From b91dc336cc7e0c9843f87ad0ba4cdbc821d69549 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Fri, 3 Aug 2007 15:54:36 -0700 Subject: [XTENSA] fix wrong usage of __init and __initdata in traps.c A variable was defined with __init instead of __initdata and the function accessing that variable wasn't initialized with __init. Signed-off-by: Chris Zankel --- arch/xtensa/kernel/traps.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'arch') diff --git a/arch/xtensa/kernel/traps.c b/arch/xtensa/kernel/traps.c index c5e62f9d9f5..8be99c777d9 100644 --- a/arch/xtensa/kernel/traps.c +++ b/arch/xtensa/kernel/traps.c @@ -83,7 +83,7 @@ typedef struct { void* handler; } dispatch_init_table_t; -dispatch_init_table_t __init dispatch_init_table[] = { +static dispatch_init_table_t __initdata dispatch_init_table[] = { { EXCCAUSE_ILLEGAL_INSTRUCTION, 0, do_illegal_instruction}, { EXCCAUSE_SYSTEM_CALL, KRNL, fast_syscall_kernel }, @@ -305,7 +305,7 @@ do_debug(struct pt_regs *regs) #define set_handler(idx,handler) (exc_table[idx] = (unsigned long) (handler)) -void trap_init(void) +void __init trap_init(void) { int i; -- cgit v1.2.3 From b2444d34a0e727a80fee4b725bc5ecb9d0f206f0 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Sat, 4 Aug 2007 09:23:54 -0700 Subject: [XTENSA] Add freestanding option to CFLAGS We also need to set the freestanding option for GCC in the CFLAGS. Signed-off-by: Chris Zankel --- arch/xtensa/Makefile | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) (limited to 'arch') diff --git a/arch/xtensa/Makefile b/arch/xtensa/Makefile index 95f836db38f..acf05be2492 100644 --- a/arch/xtensa/Makefile +++ b/arch/xtensa/Makefile @@ -27,7 +27,12 @@ platform-$(CONFIG_XTENSA_PLATFORM_ISS) := iss PLATFORM = $(platform-y) export PLATFORM -CFLAGS += -pipe -mlongcalls +# temporarily until string.h is fixed +cflags-y += -ffreestanding + +cflags-y += -pipe -mlongcalls + +CFLAGS += $(cflags-y) KBUILD_DEFCONFIG := iss_defconfig -- cgit v1.2.3 From 73089cbfdf0c69e061a4fa90d614679e630c6727 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Sat, 4 Aug 2007 09:27:30 -0700 Subject: [XTENSA] Move pre-initialized structures to init_task.c Move all initialization structures for the initial task to it's own file. Signed-off-by: Chris Zankel --- arch/xtensa/kernel/Makefile | 2 +- arch/xtensa/kernel/init_task.c | 38 ++++++++++++++++++++++++++++++++++++++ arch/xtensa/kernel/process.c | 15 +-------------- 3 files changed, 40 insertions(+), 15 deletions(-) create mode 100644 arch/xtensa/kernel/init_task.c (limited to 'arch') diff --git a/arch/xtensa/kernel/Makefile b/arch/xtensa/kernel/Makefile index 71f733c4f66..0e3ac686c17 100644 --- a/arch/xtensa/kernel/Makefile +++ b/arch/xtensa/kernel/Makefile @@ -7,7 +7,7 @@ extra-y := head.o vmlinux.lds obj-y := align.o entry.o irq.o coprocessor.o process.o ptrace.o semaphore.o \ setup.o signal.o syscall.o time.o traps.o vectors.o platform.o \ - pci-dma.o + pci-dma.o init_task.o ## windowspill.o diff --git a/arch/xtensa/kernel/init_task.c b/arch/xtensa/kernel/init_task.c new file mode 100644 index 00000000000..021b4f46ff9 --- /dev/null +++ b/arch/xtensa/kernel/init_task.c @@ -0,0 +1,38 @@ +/* + * arch/xtensa/kernel/init_task.c + * + * Xtensa Processor version. + * + * This file is subject to the terms and conditions of the GNU General Public + * License. See the file "COPYING" in the main directory of this archive + * for more details. + * + * Copyright (C) 2007 Tensilica Inc. + * + * Chris Zankel + */ + +#include +#include +#include +#include +#include +#include + +#include + +static struct fs_struct init_fs = INIT_FS; +static struct files_struct init_files = INIT_FILES; +static struct signal_struct init_signals = INIT_SIGNALS(init_signals); +static struct sighand_struct init_sighand = INIT_SIGHAND(init_sighand); +struct mm_struct init_mm = INIT_MM(init_mm); + +EXPORT_SYMBOL(init_mm); + +union thread_union init_thread_union + __attribute__((__section__(".data.init_task"))) = +{ INIT_THREAD_INFO(init_task) }; + +struct task_struct init_task = INIT_TASK(init_task); + +EXPORT_SYMBOL(init_task); diff --git a/arch/xtensa/kernel/process.c b/arch/xtensa/kernel/process.c index ce758bab95b..f53d7bd9dfb 100644 --- a/arch/xtensa/kernel/process.c +++ b/arch/xtensa/kernel/process.c @@ -30,6 +30,7 @@ #include #include #include +#include #include #include @@ -45,20 +46,6 @@ extern void ret_from_fork(void); -static struct fs_struct init_fs = INIT_FS; -static struct files_struct init_files = INIT_FILES; -static struct signal_struct init_signals = INIT_SIGNALS(init_signals); -static struct sighand_struct init_sighand = INIT_SIGHAND(init_sighand); -struct mm_struct init_mm = INIT_MM(init_mm); -EXPORT_SYMBOL(init_mm); - -union thread_union init_thread_union - __attribute__((__section__(".data.init_task"))) = -{ INIT_THREAD_INFO(init_task) }; - -struct task_struct init_task = INIT_TASK(init_task); -EXPORT_SYMBOL(init_task); - struct task_struct *current_set[NR_CPUS] = {&init_task, }; void (*pm_power_off)(void) = NULL; -- cgit v1.2.3 From 787a22d1d284b21ad810fd0bedbdefb329f31cd2 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Sat, 4 Aug 2007 14:31:04 -0700 Subject: [XTENSA] Move string-io functions to io.c from pci.c The string-io functions (outs{bwl}, ins{bwl}) are independent from the PCI option and should be in a separate file. Signed-off-by: Chris Zankel --- arch/xtensa/kernel/Makefile | 2 +- arch/xtensa/kernel/io.c | 75 +++++++++++++++++++++++++++++++++++++++++++++ arch/xtensa/kernel/pci.c | 69 ----------------------------------------- 3 files changed, 76 insertions(+), 70 deletions(-) create mode 100644 arch/xtensa/kernel/io.c (limited to 'arch') diff --git a/arch/xtensa/kernel/Makefile b/arch/xtensa/kernel/Makefile index 0e3ac686c17..f582d6a24ec 100644 --- a/arch/xtensa/kernel/Makefile +++ b/arch/xtensa/kernel/Makefile @@ -7,7 +7,7 @@ extra-y := head.o vmlinux.lds obj-y := align.o entry.o irq.o coprocessor.o process.o ptrace.o semaphore.o \ setup.o signal.o syscall.o time.o traps.o vectors.o platform.o \ - pci-dma.o init_task.o + pci-dma.o init_task.o io.o ## windowspill.o diff --git a/arch/xtensa/kernel/io.c b/arch/xtensa/kernel/io.c new file mode 100644 index 00000000000..5b65269b1d2 --- /dev/null +++ b/arch/xtensa/kernel/io.c @@ -0,0 +1,75 @@ +/* + * arch/xtensa/io.c + * + * IO primitives + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License as published by the + * Free Software Foundation; either version 2 of the License, or (at your + * option) any later version. + * + * Copied from sparc. + * + * Chris Zankel + * + */ + +#include +#include + +void outsb(unsigned long addr, const void *src, unsigned long count) { + while (count) { + count -= 1; + writeb(*(const char *)src, addr); + src += 1; + addr += 1; + } +} + +void outsw(unsigned long addr, const void *src, unsigned long count) { + while (count) { + count -= 2; + writew(*(const short *)src, addr); + src += 2; + addr += 2; + } +} + +void outsl(unsigned long addr, const void *src, unsigned long count) { + while (count) { + count -= 4; + writel(*(const long *)src, addr); + src += 4; + addr += 4; + } +} + +void insb(unsigned long addr, void *dst, unsigned long count) { + while (count) { + count -= 1; + *(unsigned char *)dst = readb(addr); + dst += 1; + addr += 1; + } +} + +void insw(unsigned long addr, void *dst, unsigned long count) { + while (count) { + count -= 2; + *(unsigned short *)dst = readw(addr); + dst += 2; + addr += 2; + } +} + +void insl(unsigned long addr, void *dst, unsigned long count) { + while (count) { + count -= 4; + /* + * XXX I am sure we are in for an unaligned trap here. + */ + *(unsigned long *)dst = readl(addr); + dst += 4; + addr += 4; + } +} diff --git a/arch/xtensa/kernel/pci.c b/arch/xtensa/kernel/pci.c index 77deae5290f..b7c073484e0 100644 --- a/arch/xtensa/kernel/pci.c +++ b/arch/xtensa/kernel/pci.c @@ -394,72 +394,3 @@ int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma, return ret; } - -/* - * This probably belongs here rather than ioport.c because - * we do not want this crud linked into SBus kernels. - * Also, think for a moment about likes of floppy.c that - * include architecture specific parts. They may want to redefine ins/outs. - * - * We do not use horrible macros here because we want to - * advance pointer by sizeof(size). - */ -void outsb(unsigned long addr, const void *src, unsigned long count) { - while (count) { - count -= 1; - writeb(*(const char *)src, addr); - src += 1; - addr += 1; - } -} - -void outsw(unsigned long addr, const void *src, unsigned long count) { - while (count) { - count -= 2; - writew(*(const short *)src, addr); - src += 2; - addr += 2; - } -} - -void outsl(unsigned