xref: /openbmc/linux/arch/powerpc/mm/book3s32/mmu.c (revision abade675e02e1b73da0c20ffaf08fbe309038298)
1 /*
2  * This file contains the routines for handling the MMU on those
3  * PowerPC implementations where the MMU substantially follows the
4  * architecture specification.  This includes the 6xx, 7xx, 7xxx,
5  * and 8260 implementations but excludes the 8xx and 4xx.
6  *  -- paulus
7  *
8  *  Derived from arch/ppc/mm/init.c:
9  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
10  *
11  *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
12  *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
13  *    Copyright (C) 1996 Paul Mackerras
14  *
15  *  Derived from "arch/i386/mm/init.c"
16  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
17  *
18  *  This program is free software; you can redistribute it and/or
19  *  modify it under the terms of the GNU General Public License
20  *  as published by the Free Software Foundation; either version
21  *  2 of the License, or (at your option) any later version.
22  *
23  */
24 
25 #include <linux/kernel.h>
26 #include <linux/mm.h>
27 #include <linux/init.h>
28 #include <linux/highmem.h>
29 #include <linux/memblock.h>
30 
31 #include <asm/prom.h>
32 #include <asm/mmu.h>
33 #include <asm/machdep.h>
34 #include <asm/code-patching.h>
35 #include <asm/sections.h>
36 
37 #include <mm/mmu_decl.h>
38 
39 struct hash_pte *Hash;
40 static unsigned long Hash_size, Hash_mask;
41 unsigned long _SDR1;
42 static unsigned int hash_mb, hash_mb2;
43 
44 struct ppc_bat BATS[8][2];	/* 8 pairs of IBAT, DBAT */
45 
46 struct batrange {		/* stores address ranges mapped by BATs */
47 	unsigned long start;
48 	unsigned long limit;
49 	phys_addr_t phys;
50 } bat_addrs[8];
51 
52 /*
53  * Return PA for this VA if it is mapped by a BAT, or 0
54  */
55 phys_addr_t v_block_mapped(unsigned long va)
56 {
57 	int b;
58 	for (b = 0; b < ARRAY_SIZE(bat_addrs); ++b)
59 		if (va >= bat_addrs[b].start && va < bat_addrs[b].limit)
60 			return bat_addrs[b].phys + (va - bat_addrs[b].start);
61 	return 0;
62 }
63 
64 /*
65  * Return VA for a given PA or 0 if not mapped
66  */
67 unsigned long p_block_mapped(phys_addr_t pa)
68 {
69 	int b;
70 	for (b = 0; b < ARRAY_SIZE(bat_addrs); ++b)
71 		if (pa >= bat_addrs[b].phys
72 	    	    && pa < (bat_addrs[b].limit-bat_addrs[b].start)
73 		              +bat_addrs[b].phys)
74 			return bat_addrs[b].start+(pa-bat_addrs[b].phys);
75 	return 0;
76 }
77 
78 static int find_free_bat(void)
79 {
80 	int b;
81 
82 	if (cpu_has_feature(CPU_FTR_601)) {
83 		for (b = 0; b < 4; b++) {
84 			struct ppc_bat *bat = BATS[b];
85 
86 			if (!(bat[0].batl & 0x40))
87 				return b;
88 		}
89 	} else {
90 		int n = mmu_has_feature(MMU_FTR_USE_HIGH_BATS) ? 8 : 4;
91 
92 		for (b = 0; b < n; b++) {
93 			struct ppc_bat *bat = BATS[b];
94 
95 			if (!(bat[1].batu & 3))
96 				return b;
97 		}
98 	}
99 	return -1;
100 }
101 
102 /*
103  * This function calculates the size of the larger block usable to map the
104  * beginning of an area based on the start address and size of that area:
105  * - max block size is 8M on 601 and 256 on other 6xx.
106  * - base address must be aligned to the block size. So the maximum block size
107  *   is identified by the lowest bit set to 1 in the base address (for instance
108  *   if base is 0x16000000, max size is 0x02000000).
109  * - block size has to be a power of two. This is calculated by finding the
110  *   highest bit set to 1.
111  */
112 static unsigned int block_size(unsigned long base, unsigned long top)
113 {
114 	unsigned int max_size = (cpu_has_feature(CPU_FTR_601) ? 8 : 256) << 20;
115 	unsigned int base_shift = (ffs(base) - 1) & 31;
116 	unsigned int block_shift = (fls(top - base) - 1) & 31;
117 
118 	return min3(max_size, 1U << base_shift, 1U << block_shift);
119 }
120 
121 /*
122  * Set up one of the IBAT (block address translation) register pairs.
