 volatile u_int stopped_cpus;
64 cpumask_t smp_active_mask = 1; /* which cpus are ready for IPIs etc? */
65 static int boot_address;
66 static cpumask_t smp_startup_mask = 1; /* which cpus have been started */
67 int mp_naps; /* # of Applications processors */
70 /* function prototypes XXX these should go elsewhere */
71 void bootstrap_idle(void);
72 void single_cpu_ipi(int, int, int);
73 void selected_cpu_ipi(u_int, int, int);
75 void ipi_handler(int);
80 /* AP uses this during bootstrap. Do not staticize. */
85 /* XXX these need to go into the appropriate header file */
86 static int start_all_aps(u_int);
87 void init_secondary(void);
88 void *start_ap(void *);
91 * Get SMP fully working before we start initializing devices.
98 cpumask_t ncpus_mask = 0;
100 for (i = 1; i <= ncpus; i++)
101 ncpus_mask |= (1 << i);
105 kprintf("Finish MP startup\n");
107 /* build our map of 'other' CPUs */
108 mycpu->gd_other_cpus = smp_startup_mask & ~(1 << mycpu->gd_cpuid);
111 * Let the other cpu's finish initializing and build their map
115 while (smp_active_mask != smp_startup_mask)
118 while (try_mplock() == 0)
121 kprintf("Active CPU Mask: %08x\n", smp_active_mask);
124 SYSINIT(finishsmp, SI_BOOT2_FINISH_SMP, SI_ORDER_FIRST, ap_finish, NULL)
128 start_ap(void *arg __unused)
134 return(NULL); /* NOTREACHED */
137 /* storage for AP thread IDs */
138 pthread_t ap_tids[MAXCPU];
145 /* XXX testing 2 cpus */
150 /* ncpus2 -- ncpus rounded down to the nearest power of 2 */
151 for (shift = 0; (1 << shift) <= ncpus; ++shift)
154 ncpus2_shift = shift;
156 ncpus2_mask = ncpus2 - 1;
158 /* ncpus_fit -- ncpus rounded up to the nearest power of 2 */
159 if ((1 << shift) < ncpus)
161 ncpus_fit = 1 << shift;
162 ncpus_fit_mask = ncpus_fit - 1;
165 * cpu0 initialization
167 mycpu->gd_ipiq = (void *)kmem_alloc(&kernel_map,
168 sizeof(lwkt_ipiq) * ncpus);
169 bzero(mycpu->gd_ipiq, sizeof(lwkt_ipiq) * ncpus);
174 start_all_aps(boot_address);
183 kprintf("DragonFly/MP: Multiprocessor\n");
184 kprintf(" cpu0 (BSP)\n");
186 for (x = 1; x <= mp_naps; ++x)
187 kprintf(" cpu%d (AP)\n", x);
191 forward_fastint_remote(void *arg)
193 panic("XXX forward_fastint_remote()");
197 cpu_send_ipiq(int dcpu)
199 if ((1 << dcpu) & smp_active_mask)
200 if (pthread_kill(ap_tids[dcpu], SIGUSR1) != 0)
201 panic("pthread_kill failed in cpu_send_ipiq");
203 panic("XXX cpu_send_ipiq()");
215 single_cpu_ipi(int cpu, int vector, int delivery_mode)
217 kprintf("XXX single_cpu_ipi\n");
221 selected_cpu_ipi(u_int target, int vector, int delivery_mode)
225 int n = bsfl(target);
227 single_cpu_ipi(n, vector, delivery_mode);
235 map &= smp_active_mask;
241 if (pthread_kill(ap_tids[n], SIGSTOP) != 0)
242 panic("stop_cpus: pthread_kill failed");
246 panic("XXX stop_cpus()");
253 restart_cpus(u_int map)
255 map &= smp_active_mask;
261 if (pthread_kill(ap_tids[n], SIGCONT) != 0)
262 panic("restart_cpus: pthread_kill failed");
266 panic("XXX restart_cpus()");
276 * Adjust smp_startup_mask to signal the BSP that we have started
277 * up successfully. Note that we do not yet hold the BGL. The BSP
278 * is waiting for our signal.
280 * We can't set our bit in smp_active_mask yet because we are holding
281 * interrupts physically disabled and remote cpus could deadlock
282 * trying to send us an IPI.
284 smp_startup_mask |= 1 << mycpu->gd_cpuid;
288 * Interlock for finalization. Wait until mp_finish is non-zero,
289 * then get the MP lock.
291 * Note: We are in a critical section.
293 * Note: We have to synchronize td_mpcount to our desired MP state
294 * before calling cpu_try_mplock().
296 * Note: we are the idle thread, we can only spin.
