board_bwct.c revision 177883
1/*-
2 * Copyright (c) 1994-1998 Mark Brinicombe.
3 * Copyright (c) 1994 Brini.
4 * All rights reserved.
5 *
6 * This code is derived from software written for Brini by Mark Brinicombe
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. All advertising materials mentioning features or use of this software
17 *    must display the following acknowledgement:
18 *      This product includes software developed by Brini.
19 * 4. The name of the company nor the name of the author may be used to
20 *    endorse or promote products derived from this software without specific
21 *    prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY BRINI ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26 * IN NO EVENT SHALL BRINI OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
27 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
28 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
29 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
34 *
35 * RiscBSD kernel project
36 *
37 * machdep.c
38 *
39 * Machine dependant functions for kernel setup
40 *
41 * This file needs a lot of work.
42 *
43 * Created      : 17/09/94
44 */
45
46#include "opt_msgbuf.h"
47#include "opt_at91.h"
48
49#include <sys/cdefs.h>
50__FBSDID("$FreeBSD: head/sys/arm/at91/kb920x_machdep.c 177883 2008-04-03 16:44:50Z imp $");
51
52#define _ARM32_BUS_DMA_PRIVATE
53#include <sys/param.h>
54#include <sys/systm.h>
55#include <sys/sysproto.h>
56#include <sys/signalvar.h>
57#include <sys/imgact.h>
58#include <sys/kernel.h>
59#include <sys/ktr.h>
60#include <sys/linker.h>
61#include <sys/lock.h>
62#include <sys/malloc.h>
63#include <sys/mutex.h>
64#include <sys/pcpu.h>
65#include <sys/proc.h>
66#include <sys/ptrace.h>
67#include <sys/cons.h>
68#include <sys/bio.h>
69#include <sys/bus.h>
70#include <sys/buf.h>
71#include <sys/exec.h>
72#include <sys/kdb.h>
73#include <sys/msgbuf.h>
74#include <machine/reg.h>
75#include <machine/cpu.h>
76
77#include <vm/vm.h>
78#include <vm/pmap.h>
79#include <vm/vm_object.h>
80#include <vm/vm_page.h>
81#include <vm/vm_pager.h>
82#include <vm/vm_map.h>
83#include <vm/vnode_pager.h>
84#include <machine/pmap.h>
85#include <machine/vmparam.h>
86#include <machine/pcb.h>
87#include <machine/undefined.h>
88#include <machine/machdep.h>
89#include <machine/metadata.h>
90#include <machine/armreg.h>
91#include <machine/bus.h>
92#include <sys/reboot.h>
93
94#include <arm/at91/at91rm92reg.h>
95#include <arm/at91/at91_piovar.h>
96#include <arm/at91/at91_pio_rm9200.h>
97
98#define KERNEL_PT_SYS		0	/* Page table for mapping proc0 zero page */
99#define KERNEL_PT_KERN		1
100#define KERNEL_PT_KERN_NUM	22
101#define KERNEL_PT_AFKERNEL	KERNEL_PT_KERN + KERNEL_PT_KERN_NUM	/* L2 table for mapping after kernel */
102#define	KERNEL_PT_AFKERNEL_NUM	5
103
104/* this should be evenly divisable by PAGE_SIZE / L2_TABLE_SIZE_REAL (or 4) */
105#define NUM_KERNEL_PTS		(KERNEL_PT_AFKERNEL + KERNEL_PT_AFKERNEL_NUM)
106
107/* Define various stack sizes in pages */
108#define IRQ_STACK_SIZE	1
109#define ABT_STACK_SIZE	1
110#define UND_STACK_SIZE	1
111
112extern u_int data_abort_handler_address;
113extern u_int prefetch_abort_handler_address;
114extern u_int undefined_handler_address;
115
116struct pv_addr kernel_pt_table[NUM_KERNEL_PTS];
117
118extern void *_end;
119
120extern int *end;
121
122struct pcpu __pcpu;
123struct pcpu *pcpup = &__pcpu;
124
125/* Physical and virtual addresses for some global pages */
126
127vm_paddr_t phys_avail[10];
128vm_paddr_t dump_avail[4];
129vm_offset_t physical_pages;
130
131struct pv_addr systempage;
132struct pv_addr msgbufpv;
133struct pv_addr irqstack;
134struct pv_addr undstack;
135struct pv_addr abtstack;
136struct pv_addr kernelstack;
137
138static struct trapframe proc0_tf;
139
140/* Static device mappings. */
141static const struct pmap_devmap kb920x_devmap[] = {
142	/*
143	 * Map the on-board devices VA == PA so that we can access them
144	 * with the MMU on or off.
145	 */
146	{
147		/*
148		 * This at least maps the interrupt controller, the UART
149		 * and the timer. Other devices should use newbus to
150		 * map their memory anyway.
