board_bwct.c revision 163709
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_ddb.h"
48#include "opt_at91.h"
49
50#include <sys/cdefs.h>
51__FBSDID("$FreeBSD: head/sys/arm/at91/kb920x_machdep.c 163709 2006-10-26 21:42:22Z jb $");
52
53#define _ARM32_BUS_DMA_PRIVATE
54#include <sys/param.h>
55#include <sys/systm.h>
56#include <sys/sysproto.h>
57#include <sys/signalvar.h>
58#include <sys/imgact.h>
59#include <sys/kernel.h>
60#include <sys/ktr.h>
61#include <sys/linker.h>
62#include <sys/lock.h>
63#include <sys/malloc.h>
64#include <sys/mutex.h>
65#include <sys/pcpu.h>
66#include <sys/proc.h>
67#include <sys/ptrace.h>
68#include <sys/cons.h>
69#include <sys/bio.h>
70#include <sys/bus.h>
71#include <sys/buf.h>
72#include <sys/exec.h>
73#include <sys/kdb.h>
74#include <sys/msgbuf.h>
75#include <machine/reg.h>
76#include <machine/cpu.h>
77
78#include <vm/vm.h>
79#include <vm/pmap.h>
80#include <vm/vm.h>
81#include <vm/vm_object.h>
82#include <vm/vm_page.h>
83#include <vm/vm_pager.h>
84#include <vm/vm_map.h>
85#include <vm/vnode_pager.h>
86#include <machine/pmap.h>
87#include <machine/vmparam.h>
88#include <machine/pcb.h>
89#include <machine/undefined.h>
90#include <machine/machdep.h>
91#include <machine/metadata.h>
92#include <machine/armreg.h>
93#include <machine/bus.h>
94#include <sys/reboot.h>
95
96#include <arm/at91/at91rm92reg.h>
97#include <arm/at91/at91_piovar.h>
98#include <arm/at91/at91_pio_rm9200.h>
99
100#define KERNEL_PT_SYS		0	/* Page table for mapping proc0 zero page */
101#define KERNEL_PT_KERN		1
102#define KERNEL_PT_KERN_NUM	22
103#define KERNEL_PT_AFKERNEL	KERNEL_PT_KERN + KERNEL_PT_KERN_NUM	/* L2 table for mapping after kernel */
104#define	KERNEL_PT_AFKERNEL_NUM	5
105
106/* this should be evenly divisable by PAGE_SIZE / L2_TABLE_SIZE_REAL (or 4) */
107#define NUM_KERNEL_PTS		(KERNEL_PT_AFKERNEL + KERNEL_PT_AFKERNEL_NUM)
108
109/* Define various stack sizes in pages */
110#define IRQ_STACK_SIZE	1
111#define ABT_STACK_SIZE	1
112#define UND_STACK_SIZE	1
113
114extern u_int data_abort_handler_address;
115extern u_int prefetch_abort_handler_address;
116extern u_int undefined_handler_address;
117
118struct pv_addr kernel_pt_table[NUM_KERNEL_PTS];
119
120extern void *_end;
121
122extern int *end;
123
124struct pcpu __pcpu;
125struct pcpu *pcpup = &__pcpu;
126
127/* Physical and virtual addresses for some global pages */
128
129vm_paddr_t phys_avail[10];
130vm_paddr_t dump_avail[4];
131vm_offset_t physical_pages;
132vm_offset_t clean_sva, clean_eva;
133
134struct pv_addr systempage;
135struct pv_addr msgbufpv;
136struct pv_addr irqstack;
137struct pv_addr undstack;
138struct pv_addr abtstack;
139struct pv_addr kernelstack;
140struct pv_addr minidataclean;
141
142static struct trapframe proc0_tf;
143
144/* Static device mappings. */
145static const struct pmap_devmap kb920x_devmap[] = {
146	/*
147	 * Map the on-board devices VA == PA so that we can access them
148	 * with the MMU on or off.
149	 */
150	{
151		/*
152		 * This at least maps the interrupt controller, the UART
153		 * and the timer. Other devices should use newbus to
154		 * map their memory anyway.
