Skip to content

Commit 82a1a1f

Browse files
PanQLpalmer-dabbelt
PanQL
authored and
palmer-dabbelt
committed
riscv: mm: support Svnapot in hugetlb page
Svnapot can be used to support 64KB hugetlb page, so it can become a new option when using hugetlbfs. Add a basic implementation of hugetlb page, and support 64KB as a size in it by using Svnapot. For test, boot kernel with command line contains "default_hugepagesz=64K hugepagesz=64K hugepages=20" and run a simple test like this: tools/testing/selftests/vm/map_hugetlb 1 16 And it should be passed. Signed-off-by: Qinglin Pan <panqinglin00@gmail.com> Reviewed-by: Andrew Jones <ajones@ventanamicro.com> Link: https://lore.kernel.org/r/20230209131647.17245-3-panqinglin00@gmail.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
1 parent 23ad288 commit 82a1a1f

3 files changed

Lines changed: 335 additions & 2 deletions

File tree

‎arch/riscv/Kconfig‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -44,7 +44,7 @@ config RISCV
4444
select ARCH_USE_QUEUED_RWLOCKS
4545
select ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT if MMU
4646
select ARCH_WANT_FRAME_POINTERS
47-
select ARCH_WANT_GENERAL_HUGETLB
47+
select ARCH_WANT_GENERAL_HUGETLB if !RISCV_ISA_SVNAPOT
4848
select ARCH_WANT_HUGE_PMD_SHARE if 64BIT
4949
select ARCH_WANTS_THP_SWAP if HAVE_ARCH_TRANSPARENT_HUGEPAGE
5050
select BINFMT_FLAT_NO_DATA_START_OFFSET if !MMU

‎arch/riscv/include/asm/hugetlb.h‎

Lines changed: 33 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,6 @@
22
#ifndef _ASM_RISCV_HUGETLB_H
33
#define _ASM_RISCV_HUGETLB_H
44

5-
#include <asm-generic/hugetlb.h>
65
#include <asm/page.h>
76

87
static inline void arch_clear_hugepage_flags(struct page *page)
@@ -11,4 +10,37 @@ static inline void arch_clear_hugepage_flags(struct page *page)
1110
}
1211
#define arch_clear_hugepage_flags arch_clear_hugepage_flags
1312

13+
#ifdef CONFIG_RISCV_ISA_SVNAPOT
14+
#define __HAVE_ARCH_HUGE_PTE_CLEAR
15+
void huge_pte_clear(struct mm_struct *mm, unsigned long addr,
16+
pte_t *ptep, unsigned long sz);
17+
18+
#define __HAVE_ARCH_HUGE_SET_HUGE_PTE_AT
19+
void set_huge_pte_at(struct mm_struct *mm,
20+
unsigned long addr, pte_t *ptep, pte_t pte);
21+
22+
#define __HAVE_ARCH_HUGE_PTEP_GET_AND_CLEAR
23+
pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
24+
unsigned long addr, pte_t *ptep);
25+
26+
#define __HAVE_ARCH_HUGE_PTEP_CLEAR_FLUSH
27+
pte_t huge_ptep_clear_flush(struct vm_area_struct *vma,
28+
unsigned long addr, pte_t *ptep);
29+
30+
#define __HAVE_ARCH_HUGE_PTEP_SET_WRPROTECT
31+
void huge_ptep_set_wrprotect(struct mm_struct *mm,
32+
unsigned long addr, pte_t *ptep);
33+
34+
#define __HAVE_ARCH_HUGE_PTEP_SET_ACCESS_FLAGS
35+
int huge_ptep_set_access_flags(struct vm_area_struct *vma,
36+
unsigned long addr, pte_t *ptep,
37+
pte_t pte, int dirty);
38+
39+
pte_t arch_make_huge_pte(pte_t entry, unsigned int shift, vm_flags_t flags);
40+
#define arch_make_huge_pte arch_make_huge_pte
41+
42+
#endif /*CONFIG_RISCV_ISA_SVNAPOT*/
43+
44+
#include <asm-generic/hugetlb.h>
45+
1446
#endif /* _ASM_RISCV_HUGETLB_H */

‎arch/riscv/mm/hugetlbpage.c‎

Lines changed: 301 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,305 @@
22
#include <linux/hugetlb.h>
33
#include <linux/err.h>
44

