Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge tag 'loongarch-fixes-6.12-1' of git://git.kernel.org/pub/scm/linux/kernel/git/chenhuacai/linux-loongson
Pull LoongArch fixes from Huacai Chen:
"Get correct cores_per_package for SMT systems, enable IRQ if do_ale()
triggered in irq-enabled context, and fix some bugs about vDSO, memory
managenent, hrtimer in KVM, etc"
* tag 'loongarch-fixes-6.12-1' of git://git.kernel.org/pub/scm/linux/kernel/git/chenhuacai/linux-loongson:
LoongArch: KVM: Mark hrtimer to expire in hard interrupt context
LoongArch: Make KASAN usable for variable cpu_vabits
LoongArch: Set initial pte entry with PAGE_GLOBAL for kernel space
LoongArch: Don't crash in stack_top() for tasks without vDSO
LoongArch: Set correct size for vDSO code mapping
LoongArch: Enable IRQ if do_ale() triggered in irq-enabled context
LoongArch: Get correct cores_per_package for SMT systems
LoongArch: Use "Exception return address" to comment ERA
+84
-47
+4
arch/loongarch/include/asm/bootinfo.h
+4
arch/loongarch/include/asm/bootinfo.h
···
26
26
27
27
#define NR_WORDS DIV_ROUND_UP(NR_CPUS, BITS_PER_LONG)
28
28
29
+
/*
30
+
* The "core" of cores_per_node and cores_per_package stands for a
31
+
* logical core, which means in a SMT system it stands for a thread.
32
+
*/
29
33
struct loongson_system_configuration {
30
34
int nr_cpus;
31
35
int nr_nodes;
+1
-1
arch/loongarch/include/asm/kasan.h
+1
-1
arch/loongarch/include/asm/kasan.h
···
16
16
#define XRANGE_SHIFT (48)
17
17
18
18
/* Valid address length */
19
-
#define XRANGE_SHADOW_SHIFT (PGDIR_SHIFT + PAGE_SHIFT - 3)
19
+
#define XRANGE_SHADOW_SHIFT min(cpu_vabits, VA_BITS)
20
20
/* Used for taking out the valid address */
21
21
#define XRANGE_SHADOW_MASK GENMASK_ULL(XRANGE_SHADOW_SHIFT - 1, 0)
22
22
/* One segment whole address space size */
+1
-1
arch/loongarch/include/asm/loongarch.h
+1
-1
arch/loongarch/include/asm/loongarch.h
···
250
250
#define CSR_ESTAT_IS_WIDTH 15
251
251
#define CSR_ESTAT_IS (_ULCAST_(0x7fff) << CSR_ESTAT_IS_SHIFT)
252
252
253
-
#define LOONGARCH_CSR_ERA 0x6 /* ERA */
253
+
#define LOONGARCH_CSR_ERA 0x6 /* Exception return address */
254
254
255
255
#define LOONGARCH_CSR_BADV 0x7 /* Bad virtual address */
256
256
+11
arch/loongarch/include/asm/pgalloc.h
+11
arch/loongarch/include/asm/pgalloc.h
···
10
10
11
11
#define __HAVE_ARCH_PMD_ALLOC_ONE
12
12
#define __HAVE_ARCH_PUD_ALLOC_ONE
13
+
#define __HAVE_ARCH_PTE_ALLOC_ONE_KERNEL
13
14
#include <asm-generic/pgalloc.h>
14
15
15
16
static inline void pmd_populate_kernel(struct mm_struct *mm,
···
44
43
extern void pagetable_init(void);
45
44
46
45
extern pgd_t *pgd_alloc(struct mm_struct *mm);
46
+
47
+
static inline pte_t *pte_alloc_one_kernel(struct mm_struct *mm)
48
+
{
49
+
pte_t *pte = __pte_alloc_one_kernel(mm);
50
+
51
+
if (pte)
52
+
kernel_pte_init(pte);
53
+
54
+
return pte;
55
+
}
47
56
48
57
#define __pte_free_tlb(tlb, pte, address) \
49
58
do { \
+7
-28
arch/loongarch/include/asm/pgtable.h
+7
-28
arch/loongarch/include/asm/pgtable.h
···
269
269
extern void pgd_init(void *addr);
270
270
extern void pud_init(void *addr);
271
271
extern void pmd_init(void *addr);
272
+
extern void kernel_pte_init(void *addr);
272
273
273
274
/*
274
275
* Encode/decode swap entries and swap PTEs. Swap PTEs are all PTEs that
···
326
325
{
327
326
WRITE_ONCE(*ptep, pteval);
328
327
329
-
if (pte_val(pteval) & _PAGE_GLOBAL) {
330
-
pte_t *buddy = ptep_buddy(ptep);
331
-
/*
332
-
* Make sure the buddy is global too (if it's !none,
333
-
* it better already be global)
334
-
*/
335
-
if (pte_none(ptep_get(buddy))) {
336
328
#ifdef CONFIG_SMP
337
-
/*
338
-
* For SMP, multiple CPUs can race, so we need
339
-
* to do this atomically.