155		 */
156		0xdff00000,
157		0xfff00000,
158		0x100000,
159		VM_PROT_READ|VM_PROT_WRITE,
160		PTE_NOCACHE,
161	},
162	/*
163	 * We can't just map the OHCI registers VA == PA, because
164	 * AT91RM92_OHCI_BASE belongs to the userland address space.
165	 * We could just choose a different virtual address, but a better
166	 * solution would probably be to just use pmap_mapdev() to allocate
167	 * KVA, as we don't need the OHCI controller before the vm
168	 * initialization is done. However, the AT91 resource allocation
169	 * system doesn't know how to use pmap_mapdev() yet.
170	 */
171#if 0
172	{
173		/*
174		 * Add the ohci controller, and anything else that might be
175		 * on this chip select for a VA/PA mapping.
176		 */
177		AT91RM92_OHCI_BASE,
178		AT91RM92_OHCI_BASE,
179		AT91RM92_OHCI_SIZE,
180		VM_PROT_READ|VM_PROT_WRITE,
181		PTE_NOCACHE,
182	},
183#endif
184	{
185		0,
186		0,
187		0,
188		0,
189		0,
190	}
191};
192
193#define SDRAM_START 0xa0000000
194
195#ifdef DDB
196extern vm_offset_t ksym_start, ksym_end;
197#endif
198
199static long
200ramsize(void)
201{
202	uint32_t *SDRAMC = (uint32_t *)(AT91RM92_BASE + AT91RM92_SDRAMC_BASE);
203	uint32_t cr, mr;
204	int banks, rows, cols, bw;
205
206	cr = SDRAMC[AT91RM92_SDRAMC_CR / 4];
207	mr = SDRAMC[AT91RM92_SDRAMC_MR / 4];
208	bw = (mr & AT91RM92_SDRAMC_MR_DBW_16) ? 1 : 2;
209	banks = (cr & AT91RM92_SDRAMC_CR_NB_4) ? 2 : 1;
210	rows = ((cr & AT91RM92_SDRAMC_CR_NR_MASK) >> 2) + 11;
211	cols = (cr & AT91RM92_SDRAMC_CR_NC_MASK) + 8;
212	return (1 << (cols + rows + banks + bw));
213}
214
215static long
216board_init(void)
217{
218	/*
219	 * Since the USART supprots RS-485 multidrop mode, it allows the
220	 * TX pins to float.  However, for RS-232 operations, we don't want
221	 * these pins to float.  Instead, they should be pulled up to avoid
222	 * mismatches.  Linux does something similar when it configures the
223	 * TX lines.  This implies that we also allow the RX lines to float
224	 * rather than be in the state they are left in by the boot loader.
225	 * Since they are input pins, I think that this is the right thing
226	 * to do.
227	 */
228
229	/* PIOA's A periph: Turn USART 0 and 2's TX/RX pins */
230	at91_pio_use_periph_a(AT91RM92_PIOA_BASE,
231	    AT91C_PA18_RXD0 | AT91C_PA22_RXD2, 0);
232	at91_pio_use_periph_a(AT91RM92_PIOA_BASE,
233	    AT91C_PA17_TXD0 | AT91C_PA23_TXD2, 1);
234	/* PIOA's B periph: Turn USART 3's TX/RX pins */
235	at91_pio_use_periph_b(AT91RM92_PIOA_BASE, AT91C_PA6_RXD3, 0);
236	at91_pio_use_periph_b(AT91RM92_PIOA_BASE, AT91C_PA5_TXD3, 1);
237#ifdef AT91_TSC
238	/* We're using TC0's A1 and A2 input */
239	at91_pio_use_periph_b(AT91RM92_PIOA_BASE,
240	    AT91C_PA19_TIOA1 | AT91C_PA21_TIOA2, 0);
241#endif
242	/* PIOB's A periph: Turn USART 1's TX/RX pins */
243	at91_pio_use_periph_a(AT91RM92_PIOB_BASE, AT91C_PB21_RXD1, 0);
244	at91_pio_use_periph_a(AT91RM92_PIOB_BASE, AT91C_PB20_TXD1, 1);
245
246	/* Pin assignment */
247#ifdef AT91_TSC
248	/* Assert PA24 low -- talk to rubidium */
249	at91_pio_use_gpio(AT91RM92_PIOA_BASE, AT91C_PIO_PA24);
250	at91_pio_gpio_output(AT91RM92_PIOA_BASE, AT91C_PIO_PA24, 0);
251	at91_pio_gpio_clear(AT91RM92_PIOA_BASE, AT91C_PIO_PA24);