151		 */
152		0xdff00000,
153		0xfff00000,
154		0x100000,
155		VM_PROT_READ|VM_PROT_WRITE,
156		PTE_NOCACHE,
157	},
158	/*
159	 * We can't just map the OHCI registers VA == PA, because
160	 * AT91RM92_OHCI_BASE belongs to the userland address space.
161	 * We could just choose a different virtual address, but a better
162	 * solution would probably be to just use pmap_mapdev() to allocate
163	 * KVA, as we don't need the OHCI controller before the vm
164	 * initialization is done. However, the AT91 resource allocation
165	 * system doesn't know how to use pmap_mapdev() yet.
166	 */
167#if 1
168	{
169		/*
170		 * Add the ohci controller, and anything else that might be
171		 * on this chip select for a VA/PA mapping.
172		 */
173		AT91RM92_OHCI_BASE,
174		AT91RM92_OHCI_PA_BASE,
175		AT91RM92_OHCI_SIZE,
176		VM_PROT_READ|VM_PROT_WRITE,
177		PTE_NOCACHE,
178	},
179#endif
180	{
181		0,
182		0,
183		0,
184		0,
185		0,
186	}
187};
188
189static long
190ramsize(void)
191{
192	uint32_t *SDRAMC = (uint32_t *)(AT91RM92_BASE + AT91RM92_SDRAMC_BASE);
193	uint32_t cr, mr;
194	int banks, rows, cols, bw;
195
196	cr = SDRAMC[AT91RM92_SDRAMC_CR / 4];
197	mr = SDRAMC[AT91RM92_SDRAMC_MR / 4];
198	bw = (mr & AT91RM92_SDRAMC_MR_DBW_16) ? 1 : 2;
199	banks = (cr & AT91RM92_SDRAMC_CR_NB_4) ? 2 : 1;
200	rows = ((cr & AT91RM92_SDRAMC_CR_NR_MASK) >> 2) + 11;
201	cols = (cr & AT91RM92_SDRAMC_CR_NC_MASK) + 8;
202	return (1 << (cols + rows + banks + bw));
203}
204
205static long
206board_init(void)
207{
208	/*
209	 * Since the USART supports RS-485 multidrop mode, it allows the
210	 * TX pins to float.  However, for RS-232 operations, we don't want
211	 * these pins to float.  Instead, they should be pulled up to avoid
212	 * mismatches.  Linux does something similar when it configures the
213	 * TX lines.  This implies that we also allow the RX lines to float
214	 * rather than be in the state they are left in by the boot loader.
215	 * Since they are input pins, I think that this is the right thing
216	 * to do.
217	 */
218
219	/* PIOA's A periph: Turn USART 0 and 2's TX/RX pins */
220	at91_pio_use_periph_a(AT91RM92_PIOA_BASE,
221	    AT91C_PA18_RXD0 | AT91C_PA22_RXD2, 0);
222	at91_pio_use_periph_a(AT91RM92_PIOA_BASE,
223	    AT91C_PA17_TXD0 | AT91C_PA23_TXD2, 1);
224	/* PIOA's B periph: Turn USART 3's TX/RX pins */
225	at91_pio_use_periph_b(AT91RM92_PIOA_BASE, AT91C_PA6_RXD3, 0);
226	at91_pio_use_periph_b(AT91RM92_PIOA_BASE, AT91C_PA5_TXD3, 1);
227#ifdef AT91_TSC
228	/* We're using TC0's A1 and A2 input */
229	at91_pio_use_periph_b(AT91RM92_PIOA_BASE,
230	    AT91C_PA19_TIOA1 | AT91C_PA21_TIOA2, 0);
231#endif
232	/* PIOB's A periph: Turn USART 1's TX/RX pins */
233	at91_pio_use_periph_a(AT91RM92_PIOB_BASE, AT91C_PB21_RXD1, 0);
234	at91_pio_use_periph_a(AT91RM92_PIOB_BASE, AT91C_PB20_TXD1, 1);
235
236	/* Pin assignment */
237#ifdef AT91_TSC
238	/* Assert PA24 low -- talk to rubidium */
239	at91_pio_use_gpio(AT91RM92_PIOA_BASE, AT91C_PIO_PA24);
240	at91_pio_gpio_output(AT91RM92_PIOA_BASE, AT91C_PIO_PA24, 0);
241	at91_pio_gpio_clear(AT91RM92_PIOA_BASE, AT91C_PIO_PA24);
242	at91_pio_use_gpio(AT91RM92_PIOB_BASE,
243	    AT91C_PIO_PB16 | AT91C_PIO_PB17 | AT91C_PIO_PB18 | AT91C_PIO_PB19);
244#endif
245
246	return (ramsize());
247}
248
249void *
250initarm(void *arg, void *arg2)
251{
252	struct pv_addr  kernel_l1pt;
253	int loop, i;
254	u_int l1pagetable;
255	vm_offset_t freemempos;
256	vm_offset_t afterkern;
257	uint32_t memsize;
258	vm_offset_t lastaddr;
259
260	set_cpufuncs();
261	lastaddr = fake_preload_metadata();
262	pcpu_init(pcpup, 0, sizeof(struct pcpu));
263	PCPU_SET(curthread, &thread0);
264
265	freemempos = (lastaddr + PAGE_MASK) & ~PAGE_MASK;
266	/* Define a macro to simplify memory allocation */
267#define valloc_pages(var, np)                   \
268	alloc_pages((var).pv_va, (np));         \
269	(var).pv_pa = (var).pv_va + (KERNPHYSADDR - KERNVIRTADDR);
270
271#define alloc_pages(var, np)			\