340
-
*/
341
-
__asm__ __volatile__(
342
-
__AMOR "$zero, %[global], %[buddy] \n"
343
-
: [buddy] "+ZB" (buddy->pte)
344
-
: [global] "r" (_PAGE_GLOBAL)
345
-
: "memory");
346
-
347
-
DBAR(0b11000); /* o_wrw = 0b11000 */
348
-
#else /* !CONFIG_SMP */
349
-
WRITE_ONCE(*buddy, __pte(pte_val(ptep_get(buddy)) | _PAGE_GLOBAL));
350
-
#endif /* CONFIG_SMP */
351
-
}
352
-
}
329
+
if (pte_val(pteval) & _PAGE_GLOBAL)
330
+
DBAR(0b11000); /* o_wrw = 0b11000 */
331
+
#endif
353
332
}
354
333
355
334
static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
356
335
{
357
-
/* Preserve global status for the pair */
358
-
if (pte_val(ptep_get(ptep_buddy(ptep))) & _PAGE_GLOBAL)
359
-
set_pte(ptep, __pte(_PAGE_GLOBAL));
360
-
else
361
-
set_pte(ptep, __pte(0));
336
+
pte_t pte = ptep_get(ptep);
337
+
pte_val(pte) &= _PAGE_GLOBAL;
338
+
set_pte(ptep, pte);
362
339
}
363
340
364
341
#define PGD_T_LOG2 (__builtin_ffs(sizeof(pgd_t)) - 1)
+8
-6
arch/loongarch/kernel/process.c
+8
-6
arch/loongarch/kernel/process.c
···
293
293
{
294
294
unsigned long top = TASK_SIZE & PAGE_MASK;
295
295
296
-
/* Space for the VDSO & data page */
297
-
top -= PAGE_ALIGN(current->thread.vdso->size);
298
-
top -= VVAR_SIZE;
296
+
if (current->thread.vdso) {
297
+
/* Space for the VDSO & data page */
298
+
top -= PAGE_ALIGN(current->thread.vdso->size);
299
+
top -= VVAR_SIZE;
299
300
300
-
/* Space to randomize the VDSO base */
301
-
if (current->flags & PF_RANDOMIZE)
302
-
top -= VDSO_RANDOMIZE_SIZE;
301
+
/* Space to randomize the VDSO base */
302
+
if (current->flags & PF_RANDOMIZE)
303
+
top -= VDSO_RANDOMIZE_SIZE;
304
+
}
303
305
304
306
return top;
305
307
}
+2
-1
arch/loongarch/kernel/setup.c
+2
-1
arch/loongarch/kernel/setup.c
···
55
55
#define SMBIOS_FREQHIGH_OFFSET 0x17
56
56
#define SMBIOS_FREQLOW_MASK 0xFF
57
57
#define SMBIOS_CORE_PACKAGE_OFFSET 0x23
58
+
#define SMBIOS_THREAD_PACKAGE_OFFSET 0x25
58
59
#define LOONGSON_EFI_ENABLE (1 << 3)
59
60
60
61
unsigned long fw_arg0, fw_arg1, fw_arg2;
···
126
125
cpu_clock_freq = freq_temp * 1000000;
127
126
128
127
loongson_sysconf.cpuname = (void *)dmi_string_parse(dm, dmi_data[16]);
129
-
loongson_sysconf.cores_per_package = *(dmi_data + SMBIOS_CORE_PACKAGE_OFFSET);
128
+
loongson_sysconf.cores_per_package = *(dmi_data + SMBIOS_THREAD_PACKAGE_OFFSET);