5+
#ifdef CONFIG_RISCV_ISA_SVNAPOT
6+
pte_t *huge_pte_alloc(struct mm_struct *mm,
7+
struct vm_area_struct *vma,
8+
unsigned long addr,
9+
unsigned long sz)
10+
{
11+
unsigned long order;
12+
pte_t *pte = NULL;
13+
pgd_t *pgd;
14+
p4d_t *p4d;
15+
pud_t *pud;
16+
pmd_t *pmd;
17+
18+
pgd = pgd_offset(mm, addr);
19+
p4d = p4d_alloc(mm, pgd, addr);
20+
if (!p4d)
21+
return NULL;
22+
23+
pud = pud_alloc(mm, p4d, addr);
24+
if (!pud)
25+
return NULL;
26+
27+
if (sz == PUD_SIZE) {
28+
pte = (pte_t *)pud;
29+
goto out;
30+
}
31+
32+
if (sz == PMD_SIZE) {
33+
if (want_pmd_share(vma, addr) && pud_none(*pud))
34+
pte = huge_pmd_share(mm, vma, addr, pud);
35+
else
36+
pte = (pte_t *)pmd_alloc(mm, pud, addr);
37+
goto out;
38+
}
39+
40+
pmd = pmd_alloc(mm, pud, addr);
41+
if (!pmd)
42+
return NULL;
43+
44+
for_each_napot_order(order) {
45+
if (napot_cont_size(order) == sz) {
46+
pte = pte_alloc_map(mm, pmd, addr & napot_cont_mask(order));
47+
break;
48+
}
49+
}
50+
51+
out:
52+
WARN_ON_ONCE(pte && pte_present(*pte) && !pte_huge(*pte));
53+
return pte;
54+
}
55+
56+
pte_t *huge_pte_offset(struct mm_struct *mm,
57+
unsigned long addr,
58+
unsigned long sz)
59+
{
60+
unsigned long order;
61+
pte_t *pte = NULL;
62+
pgd_t *pgd;
63+
p4d_t *p4d;
64+
pud_t *pud;
65+
pmd_t *pmd;
66+
67+
pgd = pgd_offset(mm, addr);
68+
if (!pgd_present(*pgd))
69+
return NULL;
70+
71+
p4d = p4d_offset(pgd, addr);
72+
if (!p4d_present(*p4d))
73+
return NULL;
74+
75+
pud = pud_offset(p4d, addr);
76+
if (sz == PUD_SIZE)
77+
/* must be pud huge, non-present or none */
78+
return (pte_t *)pud;
79+
80+
if (!pud_present(*pud))
81+
return NULL;
82+
83+
pmd = pmd_offset(pud, addr);
84+
if (sz == PMD_SIZE)
85+
/* must be pmd huge, non-present or none */
86+
return (pte_t *)pmd;
87+
88+
if (!pmd_present(*pmd))
89+
return NULL;
90+
91+
for_each_napot_order(order) {
92+
if (napot_cont_size(order) == sz) {
93+
pte = pte_offset_kernel(pmd, addr & napot_cont_mask(order));
94+
break;
95+
}
96+
}
97+
return pte;
98+
}
99+
100+
static pte_t get_clear_contig(struct mm_struct *mm,
101+
unsigned long addr,
102+
pte_t *ptep,
103+
unsigned long pte_num)
104+
{
105+
pte_t orig_pte = ptep_get(ptep);
106+
unsigned long i;
107+
108+
for (i = 0; i < pte_num; i++, addr += PAGE_SIZE, ptep++) {
109+
pte_t pte = ptep_get_and_clear(mm, addr, ptep);
110+
111+
if (pte_dirty(pte))
112+
orig_pte = pte_mkdirty(orig_pte);
113+
114+
if (pte_young(pte))
115+
orig_pte = pte_mkyoung(orig_pte);
116+
}
117+
118+
return orig_pte;
119+
}
120+
121+
static pte_t get_clear_contig_flush(struct mm_struct *mm,
122+
unsigned long addr,
123+
pte_t *ptep,
124+
unsigned long pte_num)
125+
{
126+
pte_t orig_pte = get_clear_contig(mm, addr, ptep, pte_num);
127+
struct vm_area_struct vma = TLB_FLUSH_VMA(mm, 0);
128+
bool valid = !pte_none(orig_pte);
129+
130+
if (valid)
131+
flush_tlb_range(&vma, addr, addr + (PAGE_SIZE * pte_num));
132+
133+
return orig_pte;
134+
}
135+
136+
pte_t arch_make_huge_pte(pte_t entry, unsigned int shift, vm_flags_t flags)
137+
{
138+
unsigned long order;
139+
140+
for_each_napot_order(order) {
141+
if (shift == napot_cont_shift(order)) {
142+
entry = pte_mknapot(entry, order);
143+
break;
144+
}
145+
}
146+
if (order == NAPOT_ORDER_MAX)
147+
entry = pte_mkhuge(entry);
148+
149+
return entry;
150+
}
151+
152+
void set_huge_pte_at(struct mm_struct *mm,
153+
unsigned long addr,
154+
pte_t *ptep,
155+
pte_t pte)
156+
{
157+
int i, pte_num;
158+
159+
if (!pte_napot(pte)) {
160+
set_pte_at(mm, addr, ptep, pte);
161+
return;
162+
}
163+
164+
pte_num = napot_pte_num(napot_cont_order(pte));
165+
for (i = 0; i < pte_num; i++, ptep++, addr += PAGE_SIZE)
166+