272	(var) = freemempos;		\
273	freemempos += (np * PAGE_SIZE);		\
274	memset((char *)(var), 0, ((np) * PAGE_SIZE));
275
276	while (((freemempos - L1_TABLE_SIZE) & (L1_TABLE_SIZE - 1)) != 0)
277		freemempos += PAGE_SIZE;
278	valloc_pages(kernel_l1pt, L1_TABLE_SIZE / PAGE_SIZE);
279	for (loop = 0; loop < NUM_KERNEL_PTS; ++loop) {
280		if (!(loop % (PAGE_SIZE / L2_TABLE_SIZE_REAL))) {
281			valloc_pages(kernel_pt_table[loop],
282			    L2_TABLE_SIZE / PAGE_SIZE);
283		} else {
284			kernel_pt_table[loop].pv_va = freemempos -
285			    (loop % (PAGE_SIZE / L2_TABLE_SIZE_REAL)) *
286			    L2_TABLE_SIZE_REAL;
287			kernel_pt_table[loop].pv_pa =
288			    kernel_pt_table[loop].pv_va - KERNVIRTADDR +
289			    KERNPHYSADDR;
290		}
291		i++;
292	}
293	/*
294	 * Allocate a page for the system page mapped to V0x00000000
295	 * This page will just contain the system vectors and can be
296	 * shared by all processes.
297	 */
298	valloc_pages(systempage, 1);
299
300	/* Allocate stacks for all modes */
301	valloc_pages(irqstack, IRQ_STACK_SIZE);
302	valloc_pages(abtstack, ABT_STACK_SIZE);
303	valloc_pages(undstack, UND_STACK_SIZE);
304	valloc_pages(kernelstack, KSTACK_PAGES);
305	valloc_pages(msgbufpv, round_page(MSGBUF_SIZE) / PAGE_SIZE);
306	/*
307	 * Now we start construction of the L1 page table
308	 * We start by mapping the L2 page tables into the L1.
309	 * This means that we can replace L1 mappings later on if necessary
310	 */
311	l1pagetable = kernel_l1pt.pv_va;
312
313	/* Map the L2 pages tables in the L1 page table */
314	pmap_link_l2pt(l1pagetable, ARM_VECTORS_HIGH,
315	    &kernel_pt_table[KERNEL_PT_SYS]);
316	for (i = 0; i < KERNEL_PT_KERN_NUM; i++)
317		pmap_link_l2pt(l1pagetable, KERNBASE + i * 0x100000,
318		    &kernel_pt_table[KERNEL_PT_KERN + i]);
319	pmap_map_chunk(l1pagetable, KERNBASE, PHYSADDR,
320	   (((uint32_t)lastaddr - KERNBASE) + PAGE_SIZE) & ~(PAGE_SIZE - 1),
321	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
322	afterkern = round_page((lastaddr + L1_S_SIZE) & ~(L1_S_SIZE
323	    - 1));
324	for (i = 0; i < KERNEL_PT_AFKERNEL_NUM; i++) {
325		pmap_link_l2pt(l1pagetable, afterkern + i * 0x00100000,
326		    &kernel_pt_table[KERNEL_PT_AFKERNEL + i]);
327	}
328
329	/* Map the vector page. */
330	pmap_map_entry(l1pagetable, ARM_VECTORS_HIGH, systempage.pv_pa,
331	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
332	/* Map the stack pages */
333	pmap_map_chunk(l1pagetable, irqstack.pv_va, irqstack.pv_pa,
334	    IRQ_STACK_SIZE * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
335	pmap_map_chunk(l1pagetable, abtstack.pv_va, abtstack.pv_pa,
336	    ABT_STACK_SIZE * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
337	pmap_map_chunk(l1pagetable, undstack.pv_va, undstack.pv_pa,
338	    UND_STACK_SIZE * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
339	pmap_map_chunk(l1pagetable, kernelstack.pv_va, kernelstack.pv_pa,
340	    KSTACK_PAGES * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
341
342	pmap_map_chunk(l1pagetable, kernel_l1pt.pv_va, kernel_l1pt.pv_pa,
343	    L1_TABLE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
344	pmap_map_chunk(l1pagetable, msgbufpv.pv_va, msgbufpv.pv_pa,
345	    MSGBUF_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
346
347
348	for (loop = 0; loop < NUM_KERNEL_PTS; ++loop) {
349		pmap_map_chunk(l1pagetable, kernel_pt_table[loop].pv_va,
350		    kernel_pt_table[loop].pv_pa, L2_TABLE_SIZE,
351		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
352	}
353
354	pmap_devmap_bootstrap(l1pagetable, kb920x_devmap);
355	cpu_domains((DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2)) | DOMAIN_CLIENT);
356	setttb(kernel_l1pt.pv_pa);
357	cpu_tlb_flushID();
358	cpu_domains(DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2));
359	cninit();
360	memsize = board_init();
361	physmem = memsize / PAGE_SIZE;
362
363	/*
364	 * Pages were allocated during the secondary bootstrap for the
365	 * stacks for different CPU modes.