130
129
131
130
pr_info("CpuClock = %llu\n", cpu_clock_freq);
132
131
}
+5
arch/loongarch/kernel/traps.c
+5
arch/loongarch/kernel/traps.c
···
555
555
#else
556
556
unsigned int *pc;
557
557
558
+
if (regs->csr_prmd & CSR_PRMD_PIE)
559
+
local_irq_enable();
560
+
558
561
perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS, 1, regs, regs->csr_badvaddr);
559
562
560
563
/*
···
582
579
die_if_kernel("Kernel ale access", regs);
583
580
force_sig_fault(SIGBUS, BUS_ADRALN, (void __user *)regs->csr_badvaddr);
584
581
out:
582
+
if (regs->csr_prmd & CSR_PRMD_PIE)
583
+
local_irq_disable();
585
584
#endif
586
585
irqentry_exit(regs, state);
587
586
}
+4
-4
arch/loongarch/kernel/vdso.c
+4
-4
arch/loongarch/kernel/vdso.c
···
34
34
struct loongarch_vdso_data vdata;
35
35
} loongarch_vdso_data __page_aligned_data;
36
36
37
-
static struct page *vdso_pages[] = { NULL };
38
37
struct vdso_data *vdso_data = generic_vdso_data.data;
39
38
struct vdso_pcpu_data *vdso_pdata = loongarch_vdso_data.vdata.pdata;
40
39
struct vdso_rng_data *vdso_rng_data = &loongarch_vdso_data.vdata.rng_data;
···
84
85
85
86
struct loongarch_vdso_info vdso_info = {
86
87
.vdso = vdso_start,
87
-
.size = PAGE_SIZE,
88
88
.code_mapping = {
89
89
.name = "[vdso]",
90
-
.pages = vdso_pages,
91
90
.mremap = vdso_mremap,
92
91
},
93
92
.data_mapping = {
···
100
103
unsigned long i, cpu, pfn;
101
104
102
105
BUG_ON(!PAGE_ALIGNED(vdso_info.vdso));
103
-
BUG_ON(!PAGE_ALIGNED(vdso_info.size));
104
106
105
107
for_each_possible_cpu(cpu)
106
108
vdso_pdata[cpu].node = cpu_to_node(cpu);
109
+
110
+
vdso_info.size = PAGE_ALIGN(vdso_end - vdso_start);
111
+
vdso_info.code_mapping.pages =
112
+
kcalloc(vdso_info.size / PAGE_SIZE, sizeof(struct page *), GFP_KERNEL);
107
113
108
114
pfn = __phys_to_pfn(__pa_symbol(vdso_info.vdso));
109
115
for (i = 0; i < vdso_info.size / PAGE_SIZE; i++)
+4
-3
arch/loongarch/kvm/timer.c
+4
-3
arch/loongarch/kvm/timer.c
···
161
161
if (kvm_vcpu_is_blocking(vcpu)) {
162
162
163
163
/*
164
-
* HRTIMER_MODE_PINNED is suggested since vcpu may run in
165
-
* the same physical cpu in next time
164
+
* HRTIMER_MODE_PINNED_HARD is suggested since vcpu may run in
165
+
* the same physical cpu in next time, and the timer should run
166
+
* in hardirq context even in the PREEMPT_RT case.
