xref: /linux/arch/riscv/mm/tlbflush.c (revision 1e525507)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/mm.h>
4 #include <linux/smp.h>
5 #include <linux/sched.h>
6 #include <linux/hugetlb.h>
7 #include <asm/sbi.h>
8 #include <asm/mmu_context.h>
9 
10 static inline void local_flush_tlb_all_asid(unsigned long asid)
11 {
12 	if (asid != FLUSH_TLB_NO_ASID)
13 		__asm__ __volatile__ ("sfence.vma x0, %0"
14 				:
15 				: "r" (asid)
16 				: "memory");
17 	else
18 		local_flush_tlb_all();
19 }
20 
21 static inline void local_flush_tlb_page_asid(unsigned long addr,
22 		unsigned long asid)
23 {
24 	if (asid != FLUSH_TLB_NO_ASID)
25 		__asm__ __volatile__ ("sfence.vma %0, %1"
26 				:
27 				: "r" (addr), "r" (asid)
28 				: "memory");
29 	else
30 		local_flush_tlb_page(addr);
31 }
32 
33 /*
34  * Flush entire TLB if number of entries to be flushed is greater
35  * than the threshold below.
36  */
37 static unsigned long tlb_flush_all_threshold __read_mostly = 64;
38 
39 static void local_flush_tlb_range_threshold_asid(unsigned long start,
40 						 unsigned long size,
41 						 unsigned long stride,
42 						 unsigned long asid)
43 {
44 	unsigned long nr_ptes_in_range = DIV_ROUND_UP(size, stride);
45 	int i;
46 
47 	if (nr_ptes_in_range > tlb_flush_all_threshold) {
48 		local_flush_tlb_all_asid(asid);
49 		return;
50 	}
51 
52 	for (i = 0; i < nr_ptes_in_range; ++i) {
53 		local_flush_tlb_page_asid(start, asid);
54 		start += stride;
55 	}
56 }
57 
58 static inline void local_flush_tlb_range_asid(unsigned long start,
59 		unsigned long size, unsigned long stride, unsigned long asid)
60 {
61 	if (size <= stride)
62 		local_flush_tlb_page_asid(start, asid);
63 	else if (size == FLUSH_TLB_MAX_SIZE)
64 		local_flush_tlb_all_asid(asid);
65 	else
66 		local_flush_tlb_range_threshold_asid(start, size, stride, asid);
67 }
68 
69 /* Flush a range of kernel pages without broadcasting */
70 void local_flush_tlb_kernel_range(unsigned long start, unsigned long end)
71 {
72 	local_flush_tlb_range_asid(start, end - start, PAGE_SIZE, FLUSH_TLB_NO_ASID);
73 }
74 
75 static void __ipi_flush_tlb_all(void *info)
76 {
77 	local_flush_tlb_all();
78 }
79 
80 void flush_tlb_all(void)
81 {
82 	if (riscv_use_ipi_for_rfence())
83 		on_each_cpu(__ipi_flush_tlb_all, NULL, 1);
84 	else
85 		sbi_remote_sfence_vma_asid(NULL, 0, FLUSH_TLB_MAX_SIZE, FLUSH_TLB_NO_ASID);
86 }
87 
88 struct flush_tlb_range_data {
89 	unsigned long asid;
90 	unsigned long start;
91 	unsigned long size;
92 	unsigned long stride;
93 };
94 
95 static void __ipi_flush_tlb_range_asid(void *info)
96 {
97 	struct flush_tlb_range_data *d = info;
98 
99 	local_flush_tlb_range_asid(d->start, d->size, d->stride, d->asid);
100 }
101 
102 static void __flush_tlb_range(const struct cpumask *cmask, unsigned long asid,
103 			      unsigned long start, unsigned long size,
104 			      unsigned long stride)
105 {
106 	struct flush_tlb_range_data ftd;
107 	bool broadcast;
108 
109 	if (cpumask_empty(cmask))
110 		return;
111 
112 	if (cmask != cpu_online_mask) {
113 		unsigned int cpuid;
114 
115 		cpuid = get_cpu();
116 		/* check if the tlbflush needs to be sent to other CPUs */
117 		broadcast = cpumask_any_but(cmask, cpuid) < nr_cpu_ids;
118 	} else {
119 		broadcast = true;
120 	}
121 
122 	if (broadcast) {
123 		if (riscv_use_ipi_for_rfence()) {
124 			ftd.asid = asid;
125 			ftd.start = start;
126 			ftd.size = size;
127 			ftd.stride = stride;
128 			on_each_cpu_mask(cmask,
129 					 __ipi_flush_tlb_range_asid,
130 					 &ftd, 1);
131 		} else
132 			sbi_remote_sfence_vma_asid(cmask,
133 						   start, size, asid);
134 	} else {
135 		local_flush_tlb_range_asid(start, size, stride, asid);
136 	}
137 
138 	if (cmask != cpu_online_mask)
139 		put_cpu();
140 }
141 
142 static inline unsigned long get_mm_asid(struct mm_struct *mm)
143 {
144 	return static_branch_unlikely(&use_asid_allocator) ?
