2 * Page table handling routines for radix page table.
4 * Copyright 2015-2016, Aneesh Kumar K.V, IBM Corporation.
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.
11 #include <linux/sched/mm.h>
12 #include <linux/memblock.h>
13 #include <linux/of_fdt.h>
16 #include <asm/pgtable.h>
17 #include <asm/pgalloc.h>
19 #include <asm/machdep.h>
21 #include <asm/firmware.h>
22 #include <asm/powernv.h>
23 #include <asm/sections.h>
24 #include <asm/trace.h>
26 #include <trace/events/thp.h>
28 unsigned int mmu_pid_bits
;
29 unsigned int mmu_base_pid
;
31 static int native_register_process_table(unsigned long base
, unsigned long pg_sz
,
32 unsigned long table_size
)
34 unsigned long patb1
= base
| table_size
| PATB_GR
;
36 partition_tb
->patb1
= cpu_to_be64(patb1
);
40 static __ref
void *early_alloc_pgtable(unsigned long size
)
44 pt
= __va(memblock_alloc_base(size
, size
, MEMBLOCK_ALLOC_ANYWHERE
));
50 int radix__map_kernel_page(unsigned long ea
, unsigned long pa
,
52 unsigned int map_page_size
)
59 * Make sure task size is correct as per the max adddr
61 BUILD_BUG_ON(TASK_SIZE_USER64
> RADIX_PGTABLE_RANGE
);
62 if (slab_is_available()) {
63 pgdp
= pgd_offset_k(ea
);
64 pudp
= pud_alloc(&init_mm
, pgdp
, ea
);
67 if (map_page_size
== PUD_SIZE
) {
71 pmdp
= pmd_alloc(&init_mm
, pudp
, ea
);
74 if (map_page_size
== PMD_SIZE
) {
75 ptep
= pmdp_ptep(pmdp
);
78 ptep
= pte_alloc_kernel(pmdp
, ea
);
82 pgdp
= pgd_offset_k(ea
);
83 if (pgd_none(*pgdp
)) {
84 pudp
= early_alloc_pgtable(PUD_TABLE_SIZE
);
86 pgd_populate(&init_mm
, pgdp
, pudp
);
88 pudp
= pud_offset(pgdp
, ea
);
89 if (map_page_size
== PUD_SIZE
) {
93 if (pud_none(*pudp
)) {
94 pmdp
= early_alloc_pgtable(PMD_TABLE_SIZE
);
96 pud_populate(&init_mm
, pudp
, pmdp
);
98 pmdp
= pmd_offset(pudp
, ea
);
99 if (map_page_size
== PMD_SIZE
) {
100 ptep
= pmdp_ptep(pmdp
);
103 if (!pmd_present(*pmdp
)) {
104 ptep
= early_alloc_pgtable(PAGE_SIZE
);
105 BUG_ON(ptep
== NULL
);
106 pmd_populate_kernel(&init_mm
, pmdp
, ptep
);
108 ptep
= pte_offset_kernel(pmdp
, ea
);
112 set_pte_at(&init_mm
, ea
, ptep
, pfn_pte(pa
>> PAGE_SHIFT
, flags
));
117 #ifdef CONFIG_STRICT_KERNEL_RWX
118 void radix__change_memory_range(unsigned long start
, unsigned long end
,
127 start
= ALIGN_DOWN(start
, PAGE_SIZE
);
128 end
= PAGE_ALIGN(end
); // aligns up
130 pr_debug("Changing flags on range %lx-%lx removing 0x%lx\n",
133 for (idx
= start
; idx
< end
; idx
+= PAGE_SIZE
) {
134 pgdp
= pgd_offset_k(idx
);
135 pudp
= pud_alloc(&init_mm
, pgdp
, idx
);
138 if (pud_huge(*pudp
)) {
139 ptep
= (pte_t
*)pudp
;
142 pmdp
= pmd_alloc(&init_mm
, pudp
, idx
);
145 if (pmd_huge(*pmdp
)) {
146 ptep
= pmdp_ptep(pmdp
);
149 ptep
= pte_alloc_kernel(pmdp
, idx
);
153 radix__pte_update(&init_mm
, idx
, ptep
, clear
, 0, 0);
156 radix__flush_tlb_kernel_range(start
, end
);
159 void radix__mark_rodata_ro(void)
161 unsigned long start
, end
;
163 start
= (unsigned long)_stext
;
164 end
= (unsigned long)__init_begin
;
166 radix__change_memory_range(start
