]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - arch/powerpc/mm/pgtable-radix.c
powerpc/mm/radix: Update PID switch sequence
[mirror_ubuntu-bionic-kernel.git] / arch / powerpc / mm / pgtable-radix.c
CommitLineData
2bfd65e4
AK
1/*
2 * Page table handling routines for radix page table.
3 *
4 * Copyright 2015-2016, Aneesh Kumar K.V, IBM Corporation.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11#include <linux/sched.h>
12#include <linux/memblock.h>
13#include <linux/of_fdt.h>
14
15#include <asm/pgtable.h>
16#include <asm/pgalloc.h>
17#include <asm/dma.h>
18#include <asm/machdep.h>
19#include <asm/mmu.h>
20#include <asm/firmware.h>
21
bde3eb62
AK
22#include <trace/events/thp.h>
23
2bfd65e4
AK
24static int native_update_partition_table(u64 patb1)
25{
26 partition_tb->patb1 = cpu_to_be64(patb1);
27 return 0;
28}
29
30static __ref void *early_alloc_pgtable(unsigned long size)
31{
32 void *pt;
33
34 pt = __va(memblock_alloc_base(size, size, MEMBLOCK_ALLOC_ANYWHERE));
35 memset(pt, 0, size);
36
37 return pt;
38}
39
40int radix__map_kernel_page(unsigned long ea, unsigned long pa,
41 pgprot_t flags,
42 unsigned int map_page_size)
43{
44 pgd_t *pgdp;
45 pud_t *pudp;
46 pmd_t *pmdp;
47 pte_t *ptep;
48 /*
49 * Make sure task size is correct as per the max adddr
50 */
51 BUILD_BUG_ON(TASK_SIZE_USER64 > RADIX_PGTABLE_RANGE);
52 if (slab_is_available()) {
53 pgdp = pgd_offset_k(ea);
54 pudp = pud_alloc(&init_mm, pgdp, ea);
55 if (!pudp)
56 return -ENOMEM;
57 if (map_page_size == PUD_SIZE) {
58 ptep = (pte_t *)pudp;
59 goto set_the_pte;
60 }
61 pmdp = pmd_alloc(&init_mm, pudp, ea);
62 if (!pmdp)
63 return -ENOMEM;
64 if (map_page_size == PMD_SIZE) {
65 ptep = (pte_t *)pudp;
66 goto set_the_pte;
67 }
68 ptep = pte_alloc_kernel(pmdp, ea);
69 if (!ptep)
70 return -ENOMEM;
71 } else {
72 pgdp = pgd_offset_k(ea);
73 if (pgd_none(*pgdp)) {
74 pudp = early_alloc_pgtable(PUD_TABLE_SIZE);
75 BUG_ON(pudp == NULL);
76 pgd_populate(&init_mm, pgdp, pudp);
77 }
78 pudp = pud_offset(pgdp, ea);
79 if (map_page_size == PUD_SIZE) {
80 ptep = (pte_t *)pudp;
81 goto set_the_pte;
82 }
83 if (pud_none(*pudp)) {
84 pmdp = early_alloc_pgtable(PMD_TABLE_SIZE);
85 BUG_ON(pmdp == NULL);
86 pud_populate(&init_mm, pudp, pmdp);
87 }
88 pmdp = pmd_offset(pudp, ea);
89 if (map_page_size == PMD_SIZE) {
90 ptep = (pte_t *)pudp;
91 goto set_the_pte;
92 }
93 if (!pmd_present(*pmdp)) {
94 ptep = early_alloc_pgtable(PAGE_SIZE);
95 BUG_ON(ptep == NULL);
96 pmd_populate_kernel(&init_mm, pmdp, ptep);
97 }
98 ptep = pte_offset_kernel(pmdp, ea);
99 }
100
101set_the_pte:
102 set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT, flags));
103 smp_wmb();
104 return 0;
105}
106
107static void __init radix_init_pgtable(void)