123  * The parameters are not checked; in particular size must be a power
124  * of 2 between 128k and 256M.
125  * Only for 603+ ...
126  */
127 static void setibat(int index, unsigned long virt, phys_addr_t phys,
128 		    unsigned int size, pgprot_t prot)
129 {
130 	unsigned int bl = (size >> 17) - 1;
131 	int wimgxpp;
132 	struct ppc_bat *bat = BATS[index];
133 	unsigned long flags = pgprot_val(prot);
134 
135 	if (!cpu_has_feature(CPU_FTR_NEED_COHERENT))
136 		flags &= ~_PAGE_COHERENT;
137 
138 	wimgxpp = (flags & _PAGE_COHERENT) | (_PAGE_EXEC ? BPP_RX : BPP_XX);
139 	bat[0].batu = virt | (bl << 2) | 2; /* Vs=1, Vp=0 */
140 	bat[0].batl = BAT_PHYS_ADDR(phys) | wimgxpp;
141 	if (flags & _PAGE_USER)
142 		bat[0].batu |= 1;	/* Vp = 1 */
143 }
144 
145 static void clearibat(int index)
146 {
147 	struct ppc_bat *bat = BATS[index];
148 
149 	bat[0].batu = 0;
150 	bat[0].batl = 0;
151 }
152 
153 static unsigned long __init __mmu_mapin_ram(unsigned long base, unsigned long top)
154 {
155 	int idx;
156 
157 	while ((idx = find_free_bat()) != -1 && base != top) {
158 		unsigned int size = block_size(base, top);
159 
160 		if (size < 128 << 10)
161 			break;
162 		setbat(idx, PAGE_OFFSET + base, base, size, PAGE_KERNEL_X);
163 		base += size;
164 	}
165 
166 	return base;
167 }
168 
169 unsigned long __init mmu_mapin_ram(unsigned long base, unsigned long top)
170 {
171 	unsigned long done;
172 	unsigned long border = (unsigned long)__init_begin - PAGE_OFFSET;
173 
174 	if (__map_without_bats) {
175 		pr_debug("RAM mapped without BATs\n");
176 		return base;
177 	}
178 
179 	if (!strict_kernel_rwx_enabled() || base >= border || top <= border)
180 		return __mmu_mapin_ram(base, top);
181 
182 	done = __mmu_mapin_ram(base, border);
183 	if (done != border)
184 		return done;
185 
186 	return __mmu_mapin_ram(border, top);
187 }
188 
189 void mmu_mark_initmem_nx(void)
190 {
191 	int nb = mmu_has_feature(MMU_FTR_USE_HIGH_BATS) ? 8 : 4;
192 	int i;
193 	unsigned long base = (unsigned long)_stext - PAGE_OFFSET;
194 	unsigned long top = (unsigned long)_etext - PAGE_OFFSET;
195 	unsigned long size;
196 
197 	if (cpu_has_feature(CPU_FTR_601))
198 		return;
199 
200 	for (i = 0; i < nb - 1 && base < top && top - base > (128 << 10);) {
201 		size = block_size(base, top);
202 		setibat(i++, PAGE_OFFSET + base, base, size, PAGE_KERNEL_TEXT);
203 		base += size;
204 	}
205 	if (base < top) {
206 		size = block_size(base, top);
207 		size = max(size, 128UL << 10);
208 		if ((top - base) > size) {
209 			if (strict_kernel_rwx_enabled())
210 				pr_warn("Kernel _etext not properly aligned\n");
211 			size <<= 1;
212 		}
213 		setibat(i++, PAGE_OFFSET + base, base, size, PAGE_KERNEL_TEXT);
214 		base += size;
215 	}
216 	for (; i < nb; i++)
217 		clearibat(i);
218 
219 	update_bats();
220 
221 	for (i = TASK_SIZE >> 28; i < 16; i++) {
222 		/* Do not set NX on VM space for modules */
223 		if (IS_ENABLED(CONFIG_MODULES) &&
224 		    (VMALLOC_START & 0xf0000000) == i << 28)
225 			break;
226 		mtsrin(mfsrin(i << 28) | 0x10000000, i << 28);
227 	}
228 }
229 
230 void mmu_mark_rodata_ro(void)
231 {
232 	int nb = mmu_has_feature(MMU_FTR_USE_HIGH_BATS) ? 8 : 4;
233 	int i;
234 
235 	if (cpu_has_feature(CPU_FTR_601))
236 		return;
237 
238 	for (i = 0; i < nb; i++) {
239 		struct ppc_bat *bat = BATS[i];
240 
241 		if (bat_addrs[i].start < (unsigned long)__init_begin)
242 			bat[1].batl = (bat[1].batl & ~BPP_RW) | BPP_RX;
243 	}
244 
245 	update_bats();
246 }
247 
248 /*
249  * Set up one of the I/D BAT (block address translation) register pairs.