, end
, _PAGE_WRITE
);
169 void radix__mark_initmem_nx(void)
171 unsigned long start
= (unsigned long)__init_begin
;
172 unsigned long end
= (unsigned long)__init_end
;
174 radix__change_memory_range(start
, end
, _PAGE_EXEC
);
176 #endif /* CONFIG_STRICT_KERNEL_RWX */
178 static inline void __meminit
print_mapping(unsigned long start
,
185 pr_info("Mapped range 0x%lx - 0x%lx with 0x%lx\n", start
, end
, size
);
188 static int __meminit
create_physical_mapping(unsigned long start
,
191 unsigned long vaddr
, addr
, mapping_size
= 0;
193 unsigned long max_mapping_size
;
194 #ifdef CONFIG_STRICT_KERNEL_RWX
195 int split_text_mapping
= 1;
197 int split_text_mapping
= 0;
200 start
= _ALIGN_UP(start
, PAGE_SIZE
);
201 for (addr
= start
; addr
< end
; addr
+= mapping_size
) {
202 unsigned long gap
, previous_size
;
206 previous_size
= mapping_size
;
207 max_mapping_size
= PUD_SIZE
;
210 if (IS_ALIGNED(addr
, PUD_SIZE
) && gap
>= PUD_SIZE
&&
211 mmu_psize_defs
[MMU_PAGE_1G
].shift
&&
212 PUD_SIZE
<= max_mapping_size
)
213 mapping_size
= PUD_SIZE
;
214 else if (IS_ALIGNED(addr
, PMD_SIZE
) && gap
>= PMD_SIZE
&&
215 mmu_psize_defs
[MMU_PAGE_2M
].shift
)
216 mapping_size
= PMD_SIZE
;
218 mapping_size
= PAGE_SIZE
;
220 if (split_text_mapping
&& (mapping_size
== PUD_SIZE
) &&
221 (addr
<= __pa_symbol(__init_begin
)) &&
222 (addr
+ mapping_size
) >= __pa_symbol(_stext
)) {
223 max_mapping_size
= PMD_SIZE
;
227 if (split_text_mapping
&& (mapping_size
== PMD_SIZE
) &&
228 (addr
<= __pa_symbol(__init_begin
)) &&
229 (addr
+ mapping_size
) >= __pa_symbol(_stext
))
230 mapping_size
= PAGE_SIZE
;
232 if (mapping_size
!= previous_size
) {
233 print_mapping(start
, addr
, previous_size
);
237 vaddr
= (unsigned long)__va(addr
);
239 if (overlaps_kernel_text(vaddr
, vaddr
+ mapping_size
) ||
240 overlaps_interrupt_vector_text(vaddr
, vaddr
+ mapping_size
))
241 prot
= PAGE_KERNEL_X
;
245 rc
= radix__map_kernel_page(vaddr
, addr
, prot
, mapping_size
);
250 print_mapping(start
, addr
, mapping_size
);
254 static void __init
radix_init_pgtable(void)
256 unsigned long rts_field
;
257 struct memblock_region
*reg
;
259 /* We don't support slb for radix */
262 * Create the linear mapping, using standard page size for now
264 for_each_memblock(memory
, reg
)
265 WARN_ON(create_physical_mapping(reg
->base
,
266 reg
->base
+ reg
->size
));
268 /* Find out how many PID bits are supported */
269 if (cpu_has_feature(CPU_FTR_HVMODE
)) {
272 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
274 * When KVM is possible, we only use the top half of the
275 * PID space to avoid collisions between host and guest PIDs
276 * which can cause problems due to prefetch when exiting the
279 mmu_base_pid
= 1 << (mmu_pid_bits
- 1);
284 /* The guest uses the bottom half of the PID space */
291 * Allocate Partition table and process table for the
294 BUG_ON(PRTB_SIZE_SHIFT
> 36);
295 process_tb
= early_alloc_pgtable(1UL << PRTB_SIZE_SHIFT
);
297 * Fill in the process table.
