xref: /linux/arch/powerpc/platforms/pseries/setup.c (revision cc53291521701f9c7c7265bbb3c140563174d8b2)
1 /*
2  *  64-bit pSeries and RS/6000 setup code.
3  *
4  *  Copyright (C) 1995  Linus Torvalds
5  *  Adapted from 'alpha' version by Gary Thomas
6  *  Modified by Cort Dougan (cort@cs.nmt.edu)
7  *  Modified by PPC64 Team, IBM Corp
8  *
9  * This program is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU General Public License
11  * as published by the Free Software Foundation; either version
12  * 2 of the License, or (at your option) any later version.
13  */
14 
15 /*
16  * bootup setup stuff..
17  */
18 
19 #undef DEBUG
20 
21 #include <linux/config.h>
22 #include <linux/cpu.h>
23 #include <linux/errno.h>
24 #include <linux/sched.h>
25 #include <linux/kernel.h>
26 #include <linux/mm.h>
27 #include <linux/stddef.h>
28 #include <linux/unistd.h>
29 #include <linux/slab.h>
30 #include <linux/user.h>
31 #include <linux/a.out.h>
32 #include <linux/tty.h>
33 #include <linux/major.h>
34 #include <linux/interrupt.h>
35 #include <linux/reboot.h>
36 #include <linux/init.h>
37 #include <linux/ioport.h>
38 #include <linux/console.h>
39 #include <linux/pci.h>
40 #include <linux/utsname.h>
41 #include <linux/adb.h>
42 #include <linux/module.h>
43 #include <linux/delay.h>
44 #include <linux/irq.h>
45 #include <linux/seq_file.h>
46 #include <linux/root_dev.h>
47 
48 #include <asm/mmu.h>
49 #include <asm/processor.h>
50 #include <asm/io.h>
51 #include <asm/pgtable.h>
52 #include <asm/prom.h>
53 #include <asm/rtas.h>
54 #include <asm/pci-bridge.h>
55 #include <asm/iommu.h>
56 #include <asm/dma.h>
57 #include <asm/machdep.h>
58 #include <asm/irq.h>
59 #include <asm/kexec.h>
60 #include <asm/time.h>
61 #include <asm/nvram.h>
62 #include "xics.h"
63 #include <asm/firmware.h>
64 #include <asm/pmc.h>
65 #include <asm/mpic.h>
66 #include <asm/ppc-pci.h>
67 #include <asm/i8259.h>
68 #include <asm/udbg.h>
69 #include <asm/smp.h>
70 
71 #include "plpar_wrappers.h"
72 
73 #ifdef DEBUG
74 #define DBG(fmt...) udbg_printf(fmt)
75 #else
76 #define DBG(fmt...)
77 #endif
78 
79 extern void find_udbg_vterm(void);
80 
81 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
82 
83 extern void pSeries_system_reset_exception(struct pt_regs *regs);
84 extern int pSeries_machine_check_exception(struct pt_regs *regs);
85 
86 static void pseries_shared_idle(void);
87 static void pseries_dedicated_idle(void);
88 
89 struct mpic *pSeries_mpic;
90 
91 void pSeries_show_cpuinfo(struct seq_file *m)
92 {
93 	struct device_node *root;
94 	const char *model = "";
95 
96 	root = of_find_node_by_path("/");
97 	if (root)
98 		model = get_property(root, "model", NULL);
99 	seq_printf(m, "machine\t\t: CHRP %s\n", model);
100 	of_node_put(root);
101 }
102 
103 /* Initialize firmware assisted non-maskable interrupts if
104  * the firmware supports this feature.