set_pte_at(mm, addr, ptep, pte);
167+
}
168+
169+
int huge_ptep_set_access_flags(struct vm_area_struct *vma,
170+
unsigned long addr,
171+
pte_t *ptep,
172+
pte_t pte,
173+
int dirty)
174+
{
175+
struct mm_struct *mm = vma->vm_mm;
176+
unsigned long order;
177+
pte_t orig_pte;
178+
int i, pte_num;
179+
180+
if (!pte_napot(pte))
181+
return ptep_set_access_flags(vma, addr, ptep, pte, dirty);
182+
183+
order = napot_cont_order(pte);
184+
pte_num = napot_pte_num(order);
185+
ptep = huge_pte_offset(mm, addr, napot_cont_size(order));
186+
orig_pte = get_clear_contig_flush(mm, addr, ptep, pte_num);
187+
188+
if (pte_dirty(orig_pte))
189+
pte = pte_mkdirty(pte);
190+
191+
if (pte_young(orig_pte))
192+
pte = pte_mkyoung(pte);
193+
194+
for (i = 0; i < pte_num; i++, addr += PAGE_SIZE, ptep++)
195+
set_pte_at(mm, addr, ptep, pte);
196+
197+
return true;
198+
}
199+
200+
pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
201+
unsigned long addr,
202+
pte_t *ptep)
203+
{
204+
pte_t orig_pte = ptep_get(ptep);
205+
int pte_num;
206+
207+
if (!pte_napot(orig_pte))
208+
return ptep_get_and_clear(mm, addr, ptep);
209+
210+
pte_num = napot_pte_num(napot_cont_order(orig_pte));
211+
212+
return get_clear_contig(mm, addr, ptep, pte_num);
213+
}
214+
215+
void huge_ptep_set_wrprotect(struct mm_struct *mm,
216+
unsigned long addr,
217+
pte_t *ptep)
218+
{
219+
pte_t pte = ptep_get(ptep);
220+
unsigned long order;
221+
int i, pte_num;
222+
223+
if (!pte_napot(pte)) {
224+
ptep_set_wrprotect(mm, addr, ptep);
225+
return;
226+
}
227+
228+
order = napot_cont_order(pte);
229+
pte_num = napot_pte_num(order);
230+
ptep = huge_pte_offset(mm, addr, napot_cont_size(order));
231+
232+
for (i = 0; i < pte_num; i++, addr += PAGE_SIZE, ptep++)
233+
ptep_set_wrprotect(mm, addr, ptep);
234+
}
235+
236+
pte_t huge_ptep_clear_flush(struct vm_area_struct *vma,
237+
unsigned long addr,
238+
pte_t *ptep)
239+
{
240+
pte_t pte = ptep_get(ptep);
241+
int pte_num;
242+
243+
if (!pte_napot(pte))
244+
return ptep_clear_flush(vma, addr, ptep);
245+
246+
pte_num = napot_pte_num(napot_cont_order(pte));
247+
248+
return get_clear_contig_flush(vma->vm_mm, addr, ptep, pte_num);
249+
}
250+
251+
void huge_pte_clear(struct mm_struct *mm,
252+
unsigned long addr,
253+
pte_t *ptep,
254+
unsigned long sz)
255+
{
256+
pte_t pte = READ_ONCE(*ptep);
257+
int i, pte_num;
258+
259+
if (!pte_napot(pte)) {
260+
pte_clear(mm, addr, ptep);
261+
return;
262+
}
263+
264+
pte_num = napot_pte_num(napot_cont_order(pte));
265+
for (i = 0; i < pte_num; i++, addr += PAGE_SIZE, ptep++)
266+
pte_clear(mm, addr, ptep);
267+
}
268+
269+
static __init bool is_napot_size(unsigned long size)
270+
{
271+
unsigned long order;
272+
273+
if (!has_svnapot())
274+
return false;
275+
276+
for_each_napot_order(order) {
277+
if (size == napot_cont_size(order))
278+
return true;
279+
}
280+
return false;
281+
}
282+
283+
static __init int napot_hugetlbpages_init(void)
284+
{
285+
if (has_svnapot()) {
286+
unsigned long order;
287+
288+
for_each_napot_order(order)
289+
hugetlb_add_hstate(order);
290+
}
291+
return 0;
292+
}
293+
arch_initcall(napot_hugetlbpages_init);
294+
295+
#else
296+
297+
static __init bool is_napot_size(unsigned long size)
298+
{
299+
return false;
300+
}
301+
302+
#endif /*CONFIG_RISCV_ISA_SVNAPOT*/
303+
5304
int pud_huge(pud_t pud)
6305
{
7306
return pud_leaf(pud);
@@ -18,6 +317,8 @@ bool __init arch_hugetlb_valid_size(unsigned long size)
18317
return true;
19318
else if (IS_ENABLED(CONFIG_64BIT) && size == PUD_SIZE)
20319
return true;
320+
else if (is_napot_size(size))
321+
return true;
21322
else
22323
return false;
23324
}

0 commit comments

Comments
 (0)