250  * The parameters are not checked; in particular size must be a power
251  * of 2 between 128k and 256M.
252  * On 603+, only set IBAT when _PAGE_EXEC is set
253  */
254 void __init setbat(int index, unsigned long virt, phys_addr_t phys,
255 		   unsigned int size, pgprot_t prot)
256 {
257 	unsigned int bl;
258 	int wimgxpp;
259 	struct ppc_bat *bat = BATS[index];
260 	unsigned long flags = pgprot_val(prot);
261 
262 	if ((flags & _PAGE_NO_CACHE) ||
263 	    (cpu_has_feature(CPU_FTR_NEED_COHERENT) == 0))
264 		flags &= ~_PAGE_COHERENT;
265 
266 	bl = (size >> 17) - 1;
267 	if (PVR_VER(mfspr(SPRN_PVR)) != 1) {
268 		/* 603, 604, etc. */
269 		/* Do DBAT first */
270 		wimgxpp = flags & (_PAGE_WRITETHRU | _PAGE_NO_CACHE
271 				   | _PAGE_COHERENT | _PAGE_GUARDED);
272 		wimgxpp |= (flags & _PAGE_RW)? BPP_RW: BPP_RX;
273 		bat[1].batu = virt | (bl << 2) | 2; /* Vs=1, Vp=0 */
274 		bat[1].batl = BAT_PHYS_ADDR(phys) | wimgxpp;
275 		if (flags & _PAGE_USER)
276 			bat[1].batu |= 1; 	/* Vp = 1 */
277 		if (flags & _PAGE_GUARDED) {
278 			/* G bit must be zero in IBATs */
279 			flags &= ~_PAGE_EXEC;
280 		}
281 		if (flags & _PAGE_EXEC)
282 			bat[0] = bat[1];
283 		else
284 			bat[0].batu = bat[0].batl = 0;
285 	} else {
286 		/* 601 cpu */
287 		if (bl > BL_8M)
288 			bl = BL_8M;
289 		wimgxpp = flags & (_PAGE_WRITETHRU | _PAGE_NO_CACHE
290 				   | _PAGE_COHERENT);
291 		wimgxpp |= (flags & _PAGE_RW)?
292 			((flags & _PAGE_USER)? PP_RWRW: PP_RWXX): PP_RXRX;
293 		bat->batu = virt | wimgxpp | 4;	/* Ks=0, Ku=1 */
294 		bat->batl = phys | bl | 0x40;	/* V=1 */
295 	}
296 
297 	bat_addrs[index].start = virt;
298 	bat_addrs[index].limit = virt + ((bl + 1) << 17) - 1;
299 	bat_addrs[index].phys = phys;
300 }
301 
302 /*
303  * Preload a translation in the hash table
304  */
305 void hash_preload(struct mm_struct *mm, unsigned long ea,
306 		  bool is_exec, unsigned long trap)
307 {
308 	pmd_t *pmd;
309 
310 	if (!Hash)
311 		return;
312 	pmd = pmd_offset(pud_offset(pgd_offset(mm, ea), ea), ea);
313 	if (!pmd_none(*pmd))
314 		add_hash_page(mm->context.id, ea, pmd_val(*pmd));
315 }
316 
317 /*
318  * Initialize the hash table and patch the instructions in hashtable.S.
319  */
320 void __init MMU_init_hw(void)
321 {
322 	unsigned int n_hpteg, lg_n_hpteg;
323 
324 	if (!mmu_has_feature(MMU_FTR_HPTE_TABLE))
325 		return;
326 
327 	if ( ppc_md.progress ) ppc_md.progress("hash:enter", 0x105);
328 
329 #define LG_HPTEG_SIZE	6		/* 64 bytes per HPTEG */
330 #define SDR1_LOW_BITS	((n_hpteg - 1) >> 10)
331 #define MIN_N_HPTEG	1024		/* min 64kB hash table */
332 
333 	/*
334 	 * Allow 1 HPTE (1/8 HPTEG) for each page of memory.
335 	 * This is less than the recommended amount, but then
336 	 * Linux ain't AIX.
337 	 */
338 	n_hpteg = total_memory / (PAGE_SIZE * 8);
339 	if (n_hpteg < MIN_N_HPTEG)
340 		n_hpteg = MIN_N_HPTEG;
341 	lg_n_hpteg = __ilog2(n_hpteg);
342 	if (n_hpteg & (n_hpteg - 1)) {
343 		++lg_n_hpteg;		/* round up if not power of 2 */
344 		n_hpteg = 1 << lg_n_hpteg;
345 	}
346 	Hash_size = n_hpteg << LG_HPTEG_SIZE;
347 
348 	/*
349 	 * Find some memory for the hash table.
350 	 */
351 	if ( ppc_md.progress ) ppc_md.progress("hash:find piece", 0x322);
352 	Hash = memblock_alloc(Hash_size, Hash_size);
353 	if (!Hash)
354 		panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
355 		      __func__, Hash_size, Hash_size);
356 	_SDR1 = __pa(Hash) | SDR1_LOW_BITS;
357 
358 	pr_info("Total memory = %lldMB; using %ldkB for hash table\n",
359 		(unsigned long long)(total_memory >> 20), Hash_size >> 10);
360 
361 
362 	Hash_mask = n_hpteg - 1;
363 	hash_mb2 = hash_mb = 32 - LG_HPTEG_SIZE - lg_n_hpteg;
364 	if (lg_n_hpteg > 16)
365 		hash_mb2 = 16 - LG_HPTEG_SIZE;
366 }
367 
368 void __init MMU_init_hw_patch(void)
369 {
370 	unsigned int hmask = Hash_mask >> (16 - LG_HPTEG_SIZE);
371 
372 	if (ppc_md.progress)
373 		ppc_md.progress("hash:patch", 0x345);
374 	if (ppc_md.progress)
375 		ppc_md.progress("hash:done", 0x205);
376 
377 	/* WARNING: Make sure nothing can trigger a KASAN check past this point */
378 
379 	/*
380 	 * Patch up the instructions in hashtable.S:create_hpte
381 	 */
382 	modify_instruction_site(&patch__hash_page_A0, 0xffff,
383 				((unsigned int)Hash - PAGE_OFFSET) >> 16);
384 	modify_instruction_site(&patch__hash_page_A1, 0x7c0, hash_mb << 6);
385 	modify_instruction_site(&patch__hash_page_A2, 0x7c0, hash_mb2 << 6);
386 	modify_instruction_site(&patch__hash_page_B, 0xffff, hmask);
387 	modify_instruction_site(&patch__hash_page_C, 0xffff, hmask);
388 
389 	/*
390 	 * Patch up the instructions in hashtable.S:flush_hash_page
391 	 */
392 	modify_instruction_site(&patch__flush_hash_A0, 0xffff,
393 				((unsigned int)Hash - PAGE_OFFSET) >> 16);
394 	modify_instruction_site(&patch__flush_hash_A1, 0x7c0, hash_mb << 6);
395 	modify_instruction_site(&patch__flush_hash_A2, 0x7c0, hash_mb2 << 6);
396 	modify_instruction_site(&patch__flush_hash_B, 0xffff, hmask);
397 }
398 
399 void setup_initial_memory_limit(phys_addr_t first_memblock_base,
400 				phys_addr_t first_memblock_size)
401 {
402 	/* We don't currently support the first MEMBLOCK not mapping 0
403 	 * physical on those processors
404 	 */
405 	BUG_ON(first_memblock_base != 0);
406 
407 	/* 601 can only access 16MB at the moment */
408 	if (PVR_VER(mfspr(SPRN_PVR)) == 1)
409 		memblock_set_current_limit(min_t(u64, first_memblock_size, 0x01000000));
410 	else /* Anything else has 256M mapped */
411 		memblock_set_current_limit(min_t(u64, first_memblock_size, 0x10000000));
412 }
413 
414 void __init print_system_hash_info(void)
415 {
416 	pr_info("Hash_size         = 0x%lx\n", Hash_size);
417 	if (Hash_mask)
418 		pr_info("Hash_mask         = 0x%lx\n", Hash_mask);
419 }
420 
421 #ifdef CONFIG_PPC_KUEP
422 void __init setup_kuep(bool disabled)
423 {
424 	pr_info("Activating Kernel Userspace Execution Prevention\n");
425 
426 	if (cpu_has_feature(CPU_FTR_601))
427 		pr_warn("KUEP is not working on powerpc 601 (No NX bit in Seg Regs)\n");
428 
429 	if (disabled)
430 		pr_warn("KUEP cannot be disabled yet on 6xx when compiled in\n");
431 }
432 #endif
433 
434 #ifdef CONFIG_PPC_KUAP
435 void __init setup_kuap(bool disabled)
436 {
437 	pr_info("Activating Kernel Userspace Access Protection\n");
438 
439 	if (disabled)
440 		pr_warn("KUAP cannot be disabled yet on 6xx when compiled in\n");
441 }
442 #endif
443