299 rts_field
= radix__get_tree_size();
300 process_tb
->prtb0
= cpu_to_be64(rts_field
| __pa(init_mm
.pgd
) | RADIX_PGD_INDEX_SIZE
);
302 * Fill in the partition table. We are suppose to use effective address
303 * of process table here. But our linear mapping also enable us to use
304 * physical address here.
306 register_process_table(__pa(process_tb
), 0, PRTB_SIZE_SHIFT
- 12);
307 pr_info("Process table %p and radix root for kernel: %p\n", process_tb
, init_mm
.pgd
);
308 asm volatile("ptesync" : : : "memory");
309 asm volatile(PPC_TLBIE_5(%0,%1,2,1,1) : :
310 "r" (TLBIEL_INVAL_SET_LPID
), "r" (0));
311 asm volatile("eieio; tlbsync; ptesync" : : : "memory");
312 trace_tlbie(0, 0, TLBIEL_INVAL_SET_LPID
, 0, 2, 1, 1);
315 static void __init
radix_init_partition_table(void)
317 unsigned long rts_field
, dw0
;
319 mmu_partition_table_init();
320 rts_field
= radix__get_tree_size();
321 dw0
= rts_field
| __pa(init_mm
.pgd
) | RADIX_PGD_INDEX_SIZE
| PATB_HR
;
322 mmu_partition_table_set_entry(0, dw0
, 0);
324 pr_info("Initializing Radix MMU\n");
325 pr_info("Partition table %p\n", partition_tb
);
328 void __init
radix_init_native(void)
330 register_process_table
= native_register_process_table
;
333 static int __init
get_idx_from_shift(unsigned int shift
)
354 static int __init
radix_dt_scan_page_sizes(unsigned long node
,
355 const char *uname
, int depth
,
362 const char *type
= of_get_flat_dt_prop(node
, "device_type", NULL
);
364 /* We are scanning "cpu" nodes only */
365 if (type
== NULL
|| strcmp(type
, "cpu") != 0)
368 /* Find MMU PID size */
369 prop
= of_get_flat_dt_prop(node
, "ibm,mmu-pid-bits", &size
);
370 if (prop
&& size
== 4)
371 mmu_pid_bits
= be32_to_cpup(prop
);
373 /* Grab page size encodings */
374 prop
= of_get_flat_dt_prop(node
, "ibm,processor-radix-AP-encodings", &size
);
378 pr_info("Page sizes from device-tree:\n");
379 for (; size
>= 4; size
-= 4, ++prop
) {
381 struct mmu_psize_def
*def
;
383 /* top 3 bit is AP encoding */
384 shift
= be32_to_cpu(prop
[0]) & ~(0xe << 28);
385 ap
= be32_to_cpu(prop
[0]) >> 29;
386 pr_info("Page size shift = %d AP=0x%x\n", shift
, ap
);
388 idx
= get_idx_from_shift(shift
);
392 def
= &mmu_psize_defs
[idx
];
398 cur_cpu_spec
->mmu_features
&= ~MMU_FTR_NO_SLBIE_B
;
402 void __init
radix__early_init_devtree(void)
407 * Try to find the available page sizes in the device-tree
409 rc
= of_scan_flat_dt(radix_dt_scan_page_sizes
, NULL
);
410 if (rc
!= 0) /* Found */
413 * let's assume we have page 4k and 64k support
415 mmu_psize_defs
[MMU_PAGE_4K
].shift
= 12;
416 mmu_psize_defs
[MMU_PAGE_4K
].ap
= 0x0;
418 mmu_psize_defs
[MMU_PAGE_64K
].shift
= 16;
419 mmu_psize_defs
[MMU_PAGE_64K
].ap
= 0x5;
421 #ifdef CONFIG_SPARSEMEM_VMEMMAP