298 * Note: The load fence is memory volatile and prevents the compiler
299 * from improperly caching mp_finish, and the cpu from improperly
303 while (mp_finish == 0) {
306 ++curthread->td_mpcount;
307 while (cpu_try_mplock() == 0)
310 /* BSP may have changed PTD while we're waiting for the lock */
313 /* Build our map of 'other' CPUs. */
314 mycpu->gd_other_cpus = smp_startup_mask & ~(1 << mycpu->gd_cpuid);
316 kprintf("SMP: AP CPU #%d Launched!\n", mycpu->gd_cpuid);
319 /* Set memory range attributes for this CPU to match the BSP */
322 * Once we go active we must process any IPIQ messages that may
323 * have been queued, because no actual IPI will occur until we
324 * set our bit in the smp_active_mask. If we don't the IPI
325 * message interlock could be left set which would also prevent
328 * The idle loop doesn't expect the BGL to be held and while
329 * lwkt_switch() normally cleans things up this is a special case
330 * because we returning almost directly into the idle loop.
332 * The idle thread is never placed on the runq, make sure
333 * nothing we've done put it there.
335 KKASSERT(curthread->td_mpcount == 1);
336 smp_active_mask |= 1 << mycpu->gd_cpuid;
338 mdcpu->gd_fpending = 0;
339 mdcpu->gd_ipending = 0;
340 initclocks_pcpu(); /* clock interrupts (via IPIs) */
344 * Releasing the mp lock lets the BSP finish up the SMP init
347 KKASSERT((curthread->td_flags & TDF_RUNQ) == 0);
354 struct mdglobaldata *md;
355 struct privatespace *ps;
357 ps = &CPU_prvspace[myid];
359 KKASSERT(ps->mdglobaldata.mi.gd_prvspace == ps);
362 * Setup the %gs for cpu #n. The mycpu macro works after this
365 tls_set_fs(&CPU_prvspace[myid], sizeof(struct privatespace));
367 md = mdcpu; /* loaded through %fs:0 (mdglobaldata.mi.gd_prvspace)*/
369 md->gd_common_tss.tss_esp0 = 0; /* not used until after switch */
370 md->gd_common_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
371 md->gd_common_tss.tss_ioopt = (sizeof md->gd_common_tss) << 16;
374 * Set to a known state:
375 * Set by mpboot.s: CR0_PG, CR0_PE
376 * Set by cpu_setregs: CR0_NE, CR0_MP, CR0_TS, CR0_WP, CR0_AM
381 start_all_aps(u_int boot_addr)
384 struct mdglobaldata *gd;
385 struct privatespace *ps;
389 struct lwp_params params;
393 * needed for ipis to initial thread
394 * FIXME: rename ap_tids?
396 ap_tids[0] = pthread_self();
398 for (x = 1; x <= mp_naps; x++)
400 /* Allocate space for the CPU's private space. */
401 va = (vm_offset_t)&CPU_prvspace[x];
402 for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) {
403 va =(vm_offset_t)&CPU_prvspace[x].mdglobaldata + i;
404 m = vm_page_alloc(&kernel_object, va, VM_ALLOC_SYSTEM);
405 pmap_kenter_quick(va, m->phys_addr);
408 for (i = 0; i < sizeof(CPU_prvspace[x].idlestack); i += PAGE_SIZE) {
409 va =(vm_offset_t)&CPU_prvspace[x].idlestack + i;
410 m = vm_page_alloc(&kernel_object, va, VM_ALLOC_SYSTEM);
411 pmap_kenter_quick(va, m->phys_addr);
414 gd = &CPU_prvspace[x].mdglobaldata; /* official location */
415 bzero(gd, sizeof(*gd));
416 gd->mi.gd_prvspace = ps = &CPU_prvspace[x];
418 /* prime data page for it to use */
419 mi_gdinit(&gd->mi, x);
423 gd->gd_CMAP1 = pmap_kpte((vm_offset_t)CPU_prvspace[x].CPAGE1);
424 gd->gd_CMAP2 = pmap_kpte((vm_offset_t)CPU_prvspace[x].CPAGE2);
425 gd->gd_CMAP3 = pmap_kpte((vm_offset_t)CPU_prvspace[x].CPAGE3);
426 gd->gd_PMAP1 = pmap_kpte((vm_offset_t)CPU_prvspace[x].PPAGE1);
427 gd->gd_CADDR1 = ps->CPAGE1;
428 gd->gd_CADDR2 = ps->CPAGE2;
429 gd->gd_CADDR3 = ps->CPAGE3;
430 gd->gd_PADDR1 = (vpte_t *)ps->PPAGE1;
433 gd->mi.gd_ipiq = (void *)kmem_alloc(&kernel_map, sizeof(lwkt_ipiq) * (mp_naps + 1));
434 bzero(gd->mi.gd_ipiq, sizeof(lwkt_ipiq) * (mp_naps + 1));
437 * Setup the AP boot stack
439 bootSTK = &ps->idlestack[UPAGES*PAGE_SIZE/2];
443 * Setup the AP's lwp, this is the 'cpu'
445 pthread_create(&ap_tids[x], NULL, start_ap, NULL);
447 while((smp_startup_mask & (1 << x)) == 0)
448 cpu_lfence(); /* XXX spin until the AP has started */
450 /* XXX hack, sleep for a second to let the APs start up */