From 01858d1b0b406307626bbc01238391b06aae2c20 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Mon, 6 Aug 2007 23:57:57 -0700 Subject: [PATCH] [XTENSA] Add support for executable/non-executable feature in the mmu Newer processor versions starting with Xtensa6/LX2 support an 'executable' bit for memory pages. This bit replaces the 'valid' bit, so it must be always set to one for older processor versions. To mark a page invalid, we now set the cache-attributes to b11, which is backward compatible. Signed-off-by: Chris Zankel --- arch/xtensa/kernel/entry.S | 42 ++++--- include/asm-xtensa/pgtable.h | 213 +++++++++++++++++------------------ 2 files changed, 129 insertions(+), 126 deletions(-) diff --git a/arch/xtensa/kernel/entry.S b/arch/xtensa/kernel/entry.S index 8dc7a2c26ff9..65741e3368e7 100644 --- a/arch/xtensa/kernel/entry.S +++ b/arch/xtensa/kernel/entry.S @@ -1572,10 +1572,12 @@ ENTRY(fast_second_level_miss) l32i a0, a1, TASK_MM # tsk->mm beqz a0, 9f -8: rsr a1, EXCVADDR # fault address - _PGD_OFFSET(a0, a1, a1) + + /* We deliberately destroy a3 that holds the exception table. */ + +8: rsr a3, EXCVADDR # fault address + _PGD_OFFSET(a0, a3, a1) l32i a0, a0, 0 # read pmdval - //beqi a0, _PAGE_USER, 2f beqz a0, 2f /* Read ptevaddr and convert to top of page-table page. @@ -1596,20 +1598,34 @@ ENTRY(fast_second_level_miss) extui a1, a0, 0, PAGE_SHIFT # ... & PAGE_MASK xor a0, a0, a1 - - movi a1, PAGE_DIRECTORY + movi a1, _PAGE_DIRECTORY or a0, a0, a1 # ... | PAGE_DIRECTORY + /* + * We utilize all three wired-ways (7-9( to hold pmd translations. + * Memory regions are mapped to the DTLBs according to bits 28 and 29. + * This allows to map the three most common regions to three different + * DTLBs: + * 0,1 -> way 7 program (0040.0000) and virtual (c000.0000) + * 2 -> way 8 shared libaries (2000.0000) + * 3 -> way 0 stack (3000.0000) + */ + + extui a3, a3, 28, 2 # addr. bit 28 and 29 0,1,2,3 rsr a1, PTEVADDR + addx2 a3, a3, a3 # -> 0,3,6,9 srli a1, a1, PAGE_SHIFT + extui a3, a3, 2, 2 # -> 0,0,1,2 slli a1, a1, PAGE_SHIFT # ptevaddr & PAGE_MASK - addi a1, a1, DTLB_WAY_PGD # ... + way_number + addi a3, a3, DTLB_WAY_PGD + add a1, a1, a3 # ... + way_number - wdtlb a0, a1 +3: wdtlb a0, a1 dsync /* Exit critical section. */ +4: movi a3, exc_table # restore a3 movi a0, 0 s32i a0, a3, EXC_TABLE_FIXUP @@ -1638,6 +1654,7 @@ ENTRY(fast_second_level_miss) 2: /* Invalid PGD, default exception handling */ + movi a3, exc_table rsr a1, DEPC xsr a3, EXCSAVE_1 s32i a1, a2, PT_AREG2 @@ -1682,15 +1699,15 @@ ENTRY(fast_store_prohibited) 8: rsr a1, EXCVADDR # fault address _PGD_OFFSET(a0, a1, a4) l32i a0, a0, 0 - //beqi a0, _PAGE_USER, 2f # FIXME use _PAGE_INVALID beqz a0, 2f + /* Note that we assume _PAGE_WRITABLE_BIT is only set if pte is valid.*/ + _PTE_OFFSET(a0, a1, a4) l32i a4, a0, 0 # read pteval - movi a1, _PAGE_VALID | _PAGE_RW - bnall a4, a1, 2f + bbci.l a4, _PAGE_WRITABLE_BIT, 2f - movi a1, _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_WRENABLE + movi a1, _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_HW_WRITE or a4, a4, a1 rsr a1, EXCVADDR s32i a4, a0, 0 @@ -1700,10 +1717,7 @@ ENTRY(fast_store_prohibited) dhwb a0, 0 #endif pdtlb a0, a1 - beqz a0, 1f - idtlb a0 // FIXME do we need this? wdtlb a4, a0 -1: /* Exit critical section. */ diff --git a/include/asm-xtensa/pgtable.h b/include/asm-xtensa/pgtable.h index 06850f3b26a7..667a6c46b5a1 