long addr, const void *src, unsigned long count) { - while (count) { - count -= 4; - writel(*(const long *)src, addr); - src += 4; - addr += 4; - } -} - -void insb(unsigned long addr, void *dst, unsigned long count) { - while (count) { - count -= 1; - *(unsigned char *)dst = readb(addr); - dst += 1; - addr += 1; - } -} - -void insw(unsigned long addr, void *dst, unsigned long count) { - while (count) { - count -= 2; - *(unsigned short *)dst = readw(addr); - dst += 2; - addr += 2; - } -} - -void insl(unsigned long addr, void *dst, unsigned long count) { - while (count) { - count -= 4; - /* - * XXX I am sure we are in for an unaligned trap here. - */ - *(unsigned long *)dst = readl(addr); - dst += 4; - addr += 4; - } -} - - - -- cgit v1.2.3 From bc671aa9838f234ccfc794a77325628f1e41e083 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Sun, 5 Aug 2007 10:22:58 -0700 Subject: [XTENSA] Fix fadvise64_64 Xtensa passes long long arguments in a even/odd register pair, so we also need to shuffle the arguments when passed through the system call to avoid an empty argument register. Signed-off-by: Chris Zankel --- arch/xtensa/kernel/syscall.c | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'arch') diff --git a/arch/xtensa/kernel/syscall.c b/arch/xtensa/kernel/syscall.c index fe3834bc1db..f3e16efcd47 100644 --- a/arch/xtensa/kernel/syscall.c +++ b/arch/xtensa/kernel/syscall.c @@ -93,3 +93,8 @@ asmlinkage long xtensa_shmat(int shmid, char __user *shmaddr, int shmflg) return (long)ret; } +asmlinkage long xtensa_fadvise64_64(int fd, int advice, unsigned long long offset, unsigned long long len) +{ + return sys_fadvise64_64(fd, offset, len, advice); +} + -- cgit v1.2.3 From 2b8aea74e78e977b1f9987e23e3e59f3ef4359f4 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Sun, 5 Aug 2007 10:26:30 -0700 Subject: [XTENSA] Fix timer instabilities. The timer code could have missed a tick. Signed-off-by: Chris Zankel --- arch/xtensa/kernel/time.c | 46 +++++++++++++++++++++++++++++----------------- 1 file changed, 29 insertions(+), 17 deletions(-) (limited to 'arch') diff --git a/arch/xtensa/kernel/time.c b/arch/xtensa/kernel/time.c index 22949be4a5d..60d29fe0b1b 100644 --- a/arch/xtensa/kernel/time.c +++ b/arch/xtensa/kernel/time.c @@ -32,12 +32,20 @@ EXPORT_SYMBOL(rtc_lock); #ifdef CONFIG_XTENSA_CALIBRATE_CCOUNT unsigned long ccount_per_jiffy; /* per 1/HZ */ -unsigned long ccount_nsec; /* nsec per ccount increment */ +unsigned long nsec_per_ccount; /* nsec per ccount increment */ #endif -unsigned int last_ccount_stamp; static long last_rtc_update = 0; +/* + * Scheduler clock - returns current tim in nanosec units. + */ + +unsigned long long sched_clock(void) +{ + return (unsigned long long)jiffies * (1000000000 / HZ); +} + static irqreturn_t timer_interrupt(int irq, void *dev_id); static struct irqaction timer_irqaction = { .handler = timer_interrupt, @@ -69,7 +77,6 @@ void __init time_init(void) xtime.tv_nsec = 0; last_rtc_update = xtime.tv_sec = sec_n; - last_ccount_stamp = get_ccount(); set_normalized_timespec(&wall_to_monotonic, -xtime.tv_sec, -xtime.tv_nsec); @@ -85,7 +92,7 @@ int do_settimeofday(struct timespec *tv) { time_t wtm_sec, sec = tv->tv_sec; long wtm_nsec, nsec = tv->tv_nsec; - unsigned long ccount; + unsigned long delta; if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC) return -EINVAL; @@ -97,8 +104,10 @@ int do_settimeofday(struct timespec *tv) * wall time. Discover what correction gettimeofday() would have * made, and then undo it! */ - ccount = get_ccount(); - nsec -= (ccount - last_ccount_stamp) * CCOUNT_NSEC; + + delta = CCOUNT_PER_JIFFY; + delta += get_ccount() - get_linux_timer(); + nsec -= delta * NSEC_PER_CCOUNT; wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec); wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec); @@ -117,17 +126,21 @@ EXPORT_SYMBOL(do_settimeofday); void do_gettimeofday(struct timeval *tv) { unsigned long flags; - unsigned long sec, usec, delta, seq; + unsigned long volatile sec, usec, delta, seq; do { seq = read_seqbegin_irqsave(&xtime_lock, flags); - delta = get_ccount() - last_ccount_stamp; sec = xtime.tv_sec; usec = (xtime.tv_nsec / NSEC_PER_USEC); + + delta = get_linux_timer() - get_ccount(); + } while (read_seqretry_irqrestore(&xtime_lock, seq, flags)); - usec += (delta * CCOUNT_NSEC) / NSEC_PER_USEC; + usec += (((unsigned long) CCOUNT_PER_JIFFY - delta) + * (unsigned long) NSEC_PER_CCOUNT) / NSEC_PER_USEC; + for (; usec >= 1000000; sec++, usec -= 1000000) ; @@ -158,9 +171,12 @@ again: write_seqlock(&xtime_lock); - last_ccount_stamp = next; + do_timer(1); /* Linux handler in kernel/timer.c */ + + /* Note that writing CCOMPARE clears the interrupt. */ + next += CCOUNT_PER_JIFFY; - do_timer (1); /* Linux handler in kernel/timer.c */ + set_linux_timer(next); if (ntp_synced() && xtime.tv_sec - last_rtc_update >= 659 && @@ -175,19 +191,15 @@ again: write_sequnlock(&xtime_lock); } - /* NOTE: writing CCOMPAREn clears the interrupt. */ + /* Allow platform to do something useful (Wdog). */ - set_linux_timer (next); + platform_heartbeat(); /* Make sure we didn't miss any tick... */ if ((signed long)(get_ccount() - next) > 0) goto again; - /* Allow platform to do something useful (Wdog). */ - - platform_heartbeat(); - return IRQ_HANDLED; } -- cgit v1.2.3 From 5c1c8085b5dc30ae4ff0ee54039e387ed59262bf Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Sun, 5 Aug 2007 11:51:57 -0700 Subject: [XTENSA] Initialize semaphore_wake_lock Initialize semaphore_wake_lock. Signed-off-by: Chris Zankel --- arch/xtensa/kernel/semaphore.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'arch') diff --git a/arch/xtensa/kernel/semaphore.c b/arch/xtensa/kernel/semaphore.c index d40f4b1b75a..995c6410ae1 100644 --- a/arch/xtensa/kernel/semaphore.c +++ b/arch/xtensa/kernel/semaphore.c @@ -100,7 +100,7 @@ static __inline__ int waking_non_zero_trylock(struct semaphore *sem) return ret; } -spinlock_t semaphore_wake_lock; +DEFINE_SPINLOCK(semaphore_wake_lock); /* * Semaphores are implemented using a two-way counter: -- cgit v1.2.3 From 01858d1b0b406307626bbc01238391b06aae2c20 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Mon, 6 Aug 2007 23:57:57 -0700 Subject: [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 ++++++++++++++++++++++++++++-------------- 1 file changed, 28 insertions(+), 14 deletions(-) (limited to 'arch') diff --git a/arch/xtensa/kernel/entry.S b/arch/xtensa/kernel/entry.S index 8dc7a2c26ff..65741e3368e 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. */ -- cgit v1.2.3 From ff6fd469885aafa5ec387babcb6537f3c00d6df0 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Tue, 14 Aug 2007 13:02:06 -0700 Subject: [XTENSA] Add kernel module support Add kernel module support. Signed-off-by: Chris Zankel --- arch/xtensa/kernel/module.c | 195 ++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 177 insertions(+), 18 deletions(-) (limited to 'arch') diff --git a/arch/xtensa/kernel/module.c b/arch/xtensa/kernel/module.c index 2ea1755a085..ddf14dcf2ad 100644 --- a/arch/xtensa/kernel/module.c +++ b/arch/xtensa/kernel/module.c @@ -7,7 +7,7 @@ * License. See the file "COPYING" in the main directory of this archive * for more details. * - * Copyright (C) 2001 - 2005 Tensilica Inc. + * Copyright (C) 2001 - 2006 Tensilica Inc. * * Chris Zankel * @@ -22,57 +22,216 @@ #include #include -LIST_HEAD(module_buf_list); +#undef DEBUG_RELOCATE void *module_alloc(unsigned long size) { - panic("module_alloc not implemented"); + if (size == 0) + return NULL; + return vmalloc(size); } void module_free(struct module *mod, void *module_region) { - panic("module_free not implemented"); + vfree(module_region); + /* FIXME: If module_region == mod->init_region, trim exception + table entries. */ } int module_frob_arch_sections(Elf32_Ehdr *hdr, Elf32_Shdr *sechdrs, char *secstrings, - struct module *me) + struct module *mod) { - panic("module_frob_arch_sections not implemented"); + return 0; +} + +static int +decode_calln_opcode (unsigned char *location) +{ +#ifdef __XTENSA_EB__ + return (location[0] & 0xf0) == 0x50; +#endif +#ifdef __XTENSA_EL__ + return (location[0] & 0xf) == 0x5; +#endif +} + +static int +decode_l32r_opcode (unsigned char *location) +{ +#ifdef __XTENSA_EB__ + return (location[0] & 0xf0) == 0x10; +#endif +#ifdef __XTENSA_EL__ + return (location[0] & 0xf) == 0x1; +#endif } int apply_relocate(Elf32_Shdr *sechdrs, const char *strtab, unsigned int symindex, unsigned int relsec, - struct module *module) + struct module *mod) { - panic ("apply_relocate not implemented"); + printk(KERN_ERR "module %s: REL RELOCATION unsupported\n", + mod->name); + return -ENOEXEC; + } int