366	 * We must now set the r13 registers in the different CPU modes to
367	 * point to these stacks.
368	 * Since the ARM stacks use STMFD etc. we must set r13 to the top end
369	 * of the stack memory.
370	 */
371
372	cpu_control(CPU_CONTROL_MMU_ENABLE, CPU_CONTROL_MMU_ENABLE);
373	set_stackptr(PSR_IRQ32_MODE,
374	    irqstack.pv_va + IRQ_STACK_SIZE * PAGE_SIZE);
375	set_stackptr(PSR_ABT32_MODE,
376	    abtstack.pv_va + ABT_STACK_SIZE * PAGE_SIZE);
377	set_stackptr(PSR_UND32_MODE,
378	    undstack.pv_va + UND_STACK_SIZE * PAGE_SIZE);
379
380
381
382	/*
383	 * We must now clean the cache again....
384	 * Cleaning may be done by reading new data to displace any
385	 * dirty data in the cache. This will have happened in setttb()
386	 * but since we are boot strapping the addresses used for the read
387	 * may have just been remapped and thus the cache could be out
388	 * of sync. A re-clean after the switch will cure this.
389	 * After booting there are no gross reloations of the kernel thus
390	 * this problem will not occur after initarm().
391	 */
392	cpu_idcache_wbinv_all();
393
394	/* Set stack for exception handlers */
395
396	data_abort_handler_address = (u_int)data_abort_handler;
397	prefetch_abort_handler_address = (u_int)prefetch_abort_handler;
398	undefined_handler_address = (u_int)undefinedinstruction_bounce;
399	undefined_init();
400
401	proc_linkup0(&proc0, &thread0);
402	thread0.td_kstack = kernelstack.pv_va;
403	thread0.td_pcb = (struct pcb *)
404		(thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
405	thread0.td_pcb->pcb_flags = 0;
406	thread0.td_frame = &proc0_tf;
407	pcpup->pc_curpcb = thread0.td_pcb;
408
409	arm_vector_init(ARM_VECTORS_HIGH, ARM_VEC_ALL);
410
411	pmap_curmaxkvaddr = afterkern + 0x100000 * (KERNEL_PT_KERN_NUM - 1);
412	/*
413	 * ARM_USE_SMALL_ALLOC uses dump_avail, so it must be filled before
414	 * calling pmap_bootstrap.
415	 */
416	dump_avail[0] = PHYSADDR;
417	dump_avail[1] = PHYSADDR + memsize;
418	dump_avail[2] = 0;
419	dump_avail[3] = 0;
420
421	pmap_bootstrap(freemempos,
422	    KERNVIRTADDR + 3 * memsize,
423	    &kernel_l1pt);
424	msgbufp = (void*)msgbufpv.pv_va;
425	msgbufinit(msgbufp, MSGBUF_SIZE);
426	mutex_init();
427
428	i = 0;
429
430#if PHYSADDR != KERNPHYSADDR
431	phys_avail[i++] = PHYSADDR;
432	phys_avail[i++] = KERNPHYSADDR;
433#endif
434	phys_avail[i++] = virtual_avail - KERNVIRTADDR + KERNPHYSADDR;
435	phys_avail[i++] = PHYSADDR + memsize;
436	phys_avail[i++] = 0;
437	phys_avail[i++] = 0;
438	/* Do basic tuning, hz etc */
439	init_param1();
440	init_param2(physmem);
441	kdb_init();
442	return ((void *)(kernelstack.pv_va + USPACE_SVC_STACK_TOP -
443	    sizeof(struct pcb)));
444}
445