105  */
106 static void __init fwnmi_init(void)
107 {
108 	unsigned long system_reset_addr, machine_check_addr;
109 
110 	int ibm_nmi_register = rtas_token("ibm,nmi-register");
111 	if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
112 		return;
113 
114 	/* If the kernel's not linked at zero we point the firmware at low
115 	 * addresses anyway, and use a trampoline to get to the real code. */
116 	system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
117 	machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
118 
119 	if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
120 				machine_check_addr))
121 		fwnmi_active = 1;
122 }
123 
124 static void __init pSeries_init_mpic(void)
125 {
126         unsigned int *addrp;
127 	struct device_node *np;
128 	unsigned long intack = 0;
129 
130 	/* All ISUs are setup, complete initialization */
131 	mpic_init(pSeries_mpic);
132 
133 	/* Check what kind of cascade ACK we have */
134         if (!(np = of_find_node_by_name(NULL, "pci"))
135             || !(addrp = (unsigned int *)
136                  get_property(np, "8259-interrupt-acknowledge", NULL)))
137                 printk(KERN_ERR "Cannot find pci to get ack address\n");
138         else
139 		intack = addrp[prom_n_addr_cells(np)-1];
140 	of_node_put(np);
141 
142 	/* Setup the legacy interrupts & controller */
143 	i8259_init(intack, 0);
144 
145 	/* Hook cascade to mpic */
146 	mpic_setup_cascade(NUM_ISA_INTERRUPTS, i8259_irq_cascade, NULL);
147 }
148 
149 static void __init pSeries_setup_mpic(void)
150 {
151 	unsigned int *opprop;
152 	unsigned long openpic_addr = 0;
153         unsigned char senses[NR_IRQS - NUM_ISA_INTERRUPTS];
154         struct device_node *root;
155 	int irq_count;
156 
157 	/* Find the Open PIC if present */
158 	root = of_find_node_by_path("/");
159 	opprop = (unsigned int *) get_property(root, "platform-open-pic", NULL);
160 	if (opprop != 0) {
161 		int n = prom_n_addr_cells(root);
162 
163 		for (openpic_addr = 0; n > 0; --n)
164 			openpic_addr = (openpic_addr << 32) + *opprop++;
165 		printk(KERN_DEBUG "OpenPIC addr: %lx\n", openpic_addr);
166 	}
167 	of_node_put(root);
168 
169 	BUG_ON(openpic_addr == 0);
170 
171 	/* Get the sense values from OF */
172 	prom_get_irq_senses(senses, NUM_ISA_INTERRUPTS, NR_IRQS);
173 
174 	/* Setup the openpic driver */
175 	irq_count = NR_IRQS - NUM_ISA_INTERRUPTS - 4; /* leave room for IPIs */
176 	pSeries_mpic = mpic_alloc(openpic_addr, MPIC_PRIMARY,
177 				  16, 16, irq_count, /* isu size, irq offset, irq count */
178 				  NR_IRQS - 4, /* ipi offset */
179 				  senses, irq_count, /* sense & sense size */
180 				  " MPIC     ");
181 }
182 
183 static void pseries_lpar_enable_pmcs(void)
184 {
185 	unsigned long set, reset;
186 
187 	power4_enable_pmcs();
188 
189 	set = 1UL << 63;
190 	reset = 0;
191 	plpar_hcall_norets(H_PERFMON, set, reset);
192 
193 	/* instruct hypervisor to maintain PMCs */
194 	if (firmware_has_feature(FW_FEATURE_SPLPAR))
195 		get_paca()->lppaca.pmcregs_in_use = 1;
196 }
197 
198 static void __init pSeries_setup_arch(void)
199 {
200 	/* Fixup ppc_md depending on the type of interrupt controller */
201 	if (ppc64_interrupt_controller == IC_OPEN_PIC) {
202 		ppc_md.init_IRQ       = pSeries_init_mpic;
203 		ppc_md.get_irq        = mpic_get_irq;
204 		/* Allocate the mpic now, so that find_and_init_phbs() can
205 		 * fill the ISUs */
206 		pSeries_setup_mpic();
207 	} else {
208 		ppc_md.init_IRQ       = xics_init_IRQ;
209 		ppc_md.get_irq        = xics_get_irq;
210 	}
211 
212 #ifdef CONFIG_SMP
213 	smp_init_pSeries();
214 #endif
215 	/* openpic global configuration register (64-bit format). */
216 	/* openpic Interrupt Source Unit pointer (64-bit format). */
217 	/* python0 facility area (mmio) (64-bit format) REAL address. */
218 
219 	/* init to some ~sane value until calibrate_delay() runs */
220 	loops_per_jiffy = 50000000;
221 
222 	if (ROOT_DEV == 0) {
223 		printk("No ramdisk, default root is /dev/sda2\n");
224 		ROOT_DEV = Root_SDA2;
225 	}
226 
227 	fwnmi_init();
228 
229 	/* Find and initialize PCI host bridges */
230 	init_pci_config_tokens();
231 	find_and_init_phbs();
232 	eeh_init();
233 
234 	pSeries_nvram_init();
235 
236 	/* Choose an idle loop */
237 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
238 		vpa_init(boot_cpuid);
239 		if (get_paca()->lppaca.shared_proc) {
240 			printk(KERN_INFO "Using shared processor idle loop\n");
241 			ppc_md.idle_loop = pseries_shared_idle;
242 		} else {
243 			printk(KERN_INFO "Using dedicated idle loop\n");
244 			ppc_md.idle_loop = pseries_dedicated_idle;
245 		}
246 	} else {
247 		printk(KERN_INFO "Using default idle loop\n");
248 		ppc_md.idle_loop = default_idle;
249 	}
250 
251 	if (platform_is_lpar())
252 		ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
253 	else
254 		ppc_md.enable_pmcs = power4_enable_pmcs;
255 }
256 
257 static int __init pSeries_init_panel(void)
258 {
259 	/* Manually leave the kernel version on the panel. */
260 	ppc_md.progress("Linux ppc64\n", 0);
261 	ppc_md.progress(system_utsname.version, 0);
262 
263 	return 0;
264 }
265 arch_initcall(pSeries_init_panel);
266 
267 
268 /* Build up the ppc64_firmware_features bitmask field
269  * using contents of device-tree/ibm,hypertas-functions.