apply_relocate_add(Elf32_Shdr *sechdrs, const char *strtab, unsigned int symindex, unsigned int relsec, - struct module *module) + struct module *mod) { - panic("apply_relocate_add not implemented"); + unsigned int i; + Elf32_Rela *rela = (void *)sechdrs[relsec].sh_addr; + Elf32_Sym *sym; + unsigned char *location; + uint32_t value; + +#ifdef DEBUG_RELOCATE + printk("Applying relocate section %u to %u\n", relsec, + sechdrs[relsec].sh_info); +#endif + for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) { + location = (char *)sechdrs[sechdrs[relsec].sh_info].sh_addr + + rela[i].r_offset; + sym = (Elf32_Sym *)sechdrs[symindex].sh_addr + + ELF32_R_SYM(rela[i].r_info); + value = sym->st_value + rela[i].r_addend; + + switch (ELF32_R_TYPE(rela[i].r_info)) { + case R_XTENSA_NONE: + case R_XTENSA_DIFF8: + case R_XTENSA_DIFF16: + case R_XTENSA_DIFF32: + case R_XTENSA_ASM_EXPAND: + break; + + case R_XTENSA_32: + case R_XTENSA_PLT: + *(uint32_t *)location += value; + break; + + case R_XTENSA_SLOT0_OP: + if (decode_calln_opcode(location)) { + value -= ((unsigned long)location & -4) + 4; + if ((value & 3) != 0 || + ((value + (1 << 19)) >> 20) != 0) { + printk("%s: relocation out of range, " + "section %d reloc %d " + "sym '%s'\n", + mod->name, relsec, i, + strtab + sym->st_name); + return -ENOEXEC; + } + value = (signed int)value >> 2; +#ifdef __XTENSA_EB__ + location[0] = ((location[0] & ~0x3) | + ((value >> 16) & 0x3)); + location[1] = (value >> 8) & 0xff; + location[2] = value & 0xff; +#endif +#ifdef __XTENSA_EL__ + location[0] = ((location[0] & ~0xc0) | + ((value << 6) & 0xc0)); + location[1] = (value >> 2) & 0xff; + location[2] = (value >> 10) & 0xff; +#endif + } else if (decode_l32r_opcode(location)) { + value -= (((unsigned long)location + 3) & -4); + if ((value & 3) != 0 || + (signed int)value >> 18 != -1) { + printk("%s: relocation out of range, " + "section %d reloc %d " + "sym '%s'\n", + mod->name, relsec, i, + strtab + sym->st_name); + return -ENOEXEC; + } + value = (signed int)value >> 2; + +#ifdef __XTENSA_EB__ + location[1] = (value >> 8) & 0xff; + location[2] = value & 0xff; +#endif +#ifdef __XTENSA_EL__ + location[1] = value & 0xff; + location[2] = (value >> 8) & 0xff; +#endif + } + /* FIXME: Ignore any other opcodes. The Xtensa + assembler currently assumes that the linker will + always do relaxation and so all PC-relative + operands need relocations. (The assembler also + writes out the tentative PC-relative values, + assuming no link-time relaxation, so it is usually + safe to ignore the relocations.) If the + assembler's "--no-link-relax" flag can be made to + work, and if all kernel modules can be assembled + with that flag, then unexpected relocations could + be detected here. */ + break; + + case R_XTENSA_SLOT1_OP: + case R_XTENSA_SLOT2_OP: + case R_XTENSA_SLOT3_OP: + case R_XTENSA_SLOT4_OP: + case R_XTENSA_SLOT5_OP: + case R_XTENSA_SLOT6_OP: + case R_XTENSA_SLOT7_OP: + case R_XTENSA_SLOT8_OP: + case R_XTENSA_SLOT9_OP: + case R_XTENSA_SLOT10_OP: + case R_XTENSA_SLOT11_OP: + case R_XTENSA_SLOT12_OP: + case R_XTENSA_SLOT13_OP: + case R_XTENSA_SLOT14_OP: + printk("%s: unexpected FLIX relocation: %u\n", + mod->name, + ELF32_R_TYPE(rela[i].r_info)); + return -ENOEXEC; + + case R_XTENSA_SLOT0_ALT: + case R_XTENSA_SLOT1_ALT: + case R_XTENSA_SLOT2_ALT: + case R_XTENSA_SLOT3_ALT: + case R_XTENSA_SLOT4_ALT: + case R_XTENSA_SLOT5_ALT: + case R_XTENSA_SLOT6_ALT: + case R_XTENSA_SLOT7_ALT: + case R_XTENSA_SLOT8_ALT: + case R_XTENSA_SLOT9_ALT: + case R_XTENSA_SLOT10_ALT: + case R_XTENSA_SLOT11_ALT: + case R_XTENSA_SLOT12_ALT: + case R_XTENSA_SLOT13_ALT: + case R_XTENSA_SLOT14_ALT: + printk("%s: unexpected ALT relocation: %u\n", + mod->name, + ELF32_R_TYPE(rela[i].r_info)); + return -ENOEXEC; + + default: + printk("%s: unexpected relocation: %u\n", + mod->name, + ELF32_R_TYPE(rela[i].r_info)); + return -ENOEXEC; + } + } + return 0; } int module_finalize(const Elf_Ehdr *hdr, const Elf_Shdr *sechdrs, - struct module *me) + struct module *mod) { - panic ("module_finalize not implemented"); + return 0; } void module_arch_cleanup(struct module *mod) { - panic("module_arch_cleanup not implemented"); -} - -struct bug_entry *module_find_bug(unsigned long bugaddr) -{ - panic("module_find_bug not implemented"); } -- cgit v1.2.3 From 