422 if (mmu_psize_defs
[MMU_PAGE_2M
].shift
) {
424 * map vmemmap using 2M if available
426 mmu_vmemmap_psize
= MMU_PAGE_2M
;
428 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
432 static void update_hid_for_radix(void)
435 unsigned long rb
= 3UL << PPC_BITLSHIFT(53); /* IS = 3 */
437 asm volatile("ptesync": : :"memory");
438 /* prs = 0, ric = 2, rs = 0, r = 1 is = 3 */
439 asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
440 : : "r"(rb
), "i"(1), "i"(0), "i"(2), "r"(0) : "memory");
441 /* prs = 1, ric = 2, rs = 0, r = 1 is = 3 */
442 asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
443 : : "r"(rb
), "i"(1), "i"(1), "i"(2), "r"(0) : "memory");
444 asm volatile("eieio; tlbsync; ptesync; isync; slbia": : :"memory");
445 trace_tlbie(0, 0, rb
, 0, 2, 0, 1);
446 trace_tlbie(0, 0, rb
, 0, 2, 1, 1);
451 hid0
= mfspr(SPRN_HID0
);
452 hid0
|= HID0_POWER9_RADIX
;
453 mtspr(SPRN_HID0
, hid0
);
454 asm volatile("isync": : :"memory");
456 /* Wait for it to happen */
457 while (!(mfspr(SPRN_HID0
) & HID0_POWER9_RADIX
))
461 static void radix_init_amor(void)
464 * In HV mode, we init AMOR (Authority Mask Override Register) so that
465 * the hypervisor and guest can setup IAMR (Instruction Authority Mask
466 * Register), enable key 0 and set it to 1.
468 * AMOR = 0b1100 .... 0000 (Mask for key 0 is 11)
470 mtspr(SPRN_AMOR
, (3ul << 62));
473 static void radix_init_iamr(void)
478 * The IAMR should set to 0 on DD1.
480 if (cpu_has_feature(CPU_FTR_POWER9_DD1
))
486 * Radix always uses key0 of the IAMR to determine if an access is
487 * allowed. We set bit 0 (IBM bit 1) of key0, to prevent instruction
490 mtspr(SPRN_IAMR
, iamr
);
493 void __init
radix__early_init_mmu(void)
497 #ifdef CONFIG_PPC_64K_PAGES
498 /* PAGE_SIZE mappings */
499 mmu_virtual_psize
= MMU_PAGE_64K
;
501 mmu_virtual_psize
= MMU_PAGE_4K
;
504 #ifdef CONFIG_SPARSEMEM_VMEMMAP
505 /* vmemmap mapping */
506 mmu_vmemmap_psize
= mmu_virtual_psize
;
509 * initialize page table size
511 __pte_index_size
= RADIX_PTE_INDEX_SIZE
;
512 __pmd_index_size
= RADIX_PMD_INDEX_SIZE
;
513 __pud_index_size
= RADIX_PUD_INDEX_SIZE
;
514 __pgd_index_size
= RADIX_PGD_INDEX_SIZE
;
515 __pmd_cache_index
= RADIX_PMD_INDEX_SIZE
;
516 __pte_table_size
= RADIX_PTE_TABLE_SIZE
;
517 __pmd_table_size
= RADIX_PMD_TABLE_SIZE
;
518 __pud_table_size
= RADIX_PUD_TABLE_SIZE
;
519 __pgd_table_size
= RADIX_PGD_TABLE_SIZE
;
521 __pmd_val_bits
= RADIX_PMD_VAL_BITS
;
522 __pud_val_bits
= RADIX_PUD_VAL_BITS
;
523 __pgd_val_bits
= RADIX_PGD_VAL_BITS
;
525 __kernel_virt_start
= RADIX_KERN_VIRT_START
;
526 __kernel_virt_size
= RADIX_KERN_VIRT_SIZE
;
527 __vmalloc_start
= RADIX_VMALLOC_START
;
528 __vmalloc_end
= RADIX_VMALLOC_END
;
529 vmemmap
= (struct page
*)RADIX_VMEMMAP_BASE
;
530 ioremap_bot
= IOREMAP_BASE
;
533 pci_io_base
= ISA_IO_BASE
;
537 * For now radix also use the same frag size
539 __pte_frag_nr
= H_PTE_FRAG_NR
;
540 __pte_frag_size_shift
= H_PTE_FRAG_SIZE_SHIFT
;
542 if (!firmware_has_feature(FW_FEATURE_LPAR
)) {
544 if (cpu_has_feature(CPU_FTR_POWER9_DD1
))
545 update_hid_for_radix();
546 lpcr
= mfspr(SPRN_LPCR
);
547 mtspr(SPRN_LPCR
, lpcr
| LPCR_UPRT
| LPCR_HR
);
548 radix_init_partition_table();
551 radix_init_pseries();
554 memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE
);
557 radix_init_pgtable();
560 void radix__early_init_mmu_secondary(void)
564 * update partition table control register and UPRT
566 if (!firmware_has_feature(FW_FEATURE_LPAR
)) {
568 if (cpu_has_feature(CPU_FTR_POWER9_DD1
))
569 update_hid_for_radix();
571 lpcr
= mfspr(SPRN_LPCR
);
572 mtspr(SPRN_LPCR
, lpcr
| LPCR_UPRT
| LPCR_HR
);
575 __pa(partition_tb
) | (PATB_SIZE_SHIFT
- 12));
581 void radix__mmu_cleanup_all(void)
585 if (!firmware_has_feature(FW_FEATURE_LPAR
)) {
586 lpcr
= mfspr(SPRN_LPCR
);
587 mtspr(SPRN_LPCR
, lpcr
& ~LPCR_UPRT
);
589 powernv_set_nmmu_ptcr(0);
590 radix__flush_tlb_all();
594 void radix__setup_initial_memory_limit(phys_addr_t first_memblock_base
,
595 phys_addr_t first_memblock_size
)
597 /* We don't currently support the first MEMBLOCK not mapping 0
598 * physical on those processors
600 BUG_ON(first_memblock_base
!= 0);
602 * We limit the allocation that depend on ppc64_rma_size
603 * to first_memblock_size. We also clamp it to 1GB to
604 * avoid some funky things such as RTAS bugs.
606 * On radix config we really don't have a limitation
607 * on real mode access. But keeping it as above works
610 ppc64_rma_size
= min_t(u64
, first_memblock_size
, 0x40000000);
612 * Finally limit subsequent allocations. We really don't want
613 * to limit the memblock allocations to rma_size. FIXME!! should
614 * we even limit at all ?
616 memblock_set_current_limit(first_memblock_base
+ first_memblock_size
);
619 #ifdef CONFIG_MEMORY_HOTPLUG
620 static void free_pte_table(pte_t
*pte_start
, pmd_t
*pmd
)
625 for (i
= 0; i
< PTRS_PER_PTE
; i
++) {
631 pte_free_kernel(&init_mm
, pte_start
);
635 static void free_pmd_table(pmd_t
*pmd_start
, pud_t
*pud
)
640 for (i
= 0; i
< PTRS_PER_PMD
; i
++) {
646 pmd_free(&init_mm
, pmd_start
);
650 static void remove_pte_table(pte_t
*pte_start
, unsigned long addr
,
656 pte
= pte_start
+ pte_index(addr
);
657 for (; addr
< end
; addr
= next
, pte
++) {
658 next
= (addr
+ PAGE_SIZE
) & PAGE_MASK
;
662 if (!pte_present(*pte
))
665 if (!PAGE_ALIGNED(addr
) || !PAGE_ALIGNED(next
)) {
667 * The vmemmap_free() and remove_section_mapping()
668 * codepaths call us with aligned addresses.