252	at91_pio_use_gpio(AT91RM92_PIOB_BASE,
253	    AT91C_PIO_PB16 | AT91C_PIO_PB17 | AT91C_PIO_PB18 | AT91C_PIO_PB19);
254#endif
255
256	return (ramsize());
257}
258
259void *
260initarm(void *arg, void *arg2)
261{
262	struct pv_addr  kernel_l1pt;
263	int loop;
264	u_int l1pagetable;
265	vm_offset_t freemempos;
266	vm_offset_t afterkern;
267	int i;
268	uint32_t fake_preload[35];
269	uint32_t memsize;
270	vm_offset_t lastaddr;
271#ifdef DDB
272	vm_offset_t zstart = 0, zend = 0;
273#endif
274
275	i = 0;
276
277	set_cpufuncs();
278
279	fake_preload[i++] = MODINFO_NAME;
280	fake_preload[i++] = strlen("elf kernel") + 1;
281	strcpy((char*)&fake_preload[i++], "elf kernel");
282	i += 2;
283	fake_preload[i++] = MODINFO_TYPE;
284	fake_preload[i++] = strlen("elf kernel") + 1;
285	strcpy((char*)&fake_preload[i++], "elf kernel");
286	i += 2;
287	fake_preload[i++] = MODINFO_ADDR;
288	fake_preload[i++] = sizeof(vm_offset_t);
289	fake_preload[i++] = KERNBASE;
290	fake_preload[i++] = MODINFO_SIZE;
291	fake_preload[i++] = sizeof(uint32_t);
292	fake_preload[i++] = (uint32_t)&end - KERNBASE;
293#ifdef DDB
294	if (*(uint32_t *)KERNVIRTADDR == MAGIC_TRAMP_NUMBER) {
295		fake_preload[i++] = MODINFO_METADATA|MODINFOMD_SSYM;
296		fake_preload[i++] = sizeof(vm_offset_t);
297		fake_preload[i++] = *(uint32_t *)(KERNVIRTADDR + 4);
298		fake_preload[i++] = MODINFO_METADATA|MODINFOMD_ESYM;
299		fake_preload[i++] = sizeof(vm_offset_t);
300		fake_preload[i++] = *(uint32_t *)(KERNVIRTADDR + 8);
301		lastaddr = *(uint32_t *)(KERNVIRTADDR + 8);
302		zend = lastaddr;
303		zstart = *(uint32_t *)(KERNVIRTADDR + 4);
304		ksym_start = zstart;
305		ksym_end = zend;
306	} else
307#endif
308		lastaddr = (vm_offset_t)&end;
309
310	fake_preload[i++] = 0;
311	fake_preload[i] = 0;
312	preload_metadata = (void *)fake_preload;
313
314
315	pcpu_init(pcpup, 0, sizeof(struct pcpu));
316	PCPU_SET(curthread, &thread0);
317
318#define KERNEL_TEXT_BASE (KERNBASE)
319	freemempos = (lastaddr + PAGE_MASK) & ~PAGE_MASK;
320	/* Define a macro to simplify memory allocation */
321#define valloc_pages(var, np)                   \
322	alloc_pages((var).pv_va, (np));         \
323	(var).pv_pa = (var).pv_va + (KERNPHYSADDR - KERNVIRTADDR);
324
325#define alloc_pages(var, np)			\
326	(var) = freemempos;		\
327	freemempos += (np * PAGE_SIZE);		\
328	memset((char *)(var), 0, ((np) * PAGE_SIZE));
329
330	while (((freemempos - L1_TABLE_SIZE) & (L1_TABLE_SIZE - 1)) != 0)
331		freemempos += PAGE_SIZE;
332	valloc_pages(kernel_l1pt, L1_TABLE_SIZE / PAGE_SIZE);
333	for (loop = 0; loop < NUM_KERNEL_PTS; ++loop) {
334		if (!(loop % (PAGE_SIZE / L2_TABLE_SIZE_REAL))) {
335			valloc_pages(kernel_pt_table[loop],
336			    L2_TABLE_SIZE / PAGE_SIZE);
337		} else {
338			kernel_pt_table[loop].pv_va = freemempos -
339			    (loop % (PAGE_SIZE / L2_TABLE_SIZE_REAL)) *
340			    L2_TABLE_SIZE_REAL;
341			kernel_pt_table[loop].pv_pa =
342			    kernel_pt_table[loop].pv_va - KERNVIRTADDR +
343			    KERNPHYSADDR;
344		}
345		i++;
346	}
347	/*
348	 * Allocate a page for the system page mapped to V0x00000000
349	 * This page will just contain the system vectors and can be
350	 * shared by all processes.
351	 */
352	valloc_pages(systempage, 1);
353
354	/* Allocate stacks for all modes */
355	valloc_pages(irqstack, IRQ_STACK_SIZE);
356	valloc_pages(abtstack, ABT_STACK_SIZE);
357	valloc_pages(undstack, UND_STACK_SIZE);
358	valloc_pages(kernelstack, KSTACK_PAGES);
359	alloc_pages(minidataclean.pv_pa, 1);
360	valloc_pages(msgbufpv, round_page(MSGBUF_SIZE) / PAGE_SIZE);
361	/*
362	 * Now we start construction of the L1 page table
363	 * We start by mapping the L2 page tables into the L1.
364	 * This means that we can replace L1 mappings later on if necessary
365	 */
366	l1pagetable = kernel_l1pt.pv_va;
367
368	/* Map the L2 pages tables in the L1 page table */
369	pmap_link_l2pt(l1pagetable, ARM_VECTORS_HIGH,
370	    &kernel_pt_table[KERNEL_PT_SYS]);
371	for (i = 0; i < KERNEL_PT_KERN_NUM; i++)
372		pmap_link_l2pt(l1pagetable, KERNBASE + i * 0x100000,
373		    &kernel_pt_table[KERNEL_PT_KERN + i]);
374	pmap_map_chunk(l1pagetable, KERNBASE, KERNPHYSADDR,
375	   (((uint32_t)(lastaddr) - KERNBASE) + PAGE_SIZE) & ~(PAGE_SIZE - 1),
376	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
377	afterkern = round_page((lastaddr + L1_S_SIZE) & ~(L1_S_SIZE
378	    - 1));
379	for (i = 0; i < KERNEL_PT_AFKERNEL_NUM; i++) {
380		pmap_link_l2pt(l1pagetable, afterkern + i * 0x00100000,
381		    &kernel_pt_table[KERNEL_PT_AFKERNEL + i]);
382	}
383	pmap_map_entry(l1pagetable, afterkern, minidataclean.pv_pa,
384	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
385
386
387	/* Map the vector page. */
388	pmap_map_entry(l1pagetable, ARM_VECTORS_HIGH, systempage.pv_pa,
389	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
390	/* Map the stack pages */
391	pmap_map_chunk(l1pagetable, irqstack.pv_va, irqstack.pv_pa,
392	    IRQ_STACK_SIZE * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
393	pmap_map_chunk(l1pagetable, abtstack.pv_va, abtstack.pv_pa,
394	    ABT_STACK_SIZE * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
395	pmap_map_chunk(l1pagetable, undstack.pv_va, undstack.pv_pa,
396	    UND_STACK_SIZE * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
397	pmap_map_chunk(l1pagetable, kernelstack.pv_va, kernelstack.pv_pa,
398	    KSTACK_PAGES * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
399
400	pmap_map_chunk(l1pagetable, kernel_l1pt.pv_va, kernel_l1pt.pv_pa,
401	    L1_TABLE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
402	pmap_map_chunk(l1pagetable, msgbufpv.pv_va, msgbufpv.pv_pa,
403	    MSGBUF_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
404
405
406	for (loop = 0; loop < NUM_KERNEL_PTS; ++loop) {
407		pmap_map_chunk(l1pagetable, kernel_pt_table[loop].pv_va,
408		    kernel_pt_table[loop].pv_pa, L2_TABLE_SIZE,
409		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
410	}
411
412	pmap_devmap_bootstrap(l1pagetable, kb920x_devmap);
413	cpu_domains((DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2)) | DOMAIN_CLIENT);
414	setttb(kernel_l1pt.pv_pa);
415	cpu_tlb_flushID();
416	cpu_domains(DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2));
417	cninit();
418	memsize = board_init();
419	physmem = memsize / PAGE_SIZE;
420
421	/*
422	 * Pages were allocated during the secondary bootstrap for the
423	 * stacks for different CPU modes.
424	 * We must now set the r13 registers in the different CPU modes to
425	 * point to these stacks.
426	 * Since the ARM stacks use STMFD etc. we must set r13 to the top end
427	 * of the stack memory.
428	 */
429
430	cpu_control(CPU_CONTROL_MMU_ENABLE, CPU_CONTROL_MMU_ENABLE);
431	set_stackptr(PSR_IRQ32_MODE,
432	    irqstack.pv_va + IRQ_STACK_SIZE * PAGE_SIZE);
433	set_stackptr(PSR_ABT32_MODE,
434	    abtstack.pv_va + ABT_STACK_SIZE * PAGE_SIZE);
435	set_stackptr(PSR_UND32_MODE,
436	    undstack.pv_va + UND_STACK_SIZE * PAGE_SIZE);
437
438
439
440	/*
441	 * We must now clean the cache again....
442	 * Cleaning may be done by reading new data to displace any
443	 * dirty data in the cache. This will have happened in setttb()
444	 * but since we are boot strapping the addresses used for the read
445	 * may have just been remapped and thus the cache could be out
446	 * of sync. A re-clean after the switch will cure this.
447	 * After booting there are no gross reloations of the kernel thus
448	 * this problem will not occur after initarm().
449	 */
450	cpu_idcache_wbinv_all();
451
452	/* Set stack for exception handlers */
453
454	data_abort_handler_address = (u_int)data_abort_handler;
455	prefetch_abort_handler_address = (u_int)prefetch_abort_handler;
456	undefined_handler_address = (u_int)undefinedinstruction_bounce;
457	undefined_init();
458
459#ifdef KSE
460	proc_linkup(&proc0, &ksegrp0, &thread0);
461#else
462	proc_linkup(&proc0, &thread0);
463#endif
464	thread0.td_kstack = kernelstack.pv_va;
465	thread0.td_pcb = (struct pcb *)
466		(thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
467	thread0.td_pcb->pcb_flags = 0;
468	thread0.td_frame = &proc0_tf;
469	pcpup->pc_curpcb = thread0.td_pcb;
470
471	arm_vector_init(ARM_VECTORS_HIGH, ARM_VEC_ALL);
472
473	pmap_curmaxkvaddr = afterkern + 0x100000 * (KERNEL_PT_KERN_NUM - 1);
474	/*
475	 * ARM_USE_SMALL_ALLOC uses dump_avail, so it must be filled before
476	 * calling pmap_bootstrap.
477	 */
478	dump_avail[0] = KERNPHYSADDR;
479	dump_avail[1] = KERNPHYSADDR + memsize;
480	dump_avail[2] = 0;
481	dump_avail[3] = 0;
482
483	pmap_bootstrap(freemempos,
484	    KERNVIRTADDR + 3 * memsize,
485	    &kernel_l1pt);
486	msgbufp = (void*)msgbufpv.pv_va;
487	msgbufinit(msgbufp, MSGBUF_SIZE);
488	mutex_init();
489
490	i = 0;
491
492	phys_avail[0] = virtual_avail - KERNVIRTADDR + KERNPHYSADDR;
493	phys_avail[1] = KERNPHYSADDR + memsize;
494	phys_avail[2] = 0;
495	phys_avail[3] = 0;
496	/* Do basic tuning, hz etc */
497	init_param1();
498	init_param2(physmem);
499	avail_end = KERNPHYSADDR + memsize - 1;
500	kdb_init();
501	return ((void *)(kernelstack.pv_va + USPACE_SVC_STACK_TOP -
502	    sizeof(struct pcb)));
503}
504