6656920b0b50beacb6cb64cf55273cbb686e436e Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Wed, 22 Aug 2007 10:14:51 -0700 Subject: [XTENSA] Add support for cache-aliasing Add support for processors that have cache-aliasing issues, such as the Stretch S5000 processor. Cache-aliasing means that the size of the cache (for one way) is larger than the page size, thus, a page can end up in several places in cache depending on the virtual to physical translation. The method used here is to map a user page temporarily through the auto-refill way 0 and of of the DTLB. We probably will want to revisit this issue and use a better approach with kmap/kunmap. Signed-off-by: Chris Zankel --- arch/xtensa/kernel/asm-offsets.c | 13 +- arch/xtensa/kernel/entry.S | 75 +++++++++- arch/xtensa/mm/Makefile | 6 +- arch/xtensa/mm/cache.c | 256 ++++++++++++++++++++++++++++++++ arch/xtensa/mm/fault.c | 6 +- arch/xtensa/mm/init.c | 252 ++------------------------------ arch/xtensa/mm/misc.S | 306 ++++++++++++++++++++++++++++++++++----- 7 files changed, 624 insertions(+), 290 deletions(-) create mode 100644 arch/xtensa/mm/cache.c (limited to 'arch') diff --git a/arch/xtensa/kernel/asm-offsets.c b/arch/xtensa/kernel/asm-offsets.c index d0323cd6a2e..d5ffe7b6443 100644 --- a/arch/xtensa/kernel/asm-offsets.c +++ b/arch/xtensa/kernel/asm-offsets.c @@ -18,12 +18,13 @@ #include #include #include +#include + #include #include #include #define DEFINE(sym, val) asm volatile("\n->" #sym " %0 " #val : : "i" (val)) -#define BLANK() asm volatile("\n->" : : ) int main(void) { @@ -63,7 +64,6 @@ int main(void) DEFINE(PT_SIZE, sizeof(struct pt_regs)); DEFINE(PT_AREG_END, offsetof (struct pt_regs, areg[XCHAL_NUM_AREGS])); DEFINE(PT_USER_SIZE, offsetof(struct pt_regs, areg[XCHAL_NUM_AREGS])); - BLANK(); /* struct task_struct */ DEFINE(TASK_PTRACE, offsetof (struct task_struct, ptrace)); @@ -73,27 +73,26 @@ int main(void) DEFINE(TASK_THREAD, offsetof (struct task_struct, thread)); DEFINE(TASK_THREAD_INFO, offsetof (struct task_struct, stack)); DEFINE(TASK_STRUCT_SIZE, sizeof (struct task_struct)); - BLANK(); /* struct thread_info (offset from start_struct) */ DEFINE(THREAD_RA, offsetof (struct task_struct, thread.ra)); DEFINE(THREAD_SP, offsetof (struct task_struct, thread.sp)); DEFINE(THREAD_CP_SAVE, offsetof (struct task_struct, thread.cp_save)); DEFINE(THREAD_CURRENT_DS, offsetof (struct task_struct, thread.current_ds)); - BLANK(); /* struct mm_struct */ DEFINE(MM_USERS, offsetof(struct mm_struct, mm_users)); DEFINE(MM_PGD, offsetof (struct mm_struct, pgd)); DEFINE(MM_CONTEXT, offsetof (struct mm_struct, context)); - BLANK(); - DEFINE(PT_SINGLESTEP_BIT, PT_SINGLESTEP_BIT); + + /* struct page */ + DEFINE(PAGE_FLAGS, offsetof(struct page, flags)); /* constants */ DEFINE(_CLONE_VM, CLONE_VM); DEFINE(_CLONE_UNTRACED, CLONE_UNTRACED); + DEFINE(PG_ARCH_1, PG_arch_1); return 0; } - diff --git a/arch/xtensa/kernel/entry.S b/arch/xtensa/kernel/entry.S index 65741e3368e..91a689eca43 100644 --- a/arch/xtensa/kernel/entry.S +++ b/arch/xtensa/kernel/entry.S @@ -7,7 +7,7 @@ * License. See the file "COPYING" in the main directory of this archive * for more details. * - * Copyright (C) 2004-2005 by Tensilica Inc. + * Copyright (C) 2004-2007 by Tensilica Inc. * * Chris Zankel * @@ -169,7 +169,7 @@ _user_exception: * We have to save all registers up to the first '1' from * the right, except the current frame (bit 0). * Assume a2 is: 001001000110001 - * All regiser frames starting from the top fiel to the marked '1' + * All register frames starting from the top field to the marked '1' * must be saved. */ @@ -1590,7 +1590,7 @@ ENTRY(fast_second_level_miss) * The messy computation for 'pteval' above really simplifies * into the following: * - * pteval = ((pmdval - PAGE_OFFSET) & PAGE_MASK) | PAGE_KERNEL + * pteval = ((pmdval - PAGE_OFFSET) & PAGE_MASK) | PAGE_DIRECTORY */ movi a1, -PAGE_OFFSET @@ -1602,7 +1602,7 @@ ENTRY(fast_second_level_miss) or a0, a0, a1 # ... | PAGE_DIRECTORY /* - * We utilize all three wired-ways (7-9( to hold pmd translations. + * 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: @@ -1652,6 +1652,73 @@ ENTRY(fast_second_level_miss) 9: l32i a0, a1, TASK_ACTIVE_MM # unlikely case mm == 0 j 8b +#if (DCACHE_WAY_SIZE > PAGE_SIZE) + +2: /* Special case for cache aliasing. + * We (should) only get here if a clear_user_page, copy_user_page + * or