270  * Ultimately this functionality may be moved into prom.c prom_init().
271  */
272 static void __init fw_feature_init(void)
273 {
274 	struct device_node * dn;
275 	char * hypertas;
276 	unsigned int len;
277 
278 	DBG(" -> fw_feature_init()\n");
279 
280 	ppc64_firmware_features = 0;
281 	dn = of_find_node_by_path("/rtas");
282 	if (dn == NULL) {
283 		printk(KERN_ERR "WARNING ! Cannot find RTAS in device-tree !\n");
284 		goto no_rtas;
285 	}
286 
287 	hypertas = get_property(dn, "ibm,hypertas-functions", &len);
288 	if (hypertas) {
289 		while (len > 0){
290 			int i, hypertas_len;
291 			/* check value against table of strings */
292 			for(i=0; i < FIRMWARE_MAX_FEATURES ;i++) {
293 				if ((firmware_features_table[i].name) &&
294 				    (strcmp(firmware_features_table[i].name,hypertas))==0) {
295 					/* we have a match */
296 					ppc64_firmware_features |=
297 						(firmware_features_table[i].val);
298 					break;
299 				}
300 			}
301 			hypertas_len = strlen(hypertas);
302 			len -= hypertas_len +1;
303 			hypertas+= hypertas_len +1;
304 		}
305 	}
306 
307 	of_node_put(dn);
308 no_rtas:
309 
310 	DBG(" <- fw_feature_init()\n");
311 }
312 
313 
314 static  void __init pSeries_discover_pic(void)
315 {
316 	struct device_node *np;
317 	char *typep;
318 
319 	/*
320 	 * Setup interrupt mapping options that are needed for finish_device_tree
321 	 * to properly parse the OF interrupt tree & do the virtual irq mapping
322 	 */
323 	__irq_offset_value = NUM_ISA_INTERRUPTS;
324 	ppc64_interrupt_controller = IC_INVALID;
325 	for (np = NULL; (np = of_find_node_by_name(np, "interrupt-controller"));) {
326 		typep = (char *)get_property(np, "compatible", NULL);
327 		if (strstr(typep, "open-pic"))
328 			ppc64_interrupt_controller = IC_OPEN_PIC;
329 		else if (strstr(typep, "ppc-xicp"))
330 			ppc64_interrupt_controller = IC_PPC_XIC;
331 		else
332 			printk("pSeries_discover_pic: failed to recognize"
333 			       " interrupt-controller\n");
334 		break;
335 	}
336 }
337 
338 static void pSeries_mach_cpu_die(void)
339 {
340 	local_irq_disable();
341 	idle_task_exit();
342 	/* Some hardware requires clearing the CPPR, while other hardware does not
343 	 * it is safe either way
344 	 */
345 	pSeriesLP_cppr_info(0, 0);
346 	rtas_stop_self();
347 	/* Should never get here... */
348 	BUG();
349 	for(;;);
350 }
351 
352 static int pseries_set_dabr(unsigned long dabr)
353 {
354 	return plpar_hcall_norets(H_SET_DABR, dabr);
355 }
356 
357 static int pseries_set_xdabr(unsigned long dabr)
358 {
359 	/* We want to catch accesses from kernel and userspace */
360 	return plpar_hcall_norets(H_SET_XDABR, dabr,
361 			H_DABRX_KERNEL | H_DABRX_USER);
362 }
363 
364 /*
365  * Early initialization.  Relocation is on but do not reference unbolted pages
366  */
367 static void __init pSeries_init_early(void)
368 {
369 	int iommu_off = 0;
370 
371 	DBG(" -> pSeries_init_early()\n");
372 
373 	fw_feature_init();
374 
375 	if (platform_is_lpar())
376 		hpte_init_lpar();
377 	else {
378 		hpte_init_native();
379 		iommu_off = (of_chosen &&
380 			     get_property(of_chosen, "linux,iommu-off", NULL));
381 	}
382 
383 	if (platform_is_lpar())