100644 --- a/include/asm-xtensa/pgtable.h +++ b/include/asm-xtensa/pgtable.h @@ -2,10 +2,10 @@ * linux/include/asm-xtensa/pgtable.h * * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License version2 as + * it under the terms of the GNU General Public License version 2 as * published by the Free Software Foundation. * - * Copyright (C) 2001 - 2005 Tensilica Inc. + * Copyright (C) 2001 - 2007 Tensilica Inc. */ #ifndef _XTENSA_PGTABLE_H @@ -23,7 +23,7 @@ /* * The Xtensa architecture port of Linux has a two-level page table system, - * i.e. the logical three-level Linux page table layout are folded. + * i.e. the logical three-level Linux page table layout is folded. * Each task has the following memory page tables: * * PGD table (page directory), ie. 3rd-level page table: @@ -43,6 +43,7 @@ * * The individual pages are 4 kB big with special pages for the empty_zero_page. */ + #define PGDIR_SHIFT 22 #define PGDIR_SIZE (1UL << PGDIR_SHIFT) #define PGDIR_MASK (~(PGDIR_SIZE-1)) @@ -53,12 +54,10 @@ */ #define PTRS_PER_PTE 1024 #define PTRS_PER_PTE_SHIFT 10 -#define PTRS_PER_PMD 1 #define PTRS_PER_PGD 1024 #define PGD_ORDER 0 -#define PMD_ORDER 0 #define USER_PTRS_PER_PGD (TASK_SIZE/PGDIR_SIZE) -#define FIRST_USER_ADDRESS 0 +#define FIRST_USER_ADDRESS 0 #define FIRST_USER_PGD_NR (FIRST_USER_ADDRESS >> PGDIR_SHIFT) /* virtual memory area. We keep a distance to other memory regions to be @@ -86,47 +85,54 @@ * See further below for PTE layout for swapped-out pages. */ -#define _PAGE_VALID (1<<0) /* hardware: page is accessible */ -#define _PAGE_WRENABLE (1<<1) /* hardware: page is writable */ +#define _PAGE_HW_EXEC (1<<0) /* hardware: page is executable */ +#define _PAGE_HW_WRITE (1<<1) /* hardware: page is writable */ + +#define _PAGE_FILE (1<<1) /* non-linear mapping, if !present */ +#define _PAGE_PROTNONE (3<<0) /* special case for VM_PROT_NONE */ /* None of these cache modes include MP coherency: */ -#define _PAGE_NO_CACHE (0<<2) /* bypass, non-speculative */ -#if XCHAL_DCACHE_IS_WRITEBACK -# define _PAGE_WRITEBACK (1<<2) /* write back */ -# define _PAGE_WRITETHRU (2<<2) /* write through */ -#else -# define _PAGE_WRITEBACK (1<<2) /* assume write through */ -# define _PAGE_WRITETHRU (1<<2) -#endif -#define _PAGE_NOALLOC (3<<2) /* don't allocate cache,if not cached */ -#define _CACHE_MASK (3<<2) +#define _PAGE_CA_BYPASS (0<<2) /* bypass, non-speculative */ +#define _PAGE_CA_WB (1<<2) /* write-back */ +#define _PAGE_CA_WT (2<<2) /* write-through */ +#define _PAGE_CA_MASK (3<<2) +#define _PAGE_INVALID (3<<2) #define _PAGE_USER (1<<4) /* user access (ring=1) */ -#define _PAGE_KERNEL (0<<4) /* kernel access (ring=0) */ /* Software */ -#define _PAGE_RW (1<<6) /* software: page writable */ +#define _PAGE_WRITABLE_BIT 6 +#define _PAGE_WRITABLE (1<<6) /* software: page writable */ #define _PAGE_DIRTY (1<<7) /* software: page dirty */ #define _PAGE_ACCESSED (1<<8) /* software: page accessed (read) */ -#define _PAGE_FILE (1<<9) /* nonlinear file mapping*/ -#define _PAGE_CHG_MASK (PAGE_MASK | _PAGE_ACCESSED | _CACHE_MASK | _PAGE_DIRTY) -#define _PAGE_PRESENT ( _PAGE_VALID | _PAGE_WRITEBACK | _PAGE_ACCESSED) +/* On older HW revisions, we always have to set bit 0 */ +#if XCHAL_HW_VERSION_MAJOR < 2000 +# define _PAGE_VALID (1<<0) +#else +# define _PAGE_VALID 0 +#endif + +#define _PAGE_CHG_MASK (PAGE_MASK | _PAGE_ACCESSED | _PAGE_DIRTY) +#define _PAGE_PRESENT (_PAGE_VALID | _PAGE_CA_WB | _PAGE_ACCESSED) #ifdef CONFIG_MMU -# define PAGE_NONE __pgprot(_PAGE_PRESENT) -# define PAGE_SHARED __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_RW) -# define PAGE_COPY __pgprot(_PAGE_PRESENT | _PAGE_USER) -# define PAGE_READONLY __pgprot(_PAGE_PRESENT | _PAGE_USER) -# define PAGE_KERNEL __pgprot(_PAGE_PRESENT | _PAGE_KERNEL | _PAGE_WRENABLE) -# define PAGE_INVALID __pgprot(_PAGE_USER) - -# if (DCACHE_WAY_SIZE > PAGE_SIZE) -# define PAGE_DIRECTORY __pgprot(_PAGE_VALID | _PAGE_ACCESSED | _PAGE_KERNEL) -# else -# define PAGE_DIRECTORY __pgprot(_PAGE_PRESENT | _PAGE_KERNEL) -# endif +#define PAGE_NONE __pgprot(_PAGE_INVALID | _PAGE_USER | _PAGE_PROTNONE) +#define PAGE_COPY __pgprot(_PAGE_PRESENT | _PAGE_USER) +#define PAGE_COPY_EXEC __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_HW_EXEC) +#define PAGE_READONLY __pgprot(_PAGE_PRESENT | _PAGE_USER) +#define PAGE_READONLY_EXEC __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_HW_EXEC) +#define PAGE_SHARED __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_WRITABLE) +#define PAGE_SHARED_EXEC \ + __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_WRITABLE | _PAGE_HW_EXEC) +#define PAGE_KERNEL __pgprot(_PAGE_PRESENT) + +#if (DCACHE_WAY_SIZE > PAGE_SIZE) +# define _PAGE_DIRECTORY (_PAGE_VALID | _PAGE_ACCESSED | _PAGE_HW_WRITE) +#else +# define _PAGE_DIRECTORY (_PAGE_VALID|_PAGE_ACCESSED|_PAGE_HW_WRITE|_PAGE_CA_WB) +#endif #else /* no mmu */ @@ -145,23 +151,23 @@ * What follows is the closest we can get by reasonable means.. * See linux/mm/mmap.c for protection_map[] array that uses these definitions. */ -#define __P000 PAGE_NONE /* private --- */ -#define __P001 PAGE_READONLY /* private --r */ -#define __P010 PAGE_COPY /* private -w- */ -#define __P011 PAGE_COPY /* private -wr */ -#define __P100 PAGE_READONLY /* private x-- */ -#define __P101 PAGE_READONLY /* private x-r */ -#define __P110 PAGE_COPY /* private xw- */ -#define __P111 PAGE_COPY /* private xwr */ - -#define __S000 PAGE_NONE /* shared --- */ -#define __S001 PAGE_READONLY /* shared --r */ -#define __S010 PAGE_SHARED /* shared -w- */ -#define __S011 PAGE_SHARED /* shared -wr */ -#define __S100 PAGE_READONLY /* shared x-- */ -#define __S101 PAGE_READONLY /* shared x-r */ -#define __S110 PAGE_SHARED /* shared xw- */ -#define __S111 PAGE_SHARED /* shared xwr */ +#define __P000 PAGE_NONE /* private --- */ +#define __P001 PAGE_READONLY /* private --r */ +#define __P010 PAGE_COPY /* private -w- */ +#define __P011 PAGE_COPY /* private -wr */ +#define __P100 PAGE_READONLY_EXEC /* private x-- */ +#define __P101 PAGE_READONLY_EXEC /* private x-r */ +#define __P110 PAGE_COPY_EXEC /* private xw- */ +#define __P111 PAGE_COPY_EXEC /* private xwr */ + +#define __S000 PAGE_NONE /* shared --- */ +#define __S001 PAGE_READONLY /* shared --r */ +#define __S010 PAGE_SHARED /* shared -w- */ +#define __S011 PAGE_SHARED /* shared -wr */ +#define __S100 PAGE_READONLY_EXEC /* shared x-- */ +#define __S101 PAGE_READONLY_EXEC /* shared x-r */ +#define __S110 PAGE_SHARED_EXEC /* shared xw- */ +#define __S111 PAGE_SHARED_EXEC /* shared xwr */ #ifndef __ASSEMBLY__ @@ -183,35 +189,42 @@ extern pgd_t swapper_pg_dir[PAGE_SIZE/sizeof(pgd_t)]; #define pmd_page(pmd) virt_to_page(pmd_val(pmd)) /* - * The following only work if pte_present() is true. + * pte status. */ -#define pte_none(pte) (!