108{
109 int loop_count;
110 u64 base, end, start_addr;
111 unsigned long rts_field;
112 struct memblock_region *reg;
113 unsigned long linear_page_size;
114
115 /* We don't support slb for radix */
116 mmu_slb_size = 0;
117 /*
118 * Create the linear mapping, using standard page size for now
119 */
120 loop_count = 0;
121 for_each_memblock(memory, reg) {
122
123 start_addr = reg->base;
124
125redo:
126 if (loop_count < 1 && mmu_psize_defs[MMU_PAGE_1G].shift)
127 linear_page_size = PUD_SIZE;
128 else if (loop_count < 2 && mmu_psize_defs[MMU_PAGE_2M].shift)
129 linear_page_size = PMD_SIZE;
130 else
131 linear_page_size = PAGE_SIZE;
132
133 base = _ALIGN_UP(start_addr, linear_page_size);
134 end = _ALIGN_DOWN(reg->base + reg->size, linear_page_size);
135
136 pr_info("Mapping range 0x%lx - 0x%lx with 0x%lx\n",
137 (unsigned long)base, (unsigned long)end,
138 linear_page_size);
139
140 while (base < end) {
141 radix__map_kernel_page((unsigned long)__va(base),
142 base, PAGE_KERNEL_X,
143 linear_page_size);
144 base += linear_page_size;
145 }
146 /*
147 * map the rest using lower page size
148 */
149 if (end < reg->base + reg->size) {
150 start_addr = end;
151 loop_count++;
152 goto redo;
153 }
154 }
155 /*
156 * Allocate Partition table and process table for the
157 * host.
158 */
159 BUILD_BUG_ON_MSG((PRTB_SIZE_SHIFT > 23), "Process table size too large.");
160 process_tb = early_alloc_pgtable(1UL << PRTB_SIZE_SHIFT);
161 /*
162 * Fill in the process table.
2bfd65e4 163 */
b23d9c5b 164 rts_field = radix__get_tree_size();
2bfd65e4
AK
165 process_tb->prtb0 = cpu_to_be64(rts_field | __pa(init_mm.pgd) | RADIX_PGD_INDEX_SIZE);
166 /*
167 * Fill in the partition table. We are suppose to use effective address
168 * of process table here. But our linear mapping also enable us to use
169 * physical address here.
170 */
171 ppc_md.update_partition_table(__pa(process_tb) | (PRTB_SIZE_SHIFT - 12) | PATB_GR);
172 pr_info("Process table %p and radix root for kernel: %p\n", process_tb, init_mm.pgd);
173}
174
175static void __init radix_init_partition_table(void)
176{
177 unsigned long rts_field;
b23d9c5b
AK
178
179 rts_field = radix__get_tree_size();
2bfd65e4
AK
180
181 BUILD_BUG_ON_MSG((PATB_SIZE_SHIFT > 24), "Partition table size too large.");
182 partition_tb = early_alloc_pgtable(1UL << PATB_SIZE_SHIFT);
183 partition_tb->patb0 = cpu_to_be64(rts_field | __pa(init_mm.pgd) |
184 RADIX_PGD_INDEX_SIZE | PATB_HR);
56547411
AK
185 pr_info("Initializing Radix MMU\n");
186 pr_info("Partition table %p\n", partition_tb);
2bfd65e4
AK
187
188 memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
189 /*
190 * update partition table control register,
191 * 64 K size.