384 		find_udbg_vterm();
385 
386 	if (firmware_has_feature(FW_FEATURE_DABR))
387 		ppc_md.set_dabr = pseries_set_dabr;
388 	else if (firmware_has_feature(FW_FEATURE_XDABR))
389 		ppc_md.set_dabr = pseries_set_xdabr;
390 
391 	iommu_init_early_pSeries();
392 
393 	pSeries_discover_pic();
394 
395 	DBG(" <- pSeries_init_early()\n");
396 }
397 
398 
399 static int pSeries_check_legacy_ioport(unsigned int baseport)
400 {
401 	struct device_node *np;
402 
403 #define I8042_DATA_REG	0x60
404 #define FDC_BASE	0x3f0
405 
406 
407 	switch(baseport) {
408 	case I8042_DATA_REG:
409 		np = of_find_node_by_type(NULL, "8042");
410 		if (np == NULL)
411 			return -ENODEV;
412 		of_node_put(np);
413 		break;
414 	case FDC_BASE:
415 		np = of_find_node_by_type(NULL, "fdc");
416 		if (np == NULL)
417 			return -ENODEV;
418 		of_node_put(np);
419 		break;
420 	}
421 	return 0;
422 }
423 
424 /*
425  * Called very early, MMU is off, device-tree isn't unflattened
426  */
427 extern struct machdep_calls pSeries_md;
428 
429 static int __init pSeries_probe(int platform)
430 {
431 	if (platform != PLATFORM_PSERIES &&
432 	    platform != PLATFORM_PSERIES_LPAR)
433 		return 0;
434 
435 	/* if we have some ppc_md fixups for LPAR to do, do
436 	 * it here ...
437 	 */
438 
439 	return 1;
440 }
441 
442 DECLARE_PER_CPU(unsigned long, smt_snooze_delay);
443 
444 static inline void dedicated_idle_sleep(unsigned int cpu)
445 {
446 	struct paca_struct *ppaca = &paca[cpu ^ 1];
447 
448 	/* Only sleep if the other thread is not idle */
449 	if (!(ppaca->lppaca.idle)) {
450 		local_irq_disable();
451 
452 		/*
453 		 * We are about to sleep the thread and so wont be polling any
454 		 * more.
455 		 */
456 		clear_thread_flag(TIF_POLLING_NRFLAG);
457 		smp_mb__after_clear_bit();
458 
459 		/*
460 		 * SMT dynamic mode. Cede will result in this thread going
461 		 * dormant, if the partner thread is still doing work.  Thread
462 		 * wakes up if partner goes idle, an interrupt is presented, or
463 		 * a prod occurs.  Returning from the cede enables external
464 		 * interrupts.
465 		 */
466 		if (!need_resched())
467 			cede_processor();
468 		else
469 			local_irq_enable();
470 		set_thread_flag(TIF_POLLING_NRFLAG);
471 	} else {
472 		/*
473 		 * Give the HV an opportunity at the processor, since we are
474 		 * not doing any work.
475 		 */
476 		poll_pending();
477 	}
478 }
479 
480 static void pseries_dedicated_idle(void)
481 {
482 	struct paca_struct *lpaca = get_paca();
483 	unsigned int cpu = smp_processor_id();
484 	unsigned long start_snooze;
485 	unsigned long *smt_snooze_delay = &__get_cpu_var(smt_snooze_delay);
486 	set_thread_flag(TIF_POLLING_NRFLAG);
487 
488 	while (1) {
489 		/*
490 		 * Indicate to the HV that we are idle. Now would be
491 		 * a good time to find other work to dispatch.
492 		 */
493 		lpaca->lppaca.idle = 1;
494 
495 		if (!need_resched()) {
496 			start_snooze = get_tb() +
497 				*smt_snooze_delay * tb_ticks_per_usec;
498 
499 			while (!need_resched() && !cpu_is_offline(cpu)) {
500 				ppc64_runlatch_off();
501 
502 				/*
503 				 * Go into low thread priority and possibly
504 				 * low power mode.
