1 /*
2 * PowerPC64 SLB support.
3 *
4 * Copyright (C) 2004 David Gibson <dwg@au.ibm.com>, IBM
5 * Based on earlier code written by:
6 * Dave Engebretsen and Mike Corrigan {engebret|mikejc}@us.ibm.com
7 * Copyright (c) 2001 Dave Engebretsen
8 * Copyright (C) 2002 Anton Blanchard <anton@au.ibm.com>, IBM
9 *
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
15 */
16
17 #include <asm/pgtable.h>
18 #include <asm/mmu.h>
19 #include <asm/mmu_context.h>
20 #include <asm/paca.h>
21 #include <asm/cputable.h>
22 #include <asm/cacheflush.h>
23 #include <asm/smp.h>
24 #include <linux/compiler.h>
25 #include <asm/udbg.h>
26 #include <asm/code-patching.h>
27
28
29 extern void slb_allocate_realmode(unsigned long ea);
30 extern void slb_allocate_user(unsigned long ea);
31
slb_allocate(unsigned long ea)32 static void slb_allocate(unsigned long ea)
33 {
34 /* Currently, we do real mode for all SLBs including user, but
35 * that will change if we bring back dynamic VSIDs
36 */
37 slb_allocate_realmode(ea);
38 }
39
40 #define slb_esid_mask(ssize) \
41 (((ssize) == MMU_SEGSIZE_256M)? ESID_MASK: ESID_MASK_1T)
42
mk_esid_data(unsigned long ea,int ssize,unsigned long slot)43 static inline unsigned long mk_esid_data(unsigned long ea, int ssize,
44 unsigned long slot)
45 {
46 return (ea & slb_esid_mask(ssize)) | SLB_ESID_V | slot;
47 }
48
mk_vsid_data(unsigned long ea,int ssize,unsigned long flags)49 static inline unsigned long mk_vsid_data(unsigned long ea, int ssize,
50 unsigned long flags)
51 {
52 return (get_kernel_vsid(ea, ssize) << slb_vsid_shift(ssize)) | flags |
53 ((unsigned long) ssize << SLB_VSID_SSIZE_SHIFT);
54 }
55
slb_shadow_update(unsigned long ea,int ssize,unsigned long flags,unsigned long entry)56 static inline void slb_shadow_update(unsigned long ea, int ssize,
57 unsigned long flags,
58 unsigned long entry)
59 {
60 /*
61 * Clear the ESID first so the entry is not valid while we are
62 * updating it. No write barriers are needed here, provided
63 * we only update the current CPU's SLB shadow buffer.
64 */
65 get_slb_shadow()->save_area[entry].esid = 0;
66 get_slb_shadow()->save_area[entry].vsid =
67 cpu_to_be64(mk_vsid_data(ea, ssize, flags));
68 get_slb_shadow()->save_area[entry].esid =
69 cpu_to_be64(mk_esid_data(ea, ssize, entry));
70 }
71
slb_shadow_clear(unsigned long entry)72 static inline void slb_shadow_clear(unsigned long entry)
73 {
74 get_slb_shadow()->save_area[entry].esid = 0;
75 }
76
create_shadowed_slbe(unsigned long ea,int ssize,unsigned long flags,unsigned long entry)77 static inline void create_shadowed_slbe(unsigned long ea, int ssize,
78 unsigned long flags,
79 unsigned long entry)
80 {
81 /*
82 * Updating the shadow buffer before writing the SLB ensures
83 * we don't get a stale entry here if we get preempted by PHYP
84 * between these two statements.
85 */
86 slb_shadow_update(ea, ssize, flags, entry);
87
88 asm volatile("slbmte %0,%1" :
89 : "r" (mk_vsid_data(ea, ssize, flags)),
90 "r" (mk_esid_data(ea, ssize, entry))
91 : "memory" );
92 }
93
__slb_flush_and_rebolt(void)94 static void __slb_flush_and_rebolt(void)
95 {
96 /* If you change this make sure you change SLB_NUM_BOLTED
97 * and PR KVM appropriately too. */
98 unsigned long linear_llp, vmalloc_llp, lflags, vflags;
99 unsigned long ksp_esid_data, ksp_vsid_data;
100
101 linear_llp = mmu_psize_defs[mmu_linear_psize].sllp;
102 vmalloc_llp = mmu_psize_defs[mmu_vmalloc_psize].sllp;
103 lflags = SLB_VSID_KERNEL | linear_llp;
104 vflags = SLB_VSID_KERNEL | vmalloc_llp;
105
106 ksp_esid_data = mk_esid_data(get_paca()->kstack, mmu_kernel_ssize, 2);
107 if ((ksp_esid_data & ~0xfffffffUL) <= PAGE_OFFSET) {
108 ksp_esid_data &= ~SLB_ESID_V;
109 ksp_vsid_data = 0;
110 slb_shadow_clear(2);
111 } else {
112 /* Update stack entry; others don't change */
113 slb_shadow_update(get_paca()->kstack, mmu_kernel_ssize, lflags, 2);
114 ksp_vsid_data =
115 be64_to_cpu(get_slb_shadow()->save_area[2].vsid);
116 }
117
118 /* We need to do this all in asm, so we're sure we don't touch
119 * the stack between the slbia and rebolting it. */
120 asm volatile("isync\n"
121 "slbia\n"
122 /* Slot 1 - first VMALLOC segment */
123 "slbmte %0,%1\n"
124 /* Slot 2 - kernel stack */
125 "slbmte %2,%3\n"
126 "isync"
127 :: "r"(mk_vsid_data(VMALLOC_START, mmu_kernel_ssize, vflags)),
128 "r"(mk_esid_data(VMALLOC_START, mmu_kernel_ssize, 1)),
129 "r"(ksp_vsid_data),
130 "r"(ksp_esid_data)
131 : "memory");
132 }
133
slb_flush_and_rebolt(void)134 void slb_flush_and_rebolt(void)
135 {
136
137 WARN_ON(!irqs_disabled());
138
139 /*
140 * We can't take a PMU exception in the following code, so hard
141 * disable interrupts.
