1 /* $NetBSD: vm_machdep.c,v 1.102 2019/02/11 07:51:46 macallan Exp $ */
2
3 /*
4 * Copyright (c) 1996-2002 Eduardo Horvath. All rights reserved.
5 * Copyright (c) 1996
6 * The President and Fellows of Harvard College. All rights reserved.
7 * Copyright (c) 1992, 1993
8 * The Regents of the University of California. All rights reserved.
9 *
10 * This software was developed by the Computer Systems Engineering group
11 * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
12 * contributed to Berkeley.
13 *
14 * All advertising materials mentioning features or use of this software
15 * must display the following acknowledgement:
16 * This product includes software developed by the University of
17 * California, Lawrence Berkeley Laboratory.
18 * This product includes software developed by Harvard University.
19 *
20 * Redistribution and use in source and binary forms, with or without
21 * modification, are permitted provided that the following conditions
22 * are met:
23 * 1. Redistributions of source code must retain the above copyright
24 * notice, this list of conditions and the following disclaimer.
25 * 2. Redistributions in binary form must reproduce the above copyright
26 * notice, this list of conditions and the following disclaimer in the
27 * documentation and/or other materials provided with the distribution.
28 * 3. All advertising materials mentioning features or use of this software
29 * must display the following acknowledgement:
30 * This product includes software developed by Harvard University.
31 * This product includes software developed by the University of
32 * California, Berkeley and its contributors.
33 * 4. Neither the name of the University nor the names of its contributors
34 * may be used to endorse or promote products derived from this software
35 * without specific prior written permission.
36 *
37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47 * SUCH DAMAGE.
48 *
49 * @(#)vm_machdep.c 8.2 (Berkeley) 9/23/93
50 */
51
52 #include <sys/cdefs.h>
53 __KERNEL_RCSID(0, "$NetBSD: vm_machdep.c,v 1.102 2019/02/11 07:51:46 macallan Exp $");
54
55 #include "opt_multiprocessor.h"
56
57 #include <sys/param.h>
58 #include <sys/systm.h>
59 #include <sys/proc.h>
60 #include <sys/core.h>
61 #include <sys/buf.h>
62 #include <sys/exec.h>
63 #include <sys/vnode.h>
64 #include <sys/cpu.h>
65
66 #include <uvm/uvm_extern.h>
67
68 #include <machine/cpu.h>
69 #include <machine/frame.h>
70 #include <machine/pcb.h>
71 #include <machine/trap.h>
72 #include <sys/bus.h>
73
74 /*
75 * Map a user I/O request into kernel virtual address space.
76 * Note: the pages are already locked by uvm_vslock(), so we
77 * do not need to pass an access_type to pmap_enter().
78 */
79 int
vmapbuf(struct buf * bp,vsize_t len)80 vmapbuf(struct buf *bp, vsize_t len)
81 {
82 struct pmap *upmap;
83 vaddr_t uva; /* User VA (map from) */
84 vaddr_t kva; /* Kernel VA (new to) */
85 paddr_t pa; /* physical address */
86 vsize_t off;
87
88 if ((bp->b_flags & B_PHYS) == 0)
89 panic("vmapbuf");
90
91 bp->b_saveaddr = bp->b_data;
92 uva = trunc_page((vaddr_t)bp->b_data);
93 off = (vaddr_t)bp->b_data - uva;
94 len = round_page(off + len);
95 kva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
96 bp->b_data = (void *)(kva + off);
97
98 upmap = vm_map_pmap(&bp->b_proc->p_vmspace->vm_map);
99 do {
100 if (pmap_extract(upmap, uva, &pa) == FALSE)
101 panic("vmapbuf: null page frame");
102 /* Now map the page into kernel space. */
103 pmap_kenter_pa(kva, pa, VM_PROT_READ | VM_PROT_WRITE, 0);
104
105 uva += PAGE_SIZE;
106 kva += PAGE_SIZE;
107 len -= PAGE_SIZE;
108 } while (len);
109 pmap_update(pmap_kernel());
110
111 return 0;
112 }
113
114 /*
115 * Unmap a previously-mapped user I/O request.