505 				 */
506 				HMT_low();
507 				HMT_very_low();
508 
509 				if (*smt_snooze_delay != 0 &&
510 				    get_tb() > start_snooze) {
511 					HMT_medium();
512 					dedicated_idle_sleep(cpu);
513 				}
514 
515 			}
516 
517 			HMT_medium();
518 		}
519 
520 		lpaca->lppaca.idle = 0;
521 		ppc64_runlatch_on();
522 
523 		preempt_enable_no_resched();
524 		schedule();
525 		preempt_disable();
526 
527 		if (cpu_is_offline(cpu) && system_state == SYSTEM_RUNNING)
528 			cpu_die();
529 	}
530 }
531 
532 static void pseries_shared_idle(void)
533 {
534 	struct paca_struct *lpaca = get_paca();
535 	unsigned int cpu = smp_processor_id();
536 
537 	while (1) {
538 		/*
539 		 * Indicate to the HV that we are idle. Now would be
540 		 * a good time to find other work to dispatch.
541 		 */
542 		lpaca->lppaca.idle = 1;
543 
544 		while (!need_resched() && !cpu_is_offline(cpu)) {
545 			local_irq_disable();
546 			ppc64_runlatch_off();
547 
548 			/*
549 			 * Yield the processor to the hypervisor.  We return if
550 			 * an external interrupt occurs (which are driven prior
551 			 * to returning here) or if a prod occurs from another
552 			 * processor. When returning here, external interrupts
553 			 * are enabled.
554 			 *
555 			 * Check need_resched() again with interrupts disabled
556 			 * to avoid a race.
557 			 */
558 			if (!need_resched())
559 				cede_processor();
560 			else
561 				local_irq_enable();
562 
563 			HMT_medium();
564 		}
565 
566 		lpaca->lppaca.idle = 0;
567 		ppc64_runlatch_on();
568 
569 		preempt_enable_no_resched();
570 		schedule();
571 		preempt_disable();
572 
573 		if (cpu_is_offline(cpu) && system_state == SYSTEM_RUNNING)
574 			cpu_die();
575 	}
576 }
577 
578 static int pSeries_pci_probe_mode(struct pci_bus *bus)
579 {
580 	if (platform_is_lpar())
581 		return PCI_PROBE_DEVTREE;
582 	return PCI_PROBE_NORMAL;
583 }
584 
585 #ifdef CONFIG_KEXEC
586 static void pseries_kexec_cpu_down(int crash_shutdown, int secondary)
587 {
588 	/* Don't risk a hypervisor call if we're crashing */
589 	if (!crash_shutdown) {
590 		unsigned long vpa = __pa(&get_paca()->lppaca);
591 
592 		if (unregister_vpa(hard_smp_processor_id(), vpa)) {
593 			printk("VPA deregistration of cpu %u (hw_cpu_id %d) "
594 					"failed\n", smp_processor_id(),
595 					hard_smp_processor_id());
596 		}
597 	}
598 
599 	if (ppc64_interrupt_controller == IC_OPEN_PIC)
600 		mpic_teardown_this_cpu(secondary);
601 	else
602 		xics_teardown_cpu(secondary);
603 }
604 #endif
605 
606 struct machdep_calls __initdata pSeries_md = {
607 	.probe			= pSeries_probe,
608 	.setup_arch		= pSeries_setup_arch,
609 	.init_early		= pSeries_init_early,
610 	.show_cpuinfo		= pSeries_show_cpuinfo,
611 	.log_error		= pSeries_log_error,
612 	.pcibios_fixup		= pSeries_final_fixup,
613 	.pci_probe_mode		= pSeries_pci_probe_mode,
614 	.irq_bus_setup		= pSeries_irq_bus_setup,
615 	.restart		= rtas_restart,
616 	.power_off		= rtas_power_off,
617 	.halt			= rtas_halt,
618 	.panic			= rtas_os_term,
619 	.cpu_die		= pSeries_mach_cpu_die,
620 	.get_boot_time		= rtas_get_boot_time,
621 	.get_rtc_time		= rtas_get_rtc_time,
622 	.set_rtc_time		= rtas_set_rtc_time,
623 	.calibrate_decr		= generic_calibrate_decr,
624 	.progress		= rtas_progress,
625 	.check_legacy_ioport	= pSeries_check_legacy_ioport,
626 	.system_reset_exception = pSeries_system_reset_exception,
627 	.machine_check_exception = pSeries_machine_check_exception,
628 #ifdef CONFIG_KEXEC
629 	.kexec_cpu_down		= pseries_kexec_cpu_down,
630 	.machine_kexec		= default_machine_kexec,
631 	.machine_kexec_prepare	= default_machine_kexec_prepare,
632 	.machine_crash_shutdown	= default_machine_crash_shutdown,
633 #endif
634 };
635