670 WARN_ONCE(1, "%s: unaligned range\n", __func__
);
674 pte_clear(&init_mm
, addr
, pte
);
678 static void remove_pmd_table(pmd_t
*pmd_start
, unsigned long addr
,
685 pmd
= pmd_start
+ pmd_index(addr
);
686 for (; addr
< end
; addr
= next
, pmd
++) {
687 next
= pmd_addr_end(addr
, end
);
689 if (!pmd_present(*pmd
))
692 if (pmd_huge(*pmd
)) {
693 if (!IS_ALIGNED(addr
, PMD_SIZE
) ||
694 !IS_ALIGNED(next
, PMD_SIZE
)) {
695 WARN_ONCE(1, "%s: unaligned range\n", __func__
);
699 pte_clear(&init_mm
, addr
, (pte_t
*)pmd
);
703 pte_base
= (pte_t
*)pmd_page_vaddr(*pmd
);
704 remove_pte_table(pte_base
, addr
, next
);
705 free_pte_table(pte_base
, pmd
);
709 static void remove_pud_table(pud_t
*pud_start
, unsigned long addr
,
716 pud
= pud_start
+ pud_index(addr
);
717 for (; addr
< end
; addr
= next
, pud
++) {
718 next
= pud_addr_end(addr
, end
);
720 if (!pud_present(*pud
))
723 if (pud_huge(*pud
)) {
724 if (!IS_ALIGNED(addr
, PUD_SIZE
) ||
725 !IS_ALIGNED(next
, PUD_SIZE
)) {
726 WARN_ONCE(1, "%s: unaligned range\n", __func__
);
730 pte_clear(&init_mm
, addr
, (pte_t
*)pud
);
734 pmd_base
= (pmd_t
*)pud_page_vaddr(*pud
);
735 remove_pmd_table(pmd_base
, addr
, next
);
736 free_pmd_table(pmd_base
, pud
);
740 static void remove_pagetable(unsigned long start
, unsigned long end
)
742 unsigned long addr
, next
;
746 spin_lock(&init_mm
.page_table_lock
);
748 for (addr
= start
; addr
< end
; addr
= next
) {
749 next
= pgd_addr_end(addr
, end
);
751 pgd
= pgd_offset_k(addr
);
752 if (!pgd_present(*pgd
))
755 if (pgd_huge(*pgd
)) {
756 if (!IS_ALIGNED(addr
, PGDIR_SIZE
) ||
757 !IS_ALIGNED(next
, PGDIR_SIZE
)) {
758 WARN_ONCE(1, "%s: unaligned range\n", __func__
);
762 pte_clear(&init_mm
, addr
, (pte_t
*)pgd
);
766 pud_base
= (pud_t
*)pgd_page_vaddr(*pgd
);
767 remove_pud_table(pud_base
, addr
, next
);
770 spin_unlock(&init_mm
.page_table_lock
);
771 radix__flush_tlb_kernel_range(start
, end
);
774 int __ref
radix__create_section_mapping(unsigned long start
, unsigned long end
)
776 return create_physical_mapping(start
, end
);
779 int radix__remove_section_mapping(unsigned long start
, unsigned long end
)
781 remove_pagetable(start
, end
);
784 #endif /* CONFIG_MEMORY_HOTPLUG */
786 #ifdef CONFIG_SPARSEMEM_VMEMMAP
787 int __meminit
radix__vmemmap_create_mapping(unsigned long start
,
788 unsigned long page_size
,
791 /* Create a PTE encoding */
792 unsigned long flags
= _PAGE_PRESENT
| _PAGE_ACCESSED
| _PAGE_KERNEL_RW
;
794 BUG_ON(radix__map_kernel_page(start
, phys
, __pgprot(flags
), page_size
));
798 #ifdef CONFIG_MEMORY_HOTPLUG
799 void radix__vmemmap_remove_mapping(unsigned long start
, unsigned long page_size
)
801 remove_pagetable(start
, start
+ page_size
);
806 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
808 unsigned long radix__pmd_hugepage_update(struct mm_struct
*mm
, unsigned long addr
,
809 pmd_t
*pmdp
, unsigned long clr
,
814 #ifdef CONFIG_DEBUG_VM
815 WARN_ON(!radix__pmd_trans_huge(*pmdp
) && !pmd_devmap(*pmdp
));
816 assert_spin_locked(&mm
->page_table_lock
);
819 old
= radix__pte_update(mm
, addr
, (pte_t
*)pmdp
, clr
, set
, 1);
820 trace_hugepage_update(addr
, old
, clr
, set
);
825 pmd_t
radix__pmdp_collapse_flush(struct vm_area_struct
*vma
, unsigned long address
,
831 VM_BUG_ON(address
& ~HPAGE_PMD_MASK
);
832 VM_BUG_ON(radix__pmd_trans_huge(*pmdp
));
833 VM_BUG_ON(pmd_devmap(*pmdp
));
835 * khugepaged calls this for normal pmd
839 /*FIXME!! Verify whether we need this kick below */
840 kick_all_cpus_sync();
841 flush_tlb_range(vma
, address
, address
+ HPAGE_PMD_SIZE
);
846 * For us pgtable_t is pte_t *. Inorder to save the deposisted
847 * page table, we consider the allocated page table as a list
848 * head. On withdraw we need to make sure we zero out the used
849 * list_head memory area.
851 void radix__pgtable_trans_huge_deposit(struct mm_struct
*mm
, pmd_t
*pmdp
,
854 struct list_head
*lh
= (struct list_head
*) pgtable
;
856 assert_spin_locked(pmd_lockptr(mm
, pmdp
));
859 if (!pmd_huge_pte(mm
, pmdp
))
862 list_add(lh
, (struct list_head
*) pmd_huge_pte(mm
, pmdp
));
863 pmd_huge_pte(mm
, pmdp
) = pgtable
;
866 pgtable_t
radix__pgtable_trans_huge_withdraw(struct mm_struct
*mm
, pmd_t
*pmdp
)
870 struct list_head
*lh
;
872 assert_spin_locked(pmd_lockptr(mm
, pmdp
));
875 pgtable
= pmd_huge_pte(mm
, pmdp
);
876 lh
= (struct list_head
*) pgtable
;
878 pmd_huge_pte(mm
, pmdp
) = NULL
;
880 pmd_huge_pte(mm
, pmdp
) = (pgtable_t
) lh
->next
;
883 ptep
= (pte_t
*) pgtable
;
891 pmd_t
radix__pmdp_huge_get_and_clear(struct mm_struct
*mm
,
892 unsigned long addr
, pmd_t
*pmdp
)
897 old
= radix__pmd_hugepage_update(mm
, addr
, pmdp
, ~0UL, 0);
898 old_pmd
= __pmd(old
);
900 * Serialize against find_linux_pte_or_hugepte which does lock-less
901 * lookup in page tables with local interrupts disabled. For huge pages
902 * it casts pmd_t to pte_t. Since format of pte_t is different from
903 * pmd_t we want to prevent transit from pmd pointing to page table
904 * to pmd pointing to huge page (and back) while interrupts are disabled.
905 * We clear pmd to possibly replace it with page table pointer in
906 * different code paths. So make sure we wait for the parallel
907 * find_linux_pte_or_hugepage to finish.
909 kick_all_cpus_sync();
913 int radix__has_transparent_hugepage(void)
915 /* For radix 2M at PMD level means thp */
916 if (mmu_psize_defs
[MMU_PAGE_2M
].shift
== PMD_SHIFT
)
920 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */