xref: /openbmc/linux/arch/powerpc/mm/book3s32/mmu.c (revision 2cf1c348)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * This file contains the routines for handling the MMU on those
4  * PowerPC implementations where the MMU substantially follows the
5  * architecture specification.  This includes the 6xx, 7xx, 7xxx,
6  * and 8260 implementations but excludes the 8xx and 4xx.
7  *  -- paulus
8  *
9  *  Derived from arch/ppc/mm/init.c:
10  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
11  *
12  *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
13  *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
14  *    Copyright (C) 1996 Paul Mackerras
15  *
16  *  Derived from "arch/i386/mm/init.c"
17  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
18  */
19 
20 #include <linux/kernel.h>
21 #include <linux/mm.h>
22 #include <linux/init.h>
23 #include <linux/highmem.h>
24 #include <linux/memblock.h>
25 
26 #include <asm/prom.h>
27 #include <asm/mmu.h>
28 #include <asm/machdep.h>
29 #include <asm/code-patching.h>
30 #include <asm/sections.h>
31 
32 #include <mm/mmu_decl.h>
33 
34 u8 __initdata early_hash[SZ_256K] __aligned(SZ_256K) = {0};
35 
36 static struct hash_pte __initdata *Hash = (struct hash_pte *)early_hash;
37 static unsigned long __initdata Hash_size, Hash_mask;
38 static unsigned int __initdata hash_mb, hash_mb2;
39 unsigned long __initdata _SDR1;
40 
41 struct ppc_bat BATS[8][2];	/* 8 pairs of IBAT, DBAT */
42 
43 static struct batrange {	/* stores address ranges mapped by BATs */
44 	unsigned long start;
45 	unsigned long limit;
46 	phys_addr_t phys;
47 } bat_addrs[8];
48 
49 #ifdef CONFIG_SMP
50 unsigned long mmu_hash_lock;
51 #endif
52 
53 /*
54  * Return PA for this VA if it is mapped by a BAT, or 0
55  */
56 phys_addr_t v_block_mapped(unsigned long va)
57 {
58 	int b;
59 	for (b = 0; b < ARRAY_SIZE(bat_addrs); ++b)
60 		if (va >= bat_addrs[b].start && va < bat_addrs[b].limit)
61 			return bat_addrs[b].phys + (va - bat_addrs[b].start);
62 	return 0;
63 }
64 
65 /*
66  * Return VA for a given PA or 0 if not mapped
67  */
68 unsigned long p_block_mapped(phys_addr_t pa)
69 {
70 	int b;
71 	for (b = 0; b < ARRAY_SIZE(bat_addrs); ++b)
72 		if (pa >= bat_addrs[b].phys
73 	    	    && pa < (bat_addrs[b].limit-bat_addrs[b].start)
74 		              +bat_addrs[b].phys)
75 			return bat_addrs[b].start+(pa-bat_addrs[b].phys);
76 	return 0;
77 }
78 
79 static int __init find_free_bat(void)
80 {
81 	int b;
82 	int n = mmu_has_feature(MMU_FTR_USE_HIGH_BATS) ? 8 : 4;
83 
84 	for (b = 0; b < n; b++) {
85 		struct ppc_bat *bat = BATS[b];
86 
87 		if (!(bat[1].batu & 3))
88 			return b;
89 	}
90 	return -1;
91 }
92 
93 /*
94  * This function calculates the size of the larger block usable to map the
95  * beginning of an area based on the start address and size of that area:
96  * - max block size is 256 on 6xx.
97  * - base address must be aligned to the block size. So the maximum block size
98  *   is identified by the lowest bit set to 1 in the base address (for instance
99  *   if base is 0x16000000, max size is 0x02000000).
100  * - block size has to be a power of two. This is calculated by finding the
101  *   highest bit set to 1.
102  */
103 static unsigned int block_size(unsigned long base, unsigned long top)
104 {
105 	unsigned int max_size = SZ_256M;
106 	unsigned int base_shift = (ffs(base) - 1) & 31;
107 	unsigned int block_shift = (fls(top - base) - 1) & 31;
108 
109 	return min3(max_size, 1U << base_shift, 1U << block_shift);
110 }
111 
112 /*
113  * Set up one of the IBAT (block address translation) register pairs.
114  * The parameters are not checked; in particular size must be a power
115  * of 2 between 128k and 256M.
116  */
117 static void setibat(int index, unsigned long virt, phys_addr_t phys,
118 		    unsigned int size, pgprot_t prot)
119 {
120 	unsigned int bl = (size >> 17) - 1;
121 	int wimgxpp;
122 	struct ppc_bat *bat = BATS[index];
123 	unsigned long flags = pgprot_val(prot);
124 
125 	if (!cpu_has_feature(CPU_FTR_NEED_COHERENT))
126 		flags &= ~_PAGE_COHERENT;
127 
128 	wimgxpp = (flags & _PAGE_COHERENT) | (_PAGE_EXEC ? BPP_RX : BPP_XX);
129 	bat[0].batu = virt | (bl << 2) | 2; /* Vs=1, Vp=0 */
130 	bat[0].batl = BAT_PHYS_ADDR(phys) | wimgxpp;
131 	if (flags & _PAGE_USER)
132 		bat[0].batu |= 1;	/* Vp = 1 */
133 }
134 
135 static void clearibat(int index)
136 {
137 	struct ppc_bat *bat = BATS[index];
138 
139 	bat[0].batu = 0;
140 	bat[0].batl = 0;
141 }
142 
143 static unsigned long __init __mmu_mapin_ram(unsigned long base, unsigned long top)
144 {
145 	int idx;
146 
147 	while ((idx = find_free_bat()) != -1 && base != top) {
148 		unsigned int size = block_size(base, top);
149 
150 		if (size < 128 << 10)
151 			break;
152 		setbat(idx, PAGE_OFFSET + base, base, size, PAGE_KERNEL_X);
153 		base += size;
154 	}
155 
156 	return base;
157 }
158 
159 unsigned long __init mmu_mapin_ram(unsigned long base, unsigned long top)
160 {
161 	unsigned long done;
162 	unsigned long border = (unsigned long)__init_begin - PAGE_OFFSET;
163 
164 
165 	if (debug_pagealloc_enabled_or_kfence() || __map_without_bats) {
166 		pr_debug_once("Read-Write memory mapped without BATs\n");
167 		if (base >= border)
168 			return base;
169 		if (top >= border)
170 			top = border;
171 	}
172 
173 	if (!strict_kernel_rwx_enabled() || base >= border || top <= border)
174 		return __mmu_mapin_ram(base, top);
175 
176 	done = __mmu_mapin_ram(base, border);
177 	if (done != border)
178 		return done;
179 
180 	return __mmu_mapin_ram(border, top);
181 }
182 
183 static bool is_module_segment(unsigned long addr)
184 {
185 	if (!IS_ENABLED(CONFIG_MODULES))
186 		return false;
187 	if (addr < ALIGN_DOWN(MODULES_VADDR, SZ_256M))
188 		return false;
189 	if (addr > ALIGN(MODULES_END, SZ_256M) - 1)
190 		return false;
191 	return true;
192 }
193 
194 void mmu_mark_initmem_nx(void)
195 {
196 	int nb = mmu_has_feature(MMU_FTR_USE_HIGH_BATS) ? 8 : 4;
197 	int i;
198 	unsigned long base = (unsigned long)_stext - PAGE_OFFSET;
199 	unsigned long top = ALIGN((unsigned long)_etext - PAGE_OFFSET, SZ_128K);
200 	unsigned long border = (unsigned long)__init_begin - PAGE_OFFSET;
201 	unsigned long size;
202 
203 	for (i = 0; i < nb - 1 && base < top;) {
204 		size = block_size(base, top);
205 		setibat(i++, PAGE_OFFSET + base, base, size, PAGE_KERNEL_TEXT);
206 		base += size;
207 	}
208 	if (base < top) {
209 		size = block_size(base, top);
210 		if ((top - base) > size) {
211 			size <<= 1;
212 			if (strict_kernel_rwx_enabled() && base + size > border)
213 				pr_warn("Some RW data is getting mapped X. "
214 					"Adjust CONFIG_DATA_SHIFT to avoid that.\n");
215 		}
216 		setibat(i++, PAGE_OFFSET + base, base, size, PAGE_KERNEL_TEXT);
217 		base += size;
218 	}
219 	for (; i < nb; i++)
220 		clearibat(i);
221 
222 	update_bats();
223 
224 	for (i = TASK_SIZE >> 28; i < 16; i++) {
225 		/* Do not set NX on VM space for modules */
226 		if (is_module_segment(i << 28))
227 			continue;
228 
229 		mtsr(mfsr(i << 28) | 0x10000000, i << 28);
230 	}
231 }
232 
233 void mmu_mark_rodata_ro(void)
234 {
235 	int nb = mmu_has_feature(MMU_FTR_USE_HIGH_BATS) ? 8 : 4;
236 	int i;
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;
260 	unsigned long flags = pgprot_val(prot);
261 
262 	if (index == -1)
263 		index = find_free_bat();
264 	if (index == -1) {
265 		pr_err("%s: no BAT available for mapping 0x%llx\n", __func__,
266 		       (unsigned long long)phys);
267 		return;
268 	}
269 	bat = BATS[index];
270 
271 	if ((flags & _PAGE_NO_CACHE) ||
272 	    (cpu_has_feature(CPU_FTR_NEED_COHERENT) == 0))
273 		flags &= ~_PAGE_COHERENT;
274 
275 	bl = (size >> 17) - 1;
276 	/* Do DBAT first */
277 	wimgxpp = flags & (_PAGE_WRITETHRU | _PAGE_NO_CACHE
278 			   | _PAGE_COHERENT | _PAGE_GUARDED);
279 	wimgxpp |= (flags & _PAGE_RW)? BPP_RW: BPP_RX;
280 	bat[1].batu = virt | (bl << 2) | 2; /* Vs=1, Vp=0 */
281 	bat[1].batl = BAT_PHYS_ADDR(phys) | wimgxpp;
282 	if (flags & _PAGE_USER)
283 		bat[1].batu |= 1; 	/* Vp = 1 */
284 	if (flags & _PAGE_GUARDED) {
285 		/* G bit must be zero in IBATs */
286 		flags &= ~_PAGE_EXEC;
287 	}
288 	if (flags & _PAGE_EXEC)
289 		bat[0] = bat[1];
290 	else
291 		bat[0].batu = bat[0].batl = 0;
292 
293 	bat_addrs[index].start = virt;
294 	bat_addrs[index].limit = virt + ((bl + 1) << 17) - 1;
295 	bat_addrs[index].phys = phys;
296 }
297 
298 /*
299  * Preload a translation in the hash table
300  */
301 static void hash_preload(struct mm_struct *mm, unsigned long ea)
302 {
303 	pmd_t *pmd;
304 
305 	if (!mmu_has_feature(MMU_FTR_HPTE_TABLE))
306 		return;
307 	pmd = pmd_off(mm, ea);
308 	if (!pmd_none(*pmd))
309 		add_hash_page(mm->context.id, ea, pmd_val(*pmd));
310 }
311 
312 /*
313  * This is called at the end of handling a user page fault, when the
314  * fault has been handled by updating a PTE in the linux page tables.
315  * We use it to preload an HPTE into the hash table corresponding to
316  * the updated linux PTE.
317  *
318  * This must always be called with the pte lock held.
319  */
320 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
321 		      pte_t *ptep)
322 {
323 	if (!mmu_has_feature(MMU_FTR_HPTE_TABLE))
324 		return;
325 	/*
326 	 * We don't need to worry about _PAGE_PRESENT here because we are
327 	 * called with either mm->page_table_lock held or ptl lock held
328 	 */
329 
330 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
331 	if (!pte_young(*ptep) || address >= TASK_SIZE)
332 		return;
333 
334 	/* We have to test for regs NULL since init will get here first thing at boot */
335 	if (!current->thread.regs)
336 		return;
337 
338 	/* We also avoid filling the hash if not coming from a fault */
339 	if (TRAP(current->thread.regs) != 0x300 && TRAP(current->thread.regs) != 0x400)
340 		return;
341 
342 	hash_preload(vma->vm_mm, address);
343 }
344 
345 /*
346  * Initialize the hash table and patch the instructions in hashtable.S.
347  */
348 void __init MMU_init_hw(void)
349 {
350 	unsigned int n_hpteg, lg_n_hpteg;
351 
352 	if (!mmu_has_feature(MMU_FTR_HPTE_TABLE))
353 		return;
354 
355 	if ( ppc_md.progress ) ppc_md.progress("hash:enter", 0x105);
356 
357 #define LG_HPTEG_SIZE	6		/* 64 bytes per HPTEG */
358 #define SDR1_LOW_BITS	((n_hpteg - 1) >> 10)
359 #define MIN_N_HPTEG	1024		/* min 64kB hash table */
360 
361 	/*
362 	 * Allow 1 HPTE (1/8 HPTEG) for each page of memory.