142 */
143 hard_irq_disable();
144
145 __slb_flush_and_rebolt();
146 get_paca()->slb_cache_ptr = 0;
147 }
148
slb_vmalloc_update(void)149 void slb_vmalloc_update(void)
150 {
151 unsigned long vflags;
152
153 vflags = SLB_VSID_KERNEL | mmu_psize_defs[mmu_vmalloc_psize].sllp;
154 slb_shadow_update(VMALLOC_START, mmu_kernel_ssize, vflags, 1);
155 slb_flush_and_rebolt();
156 }
157
158 /* Helper function to compare esids. There are four cases to handle.
159 * 1. The system is not 1T segment size capable. Use the GET_ESID compare.
160 * 2. The system is 1T capable, both addresses are < 1T, use the GET_ESID compare.
161 * 3. The system is 1T capable, only one of the two addresses is > 1T. This is not a match.
162 * 4. The system is 1T capable, both addresses are > 1T, use the GET_ESID_1T macro to compare.
163 */
esids_match(unsigned long addr1,unsigned long addr2)164 static inline int esids_match(unsigned long addr1, unsigned long addr2)
165 {
166 int esid_1t_count;
167
168 /* System is not 1T segment size capable. */
169 if (!mmu_has_feature(MMU_FTR_1T_SEGMENT))
170 return (GET_ESID(addr1) == GET_ESID(addr2));
171
172 esid_1t_count = (((addr1 >> SID_SHIFT_1T) != 0) +
173 ((addr2 >> SID_SHIFT_1T) != 0));
174
175 /* both addresses are < 1T */
176 if (esid_1t_count == 0)
177 return (GET_ESID(addr1) == GET_ESID(addr2));
178
179 /* One address < 1T, the other > 1T. Not a match */
180 if (esid_1t_count == 1)
181 return 0;
182
183 /* Both addresses are > 1T. */
184 return (GET_ESID_1T(addr1) == GET_ESID_1T(addr2));
185 }
186
187 /* Flush all user entries from the segment table of the current processor. */
switch_slb(struct task_struct * tsk,struct mm_struct * mm)188 void switch_slb(struct task_struct *tsk, struct mm_struct *mm)
189 {
190 unsigned long offset;
191 unsigned long slbie_data = 0;
192 unsigned long pc = KSTK_EIP(tsk);
193 unsigned long stack = KSTK_ESP(tsk);
194 unsigned long exec_base;
195
196 /*
197 * We need interrupts hard-disabled here, not just soft-disabled,
198 * so that a PMU interrupt can't occur, which might try to access
199 * user memory (to get a stack trace) and possible cause an SLB miss
200 * which would update the slb_cache/slb_cache_ptr fields in the PACA.
201 */
202 hard_irq_disable();
203 offset = get_paca()->slb_cache_ptr;
204 if (!mmu_has_feature(MMU_FTR_NO_SLBIE_B) &&
205 offset <= SLB_CACHE_ENTRIES) {
206 int i;
207 asm volatile("isync" : : : "memory");
208 for (i = 0; i < offset; i++) {
209 slbie_data = (unsigned long)get_paca()->slb_cache[i]
210 << SID_SHIFT; /* EA */
211 slbie_data |= user_segment_size(slbie_data)
212 << SLBIE_SSIZE_SHIFT;
213 slbie_data |= SLBIE_C; /* C set for user addresses */
214 asm volatile("slbie %0" : : "r" (slbie_data));
215 }
216 asm volatile("isync" : : : "memory");
217 } else {
218 __slb_flush_and_rebolt();
219 }
220
221 /* Workaround POWER5 < DD2.1 issue */
222 if (offset == 1 || offset > SLB_CACHE_ENTRIES)
223 asm volatile("slbie %0" : : "r" (slbie_data));
224
225 get_paca()->slb_cache_ptr = 0;
226 get_paca()->context = mm->context;
227
228 /*
229 * preload some userspace segments into the SLB.