the aliased cache flush functions got preemptively interrupted + * by another task. Re-establish temporary mapping to the + * TLBTEMP_BASE areas. + */ + + /* We shouldn't be in a double exception */ + + l32i a0, a2, PT_DEPC + bgeui a0, VALID_DOUBLE_EXCEPTION_ADDRESS, 2f + + /* Make sure the exception originated in the special functions */ + + movi a0, __tlbtemp_mapping_start + rsr a3, EPC_1 + bltu a3, a0, 2f + movi a0, __tlbtemp_mapping_end + bgeu a3, a0, 2f + + /* Check if excvaddr was in one of the TLBTEMP_BASE areas. */ + + movi a3, TLBTEMP_BASE_1 + rsr a0, EXCVADDR + bltu a0, a3, 2f + + addi a1, a0, -(2 << (DCACHE_ALIAS_ORDER + PAGE_SHIFT)) + bgeu a1, a3, 2f + + /* Check if we have to restore an ITLB mapping. */ + + movi a1, __tlbtemp_mapping_itlb + rsr a3, EPC_1 + sub a3, a3, a1 + + /* Calculate VPN */ + + movi a1, PAGE_MASK + and a1, a1, a0 + + /* Jump for ITLB entry */ + + bgez a3, 1f + + /* We can use up to two TLBTEMP areas, one for src and one for dst. */ + + extui a3, a0, PAGE_SHIFT + DCACHE_ALIAS_ORDER, 1 + add a1, a3, a1 + + /* PPN is in a6 for the first TLBTEMP area and in a7 for the second. */ + + mov a0, a6 + movnez a0, a7, a3 + j 3b + + /* ITLB entry. We only use dst in a6. */ + +1: witlb a6, a1 + isync + j 4b + + +#endif // DCACHE_WAY_SIZE > PAGE_SIZE + + 2: /* Invalid PGD, default exception handling */ movi a3, exc_table diff --git a/arch/xtensa/mm/Makefile b/arch/xtensa/mm/Makefile index a5aed5932d7..10aec22a8f9 100644 --- a/arch/xtensa/mm/Makefile +++ b/arch/xtensa/mm/Makefile @@ -5,9 +5,5 @@ # removes any old dependencies. DON'T put your own dependencies here # unless it's something special (ie not a .c file). # -# Note 2! The CFLAGS definition is now in the main makefile... -obj-y := init.o fault.o tlb.o misc.o -obj-m := -obj-n := -obj- := +obj-y := init.o fault.o tlb.o misc.o cache.o diff --git a/arch/xtensa/mm/cache.c b/arch/xtensa/mm/cache.c new file mode 100644 index 00000000000..9a1fa9478ae --- /dev/null +++ b/arch/xtensa/mm/cache.c @@ -0,0 +1,256 @@ +/* + * arch/xtensa/mm/cache.c + * + * This file is subject to the terms and conditions of the GNU General Public + * License. See the file "COPYING" in the main directory of this archive + * for more details. + * + * Copyright (C) 2001-2006 Tensilica Inc. + * + * Chris Zankel + * Joe Taylor + * Marc Gauthier + * + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include + +//#define printd(x...) printk(x) +#define printd(x...) do { } while(0) + +/* + * Note: + * The kernel provides one architecture bit PG_arch_1 in the page flags that + * can be used for cache coherency. + * + * I$-D$ coherency. + * + * The Xtensa architecture doesn't keep the instruction cache coherent with + * the data cache. We use the architecture bit to indicate if the caches + * are coherent. The kernel clears this bit whenever a page is added to the + * page cache. At that time, the caches might not be in sync. We, therefore, + * define this flag as 'clean' if set. + * + * D-cache aliasing. + * + * With cache aliasing, we have to always flush the cache when pages are + * unmapped (see tlb_start_vma(). So, we use this flag to indicate a dirty + * page. + * + * + * + */ + +#if (DCACHE_WAY_SIZE > PAGE_SIZE) && XCHAL_DCACHE_IS_WRITEBACK + +/* + * Any time the kernel writes to a user page cache page, or it is about to + * read from a page cache page this routine is called. + * + */ + +void flush_dcache_page(struct page *page) +{ + struct address_space *mapping = page_mapping(page); + + /* + * If we have a mapping but the page is not mapped to user-space + * yet, we simply mark this page dirty and defer flushing the + * caches until update_mmu(). + */ + + if (mapping && !mapping_mapped(mapping)) { + if (!test_bit(PG_arch_1, &page->flags)) + set_bit(PG_arch_1, &page->flags); + return; + + } else { + + unsigned long phys = page_to_phys(page); + unsigned long temp = page->index << PAGE_SHIFT; + unsigned long alias = !