363 	 * This is less than the recommended amount, but then
364 	 * Linux ain't AIX.
365 	 */
366 	n_hpteg = total_memory / (PAGE_SIZE * 8);
367 	if (n_hpteg < MIN_N_HPTEG)
368 		n_hpteg = MIN_N_HPTEG;
369 	lg_n_hpteg = __ilog2(n_hpteg);
370 	if (n_hpteg & (n_hpteg - 1)) {
371 		++lg_n_hpteg;		/* round up if not power of 2 */
372 		n_hpteg = 1 << lg_n_hpteg;
373 	}
374 	Hash_size = n_hpteg << LG_HPTEG_SIZE;
375 
376 	/*
377 	 * Find some memory for the hash table.
378 	 */
379 	if ( ppc_md.progress ) ppc_md.progress("hash:find piece", 0x322);
380 	Hash = memblock_alloc(Hash_size, Hash_size);
381 	if (!Hash)
382 		panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
383 		      __func__, Hash_size, Hash_size);
384 	_SDR1 = __pa(Hash) | SDR1_LOW_BITS;
385 
386 	pr_info("Total memory = %lldMB; using %ldkB for hash table\n",
387 		(unsigned long long)(total_memory >> 20), Hash_size >> 10);
388 
389 
390 	Hash_mask = n_hpteg - 1;
391 	hash_mb2 = hash_mb = 32 - LG_HPTEG_SIZE - lg_n_hpteg;
392 	if (lg_n_hpteg > 16)
393 		hash_mb2 = 16 - LG_HPTEG_SIZE;
394 }
395 
396 void __init MMU_init_hw_patch(void)
397 {
398 	unsigned int hmask = Hash_mask >> (16 - LG_HPTEG_SIZE);
399 	unsigned int hash = (unsigned int)Hash - PAGE_OFFSET;
400 
401 	if (!mmu_has_feature(MMU_FTR_HPTE_TABLE))
402 		return;
403 
404 	if (ppc_md.progress)
405 		ppc_md.progress("hash:patch", 0x345);
406 	if (ppc_md.progress)
407 		ppc_md.progress("hash:done", 0x205);
408 
409 	/* WARNING: Make sure nothing can trigger a KASAN check past this point */
410 
411 	/*
412 	 * Patch up the instructions in hashtable.S:create_hpte
413 	 */
414 	modify_instruction_site(&patch__hash_page_A0, 0xffff, hash >> 16);
415 	modify_instruction_site(&patch__hash_page_A1, 0x7c0, hash_mb << 6);
416 	modify_instruction_site(&patch__hash_page_A2, 0x7c0, hash_mb2 << 6);
417 	modify_instruction_site(&patch__hash_page_B, 0xffff, hmask);
418 	modify_instruction_site(&patch__hash_page_C, 0xffff, hmask);
419 
420 	/*
421 	 * Patch up the instructions in hashtable.S:flush_hash_page
422 	 */
423 	modify_instruction_site(&patch__flush_hash_A0, 0xffff, hash >> 16);
424 	modify_instruction_site(&patch__flush_hash_A1, 0x7c0, hash_mb << 6);
425 	modify_instruction_site(&patch__flush_hash_A2, 0x7c0, hash_mb2 << 6);
426 	modify_instruction_site(&patch__flush_hash_B, 0xffff, hmask);
427 }
428 
429 void setup_initial_memory_limit(phys_addr_t first_memblock_base,
430 				phys_addr_t first_memblock_size)
431 {
432 	/* We don't currently support the first MEMBLOCK not mapping 0
433 	 * physical on those processors
434 	 */
435 	BUG_ON(first_memblock_base != 0);
436 
437 	memblock_set_current_limit(min_t(u64, first_memblock_size, SZ_256M));
438 }
439 
440 void __init print_system_hash_info(void)
441 {
442 	pr_info("Hash_size         = 0x%lx\n", Hash_size);
443 	if (Hash_mask)
444 		pr_info("Hash_mask         = 0x%lx\n", Hash_mask);
445 }
446 
447 void __init early_init_mmu(void)
448 {
449 }
450