2 * TLB flush routines for radix kernels.
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.
13 #include <linux/hugetlb.h>
14 #include <linux/memblock.h>
16 #include <asm/ppc-opcode.h>
18 #include <asm/tlbflush.h>
19 #include <asm/trace.h>
20 #include <asm/cputhreads.h>
22 #define RIC_FLUSH_TLB 0
23 #define RIC_FLUSH_PWC 1
24 #define RIC_FLUSH_ALL 2
26 static inline void __tlbiel_pid(unsigned long pid, int set,
29 unsigned long rb,rs,prs,r;
31 rb = PPC_BIT(53); /* IS = 1 */
32 rb |= set << PPC_BITLSHIFT(51);
33 rs = ((unsigned long)pid) << PPC_BITLSHIFT(31);
34 prs = 1; /* process scoped */
35 r = 1; /* raidx format */
37 asm volatile(PPC_TLBIEL(%0, %4, %3, %2, %1)
38 : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
39 trace_tlbie(0, 1, rb, rs, ric, prs, r);
43 * We use 128 set in radix mode and 256 set in hpt mode.
45 static inline void _tlbiel_pid(unsigned long pid, unsigned long ric)
49 asm volatile("ptesync": : :"memory");
52 * Flush the first set of the TLB, and if we're doing a RIC_FLUSH_ALL,
53 * also flush the entire Page Walk Cache.
55 __tlbiel_pid(pid, 0, ric);
57 /* For PWC, only one flush is needed */
58 if (ric == RIC_FLUSH_PWC) {
59 asm volatile("ptesync": : :"memory");
63 /* For the remaining sets, just flush the TLB */
64 for (set = 1; set < POWER9_TLB_SETS_RADIX ; set++)
65 __tlbiel_pid(pid, set, RIC_FLUSH_TLB);
67 asm volatile("ptesync": : :"memory");
68 asm volatile(PPC_INVALIDATE_ERAT "; isync" : : :"memory");
71 static inline void _tlbie_pid(unsigned long pid, unsigned long ric)
73 unsigned long rb,rs,prs,r;
75 rb = PPC_BIT(53); /* IS = 1 */
76 rs = pid << PPC_BITLSHIFT(31);
77 prs = 1; /* process scoped */
78 r = 1; /* raidx format */
80 asm volatile("ptesync": : :"memory");
81 asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
82 : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
83 asm volatile("eieio; tlbsync; ptesync": : :"memory");
84 trace_tlbie(0, 0, rb, rs, ric, prs, r);
87 static inline void _tlbiel_va(unsigned long va, unsigned long pid,
88 unsigned long ap, unsigned long ric)
90 unsigned long rb,rs,prs,r;
92 rb = va & ~(PPC_BITMASK(52, 63));
93 rb |= ap << PPC_BITLSHIFT(58);
94 rs = pid << PPC_BITLSHIFT(31);
95 prs = 1; /* process scoped */
96 r = 1; /* raidx format */
98 asm volatile("ptesync": : :"memory");
99 asm volatile(PPC_TLBIEL(%0, %4, %3, %2, %1)
100 : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
101 asm volatile("ptesync": : :"memory");
102 trace_tlbie(0, 1, rb, rs, ric, prs, r);
105 static inline void _tlbie_va(unsigned long va, unsigned long pid,
106 unsigned long ap, unsigned long ric)
108 unsigned long rb,rs,prs,r;
110 rb = va & ~(PPC_BITMASK(52, 63));
111 rb |= ap << PPC_BITLSHIFT(58);
112 rs = pid << PPC_BITLSHIFT(31);
113 prs = 1; /* process scoped */
114 r = 1; /* raidx format */
116 asm volatile("ptesync": : :"memory");
117 asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
118 : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
119 asm volatile("eieio; tlbsync; ptesync": : :"memory");