192 */
193 mtspr(SPRN_PTCR, __pa(partition_tb) | (PATB_SIZE_SHIFT - 12));
194}
195
196void __init radix_init_native(void)
197{
198 ppc_md.update_partition_table = native_update_partition_table;
199}
200
201static int __init get_idx_from_shift(unsigned int shift)
202{
203 int idx = -1;
204
205 switch (shift) {
206 case 0xc:
207 idx = MMU_PAGE_4K;
208 break;
209 case 0x10:
210 idx = MMU_PAGE_64K;
211 break;
212 case 0x15:
213 idx = MMU_PAGE_2M;
214 break;
215 case 0x1e:
216 idx = MMU_PAGE_1G;
217 break;
218 }
219 return idx;
220}
221
222static int __init radix_dt_scan_page_sizes(unsigned long node,
223 const char *uname, int depth,
224 void *data)
225{
226 int size = 0;
227 int shift, idx;
228 unsigned int ap;
229 const __be32 *prop;
230 const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
231
232 /* We are scanning "cpu" nodes only */
233 if (type == NULL || strcmp(type, "cpu") != 0)
234 return 0;
235
236 prop = of_get_flat_dt_prop(node, "ibm,processor-radix-AP-encodings", &size);
237 if (!prop)
238 return 0;
239
240 pr_info("Page sizes from device-tree:\n");
241 for (; size >= 4; size -= 4, ++prop) {
242
243 struct mmu_psize_def *def;
244
245 /* top 3 bit is AP encoding */
246 shift = be32_to_cpu(prop[0]) & ~(0xe << 28);
247 ap = be32_to_cpu(prop[0]) >> 29;
248 pr_info("Page size sift = %d AP=0x%x\n", shift, ap);
249
250 idx = get_idx_from_shift(shift);
251 if (idx < 0)
252 continue;
253
254 def = &mmu_psize_defs[idx];
255 def->shift = shift;
256 def->ap = ap;
257 }
258
259 /* needed ? */
260 cur_cpu_spec->mmu_features &= ~MMU_FTR_NO_SLBIE_B;
261 return 1;
262}
263
264static void __init radix_init_page_sizes(void)
265{
266 int rc;
267
268 /*
269 * Try to find the available page sizes in the device-tree
270 */
271 rc = of_scan_flat_dt(radix_dt_scan_page_sizes, NULL);
272 if (rc != 0) /* Found */
273 goto found;
274 /*
275 * let's assume we have page 4k and 64k support
276 */
277 mmu_psize_defs[MMU_PAGE_4K].shift = 12;
278 mmu_psize_defs[MMU_PAGE_4K].ap = 0x0;
279
280 mmu_psize_defs[MMU_PAGE_64K].shift = 16;
281 mmu_psize_defs[MMU_PAGE_64K].ap = 0x5;
282found:
283#ifdef CONFIG_SPARSEMEM_VMEMMAP
284 if (mmu_psize_defs[MMU_PAGE_2M].shift) {
285 /*
286 * map vmemmap using 2M if available
287 */
288 mmu_vmemmap_psize = MMU_PAGE_2M;
289 }
290#endif /* CONFIG_SPARSEMEM_VMEMMAP */
291 return;
292}
293
294void __init radix__early_init_mmu(void)
295{
296 unsigned long lpcr;
2bfd65e4
AK
297
298#ifdef CONFIG_PPC_64K_PAGES
299 /* PAGE_SIZE mappings */
300 mmu_virtual_psize = MMU_PAGE_64K;
301#else
302 mmu_virtual_psize = MMU_PAGE_4K;
303#endif
304
305#ifdef CONFIG_SPARSEMEM_VMEMMAP
306 /* vmemmap mapping */
307 mmu_vmemmap_psize = mmu_virtual_psize;
308#endif