145 			atomic_long_read(&mm->context.id) & asid_mask : FLUSH_TLB_NO_ASID;
146 }
147 
148 void flush_tlb_mm(struct mm_struct *mm)
149 {
150 	__flush_tlb_range(mm_cpumask(mm), get_mm_asid(mm),
151 			  0, FLUSH_TLB_MAX_SIZE, PAGE_SIZE);
152 }
153 
154 void flush_tlb_mm_range(struct mm_struct *mm,
155 			unsigned long start, unsigned long end,
156 			unsigned int page_size)
157 {
158 	__flush_tlb_range(mm_cpumask(mm), get_mm_asid(mm),
159 			  start, end - start, page_size);
160 }
161 
162 void flush_tlb_page(struct vm_area_struct *vma, unsigned long addr)
163 {
164 	__flush_tlb_range(mm_cpumask(vma->vm_mm), get_mm_asid(vma->vm_mm),
165 			  addr, PAGE_SIZE, PAGE_SIZE);
166 }
167 
168 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
169 		     unsigned long end)
170 {
171 	unsigned long stride_size;
172 
173 	if (!is_vm_hugetlb_page(vma)) {
174 		stride_size = PAGE_SIZE;
175 	} else {
176 		stride_size = huge_page_size(hstate_vma(vma));
177 
178 		/*
179 		 * As stated in the privileged specification, every PTE in a
180 		 * NAPOT region must be invalidated, so reset the stride in that
181 		 * case.
182 		 */
183 		if (has_svnapot()) {
184 			if (stride_size >= PGDIR_SIZE)
185 				stride_size = PGDIR_SIZE;
186 			else if (stride_size >= P4D_SIZE)
187 				stride_size = P4D_SIZE;
188 			else if (stride_size >= PUD_SIZE)
189 				stride_size = PUD_SIZE;
190 			else if (stride_size >= PMD_SIZE)
191 				stride_size = PMD_SIZE;
192 			else
193 				stride_size = PAGE_SIZE;
194 		}
195 	}
196 
197 	__flush_tlb_range(mm_cpumask(vma->vm_mm), get_mm_asid(vma->vm_mm),
198 			  start, end - start, stride_size);
199 }
200 
201 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
202 {
203 	__flush_tlb_range(cpu_online_mask, FLUSH_TLB_NO_ASID,
204 			  start, end - start, PAGE_SIZE);
205 }
206 
207 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
208 void flush_pmd_tlb_range(struct vm_area_struct *vma, unsigned long start,
209 			unsigned long end)
210 {
211 	__flush_tlb_range(mm_cpumask(vma->vm_mm), get_mm_asid(vma->vm_mm),
212 			  start, end - start, PMD_SIZE);
213 }
214 #endif
215 
216 bool arch_tlbbatch_should_defer(struct mm_struct *mm)
217 {
218 	return true;
219 }
220 
221 void arch_tlbbatch_add_pending(struct arch_tlbflush_unmap_batch *batch,
222 			       struct mm_struct *mm,
223 			       unsigned long uaddr)
224 {
225 	cpumask_or(&batch->cpumask, &batch->cpumask, mm_cpumask(mm));
226 }
227 
228 void arch_flush_tlb_batched_pending(struct mm_struct *mm)
229 {
230 	flush_tlb_mm(mm);
231 }
232 
233 void arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch *batch)
234 {
235 	__flush_tlb_range(&batch->cpumask, FLUSH_TLB_NO_ASID, 0,
236 			  FLUSH_TLB_MAX_SIZE, PAGE_SIZE);
237 	cpumask_clear(&batch->cpumask);
238 }
239