120 trace_tlbie(0, 0, rb, rs, ric, prs, r);
124 * Base TLB flushing operations:
126 * - flush_tlb_mm(mm) flushes the specified mm context TLB's
127 * - flush_tlb_page(vma, vmaddr) flushes one page
128 * - flush_tlb_range(vma, start, end) flushes a range of pages
129 * - flush_tlb_kernel_range(start, end) flushes kernel pages
131 * - local_* variants of page and mm only apply to the current
134 void radix__local_flush_tlb_mm(struct mm_struct *mm)
139 pid = mm->context.id;
140 if (pid != MMU_NO_CONTEXT)
141 _tlbiel_pid(pid, RIC_FLUSH_TLB);
144 EXPORT_SYMBOL(radix__local_flush_tlb_mm);
147 static void radix__local_flush_all_mm(struct mm_struct *mm)
152 pid = mm->context.id;
153 if (pid != MMU_NO_CONTEXT)
154 _tlbiel_pid(pid, RIC_FLUSH_ALL);
157 #endif /* CONFIG_SMP */
159 void radix__local_flush_tlb_page_psize(struct mm_struct *mm, unsigned long vmaddr,
163 unsigned long ap = mmu_get_ap(psize);
166 pid = mm ? mm->context.id : 0;
167 if (pid != MMU_NO_CONTEXT)
168 _tlbiel_va(vmaddr, pid, ap, RIC_FLUSH_TLB);
172 void radix__local_flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
174 #ifdef CONFIG_HUGETLB_PAGE
175 /* need the return fix for nohash.c */
176 if (vma && is_vm_hugetlb_page(vma))
177 return __local_flush_hugetlb_page(vma, vmaddr);
179 radix__local_flush_tlb_page_psize(vma ? vma->vm_mm : NULL, vmaddr,
182 EXPORT_SYMBOL(radix__local_flush_tlb_page);
185 void radix__flush_tlb_mm(struct mm_struct *mm)
190 pid = mm->context.id;
191 if (unlikely(pid == MMU_NO_CONTEXT))
194 if (!mm_is_thread_local(mm))
195 _tlbie_pid(pid, RIC_FLUSH_TLB);
197 _tlbiel_pid(pid, RIC_FLUSH_TLB);
201 EXPORT_SYMBOL(radix__flush_tlb_mm);
203 static void radix__flush_all_mm(struct mm_struct *mm)
208 pid = mm->context.id;
209 if (unlikely(pid == MMU_NO_CONTEXT))
212 if (!mm_is_thread_local(mm))
213 _tlbie_pid(pid, RIC_FLUSH_ALL);
215 _tlbiel_pid(pid, RIC_FLUSH_ALL);
220 void radix__flush_tlb_pwc(struct mmu_gather *tlb, unsigned long addr)
222 tlb->need_flush_all = 1;
224 EXPORT_SYMBOL(radix__flush_tlb_pwc);
226 void radix__flush_tlb_page_psize(struct mm_struct *mm, unsigned long vmaddr,
230 unsigned long ap = mmu_get_ap(psize);
233 pid = mm ? mm->context.id : 0;
234 if (unlikely(pid == MMU_NO_CONTEXT))
236 if (!mm_is_thread_local(mm))
237 _tlbie_va(vmaddr, pid, ap, RIC_FLUSH_TLB);
239 _tlbiel_va(vmaddr, pid, ap, RIC_FLUSH_TLB);
244 void radix__flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
246 #ifdef CONFIG_HUGETLB_PAGE
247 if (vma && is_vm_hugetlb_page(vma))
248 return flush_hugetlb_page(vma, vmaddr);
250 radix__flush_tlb_page_psize(vma ? vma->vm_mm : NULL, vmaddr,
253 EXPORT_SYMBOL(radix__flush_tlb_page);
255 #else /* CONFIG_SMP */
256 #define radix__flush_all_mm radix__local_flush_all_mm
257 #endif /* CONFIG_SMP */
259 void radix__flush_tlb_kernel_range(unsigned long start, unsigned long end)
261 _tlbie_pid(0, RIC_FLUSH_ALL);
263 EXPORT_SYMBOL(radix__flush_tlb_kernel_range);
266 * Currently, for range flushing, we just do a full mm flush. Because
267 * we use this in code path where we don' track the page size.
