1/*
  2 * Copyright 2010 Tilera Corporation. All Rights Reserved.
  3 *
  4 *   This program is free software; you can redistribute it and/or
  5 *   modify it under the terms of the GNU General Public License
  6 *   as published by the Free Software Foundation, version 2.
  7 *
  8 *   This program is distributed in the hope that it will be useful, but
  9 *   WITHOUT ANY WARRANTY; without even the implied warranty of
 10 *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
 11 *   NON INFRINGEMENT.  See the GNU General Public License for
 12 *   more details.
 13 *
 14 * based on machine_kexec.c from other architectures in linux-2.6.18
 15 */
 16
 17#include <linux/mm.h>
 18#include <linux/kexec.h>
 19#include <linux/delay.h>
 20#include <linux/reboot.h>
 21#include <linux/errno.h>
 22#include <linux/vmalloc.h>
 23#include <linux/cpumask.h>
 24#include <linux/kernel.h>
 25#include <linux/elf.h>
 26#include <linux/highmem.h>
 27#include <linux/mmu_context.h>
 28#include <linux/io.h>
 29#include <linux/timex.h>
 30#include <asm/pgtable.h>
 31#include <asm/pgalloc.h>
 32#include <asm/cacheflush.h>
 33#include <asm/checksum.h>
 34#include <asm/tlbflush.h>
 35#include <asm/homecache.h>
 36#include <hv/hypervisor.h>
 37
 38
 39/*
 40 * This stuff is not in elf.h and is not in any other kernel include.
 41 * This stuff is needed below in the little boot notes parser to
 42 * extract the command line so we can pass it to the hypervisor.
 43 */
 44struct Elf32_Bhdr {
 45	Elf32_Word b_signature;
 46	Elf32_Word b_size;
 47	Elf32_Half b_checksum;
 48	Elf32_Half b_records;
 49};
 50#define ELF_BOOT_MAGIC		0x0E1FB007
 51#define EBN_COMMAND_LINE	0x00000004
 52#define roundupsz(X) (((X) + 3) & ~3)
 53
 54/* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
 55
 56
 57void machine_shutdown(void)
 58{
 59	/*
 60	 * Normally we would stop all the other processors here, but
 61	 * the check in machine_kexec_prepare below ensures we'll only
 62	 * get this far if we've been booted with "nosmp" on the
 63	 * command line or without CONFIG_SMP so there's nothing to do
 64	 * here (for now).
 65	 */
 66}
 67
 68void machine_crash_shutdown(struct pt_regs *regs)
 69{
 70	/*
 71	 * Cannot happen.  This type of kexec is disabled on this
 72	 * architecture (and enforced in machine_kexec_prepare below).
 73	 */
 74}
 75
 76
 77int machine_kexec_prepare(struct kimage *image)
 78{
 79	if (num_online_cpus() > 1) {
 80		pr_warn("%s: detected attempt to kexec with num_online_cpus() > 1\n",
 81			__func__);
 82		return -ENOSYS;
 83	}
 84	if (image->type != KEXEC_TYPE_DEFAULT) {
 85		pr_warn("%s: detected attempt to kexec with unsupported type: %d\n",
 86			__func__, image->type);
 87		return -ENOSYS;
 88	}
 89	return 0;
 90}
 91
 92void machine_kexec_cleanup(struct kimage *image)
 93{
 94	/*
 95	 * We did nothing in machine_kexec_prepare,
 96	 * so we have nothing to do here.
 97	 */
 98}
 99
100/*
101 * If we can find elf boot notes on this page, return the command
102 * line.  Otherwise, silently return null.  Somewhat kludgy, but no
103 * good way to do this without significantly rearchitecting the
104 * architecture-independent kexec code.
105 */
106
107static unsigned char *kexec_bn2cl(void *pg)
108{
109	struct Elf32_Bhdr *bhdrp;
110	Elf32_Nhdr *nhdrp;
111	unsigned char *desc;
112	unsigned char *command_line;
113	__sum16 csum;
114
115	bhdrp = (struct Elf32_Bhdr *) pg;
116
117	/*
118	 * This routine is invoked for every source page, so make
119	 * sure to quietly ignore every impossible page.
120	 */
121	if (bhdrp->b_signature != ELF_BOOT_MAGIC ||
122	    bhdrp->b_size > PAGE_SIZE)
123		return 0;
124
125	/*
126	 * If we get a checksum mismatch, warn with the checksum
127	 * so we can diagnose better.
128	 */
129	csum = ip_compute_csum(pg, bhdrp->b_size);
130	if (csum != 0) {
131		pr_warn("%s: bad checksum %#x (size %d)\n",
132			__func__, csum, bhdrp->b_size);
133		return 0;
134	}
135
136	nhdrp = (Elf32_Nhdr *) (bhdrp + 1);
137
138	while (nhdrp->n_type != EBN_COMMAND_LINE) {
139
140		desc = (unsigned char *) (nhdrp + 1);
141		desc += roundupsz(nhdrp->n_descsz);
142
143		nhdrp = (Elf32_Nhdr *) desc;
144
145		/* still in bounds? */
146		if ((unsigned char *) (nhdrp + 1) >
147		    ((unsigned char *) pg) + bhdrp->b_size) {
148
149			pr_info("%s: out of bounds\n", __func__);
150			return 0;
151		}
152	}
153
154	command_line = (unsigned char *) (nhdrp + 1);
155	desc = command_line;
156
157	while (*desc != '\0') {
158		desc++;
159		if (((unsigned long)desc & PAGE_MASK) != (unsigned long)pg) {
160			pr_info("%s: ran off end of page\n", __func__);
161			return 0;
162		}
163	}
164
165	return command_line;
166}
167
168static void kexec_find_and_set_command_line(struct kimage *image)
169{
170	kimage_entry_t *ptr, entry;
171
172	unsigned char *command_line = 0;
173	unsigned char *r;
174	HV_Errno hverr;
175
176	for (ptr = &image->head;
177	     (entry = *ptr) && !(entry & IND_DONE);
178	     ptr = (entry & IND_INDIRECTION) ?