166
167
*/
167
-
hrtimer_start(&vcpu->arch.swtimer, expire, HRTIMER_MODE_ABS_PINNED);
168
+
hrtimer_start(&vcpu->arch.swtimer, expire, HRTIMER_MODE_ABS_PINNED_HARD);
168
169
}
169
170
}
170
171
+1
-1
arch/loongarch/kvm/vcpu.c
+1
-1
arch/loongarch/kvm/vcpu.c
···
1457
1457
vcpu->arch.vpid = 0;
1458
1458
vcpu->arch.flush_gpa = INVALID_GPA;
1459
1459
1460
-
hrtimer_init(&vcpu->arch.swtimer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
1460
+
hrtimer_init(&vcpu->arch.swtimer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED_HARD);
1461
1461
vcpu->arch.swtimer.function = kvm_swtimer_wakeup;
1462
1462
1463
1463
vcpu->arch.handle_exit = kvm_handle_exit;
+2
arch/loongarch/mm/init.c
+2
arch/loongarch/mm/init.c
+20
arch/loongarch/mm/pgtable.c
+20
arch/loongarch/mm/pgtable.c
···
116
116
EXPORT_SYMBOL_GPL(pud_init);
117
117
#endif
118
118
119
+
void kernel_pte_init(void *addr)
120
+
{
121
+
unsigned long *p, *end;
122
+
123
+
p = (unsigned long *)addr;
124
+
end = p + PTRS_PER_PTE;
125
+
126
+
do {
127
+
p[0] = _PAGE_GLOBAL;
128
+
p[1] = _PAGE_GLOBAL;
129
+
p[2] = _PAGE_GLOBAL;
130
+
p[3] = _PAGE_GLOBAL;
131
+
p[4] = _PAGE_GLOBAL;
132
+
p += 8;
133
+
p[-3] = _PAGE_GLOBAL;
134
+
p[-2] = _PAGE_GLOBAL;
135
+
p[-1] = _PAGE_GLOBAL;
136
+
} while (p != end);
137
+
}
138
+
119
139
pmd_t mk_pmd(struct page *page, pgprot_t prot)
120
140
{
121
141
pmd_t pmd;
+2
-1
include/linux/mm.h
+2
-1
include/linux/mm.h
···
3818
3818
struct page * __populate_section_memmap(unsigned long pfn,
3819
3819
unsigned long nr_pages, int nid, struct vmem_altmap *altmap,
3820
3820
struct dev_pagemap *pgmap);
3821
-
void pmd_init(void *addr);
3822
3821
void pud_init(void *addr);
3822
+
void pmd_init(void *addr);
3823
+
void kernel_pte_init(void *addr);
3823
3824
pgd_t *vmemmap_pgd_populate(unsigned long addr, int node);
3824
3825
p4d_t *vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node);
3825
3826
pud_t *vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node);
+7
-1
mm/kasan/init.c
+7
-1
mm/kasan/init.c
···
106
106
}
107
107
}
108
108
109
+
void __weak __meminit kernel_pte_init(void *addr)
110
+
{
111
+
}
112
+
109
113
static int __ref zero_pmd_populate(pud_t *pud, unsigned long addr,
110
114
unsigned long end)
111
115
{
···
130
126
131
127
if (slab_is_available())
132
128
p = pte_alloc_one_kernel(&init_mm);
133
-
else
129
+
else {
134
130
p = early_alloc(PAGE_SIZE, NUMA_NO_NODE);
131
+
kernel_pte_init(p);
132
+
}
135
133
if (!p)
136
134
return -ENOMEM;
137
135
+5
mm/sparse-vmemmap.c
+5
mm/sparse-vmemmap.c
···
184
184
return p;
185
185
}
186
186
187
+
void __weak __meminit kernel_pte_init(void *addr)
188
+
{
189
+
}
190
+
187
191
pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
188
192
{
189
193
pmd_t *pmd = pmd_offset(pud, addr);
···
195
191
void *p = vmemmap_alloc_block_zero(PAGE_SIZE, node);
196
192
if (!p)
197
193
return NULL;
194
+
kernel_pte_init(p);
198
195
pmd_populate_kernel(&init_mm, pmd, p);
199
196
}
200
197
return pmd;