309 /*
310 * initialize page table size
311 */
312 __pte_index_size = RADIX_PTE_INDEX_SIZE;
313 __pmd_index_size = RADIX_PMD_INDEX_SIZE;
314 __pud_index_size = RADIX_PUD_INDEX_SIZE;
315 __pgd_index_size = RADIX_PGD_INDEX_SIZE;
316 __pmd_cache_index = RADIX_PMD_INDEX_SIZE;
317 __pte_table_size = RADIX_PTE_TABLE_SIZE;
318 __pmd_table_size = RADIX_PMD_TABLE_SIZE;
319 __pud_table_size = RADIX_PUD_TABLE_SIZE;
320 __pgd_table_size = RADIX_PGD_TABLE_SIZE;
321
a2f41eb9
AK
322 __pmd_val_bits = RADIX_PMD_VAL_BITS;
323 __pud_val_bits = RADIX_PUD_VAL_BITS;
324 __pgd_val_bits = RADIX_PGD_VAL_BITS;
2bfd65e4 325
d6a9996e
AK
326 __kernel_virt_start = RADIX_KERN_VIRT_START;
327 __kernel_virt_size = RADIX_KERN_VIRT_SIZE;
328 __vmalloc_start = RADIX_VMALLOC_START;
329 __vmalloc_end = RADIX_VMALLOC_END;
330 vmemmap = (struct page *)RADIX_VMEMMAP_BASE;
331 ioremap_bot = IOREMAP_BASE;
bfa37087
DS
332
333#ifdef CONFIG_PCI
334 pci_io_base = ISA_IO_BASE;
335#endif
336
5ed7ecd0
AK
337 /*
338 * For now radix also use the same frag size
339 */
340 __pte_frag_nr = H_PTE_FRAG_NR;
341 __pte_frag_size_shift = H_PTE_FRAG_SIZE_SHIFT;
d6a9996e 342
a2f41eb9 343 radix_init_page_sizes();
d6c88600
AK
344 if (!firmware_has_feature(FW_FEATURE_LPAR)) {
345 lpcr = mfspr(SPRN_LPCR);
bf16cdf4 346 mtspr(SPRN_LPCR, lpcr | LPCR_UPRT | LPCR_HR);
2bfd65e4 347 radix_init_partition_table();
d6c88600 348 }
2bfd65e4
AK
349
350 radix_init_pgtable();
351}
352
353void radix__early_init_mmu_secondary(void)
354{
355 unsigned long lpcr;
356 /*
d6c88600 357 * update partition table control register and UPRT
2bfd65e4 358 */
d6c88600
AK
359 if (!firmware_has_feature(FW_FEATURE_LPAR)) {
360 lpcr = mfspr(SPRN_LPCR);
bf16cdf4 361 mtspr(SPRN_LPCR, lpcr | LPCR_UPRT | LPCR_HR);
d6c88600 362
2bfd65e4
AK
363 mtspr(SPRN_PTCR,
364 __pa(partition_tb) | (PATB_SIZE_SHIFT - 12));
d6c88600 365 }
2bfd65e4
AK
366}
367
368void radix__setup_initial_memory_limit(phys_addr_t first_memblock_base,
369 phys_addr_t first_memblock_size)
370{
177ba7c6
AK
371 /* We don't currently support the first MEMBLOCK not mapping 0
372 * physical on those processors
373 */
374 BUG_ON(first_memblock_base != 0);
375 /*
376 * We limit the allocation that depend on ppc64_rma_size
377 * to first_memblock_size. We also clamp it to 1GB to
378 * avoid some funky things such as RTAS bugs.
379 *
380 * On radix config we really don't have a limitation
381 * on real mode access. But keeping it as above works
382 * well enough.
383 */
384 ppc64_rma_size = min_t(u64, first_memblock_size, 0x40000000);
385 /*
386 * Finally limit subsequent allocations. We really don't want
387 * to limit the memblock allocations to rma_size. FIXME!! should
388 * we even limit at all ?
389 */
2bfd65e4
AK
390 memblock_set_current_limit(first_memblock_base + first_memblock_size);
391}
d9225ad9
AK
392
393#ifdef CONFIG_SPARSEMEM_VMEMMAP
394int __meminit radix__vmemmap_create_mapping(unsigned long start,
395 unsigned long page_size,
396 unsigned long phys)
397{
398 /* Create a PTE encoding */
399 unsigned long flags = _PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_KERNEL_RW;
400
401 BUG_ON(radix__map_kernel_page(start, phys, __pgprot(flags), page_size));
402 return 0;
403}
404
405#ifdef CONFIG_MEMORY_HOTPLUG
406void radix__vmemmap_remove_mapping(unsigned long start, unsigned long page_size)
407{
408 /* FIXME!! intel does more. We should free page tables mapping vmemmap ? */
409}
410#endif
411#endif
bde3eb62
AK
412
413#ifdef CONFIG_TRANSPARENT_HUGEPAGE
414
415unsigned long radix__pmd_hugepage_update(struct mm_struct *mm, unsigned long addr,
416 pmd_t *pmdp, unsigned long clr,
417 unsigned long set)
418{
419 unsigned long old;
420
421#ifdef CONFIG_DEBUG_VM
422 WARN_ON(!radix__pmd_trans_huge(*pmdp));
423 assert_spin_locked(&mm->page_table_lock);
424#endif
425
426 old = radix__pte_update(mm, addr, (pte_t *)pmdp, clr, set, 1);
427 trace_hugepage_update(addr, old, clr, set);
428
429 return old;
430}
431
432pmd_t radix__pmdp_collapse_flush(struct vm_area_struct *vma, unsigned long address,
433 pmd_t *pmdp)
434
435{
436 pmd_t pmd;
437
438 VM_BUG_ON(address & ~HPAGE_PMD_MASK);
439 VM_BUG_ON(radix__pmd_trans_huge(*pmdp));
440 /*
441 * khugepaged calls this for normal pmd
442 */
443 pmd = *pmdp;
444 pmd_clear(pmdp);
445 /*FIXME!! Verify whether we need this kick below */
446 kick_all_cpus_sync();
447 flush_tlb_range(vma, address, address + HPAGE_PMD_SIZE);
448 return pmd;
449}
450
451/*
452 * For us pgtable_t is pte_t *. Inorder to save the deposisted
453 * page table, we consider the allocated page table as a list
454 * head. On withdraw we need to make sure we zero out the used
455 * list_head memory area.
456 */
457void radix__pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
458 pgtable_t pgtable)
459{
460 struct list_head *lh = (struct list_head *) pgtable;
461
462 assert_spin_locked(pmd_lockptr(mm, pmdp));
463
464 /* FIFO */
465 if (!pmd_huge_pte(mm, pmdp))
466 INIT_LIST_HEAD(lh);
467 else
468 list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
469 pmd_huge_pte(mm, pmdp) = pgtable;
470}
471
472pgtable_t radix__pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
473{
474 pte_t *ptep;
475 pgtable_t pgtable;
476 struct list_head *lh;
477
478 assert_spin_locked(pmd_lockptr(mm, pmdp));
479
480 /* FIFO */
481 pgtable = pmd_huge_pte(mm, pmdp);
482 lh = (struct list_head *) pgtable;
483 if (list_empty(lh))
484 pmd_huge_pte(mm, pmdp) = NULL;
485 else {
486 pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
487 list_del(lh);
488 }
489 ptep = (pte_t *) pgtable;
490 *ptep = __pte(0);
491 ptep++;
492 *ptep = __pte(0);
493 return pgtable;
494}
495
496
497pmd_t radix__pmdp_huge_get_and_clear(struct mm_struct *mm,
498 unsigned long addr, pmd_t *pmdp)
499{
500 pmd_t old_pmd;
501 unsigned long old;
502
503 old = radix__pmd_hugepage_update(mm, addr, pmdp, ~0UL, 0);
504 old_pmd = __pmd(old);
505 /*
506 * Serialize against find_linux_pte_or_hugepte which does lock-less
507 * lookup in page tables with local interrupts disabled. For huge pages
508 * it casts pmd_t to pte_t. Since format of pte_t is different from
509 * pmd_t we want to prevent transit from pmd pointing to page table
510 * to pmd pointing to huge page (and back) while interrupts are disabled.
511 * We clear pmd to possibly replace it with page table pointer in
512 * different code paths. So make sure we wait for the parallel
513 * find_linux_pte_or_hugepage to finish.
514 */
515 kick_all_cpus_sync();
516 return old_pmd;
517}
518
519int radix__has_transparent_hugepage(void)
520{
521 /* For radix 2M at PMD level means thp */
522 if (mmu_psize_defs[MMU_PAGE_2M].shift == PMD_SHIFT)
523 return 1;
524 return 0;
525}
526#endif /* CONFIG_TRANSPARENT_HUGEPAGE */