179		     phys_to_virt((entry & PAGE_MASK)) : ptr + 1) {
180
181		if ((entry & IND_SOURCE)) {
182			void *va =
183				kmap_atomic_pfn(entry >> PAGE_SHIFT);
184			r = kexec_bn2cl(va);
185			if (r) {
186				command_line = r;
187				break;
188			}
189			kunmap_atomic(va);
190		}
191	}
192
193	if (command_line != 0) {
194		pr_info("setting new command line to \"%s\"\n", command_line);
195
196		hverr = hv_set_command_line(
197			(HV_VirtAddr) command_line, strlen(command_line));
198		kunmap_atomic(command_line);
199	} else {
200		pr_info("%s: no command line found; making empty\n", __func__);
201		hverr = hv_set_command_line((HV_VirtAddr) command_line, 0);
202	}
203	if (hverr)
204		pr_warn("%s: hv_set_command_line returned error: %d\n",
205			__func__, hverr);
206}
207
208/*
209 * The kexec code range-checks all its PAs, so to avoid having it run
210 * amok and allocate memory and then sequester it from every other
211 * controller, we force it to come from controller zero.  We also
212 * disable the oom-killer since if we do end up running out of memory,
213 * that almost certainly won't help.
214 */
215struct page *kimage_alloc_pages_arch(gfp_t gfp_mask, unsigned int order)
216{
217	gfp_mask |= __GFP_THISNODE | __GFP_NORETRY;
218	return alloc_pages_node(0, gfp_mask, order);
219}
220
221/*
222 * Address range in which pa=va mapping is set in setup_quasi_va_is_pa().
223 * For tilepro, PAGE_OFFSET is used since this is the largest possbile value
224 * for tilepro, while for tilegx, we limit it to entire middle level page
225 * table which we assume has been allocated and is undoubtedly large enough.
226 */
227#ifndef __tilegx__
228#define	QUASI_VA_IS_PA_ADDR_RANGE PAGE_OFFSET
229#else
230#define	QUASI_VA_IS_PA_ADDR_RANGE PGDIR_SIZE
231#endif
232
233static void setup_quasi_va_is_pa(void)
234{
235	HV_PTE pte;
236	unsigned long i;
237
238	/*
239	 * Flush our TLB to prevent conflicts between the previous contents
240	 * and the new stuff we're about to add.
241	 */
242	local_flush_tlb_all();
243
244	/*
245	 * setup VA is PA, at least up to QUASI_VA_IS_PA_ADDR_RANGE.
246	 * Note here we assume that level-1 page table is defined by
247	 * HPAGE_SIZE.
248	 */
249	pte = hv_pte(_PAGE_KERNEL | _PAGE_HUGE_PAGE);
250	pte = hv_pte_set_mode(pte, HV_PTE_MODE_CACHE_NO_L3);
251	for (i = 0; i < (QUASI_VA_IS_PA_ADDR_RANGE >> HPAGE_SHIFT); i++) {
252		unsigned long vaddr = i << HPAGE_SHIFT;
253		pgd_t *pgd = pgd_offset(current->mm, vaddr);
254		pud_t *pud = pud_offset(pgd, vaddr);
255		pte_t *ptep = (pte_t *) pmd_offset(pud, vaddr);
256		unsigned long pfn = i << (HPAGE_SHIFT - PAGE_SHIFT);
257
258		if (pfn_valid(pfn))
259			__set_pte(ptep, pfn_pte(pfn, pte));
260	}
261}
262
263
264void machine_kexec(struct kimage *image)
265{
266	void *reboot_code_buffer;
267	pte_t *ptep;
268	void (*rnk)(unsigned long, void *, unsigned long)
269		__noreturn;
270
271	/* Mask all interrupts before starting to reboot. */
272	interrupt_mask_set_mask(~0ULL);
273
274	kexec_find_and_set_command_line(image);
275
276	/*
277	 * Adjust the home caching of the control page to be cached on
278	 * this cpu, and copy the assembly helper into the control
279	 * code page, which we map in the vmalloc area.
280	 */
281	homecache_change_page_home(image->control_code_page, 0,
282				   smp_processor_id());
283	reboot_code_buffer = page_address(image->control_code_page);
284	BUG_ON(reboot_code_buffer == NULL);
285	ptep = virt_to_pte(NULL, (unsigned long)reboot_code_buffer);
286	__set_pte(ptep, pte_mkexec(*ptep));
287	memcpy(reboot_code_buffer, relocate_new_kernel,
288	       relocate_new_kernel_size);
289	__flush_icache_range(
290		(unsigned long) reboot_code_buffer,
291		(unsigned long) reboot_code_buffer + relocate_new_kernel_size);
292
293	setup_quasi_va_is_pa();
294
295	/* now call it */
296	rnk = reboot_code_buffer;
297	(*rnk)(image->head, reboot_code_buffer, image->start);
298}