269 void radix__flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
273 struct mm_struct *mm = vma->vm_mm;
275 radix__flush_tlb_mm(mm);
277 EXPORT_SYMBOL(radix__flush_tlb_range);
279 static int radix_get_mmu_psize(int page_size)
283 if (page_size == (1UL << mmu_psize_defs[mmu_virtual_psize].shift))
284 psize = mmu_virtual_psize;
285 else if (page_size == (1UL << mmu_psize_defs[MMU_PAGE_2M].shift))
287 else if (page_size == (1UL << mmu_psize_defs[MMU_PAGE_1G].shift))
294 void radix__tlb_flush(struct mmu_gather *tlb)
297 struct mm_struct *mm = tlb->mm;
298 int page_size = tlb->page_size;
300 psize = radix_get_mmu_psize(page_size);
302 * if page size is not something we understand, do a full mm flush
304 if (psize != -1 && !tlb->fullmm && !tlb->need_flush_all)
305 radix__flush_tlb_range_psize(mm, tlb->start, tlb->end, psize);
306 else if (tlb->need_flush_all) {
307 tlb->need_flush_all = 0;
308 radix__flush_all_mm(mm);
310 radix__flush_tlb_mm(mm);
313 #define TLB_FLUSH_ALL -1UL
315 * Number of pages above which we will do a bcast tlbie. Just a
316 * number at this point copied from x86
318 static unsigned long tlb_single_page_flush_ceiling __read_mostly = 33;
320 void radix__flush_tlb_range_psize(struct mm_struct *mm, unsigned long start,
321 unsigned long end, int psize)
325 int local = mm_is_thread_local(mm);
326 unsigned long ap = mmu_get_ap(psize);
327 unsigned long page_size = 1UL << mmu_psize_defs[psize].shift;
331 pid = mm ? mm->context.id : 0;
332 if (unlikely(pid == MMU_NO_CONTEXT))
335 if (end == TLB_FLUSH_ALL ||
336 (end - start) > tlb_single_page_flush_ceiling * page_size) {
338 _tlbiel_pid(pid, RIC_FLUSH_TLB);
340 _tlbie_pid(pid, RIC_FLUSH_TLB);
343 for (addr = start; addr < end; addr += page_size) {
346 _tlbiel_va(addr, pid, ap, RIC_FLUSH_TLB);
348 _tlbie_va(addr, pid, ap, RIC_FLUSH_TLB);
354 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
355 void radix__flush_tlb_collapsed_pmd(struct mm_struct *mm, unsigned long addr)
357 int local = mm_is_thread_local(mm);
358 unsigned long ap = mmu_get_ap(mmu_virtual_psize);
359 unsigned long pid, end;
362 pid = mm ? mm->context.id : 0;
364 if (unlikely(pid == MMU_NO_CONTEXT))
367 /* 4k page size, just blow the world */
368 if (PAGE_SIZE == 0x1000) {
369 radix__flush_all_mm(mm);
374 /* Otherwise first do the PWC */
376 _tlbiel_pid(pid, RIC_FLUSH_PWC);
378 _tlbie_pid(pid, RIC_FLUSH_PWC);
380 /* Then iterate the pages */
381 end = addr + HPAGE_PMD_SIZE;
382 for (; addr < end; addr += PAGE_SIZE) {
384 _tlbiel_va(addr, pid, ap, RIC_FLUSH_TLB);
386 _tlbie_va(addr, pid, ap, RIC_FLUSH_TLB);
391 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
393 void radix__flush_tlb_lpid_va(unsigned long lpid, unsigned long gpa,
394 unsigned long page_size)
396 unsigned long rb,rs,prs,r;
398 unsigned long ric = RIC_FLUSH_TLB;
400 ap = mmu_get_ap(radix_get_mmu_psize(page_size));
401 rb = gpa & ~(PPC_BITMASK(52, 63));
402 rb |= ap << PPC_BITLSHIFT(58);
403 rs = lpid & ((1UL << 32) - 1);
404 prs = 0; /* process scoped */
405 r = 1; /* raidx format */
407 asm volatile("ptesync": : :"memory");
408 asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
409 : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
410 asm volatile("eieio; tlbsync; ptesync": : :"memory");
411 trace_tlbie(lpid, 0, rb, rs, ric, prs, r);
413 EXPORT_SYMBOL(radix__flush_tlb_lpid_va);
