11965aae3SH. Peter Anvin #ifndef _ASM_X86_MMU_CONTEXT_H 21965aae3SH. Peter Anvin #define _ASM_X86_MMU_CONTEXT_H 3bb898558SAl Viro 4bb898558SAl Viro #include <asm/desc.h> 560063497SArun Sharma #include <linux/atomic.h> 6d17d8f9dSDave Hansen #include <linux/mm_types.h> 77d06d9c9SDave Hansen #include <linux/pkeys.h> 8d17d8f9dSDave Hansen 9d17d8f9dSDave Hansen #include <trace/events/tlb.h> 10d17d8f9dSDave Hansen 11bb898558SAl Viro #include <asm/pgalloc.h> 12bb898558SAl Viro #include <asm/tlbflush.h> 13bb898558SAl Viro #include <asm/paravirt.h> 14fe3d197fSDave Hansen #include <asm/mpx.h> 15bb898558SAl Viro #ifndef CONFIG_PARAVIRT 16bb898558SAl Viro static inline void paravirt_activate_mm(struct mm_struct *prev, 17bb898558SAl Viro struct mm_struct *next) 18bb898558SAl Viro { 19bb898558SAl Viro } 20bb898558SAl Viro #endif /* !CONFIG_PARAVIRT */ 21bb898558SAl Viro 227911d3f7SAndy Lutomirski #ifdef CONFIG_PERF_EVENTS 23a6673429SAndy Lutomirski extern struct static_key rdpmc_always_available; 24a6673429SAndy Lutomirski 257911d3f7SAndy Lutomirski static inline void load_mm_cr4(struct mm_struct *mm) 267911d3f7SAndy Lutomirski { 27a833581eSPeter Zijlstra if (static_key_false(&rdpmc_always_available) || 28a6673429SAndy Lutomirski atomic_read(&mm->context.perf_rdpmc_allowed)) 297911d3f7SAndy Lutomirski cr4_set_bits(X86_CR4_PCE); 307911d3f7SAndy Lutomirski else 317911d3f7SAndy Lutomirski cr4_clear_bits(X86_CR4_PCE); 327911d3f7SAndy Lutomirski } 337911d3f7SAndy Lutomirski #else 347911d3f7SAndy Lutomirski static inline void load_mm_cr4(struct mm_struct *mm) {} 357911d3f7SAndy Lutomirski #endif 367911d3f7SAndy Lutomirski 37a5b9e5a2SAndy Lutomirski #ifdef CONFIG_MODIFY_LDT_SYSCALL 38bb898558SAl Viro /* 3937868fe1SAndy Lutomirski * ldt_structs can be allocated, used, and freed, but they are never 4037868fe1SAndy Lutomirski * modified while live. 4137868fe1SAndy Lutomirski */ 4237868fe1SAndy Lutomirski struct ldt_struct { 4337868fe1SAndy Lutomirski /* 4437868fe1SAndy Lutomirski * Xen requires page-aligned LDTs with special permissions. This is 4537868fe1SAndy Lutomirski * needed to prevent us from installing evil descriptors such as 4637868fe1SAndy Lutomirski * call gates. On native, we could merge the ldt_struct and LDT 4737868fe1SAndy Lutomirski * allocations, but it's not worth trying to optimize. 4837868fe1SAndy Lutomirski */ 4937868fe1SAndy Lutomirski struct desc_struct *entries; 50bbf79d21SBorislav Petkov unsigned int nr_entries; 5137868fe1SAndy Lutomirski }; 5237868fe1SAndy Lutomirski 53a5b9e5a2SAndy Lutomirski /* 54a5b9e5a2SAndy Lutomirski * Used for LDT copy/destruction. 55a5b9e5a2SAndy Lutomirski */ 5639a0526fSDave Hansen int init_new_context_ldt(struct task_struct *tsk, struct mm_struct *mm); 5739a0526fSDave Hansen void destroy_context_ldt(struct mm_struct *mm); 58a5b9e5a2SAndy Lutomirski #else /* CONFIG_MODIFY_LDT_SYSCALL */ 5939a0526fSDave Hansen static inline int init_new_context_ldt(struct task_struct *tsk, 60a5b9e5a2SAndy Lutomirski struct mm_struct *mm) 61a5b9e5a2SAndy Lutomirski { 62a5b9e5a2SAndy Lutomirski return 0; 63a5b9e5a2SAndy Lutomirski } 6439a0526fSDave Hansen static inline void destroy_context_ldt(struct mm_struct *mm) {} 65a5b9e5a2SAndy Lutomirski #endif 66a5b9e5a2SAndy Lutomirski 6737868fe1SAndy Lutomirski static inline void load_mm_ldt(struct mm_struct *mm) 6837868fe1SAndy Lutomirski { 69a5b9e5a2SAndy Lutomirski #ifdef CONFIG_MODIFY_LDT_SYSCALL 7037868fe1SAndy Lutomirski struct ldt_struct *ldt; 7137868fe1SAndy Lutomirski 7237868fe1SAndy Lutomirski /* lockless_dereference synchronizes with smp_store_release */ 7337868fe1SAndy Lutomirski ldt = lockless_dereference(mm->context.ldt); 7437868fe1SAndy Lutomirski 7537868fe1SAndy Lutomirski /* 7637868fe1SAndy Lutomirski * Any change to mm->context.ldt is followed by an IPI to all 7737868fe1SAndy Lutomirski * CPUs with the mm active. The LDT will not be freed until 7837868fe1SAndy Lutomirski * after the IPI is handled by all such CPUs. This means that, 7937868fe1SAndy Lutomirski * if the ldt_struct changes before we return, the values we see 8037868fe1SAndy Lutomirski * will be safe, and the new values will be loaded before we run 8137868fe1SAndy Lutomirski * any user code. 8237868fe1SAndy Lutomirski * 8337868fe1SAndy Lutomirski * NB: don't try to convert this to use RCU without extreme care. 8437868fe1SAndy Lutomirski * We would still need IRQs off, because we don't want to change 8537868fe1SAndy Lutomirski * the local LDT after an IPI loaded a newer value than the one 8637868fe1SAndy Lutomirski * that we can see. 8737868fe1SAndy Lutomirski */ 8837868fe1SAndy Lutomirski 8937868fe1SAndy Lutomirski if (unlikely(ldt)) 90bbf79d21SBorislav Petkov set_ldt(ldt->entries, ldt->nr_entries); 9137868fe1SAndy Lutomirski else 9237868fe1SAndy Lutomirski clear_LDT(); 93a5b9e5a2SAndy Lutomirski #else 94a5b9e5a2SAndy Lutomirski clear_LDT(); 95a5b9e5a2SAndy Lutomirski #endif 9673534258SAndy Lutomirski } 9773534258SAndy Lutomirski 9873534258SAndy Lutomirski static inline void switch_ldt(struct mm_struct *prev, struct mm_struct *next) 9973534258SAndy Lutomirski { 10073534258SAndy Lutomirski #ifdef CONFIG_MODIFY_LDT_SYSCALL 10173534258SAndy Lutomirski /* 10273534258SAndy Lutomirski * Load the LDT if either the old or new mm had an LDT. 10373534258SAndy Lutomirski * 10473534258SAndy Lutomirski * An mm will never go from having an LDT to not having an LDT. Two 10573534258SAndy Lutomirski * mms never share an LDT, so we don't gain anything by checking to 10673534258SAndy Lutomirski * see whether the LDT changed. There's also no guarantee that 10773534258SAndy Lutomirski * prev->context.ldt actually matches LDTR, but, if LDTR is non-NULL, 10873534258SAndy Lutomirski * then prev->context.ldt will also be non-NULL. 10973534258SAndy Lutomirski * 11073534258SAndy Lutomirski * If we really cared, we could optimize the case where prev == next 11173534258SAndy Lutomirski * and we're exiting lazy mode. Most of the time, if this happens, 11273534258SAndy Lutomirski * we don't actually need to reload LDTR, but modify_ldt() is mostly 11373534258SAndy Lutomirski * used by legacy code and emulators where we don't need this level of 11473534258SAndy Lutomirski * performance. 11573534258SAndy Lutomirski * 11673534258SAndy Lutomirski * This uses | instead of || because it generates better code. 11773534258SAndy Lutomirski */ 11873534258SAndy Lutomirski if (unlikely((unsigned long)prev->context.ldt | 11973534258SAndy Lutomirski (unsigned long)next->context.ldt)) 12073534258SAndy Lutomirski load_mm_ldt(next); 12173534258SAndy Lutomirski #endif 12237868fe1SAndy Lutomirski 12337868fe1SAndy Lutomirski DEBUG_LOCKS_WARN_ON(preemptible()); 12437868fe1SAndy Lutomirski } 12537868fe1SAndy Lutomirski 1266826c8ffSBrian Gerst static inline void enter_lazy_tlb(struct mm_struct *mm, struct task_struct *tsk) 1276826c8ffSBrian Gerst { 128c6ae41e7SAlex Shi if (this_cpu_read(cpu_tlbstate.state) == TLBSTATE_OK) 129c6ae41e7SAlex Shi this_cpu_write(cpu_tlbstate.state, TLBSTATE_LAZY); 1306826c8ffSBrian Gerst } 1316826c8ffSBrian Gerst 13239a0526fSDave Hansen static inline int init_new_context(struct task_struct *tsk, 13339a0526fSDave Hansen struct mm_struct *mm) 13439a0526fSDave Hansen { 135e8c24d3aSDave Hansen #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS 136e8c24d3aSDave Hansen if (cpu_feature_enabled(X86_FEATURE_OSPKE)) { 137e8c24d3aSDave Hansen /* pkey 0 is the default and always allocated */ 138e8c24d3aSDave Hansen mm->context.pkey_allocation_map = 0x1; 139e8c24d3aSDave Hansen /* -1 means unallocated or invalid */ 140e8c24d3aSDave Hansen mm->context.execute_only_pkey = -1; 141e8c24d3aSDave Hansen } 142e8c24d3aSDave Hansen #endif 143ccd5b323SEric Biggers return init_new_context_ldt(tsk, mm); 14439a0526fSDave Hansen } 14539a0526fSDave Hansen static inline void destroy_context(struct mm_struct *mm) 14639a0526fSDave Hansen { 14739a0526fSDave Hansen destroy_context_ldt(mm); 14839a0526fSDave Hansen } 14939a0526fSDave Hansen 15069c0319aSAndy Lutomirski extern void switch_mm(struct mm_struct *prev, struct mm_struct *next, 15169c0319aSAndy Lutomirski struct task_struct *tsk); 1526826c8ffSBrian Gerst 153078194f8SAndy Lutomirski extern void switch_mm_irqs_off(struct mm_struct *prev, struct mm_struct *next, 154078194f8SAndy Lutomirski struct task_struct *tsk); 155078194f8SAndy Lutomirski #define switch_mm_irqs_off switch_mm_irqs_off 156bb898558SAl Viro 157bb898558SAl Viro #define activate_mm(prev, next) \ 158bb898558SAl Viro do { \ 159bb898558SAl Viro paravirt_activate_mm((prev), (next)); \ 160bb898558SAl Viro switch_mm((prev), (next), NULL); \ 161bb898558SAl Viro } while (0); 162bb898558SAl Viro 1636826c8ffSBrian Gerst #ifdef CONFIG_X86_32 1646826c8ffSBrian Gerst #define deactivate_mm(tsk, mm) \ 1656826c8ffSBrian Gerst do { \ 166ccbeed3aSTejun Heo lazy_load_gs(0); \ 1676826c8ffSBrian Gerst } while (0) 1686826c8ffSBrian Gerst #else 1696826c8ffSBrian Gerst #define deactivate_mm(tsk, mm) \ 1706826c8ffSBrian Gerst do { \ 1716826c8ffSBrian Gerst load_gs_index(0); \ 1726826c8ffSBrian Gerst loadsegment(fs, 0); \ 1736826c8ffSBrian Gerst } while (0) 1746826c8ffSBrian Gerst #endif 175bb898558SAl Viro 176a1ea1c03SDave Hansen static inline void arch_dup_mmap(struct mm_struct *oldmm, 177a1ea1c03SDave Hansen struct mm_struct *mm) 178a1ea1c03SDave Hansen { 179a1ea1c03SDave Hansen paravirt_arch_dup_mmap(oldmm, mm); 180a1ea1c03SDave Hansen } 181a1ea1c03SDave Hansen 182a1ea1c03SDave Hansen static inline void arch_exit_mmap(struct mm_struct *mm) 183a1ea1c03SDave Hansen { 184a1ea1c03SDave Hansen paravirt_arch_exit_mmap(mm); 185a1ea1c03SDave Hansen } 186a1ea1c03SDave Hansen 187b0e9b09bSDave Hansen #ifdef CONFIG_X86_64 188b0e9b09bSDave Hansen static inline bool is_64bit_mm(struct mm_struct *mm) 189b0e9b09bSDave Hansen { 19097f2645fSMasahiro Yamada return !IS_ENABLED(CONFIG_IA32_EMULATION) || 191b0e9b09bSDave Hansen !(mm->context.ia32_compat == TIF_IA32); 192b0e9b09bSDave Hansen } 193b0e9b09bSDave Hansen #else 194b0e9b09bSDave Hansen static inline bool is_64bit_mm(struct mm_struct *mm) 195b0e9b09bSDave Hansen { 196b0e9b09bSDave Hansen return false; 197b0e9b09bSDave Hansen } 198b0e9b09bSDave Hansen #endif 199b0e9b09bSDave Hansen 200fe3d197fSDave Hansen static inline void arch_bprm_mm_init(struct mm_struct *mm, 201fe3d197fSDave Hansen struct vm_area_struct *vma) 202fe3d197fSDave Hansen { 203fe3d197fSDave Hansen mpx_mm_init(mm); 204fe3d197fSDave Hansen } 205fe3d197fSDave Hansen 2061de4fa14SDave Hansen static inline void arch_unmap(struct mm_struct *mm, struct vm_area_struct *vma, 2071de4fa14SDave Hansen unsigned long start, unsigned long end) 2081de4fa14SDave Hansen { 209c922228eSDave Hansen /* 210c922228eSDave Hansen * mpx_notify_unmap() goes and reads a rarely-hot 211c922228eSDave Hansen * cacheline in the mm_struct. That can be expensive 212c922228eSDave Hansen * enough to be seen in profiles. 213c922228eSDave Hansen * 214c922228eSDave Hansen * The mpx_notify_unmap() call and its contents have been 215c922228eSDave Hansen * observed to affect munmap() performance on hardware 216c922228eSDave Hansen * where MPX is not present. 217c922228eSDave Hansen * 218c922228eSDave Hansen * The unlikely() optimizes for the fast case: no MPX 219c922228eSDave Hansen * in the CPU, or no MPX use in the process. Even if 220c922228eSDave Hansen * we get this wrong (in the unlikely event that MPX 221c922228eSDave Hansen * is widely enabled on some system) the overhead of 222c922228eSDave Hansen * MPX itself (reading bounds tables) is expected to 223c922228eSDave Hansen * overwhelm the overhead of getting this unlikely() 224c922228eSDave Hansen * consistently wrong. 225c922228eSDave Hansen */ 226c922228eSDave Hansen if (unlikely(cpu_feature_enabled(X86_FEATURE_MPX))) 2271de4fa14SDave Hansen mpx_notify_unmap(mm, vma, start, end); 2281de4fa14SDave Hansen } 2291de4fa14SDave Hansen 2307d06d9c9SDave Hansen #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS 2318f62c883SDave Hansen static inline int vma_pkey(struct vm_area_struct *vma) 2328f62c883SDave Hansen { 2338f62c883SDave Hansen unsigned long vma_pkey_mask = VM_PKEY_BIT0 | VM_PKEY_BIT1 | 2348f62c883SDave Hansen VM_PKEY_BIT2 | VM_PKEY_BIT3; 2357d06d9c9SDave Hansen 2367d06d9c9SDave Hansen return (vma->vm_flags & vma_pkey_mask) >> VM_PKEY_SHIFT; 2378f62c883SDave Hansen } 2387d06d9c9SDave Hansen #else 2397d06d9c9SDave Hansen static inline int vma_pkey(struct vm_area_struct *vma) 2407d06d9c9SDave Hansen { 2417d06d9c9SDave Hansen return 0; 2427d06d9c9SDave Hansen } 2437d06d9c9SDave Hansen #endif 2448f62c883SDave Hansen 24533a709b2SDave Hansen /* 24633a709b2SDave Hansen * We only want to enforce protection keys on the current process 24733a709b2SDave Hansen * because we effectively have no access to PKRU for other 24833a709b2SDave Hansen * processes or any way to tell *which * PKRU in a threaded 24933a709b2SDave Hansen * process we could use. 25033a709b2SDave Hansen * 25133a709b2SDave Hansen * So do not enforce things if the VMA is not from the current 25233a709b2SDave Hansen * mm, or if we are in a kernel thread. 25333a709b2SDave Hansen */ 25433a709b2SDave Hansen static inline bool vma_is_foreign(struct vm_area_struct *vma) 25533a709b2SDave Hansen { 25633a709b2SDave Hansen if (!current->mm) 25733a709b2SDave Hansen return true; 25833a709b2SDave Hansen /* 25933a709b2SDave Hansen * Should PKRU be enforced on the access to this VMA? If 26033a709b2SDave Hansen * the VMA is from another process, then PKRU has no 26133a709b2SDave Hansen * relevance and should not be enforced. 26233a709b2SDave Hansen */ 26333a709b2SDave Hansen if (current->mm != vma->vm_mm) 26433a709b2SDave Hansen return true; 26533a709b2SDave Hansen 26633a709b2SDave Hansen return false; 26733a709b2SDave Hansen } 26833a709b2SDave Hansen 2691b2ee126SDave Hansen static inline bool arch_vma_access_permitted(struct vm_area_struct *vma, 270d61172b4SDave Hansen bool write, bool execute, bool foreign) 27133a709b2SDave Hansen { 272d61172b4SDave Hansen /* pkeys never affect instruction fetches */ 273d61172b4SDave Hansen if (execute) 274d61172b4SDave Hansen return true; 27533a709b2SDave Hansen /* allow access if the VMA is not one from this process */ 2761b2ee126SDave Hansen if (foreign || vma_is_foreign(vma)) 27733a709b2SDave Hansen return true; 27833a709b2SDave Hansen return __pkru_allows_pkey(vma_pkey(vma), write); 27933a709b2SDave Hansen } 28033a709b2SDave Hansen 281d6e41f11SAndy Lutomirski 282d6e41f11SAndy Lutomirski /* 283d6e41f11SAndy Lutomirski * This can be used from process context to figure out what the value of 2846c690ee1SAndy Lutomirski * CR3 is without needing to do a (slow) __read_cr3(). 285d6e41f11SAndy Lutomirski * 286d6e41f11SAndy Lutomirski * It's intended to be used for code like KVM that sneakily changes CR3 287d6e41f11SAndy Lutomirski * and needs to restore it. It needs to be used very carefully. 288d6e41f11SAndy Lutomirski */ 289d6e41f11SAndy Lutomirski static inline unsigned long __get_current_cr3_fast(void) 290d6e41f11SAndy Lutomirski { 291d6e41f11SAndy Lutomirski unsigned long cr3 = __pa(this_cpu_read(cpu_tlbstate.loaded_mm)->pgd); 292d6e41f11SAndy Lutomirski 293d6e41f11SAndy Lutomirski /* For now, be very restrictive about when this can be called. */ 2944c07f904SRoman Kagan VM_WARN_ON(in_nmi() || preemptible()); 295d6e41f11SAndy Lutomirski 2966c690ee1SAndy Lutomirski VM_BUG_ON(cr3 != __read_cr3()); 297d6e41f11SAndy Lutomirski return cr3; 298d6e41f11SAndy Lutomirski } 299d6e41f11SAndy Lutomirski 3001965aae3SH. Peter Anvin #endif /* _ASM_X86_MMU_CONTEXT_H */ 301