xref: /netbsd/sys/arch/sparc/sparc/vm_machdep.c (revision 6550d01e)
1 /*	$NetBSD: vm_machdep.c,v 1.103 2011/01/14 02:06:32 rmind Exp $ */
2 
3 /*
4  * Copyright (c) 1996
5  *	The President and Fellows of Harvard College. All rights reserved.
6  * Copyright (c) 1992, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  *
9  * This software was developed by the Computer Systems Engineering group
10  * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
11  * contributed to Berkeley.
12  *
13  * All advertising materials mentioning features or use of this software
14  * must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Lawrence Berkeley Laboratory.
17  *	This product includes software developed by Harvard University.
18  *
19  * Redistribution and use in source and binary forms, with or without
20  * modification, are permitted provided that the following conditions
21  * are met:
22  * 1. Redistributions of source code must retain the above copyright
23  *    notice, this list of conditions and the following disclaimer.
24  * 2. Redistributions in binary form must reproduce the above copyright
25  *    notice, this list of conditions and the following disclaimer in the
26  *    documentation and/or other materials provided with the distribution.
27  * 3. All advertising materials mentioning features or use of this software
28  *    must display the following acknowledgement:
29  *	This product includes software developed by Harvard University.
30  *	This product includes software developed by the University of
31  *	California, Berkeley and its contributors.
32  * 4. Neither the name of the University nor the names of its contributors
33  *    may be used to endorse or promote products derived from this software
34  *    without specific prior written permission.
35  *
36  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
37  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
38  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
39  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
40  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
41  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
42  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
43  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
44  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
45  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
46  * SUCH DAMAGE.
47  *
48  *	@(#)vm_machdep.c	8.2 (Berkeley) 9/23/93
49  */
50 
51 #include <sys/cdefs.h>
52 __KERNEL_RCSID(0, "$NetBSD: vm_machdep.c,v 1.103 2011/01/14 02:06:32 rmind Exp $");
53 
54 #include "opt_multiprocessor.h"
55 
56 #include <sys/param.h>
57 #include <sys/systm.h>
58 #include <sys/proc.h>
59 #include <sys/core.h>
60 #include <sys/malloc.h>
61 #include <sys/buf.h>
62 #include <sys/exec.h>
63 #include <sys/vnode.h>
64 #include <sys/simplelock.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 
73 #include <sparc/sparc/cpuvar.h>
74 
75 /*
76  * Map a user I/O request into kernel virtual address space.
77  * Note: the pages are already locked by uvm_vslock(), so we
78  * do not need to pass an access_type to pmap_enter().
79  */
80 void
81 vmapbuf(struct buf *bp, vsize_t len)
82 {
83 	struct pmap *upmap, *kpmap;
84 	vaddr_t uva;	/* User VA (map from) */
85 	vaddr_t kva;	/* Kernel VA (new to) */
86 	paddr_t pa; 	/* physical address */
87 	vsize_t off;
88 
89 	if ((bp->b_flags & B_PHYS) == 0)
90 		panic("vmapbuf");
91 
92 	/*
93 	 * XXX:  It might be better to round/trunc to a
94 	 * segment boundary to avoid VAC problems!
95 	 */
96 	bp->b_saveaddr = bp->b_data;
97 	uva = trunc_page((vaddr_t)bp->b_data);
98 	off = (vaddr_t)bp->b_data - uva;
99 	len = round_page(off + len);
100 	kva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA);
101 	bp->b_data = (void *)(kva + off);
102 
103 	/*
104 	 * We have to flush any write-back cache on the
105 	 * user-space mappings so our new mappings will
106 	 * have the correct contents.
107 	 */
108 	if (CACHEINFO.c_vactype != VAC_NONE)
109 		cache_flush((void *)uva, len);
110 
111 	upmap = vm_map_pmap(&bp->b_proc->p_vmspace->vm_map);
112 	kpmap = vm_map_pmap(kernel_map);
113 	do {
114 		if (pmap_extract(upmap, uva, &pa) == false)
115 			panic("vmapbuf: null page frame");
116 		/* Now map the page into kernel space. */
117 		pmap_enter(kpmap, kva, pa,
118 		    VM_PROT_READ|VM_PROT_WRITE, PMAP_WIRED);
119 		uva += PAGE_SIZE;
120 		kva += PAGE_SIZE;
121 		len -= PAGE_SIZE;
122 	} while (len);
123 	pmap_update(kpmap);
124 }
125 
126 /*
127  * Unmap a previously-mapped user I/O request.
128  */
129 void
130 vunmapbuf(struct buf *bp, vsize_t len)
131 {
132 	vaddr_t kva;
133 	vsize_t off;
134 
135 	if ((bp->b_flags & B_PHYS) == 0)
136 		panic("vunmapbuf");
137 
138 	kva = trunc_page((vaddr_t)bp->b_data);
139 	off = (vaddr_t)bp->b_data - kva;
140 	len = round_page(off + len);
141 	pmap_remove(vm_map_pmap(kernel_map), kva, kva + len);
142 	pmap_update(vm_map_pmap(kernel_map));
143 	uvm_km_free(kernel_map, kva, len, UVM_KMF_VAONLY);
144 	bp->b_data = bp->b_saveaddr;
145 	bp->b_saveaddr = NULL;
146 
147 #if 0	/* XXX: The flush above is sufficient, right? */
148 	if (CACHEINFO.c_vactype != VAC_NONE)
149 		cpuinfo.cache_flush(bp->b_data, len);
150 #endif
151 }
152 
153 
154 void
155 cpu_proc_fork(struct proc *p1, struct proc *p2)
156 {
157 
158 	p2->p_md.md_flags = p1->p_md.md_flags;
159 }
160 
161 
162 /*
163  * The offset of the topmost frame in the kernel stack.
164  */
165 #define	TOPFRAMEOFF (USPACE-sizeof(struct trapframe)-sizeof(struct frame))
166 
167 /*
168  * Finish a fork operation, with process l2 nearly set up.
169  * Copy and update the pcb and trap frame, making the child ready to run.
170  *
171  * Rig the child's kernel stack so that it will start out in
172  * lwp_trampoline() and call child_return() with l2 as an
173  * argument. This causes the newly-created child process to go
174  * directly to user level with an apparent return value of 0 from
175  * fork(), while the parent process returns normally.
176  *
177  * l1 is the process being forked; if l1 == &lwp0, we are creating
178  * a kernel thread, and the return path and argument are specified with
179  * `func' and `arg'.
180  *
181  * If an alternate user-level stack is requested (with non-zero values
182  * in both the stack and stacksize args), set up the user stack pointer
183  * accordingly.
184  */
185 void
186 cpu_lwp_fork(struct lwp *l1, struct lwp *l2,
187 	     void *stack, size_t stacksize,
188 	     void (*func)(void *), void *arg)
189 {
190 	struct pcb *opcb = lwp_getpcb(l1);
191 	struct pcb *npcb = lwp_getpcb(l2);
192 	struct trapframe *tf2;
193 	struct rwindow *rp;
194 
195 	/*
196 	 * Save all user registers to l1's stack or, in the case of
197 	 * user registers and invalid stack pointers, to opcb.
198 	 * We then copy the whole pcb to l2; when switch() selects l2
199 	 * to run, it will run at the `lwp_trampoline' stub, rather
200 	 * than returning at the copying code below.
201 	 *
202 	 * If process l1 has an FPU state, we must copy it.  If it is
203 	 * the FPU user, we must save the FPU state first.
204 	 */
205 
206 	if (l1 == curlwp) {
207 		write_user_windows();
208 		opcb->pcb_psr = getpsr();
209 	}
210 #ifdef DIAGNOSTIC
211 	else if (l1 != &lwp0)	/* XXX is this valid? */
212 		panic("cpu_lwp_fork: curlwp");
213 #endif
214 
215 	memcpy((void *)npcb, (void *)opcb, sizeof(struct pcb));
216 	if (l1->l_md.md_fpstate != NULL) {
217 		struct cpu_info *cpi;
218 		int s;
219 
220 		l2->l_md.md_fpstate = malloc(sizeof(struct fpstate),
221 		    M_SUBPROC, M_WAITOK);
222 
223 		FPU_LOCK(s);
224 		if ((cpi = l1->l_md.md_fpu) != NULL) {
225 			if (cpi->fplwp != l1)
226 				panic("FPU(%d): fplwp %p",
227 					cpi->ci_cpuid, cpi->fplwp);
228 			if (l1 == cpuinfo.fplwp)
229 				savefpstate(l1->l_md.md_fpstate);
230 #if defined(MULTIPROCESSOR)
231 			else
232 				XCALL1(ipi_savefpstate, l1->l_md.md_fpstate,
233 					1 << cpi->ci_cpuid);
234 #endif
235 		}
236 		memcpy(l2->l_md.md_fpstate, l1->l_md.md_fpstate,
237 		    sizeof(struct fpstate));
238 		FPU_UNLOCK(s);
239 	} else
240 		l2->l_md.md_fpstate = NULL;
241 
242 	l2->l_md.md_fpu = NULL;
243 
244 	/*
245 	 * Setup (kernel) stack frame that will by-pass the child
246 	 * out of the kernel. (The trap frame invariably resides at
247 	 * the tippity-top of the u. area.)