(pte_val(pte) ^ _PAGE_USER)) -#define pte_present(pte) (pte_val(pte) & _PAGE_VALID) +#define pte_none(pte) (pte_val(pte) == _PAGE_INVALID) +#define pte_present(pte) \ + (((pte_val(pte) & _PAGE_CA_MASK) != _PAGE_INVALID) \ + || ((pte_val(pte) & _PAGE_PROTNONE) == _PAGE_PROTNONE)) #define pte_clear(mm,addr,ptep) \ - do { update_pte(ptep, __pte(_PAGE_USER)); } while(0) + do { update_pte(ptep, __pte(_PAGE_INVALID)); } while(0) #define pmd_none(pmd) (!pmd_val(pmd)) #define pmd_present(pmd) (pmd_val(pmd) & PAGE_MASK) -#define pmd_clear(pmdp) do { set_pmd(pmdp, __pmd(0)); } while (0) #define pmd_bad(pmd) (pmd_val(pmd) & ~PAGE_MASK) +#define pmd_clear(pmdp) do { set_pmd(pmdp, __pmd(0)); } while (0) -/* Note: We use the _PAGE_USER bit to indicate write-protect kernel memory */ - -static inline int pte_write(pte_t pte) { return pte_val(pte) & _PAGE_RW; } +static inline int pte_write(pte_t pte) { return pte_val(pte) & _PAGE_WRITABLE; } static inline int pte_dirty(pte_t pte) { return pte_val(pte) & _PAGE_DIRTY; } static inline int pte_young(pte_t pte) { return pte_val(pte) & _PAGE_ACCESSED; } static inline int pte_file(pte_t pte) { return pte_val(pte) & _PAGE_FILE; } -static inline pte_t pte_wrprotect(pte_t pte) { pte_val(pte) &= ~(_PAGE_RW | _PAGE_WRENABLE); return pte; } -static inline pte_t pte_mkclean(pte_t pte) { pte_val(pte) &= ~_PAGE_DIRTY; return pte; } -static inline pte_t pte_mkold(pte_t pte) { pte_val(pte) &= ~_PAGE_ACCESSED; return pte; } -static inline pte_t pte_mkdirty(pte_t pte) { pte_val(pte) |= _PAGE_DIRTY; return pte; } -static inline pte_t pte_mkyoung(pte_t pte) { pte_val(pte) |= _PAGE_ACCESSED; return pte; } -static inline pte_t pte_mkwrite(pte_t pte) { pte_val(pte) |= _PAGE_RW; return pte; } +static inline pte_t pte_wrprotect(pte_t pte) + { pte_val(pte) &= ~(_PAGE_WRITABLE | _PAGE_HW_WRITE); return pte; } +static inline pte_t pte_mkclean(pte_t pte) + { pte_val(pte) &= ~(_PAGE_DIRTY | _PAGE_HW_WRITE); return pte; } +static inline pte_t pte_mkold(pte_t pte) + { pte_val(pte) &= ~_PAGE_ACCESSED; return pte; } +static inline pte_t pte_mkdirty(pte_t pte) + { pte_val(pte) |= _PAGE_DIRTY; return pte; } +static inline pte_t pte_mkyoung(pte_t pte) + { pte_val(pte) |= _PAGE_ACCESSED; return pte; } +static inline pte_t pte_mkwrite(pte_t pte) + { pte_val(pte) |= _PAGE_WRITABLE; return pte; } /* * Conversion functions: convert a page and protection to a page entry, * and a page entry and page directory to the page they refer to. */ + #define pte_pfn(pte) (pte_val(pte) >> PAGE_SHIFT) #define pte_same(a,b) (pte_val(a) == pte_val(b)) #define pte_page(x) pfn_to_page(pte_pfn(x)) @@ -231,9 +244,6 @@ static inline pte_t pte_modify(pte_t pte, pgprot_t newprot) static inline void update_pte(pte_t *ptep, pte_t pteval) { *ptep = pteval; -#if (DCACHE_WAY_SIZE > PAGE_SIZE) && XCHAL_DCACHE_IS_WRITEBACK - __asm__ __volatile__ ("memw; dhwb %0, 0; dsync" :: "a" (ptep)); -#endif } struct mm_struct; @@ -249,9 +259,6 @@ static inline void set_pmd(pmd_t *pmdp, pmd_t pmdval) { *pmdp = pmdval; -#if (DCACHE_WAY_SIZE > PAGE_SIZE) && XCHAL_DCACHE_IS_WRITEBACK - __asm__ __volatile__ ("memw; dhwb %0, 0; dsync" :: "a" (pmdp)); -#endif } struct vm_area_struct; @@ -306,52 +313,34 @@ ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, pte_t *ptep) /* * Encode and decode a swap entry. - * Each PTE in a process VM's page table is either: - * "present" -- valid and not swapped out, protection bits are meaningful; - * "not present" -- which further subdivides in these two cases: - * "none" -- no mapping at all; identified by pte_none(), set by pte_clear( - * "swapped out" -- the page is swapped out, and the SWP macros below - * are used to store swap file info in the PTE itself. - * - * In the Xtensa processor MMU, any PTE entries in user space (or anywhere - * in virtual memory that can map differently across address spaces) - * must have a correct ring value that represents the RASID field that - * is changed when switching address spaces. Eg. such PTE entries cannot - * be set to ring zero, because that can cause a (global) kernel ASID - * entry to be created in the TLBs (even with invalid cache attribute), - * potentially causing a multihit exception when going back to another - * address space that mapped the same virtual address at another ring. - * - * SO: we avoid using ring bits (_PAGE_RING_MASK) in "not present" PTEs. - * We also avoid using the _PAGE_VALID bit which must be zero for non-present - * pages. * - * We end up with the following available bits: 1..3 and 7..31. - * We don't bother with 1..3 for now (we can use them later if needed), - * and chose to allocate 6 bits for SWP_TYPE and the remaining 19 bits - * for SWP_OFFSET. At least 5 bits are needed for SWP_TYPE, because it - * is currently implemented as an index into swap_info[MAX_SWAPFILES] - * and MAX_SWAPFILES is currently defined as 32 in . - * However, for some reason all other architectures in the 2.4 kernel - * reserve either 6, 7, or 8 bits so I'll not detract from that for now. :) - * SWP_OFFSET is an offset into the swap file in page-size units, so - * with 4 kB pages, 19 bits supports a maximum swap file size of 2 GB. - * - * FIXME: 2 GB isn't very big. Other bits can be used to allow - * larger swap sizes. In the meantime, it appears relatively easy to get - * around the 2 GB limitation by simply using multiple swap files. + * Format of swap pte: + * bit 0 MBZ + * bit 1 page-file (must be zero) + * bits 2 - 3 page hw access mode (must be 11: _PAGE_INVALID) + * bits 4 - 5 ring protection (must be 01: _PAGE_USER) + * bits 6 - 10 swap type (5 bits -> 32 types) + * bits 11 - 31 swap offset / PAGE_SIZE (21 bits -> 8GB) + + * Format of file pte: + * bit 0 MBZ + * bit 1 page-file (must be one: _PAGE_FILE) + * bits 2 - 3 page hw access mode (must be 11: _PAGE_INVALID) + * bits 4 - 5 ring protection (must be 01: _PAGE_USER) + * bits 6 - 31 file offset / PAGE_SIZE */ -#define __swp_type(entry) (((entry).val >> 7) & 0x3f) -#define __swp_offset(entry) ((entry).val >> 13) -#define __swp_entry(type,offs) ((swp_entry_t) {((type) << 7) | ((offs) << 13)}) +#define __swp_type(entry) (((entry).val >> 6) & 0x1f) +#define __swp_offset(entry) ((entry).val >> 11) +#define __swp_entry(type,offs) \ + ((swp_entry_t) {((type) << 6) | ((offs) << 11) | _PAGE_INVALID}) #define __pte_to_swp_entry(pte) ((swp_entry_t) { pte_val(pte) }) #define __swp_entry_to_pte(x) ((pte_t) { (x).val }) -#define PTE_FILE_MAX_BITS 29 -#define pte_to_pgoff(pte) (pte_val(pte) >> 3) -#define pgoff_to_pte(off) ((pte_t) { ((off) << 3) | _PAGE_FILE }) - +#define PTE_FILE_MAX_BITS 28 +#define pte_to_pgoff(pte) (pte_val(pte) >> 4) +#define pgoff_to_pte(off) \ + ((pte_t) { ((off) << 4) | _PAGE_INVALID | _PAGE_FILE }) #endif /* !defined (__ASSEMBLY__) */ -- 2.20.1