415 void radix__flush_tlb_lpid(unsigned long lpid)
417 unsigned long rb,rs,prs,r;
418 unsigned long ric = RIC_FLUSH_ALL;
420 rb = 0x2 << PPC_BITLSHIFT(53); /* IS = 2 */
421 rs = lpid & ((1UL << 32) - 1);
422 prs = 0; /* partition scoped */
423 r = 1; /* raidx format */
425 asm volatile("ptesync": : :"memory");
426 asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
427 : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
428 asm volatile("eieio; tlbsync; ptesync": : :"memory");
429 trace_tlbie(lpid, 0, rb, rs, ric, prs, r);
431 EXPORT_SYMBOL(radix__flush_tlb_lpid);
433 void radix__flush_pmd_tlb_range(struct vm_area_struct *vma,
434 unsigned long start, unsigned long end)
436 radix__flush_tlb_range_psize(vma->vm_mm, start, end, MMU_PAGE_2M);
438 EXPORT_SYMBOL(radix__flush_pmd_tlb_range);
440 void radix__flush_tlb_all(void)
442 unsigned long rb,prs,r,rs;
443 unsigned long ric = RIC_FLUSH_ALL;
445 rb = 0x3 << PPC_BITLSHIFT(53); /* IS = 3 */
446 prs = 0; /* partition scoped */
447 r = 1; /* raidx format */
448 rs = 1 & ((1UL << 32) - 1); /* any LPID value to flush guest mappings */
450 asm volatile("ptesync": : :"memory");
452 * now flush guest entries by passing PRS = 1 and LPID != 0
454 asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
455 : : "r"(rb), "i"(r), "i"(1), "i"(ric), "r"(rs) : "memory");
456 trace_tlbie(0, 0, rb, rs, ric, prs, r);
458 * now flush host entires by passing PRS = 0 and LPID == 0
460 asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
461 : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(0) : "memory");
462 asm volatile("eieio; tlbsync; ptesync": : :"memory");
463 trace_tlbie(0, 0, rb, 0, ric, prs, r);
466 void radix__flush_tlb_pte_p9_dd1(unsigned long old_pte, struct mm_struct *mm,
467 unsigned long address)
470 * We track page size in pte only for DD1, So we can
471 * call this only on DD1.
473 if (!cpu_has_feature(CPU_FTR_POWER9_DD1)) {
478 if (old_pte & R_PAGE_LARGE)
479 radix__flush_tlb_page_psize(mm, address, MMU_PAGE_2M);
481 radix__flush_tlb_page_psize(mm, address, mmu_virtual_psize);
484 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
485 extern void radix_kvm_prefetch_workaround(struct mm_struct *mm)
487 unsigned int pid = mm->context.id;
489 if (unlikely(pid == MMU_NO_CONTEXT))
493 * If this context hasn't run on that CPU before and KVM is
494 * around, there's a slim chance that the guest on another
495 * CPU just brought in obsolete translation into the TLB of
496 * this CPU due to a bad prefetch using the guest PID on
497 * the way into the hypervisor.
499 * We work around this here. If KVM is possible, we check if
500 * any sibling thread is in KVM. If it is, the window may exist
501 * and thus we flush that PID from the core.
503 * A potential future improvement would be to mark which PIDs
504 * have never been used on the system and avoid it if the PID
505 * is new and the process has no other cpumask bit set.
507 if (cpu_has_feature(CPU_FTR_HVMODE) && radix_enabled()) {
508 int cpu = smp_processor_id();
509 int sib = cpu_first_thread_sibling(cpu);
512 for (; sib <= cpu_last_thread_sibling(cpu) && !flush; sib++) {
515 if (paca[sib].kvm_hstate.kvm_vcpu)
519 _tlbiel_pid(pid, RIC_FLUSH_ALL);
522 EXPORT_SYMBOL_GPL(radix_kvm_prefetch_workaround);
523 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */