diff options
Diffstat (limited to 'arch/x86/include/asm/pgtable-3level.h')
| -rw-r--r-- | arch/x86/include/asm/pgtable-3level.h | 97 |
1 files changed, 86 insertions, 11 deletions
diff --git a/arch/x86/include/asm/pgtable-3level.h b/arch/x86/include/asm/pgtable-3level.h index 177b0165ea0..81bb91b49a8 100644 --- a/arch/x86/include/asm/pgtable-3level.h +++ b/arch/x86/include/asm/pgtable-3level.h @@ -9,13 +9,13 @@ */ #define pte_ERROR(e) \ - printk("%s:%d: bad pte %p(%08lx%08lx).\n", \ + pr_err("%s:%d: bad pte %p(%08lx%08lx)\n", \ __FILE__, __LINE__, &(e), (e).pte_high, (e).pte_low) #define pmd_ERROR(e) \ - printk("%s:%d: bad pmd %p(%016Lx).\n", \ + pr_err("%s:%d: bad pmd %p(%016Lx)\n", \ __FILE__, __LINE__, &(e), pmd_val(e)) #define pgd_ERROR(e) \ - printk("%s:%d: bad pgd %p(%016Lx).\n", \ + pr_err("%s:%d: bad pgd %p(%016Lx)\n", \ __FILE__, __LINE__, &(e), pgd_val(e)) /* Rules for using set_pte: the pte being assigned *must* be @@ -31,6 +31,60 @@ static inline void native_set_pte(pte_t *ptep, pte_t pte) ptep->pte_low = pte.pte_low; } +#define pmd_read_atomic pmd_read_atomic +/* + * pte_offset_map_lock on 32bit PAE kernels was reading the pmd_t with + * a "*pmdp" dereference done by gcc. Problem is, in certain places + * where pte_offset_map_lock is called, concurrent page faults are + * allowed, if the mmap_sem is hold for reading. An example is mincore + * vs page faults vs MADV_DONTNEED. On the page fault side + * pmd_populate rightfully does a set_64bit, but if we're reading the + * pmd_t with a "*pmdp" on the mincore side, a SMP race can happen + * because gcc will not read the 64bit of the pmd atomically. To fix + * this all places running pmd_offset_map_lock() while holding the + * mmap_sem in read mode, shall read the pmdp pointer using this + * function to know if the pmd is null nor not, and in turn to know if + * they can run pmd_offset_map_lock or pmd_trans_huge or other pmd + * operations. + * + * Without THP if the mmap_sem is hold for reading, the pmd can only + * transition from null to not null while pmd_read_atomic runs. So + * we can always return atomic pmd values with this function. + * + * With THP if the mmap_sem is hold for reading, the pmd can become + * trans_huge or none or point to a pte (and in turn become "stable") + * at any time under pmd_read_atomic. We could read it really + * atomically here with a atomic64_read for the THP enabled case (and + * it would be a whole lot simpler), but to avoid using cmpxchg8b we + * only return an atomic pmdval if the low part of the pmdval is later + * found stable (i.e. pointing to a pte). And we're returning a none + * pmdval if the low part of the pmd is none. In some cases the high + * and low part of the pmdval returned may not be consistent if THP is + * enabled (the low part may point to previously mapped hugepage, + * while the high part may point to a more recently mapped hugepage), + * but pmd_none_or_trans_huge_or_clear_bad() only needs the low part + * of the pmd to be read atomically to decide if the pmd is unstable + * or not, with the only exception of when the low part of the pmd is + * zero in which case we return a none pmd. + */ +static inline pmd_t pmd_read_atomic(pmd_t *pmdp) +{ + pmdval_t ret; + u32 *tmp = (u32 *)pmdp; + + ret = (pmdval_t) (*tmp); + if (ret) { + /* + * If the low part is null, we must not read the high part + * or we can end up with a partial pmd. + */ + smp_rmb(); + ret |= ((pmdval_t)*(tmp + 1)) << 32; + } + + return (pmd_t) { ret }; +} + static inline void native_set_pte_atomic(pte_t *ptep, pte_t pte) { set_64bit((unsigned long long *)(ptep), native_pte_val(pte)); @@ -69,8 +123,6 @@ static inline void native_pmd_clear(pmd_t *pmd) static inline void pud_clear(pud_t *pudp) { - unsigned long pgd; - set_pud(pudp, __pud(0)); /* @@ -79,13 +131,10 @@ static inline void pud_clear(pud_t *pudp) * section 8.1: in PAE mode we explicitly have to flush the * TLB via cr3 if the top-level pgd is changed... * - * Make sure the pud entry we're updating is within the - * current pgd to avoid unnecessary TLB flushes. + * Currently all places where pud_clear() is called either have + * flush_tlb_mm() followed or don't need TLB flush (x86_64 code or + * pud_clear_bad()), so we don't need TLB flush here. */ - pgd = read_cr3(); - if (__pa(pudp) >= pgd && __pa(pudp) < - (pgd + sizeof(pgd_t)*PTRS_PER_PGD)) - write_cr3(pgd); } #ifdef CONFIG_SMP @@ -104,9 +153,35 @@ static inline pte_t native_ptep_get_and_clear(pte_t *ptep) #define native_ptep_get_and_clear(xp) native_local_ptep_get_and_clear(xp) #endif +#ifdef CONFIG_SMP +union split_pmd { + struct { + u32 pmd_low; + u32 pmd_high; + }; + pmd_t pmd; +}; +static inline pmd_t native_pmdp_get_and_clear(pmd_t *pmdp) +{ + union split_pmd res, *orig = (union split_pmd *)pmdp; + + /* xchg acts as a barrier before setting of the high bits */ + res.pmd_low = xchg(&orig->pmd_low, 0); + res.pmd_high = orig->pmd_high; + orig->pmd_high = 0; + + return res.pmd; +} +#else +#define native_pmdp_get_and_clear(xp) native_local_pmdp_get_and_clear(xp) +#endif + /* * Bits 0, 6 and 7 are taken in the low part of the pte, * put the 32 bits of offset into the high part. + * + * For soft-dirty tracking 11 bit is taken from + * the low part of pte as well. */ #define pte_to_pgoff(pte) ((pte).pte_high) #define pgoff_to_pte(off) \ |