116 */
117 void
vunmapbuf(struct buf * bp,vsize_t len)118 vunmapbuf(struct buf *bp, vsize_t len)
119 {
120 vaddr_t kva;
121 vsize_t off;
122
123 if ((bp->b_flags & B_PHYS) == 0)
124 panic("vunmapbuf");
125
126 kva = trunc_page((vaddr_t)bp->b_data);
127 off = (vaddr_t)bp->b_data - kva;
128 len = round_page(off + len);
129 pmap_kremove(kva, len);
130 uvm_km_free(kernel_map, kva, len, UVM_KMF_VAONLY);
131 bp->b_data = bp->b_saveaddr;
132 bp->b_saveaddr = NULL;
133 }
134
135 void
cpu_proc_fork(struct proc * p1,struct proc * p2)136 cpu_proc_fork(struct proc *p1, struct proc *p2)
137 {
138
139 p2->p_md.md_flags = p1->p_md.md_flags;
140 }
141
142
143 /*
144 * The offset of the topmost frame in the kernel stack.
145 */
146 #ifdef __arch64__
147 #define TOPFRAMEOFF (USPACE-sizeof(struct trapframe)-CC64FSZ)
148 #define STACK_OFFSET BIAS
149 #else
150 #undef trapframe
151 #define trapframe trapframe64
152 #undef rwindow
153 #define rwindow rwindow32
154 #define TOPFRAMEOFF (USPACE-sizeof(struct trapframe)-CC64FSZ)
155 #define STACK_OFFSET 0
156 #endif
157
158 #ifdef DEBUG
159 char cpu_forkname[] = "cpu_lwp_fork()";
160 #endif
161
162 /*
163 * Finish a fork operation, with lwp l2 nearly set up.
164 * Copy and update the pcb and trap frame, making the child ready to run.
165 *
166 * Rig the child's kernel stack so that it will start out in
167 * lwp_trampoline() and call child_return() with l2 as an
168 * argument. This causes the newly-created child process to go
169 * directly to user level with an apparent return value of 0 from
170 * fork(), while the parent process returns normally.
171 *
172 * l1 is the process being forked; if l1 == &lwp0, we are creating
173 * a kernel thread, and the return path and argument are specified with
174 * `func' and `arg'.
175 *
176 * If an alternate user-level stack is requested (with non-zero values
177 * in both the stack and stacksize args), set up the user stack pointer
178 * accordingly.
179 */
180 void lwp_trampoline(void);
181 void
cpu_lwp_fork(register struct lwp * l1,register struct lwp * l2,void * stack,size_t stacksize,void (* func)(void *),void * arg)182 cpu_lwp_fork(register struct lwp *l1, register struct lwp *l2, void *stack, size_t stacksize, void (*func)(void *), void *arg)
183 {
184 struct pcb *opcb = lwp_getpcb(l1);
185 struct pcb *npcb = lwp_getpcb(l2);
186 struct trapframe *tf2;
187 struct rwindow *rp;
188
189 /*
190 * Save all user registers to l1's stack or, in the case of
191 * user registers and invalid stack pointers, to opcb.
192 * We then copy the whole pcb to l2; when switch() selects l2
193 * to run, it will run at the `lwp_trampoline' stub, rather
194 * than returning at the copying code below.
195 *
196 * If process l1 has an FPU state, we must copy it. If it is
197 * the FPU user, we must save the FPU state first.
198 */
199
200 #ifdef NOTDEF_DEBUG
201 printf("cpu_lwp_fork()\n");
202 #endif
203 if (l1 == curlwp) {
204 write_user_windows();
205
206 /*
207 * We're in the kernel, so we don't really care about
208 * %ccr or %asi. We do want to duplicate %pstate and %cwp.
209 */
210 opcb->pcb_pstate = getpstate();
211 opcb->pcb_cwp = getcwp();
212 }
213 #ifdef DIAGNOSTIC
214 else if (l1 != &lwp0)
215 panic("cpu_lwp_fork: curlwp");
216 #endif
217 #ifdef DEBUG
218 /* prevent us from having NULL lastcall */
219 opcb->lastcall = cpu_forkname;
220 #else
221 opcb->lastcall = NULL;
222 #endif
223 memcpy(npcb, opcb, sizeof(struct pcb));
224 if (l1->l_md.md_fpstate) {
225 fpusave_lwp(l1, true);
226 l2->l_md.md_fpstate = pool_cache_get(fpstate_cache, PR_WAITOK);
227 memcpy(l2->l_md.md_fpstate, l1->l_md.md_fpstate,
228 sizeof(struct fpstate64));
229 } else
230 l2->l_md.md_fpstate = NULL;
231
232 /*
233 * Setup (kernel) stack frame that will by-pass the child
234 * out of the kernel. (The trap frame invariably resides at
235 * the tippity-top of the u. area.)
236 */
237 tf2 = l2->l_md.md_tf = (struct trapframe *)
238 ((long)npcb + USPACE - sizeof(*tf2));
239
240 /* Copy parent's trapframe */
241 *tf2 = *(struct trapframe *)((long)opcb + USPACE - sizeof(*tf2));
242
243 /*
244 * If specified, give the child a different stack.
245 */
246 if (stack != NULL)
247 tf2->tf_out[6] = (uint64_t)(u_long)stack + stacksize;
248
249 /*
250 * Need to create a %tstate if we are forking our first userland
251 * process - in all other cases we inherit from the parent.
252 */
253 if (l2->l_proc->p_pid == 1)
254 tf2->tf_tstate = (ASI_PRIMARY_NO_FAULT<<TSTATE_ASI_SHIFT) |
255 ((PSTATE_USER)<<TSTATE_PSTATE_SHIFT);
256
257 /*
258 * Set return values in child mode and clear condition code,
259 * in case we end up running a signal handler before returning
260 * to userland.
261 */
262 tf2->tf_out[0] = 0;
263 tf2->tf_out[1] = 1;
264 tf2->tf_tstate &= ~TSTATE_CCR;
265
266 /* Construct kernel frame to return to in cpu_switch() */
267 rp = (struct rwindow *)((u_long)npcb + TOPFRAMEOFF);
268 *rp = *(struct rwindow *)((u_long)opcb + TOPFRAMEOFF);
269
270 rp->rw_local[0] = (long)func; /* Function to call */
271 rp->rw_local[1] = (long)arg; /* and its argument */
272 rp->rw_local[2] = (long)l2; /* new lwp */
273
274 npcb->pcb_pc = (long)lwp_trampoline - 8;
275 npcb->pcb_sp = (long)rp - STACK_OFFSET;
276 }
277
278 static inline void
fpusave_cpu(bool save)279 fpusave_cpu(bool save)
280 {
281 struct lwp *l = fplwp;
282
283 if (l == NULL)
284 return;
285
286 if (save)
287 savefpstate(l->l_md.md_fpstate);
288 else
289 clearfpstate();
290
291 fplwp = NULL;
292 }
293
294 void
fpusave_lwp(struct lwp * l,bool save)295 fpusave_lwp(struct lwp *l, bool save)
296 {
297 #ifdef MULTIPROCESSOR
298 volatile struct cpu_info *ci;
299
300 if (l == fplwp) {
301 int s = intr_disable();
302 fpusave_cpu(save);
303 intr_restore(s);
304 return;
305 }
306
307 for (ci = cpus; ci != NULL; ci = ci->ci_next) {
308 int spincount, retrycount=0;
309
310 if (ci == curcpu() || !CPUSET_HAS(cpus_active, ci->ci_index))
311 continue;
312 if (ci->ci_fplwp != l)
313 continue;
314 again:
315 sparc64_send_ipi(ci->ci_cpuid, save ?
316 sparc64_ipi_save_fpstate :
317 sparc64_ipi_drop_fpstate, (uintptr_t)l, 0);
318
319 spincount = 0;
320 while (ci->ci_fplwp == l) {
321 membar_Sync();
322 spincount++;
323 if (spincount > 10000000) {
324 printf("fpusave_lwp ipi didn't (%d)\n", retrycount);
325 retrycount++;
326 goto again;
327 }
328 }
329 if (retrycount > 0) printf("spincount %d\n", spincount);
330 break;
331 }
332 #else
333 if (l == fplwp)
334 fpusave_cpu(save);
335 #endif
336 }
337
338
339 void
cpu_lwp_free(struct lwp * l,int proc)340 cpu_lwp_free(struct lwp *l, int proc)
341 {
342
343 if (l->l_md.md_fpstate != NULL)
344 fpusave_lwp(l, false);
345 }
346
347 void
cpu_lwp_free2(struct lwp * l)348 cpu_lwp_free2(struct lwp *l)
349 {
350 struct fpstate64 *fs;
351
352 if ((fs = l->l_md.md_fpstate) != NULL)
353 pool_cache_put(fpstate_cache, fs);
354 }
355
356 int
cpu_lwp_setprivate(lwp_t * l,void * addr)357 cpu_lwp_setprivate(lwp_t *l, void *addr)
358 {
359 struct trapframe *tf = l->l_md.md_tf;
360
361 tf->tf_global[7] = (uintptr_t)addr;
362
363 return 0;
364 }
365