248 	 */
249 	tf2 = l2->l_md.md_tf = (struct trapframe *)
250 			((int)npcb + USPACE - sizeof(*tf2));
251 
252 	/* Copy parent's trapframe */
253 	*tf2 = *(struct trapframe *)((int)opcb + USPACE - sizeof(*tf2));
254 
255 	/*
256 	 * If specified, give the child a different stack.
257 	 */
258 	if (stack != NULL)
259 		tf2->tf_out[6] = (u_int)stack + stacksize;
260 
261 	/*
262 	 * The fork system call always uses the old system call
263 	 * convention; clear carry and skip trap instruction as
264 	 * in syscall().
265 	 * note: lwp_trampoline() sets a fresh psr when returning
266 	 * to user mode.
267 	 */
268 	/*tf2->tf_psr &= ~PSR_C;   -* success */
269 	tf2->tf_pc = tf2->tf_npc;
270 	tf2->tf_npc = tf2->tf_pc + 4;
271 
272 	/* Set return values in child mode */
273 	tf2->tf_out[0] = 0;
274 	tf2->tf_out[1] = 1;
275 
276 	/* Construct kernel frame to return to in cpu_switch() */
277 	rp = (struct rwindow *)((u_int)npcb + TOPFRAMEOFF);
278 	/**rp = *(struct rwindow *)((u_int)opcb + TOPFRAMEOFF);*/
279 	rp->rw_local[0] = (int)func;		/* Function to call */
280 	rp->rw_local[1] = (int)arg;		/* and its argument */
281 	rp->rw_local[2] = (int)l2;		/* new LWP */
282 
283 	npcb->pcb_pc = (int)lwp_trampoline - 8;
284 	npcb->pcb_sp = (int)rp;
285 	npcb->pcb_psr &= ~PSR_CWP;	/* Run in window #0 */
286 	npcb->pcb_wim = 1;		/* Fence at window #1 */
287 }
288 
289 /*
290  * Cleanup FPU state.
291  */
292 void
293 cpu_lwp_free(struct lwp *l, int proc)
294 {
295 	struct fpstate *fs;
296 
297 	if ((fs = l->l_md.md_fpstate) != NULL) {
298 		struct cpu_info *cpi;
299 		int s;
300 
301 		FPU_LOCK(s);
302 		if ((cpi = l->l_md.md_fpu) != NULL) {
303 			if (cpi->fplwp != l)
304 				panic("FPU(%d): fplwp %p",
305 					cpi->ci_cpuid, cpi->fplwp);
306 			if (l == cpuinfo.fplwp)
307 				savefpstate(fs);
308 #if defined(MULTIPROCESSOR)
309 			else
310 				XCALL1(ipi_savefpstate, fs, 1 << cpi->ci_cpuid);
311 #endif
312 			cpi->fplwp = NULL;
313 		}
314 		l->l_md.md_fpu = NULL;
315 		FPU_UNLOCK(s);
316 	}
317 }
318 
319 void
320 cpu_lwp_free2(struct lwp *l)
321 {
322 	struct fpstate *fs;
323 
324 	if ((fs = l->l_md.md_fpstate) != NULL)
325 		free((void *)fs, M_SUBPROC);
326 }
327 
328 void
329 cpu_setfunc(struct lwp *l, void (*func)(void *), void *arg)
330 {
331 	struct pcb *pcb = lwp_getpcb(l);
332 	/*struct trapframe *tf = l->l_md.md_tf;*/
333 	struct rwindow *rp;
334 
335 	/* Construct kernel frame to return to in cpu_switch() */
336 	rp = (struct rwindow *)((u_int)pcb + TOPFRAMEOFF);
337 	rp->rw_local[0] = (int)func;		/* Function to call */
338 	rp->rw_local[1] = (int)arg;		/* and its argument */
339 
340 	pcb->pcb_pc = (int)lwp_setfunc_trampoline - 8;
341 	pcb->pcb_sp = (int)rp;
342 	pcb->pcb_psr &= ~PSR_CWP;	/* Run in window #0 */
343 	pcb->pcb_wim = 1;		/* Fence at window #1 */
344 }
345