230 * Almost all 32 and 64bit PowerPC executables are linked at
231 * 0x10000000 so it makes sense to preload this segment.
232 */
233 exec_base = 0x10000000;
234
235 if (is_kernel_addr(pc) || is_kernel_addr(stack) ||
236 is_kernel_addr(exec_base))
237 return;
238
239 slb_allocate(pc);
240
241 if (!esids_match(pc, stack))
242 slb_allocate(stack);
243
244 if (!esids_match(pc, exec_base) &&
245 !esids_match(stack, exec_base))
246 slb_allocate(exec_base);
247 }
248
patch_slb_encoding(unsigned int * insn_addr,unsigned int immed)249 static inline void patch_slb_encoding(unsigned int *insn_addr,
250 unsigned int immed)
251 {
252 int insn = (*insn_addr & 0xffff0000) | immed;
253 patch_instruction(insn_addr, insn);
254 }
255
256 extern u32 slb_compare_rr_to_size[];
257 extern u32 slb_miss_kernel_load_linear[];
258 extern u32 slb_miss_kernel_load_io[];
259 extern u32 slb_compare_rr_to_size[];
260 extern u32 slb_miss_kernel_load_vmemmap[];
261
slb_set_size(u16 size)262 void slb_set_size(u16 size)
263 {
264 if (mmu_slb_size == size)
265 return;
266
267 mmu_slb_size = size;
268 patch_slb_encoding(slb_compare_rr_to_size, mmu_slb_size);
269 }
270
slb_initialize(void)271 void slb_initialize(void)
272 {
273 unsigned long linear_llp, vmalloc_llp, io_llp;
274 unsigned long lflags, vflags;
275 static int slb_encoding_inited;
276 #ifdef CONFIG_SPARSEMEM_VMEMMAP
277 unsigned long vmemmap_llp;
278 #endif
279
280 /* Prepare our SLB miss handler based on our page size */
281 linear_llp = mmu_psize_defs[mmu_linear_psize].sllp;
282 io_llp = mmu_psize_defs[mmu_io_psize].sllp;
283 vmalloc_llp = mmu_psize_defs[mmu_vmalloc_psize].sllp;
284 get_paca()->vmalloc_sllp = SLB_VSID_KERNEL | vmalloc_llp;
285 #ifdef CONFIG_SPARSEMEM_VMEMMAP
286 vmemmap_llp = mmu_psize_defs[mmu_vmemmap_psize].sllp;
287 #endif
288 if (!slb_encoding_inited) {
289 slb_encoding_inited = 1;
290 patch_slb_encoding(slb_miss_kernel_load_linear,
291 SLB_VSID_KERNEL | linear_llp);
292 patch_slb_encoding(slb_miss_kernel_load_io,
293 SLB_VSID_KERNEL | io_llp);
294 patch_slb_encoding(slb_compare_rr_to_size,
295 mmu_slb_size);
296
297 pr_devel("SLB: linear LLP = %04lx\n", linear_llp);
298 pr_devel("SLB: io LLP = %04lx\n", io_llp);
299
300 #ifdef CONFIG_SPARSEMEM_VMEMMAP
301 patch_slb_encoding(slb_miss_kernel_load_vmemmap,
302 SLB_VSID_KERNEL | vmemmap_llp);
303 pr_devel("SLB: vmemmap LLP = %04lx\n", vmemmap_llp);
304 #endif
305 }
306
307 get_paca()->stab_rr = SLB_NUM_BOLTED;
308
309 lflags = SLB_VSID_KERNEL | linear_llp;
310 vflags = SLB_VSID_KERNEL | vmalloc_llp;
311
312 /* Invalidate the entire SLB (even slot 0) & all the ERATS */
313 asm volatile("isync":::"memory");
314 asm volatile("slbmte %0,%0"::"r" (0) : "memory");
315 asm volatile("isync; slbia; isync":::"memory");
316 create_shadowed_slbe(PAGE_OFFSET, mmu_kernel_ssize, lflags, 0);
317
318 create_shadowed_slbe(VMALLOC_START, mmu_kernel_ssize, vflags, 1);
319
320 /* For the boot cpu, we're running on the stack in init_thread_union,
321 * which is in the first segment of the linear mapping, and also
322 * get_paca()->kstack hasn't been initialized yet.
323 * For secondary cpus, we need to bolt the kernel stack entry now.
324 */
325 slb_shadow_clear(2);
326 if (raw_smp_processor_id() != boot_cpuid &&
327 (get_paca()->kstack & slb_esid_mask(mmu_kernel_ssize)) > PAGE_OFFSET)
328 create_shadowed_slbe(get_paca()->kstack,
329 mmu_kernel_ssize, lflags, 2);
330
331 asm volatile("isync":::"memory");
332 }
333