(DCACHE_ALIAS_EQ(temp, phys)); + unsigned long virt; + + /* + * Flush the page in kernel space and user space. + * Note that we can omit that step if aliasing is not + * an issue, but we do have to synchronize I$ and D$ + * if we have a mapping. + */ + + if (!alias && !mapping) + return; + + __flush_invalidate_dcache_page((long)page_address(page)); + + virt = TLBTEMP_BASE_1 + (temp & DCACHE_ALIAS_MASK); + + if (alias) + __flush_invalidate_dcache_page_alias(virt, phys); + + if (mapping) + __invalidate_icache_page_alias(virt, phys); + } + + /* There shouldn't be an entry in the cache for this page anymore. */ +} + + +/* + * For now, flush the whole cache. FIXME?? + */ + +void flush_cache_range(struct vm_area_struct* vma, + unsigned long start, unsigned long end) +{ + __flush_invalidate_dcache_all(); + __invalidate_icache_all(); +} + +/* + * Remove any entry in the cache for this page. + * + * Note that this function is only called for user pages, so use the + * alias versions of the cache flush functions. + */ + +void flush_cache_page(struct vm_area_struct* vma, unsigned long address, + unsigned long pfn) +{ + /* Note that we have to use the 'alias' address to avoid multi-hit */ + + unsigned long phys = page_to_phys(pfn_to_page(pfn)); + unsigned long virt = TLBTEMP_BASE_1 + (address & DCACHE_ALIAS_MASK); + + __flush_invalidate_dcache_page_alias(virt, phys); + __invalidate_icache_page_alias(virt, phys); +} + +#endif + +void +update_mmu_cache(struct vm_area_struct * vma, unsigned long addr, pte_t pte) +{ + unsigned long pfn = pte_pfn(pte); + struct page *page; + + if (!pfn_valid(pfn)) + return; + + page = pfn_to_page(pfn); + + /* Invalidate old entry in TLBs */ + + invalidate_itlb_mapping(addr); + invalidate_dtlb_mapping(addr); + +#if (DCACHE_WAY_SIZE > PAGE_SIZE) && XCHAL_DCACHE_IS_WRITEBACK + + if (!PageReserved(page) && test_bit(PG_arch_1, &page->flags)) { + + unsigned long vaddr = TLBTEMP_BASE_1 + (addr & DCACHE_ALIAS_MASK); + unsigned long paddr = (unsigned long) page_address(page); + unsigned long phys = page_to_phys(page); + + __flush_invalidate_dcache_page(paddr); + + __flush_invalidate_dcache_page_alias(vaddr, phys); + __invalidate_icache_page_alias(vaddr, phys); + + clear_bit(PG_arch_1, &page->flags); + } +#else + if (!PageReserved(page) && !test_bit(PG_arch_1, &page->flags) + && (vma->vm_flags & VM_EXEC) != 0) { + unsigned long vaddr = addr & PAGE_MASK; + __flush_dcache_page(vaddr); + __invalidate_icache_page(vaddr); + set_bit(PG_arch_1, &page->flags); + } +#endif +} + +/* + * access_process_vm() has called get_user_pages(), which has done a + * flush_dcache_page() on the page. + */ + +#if (DCACHE_WAY_SIZE > PAGE_SIZE) && XCHAL_DCACHE_IS_WRITEBACK + +void copy_to_user_page(struct vm_area_struct *vma, struct page *page, + unsigned long vaddr, void *dst, const void *src, + unsigned long len) +{ + unsigned long phys = page_to_phys(page); + unsigned long alias = !(DCACHE_ALIAS_EQ(vaddr, phys)); + + /* Flush and invalidate user page if aliased. */ + + if (alias) { + unsigned long temp = TLBTEMP_BASE_1 + (vaddr & DCACHE_ALIAS_MASK); + __flush_invalidate_dcache_page_alias(temp, phys); + } + + /* Copy data */ + + memcpy(dst, src, len); + + /* + * Flush and invalidate kernel page if aliased and synchronize + * data and instruction caches for executable pages. + */ + + if (alias) { + unsigned long temp = TLBTEMP_BASE_1 + (vaddr & DCACHE_ALIAS_MASK); + + __flush_invalidate_dcache_range((unsigned long) dst, len); + if ((vma->vm_flags & VM_EXEC) != 0) { + __invalidate_icache_page_alias(temp, phys); + } + + } else if ((vma->vm_flags & VM_EXEC) != 0) { + __flush_dcache_range((unsigned long)dst,len); + __invalidate_icache_range((unsigned long) dst, len); + } +} + +extern void copy_from_user_page(struct vm_area_struct *vma, struct page *page, + unsigned long vaddr, void *dst, const void *src, + unsigned long len) +{ + unsigned long phys = page_to_phys(page); + unsigned long alias = !(DCACHE_ALIAS_EQ(vaddr, phys)); + + /* + * Flush user page if aliased. + * (Note: a simply flush would be sufficient) + */ + + if (alias) { + unsigned long temp = TLBTEMP_BASE_1 + (vaddr & DCACHE_ALIAS_MASK); + __flush_invalidate_dcache_page_alias(temp, phys); + } + + memcpy(dst, src, len); +} + +#endif diff --git a/arch/xtensa/mm/fault.c b/arch/xtensa/mm/fault.c index 16004067add..45d28f217c0 100644 --- a/arch/xtensa/mm/fault.c +++ b/arch/xtensa/mm/fault.c @@ -24,6 +24,8 @@ unsigned long asid_cache = ASID_USER_FIRST; void bad_page_fault(struct pt_regs*, unsigned long, int); +#undef DEBUG_PAGE_FAULT + /* * This routine handles page faults. It determines the address, * and the problem, and then passes it off to one of the appropriate @@ -64,7 +66,7 @@ void do_page_fault(struct pt_regs *regs) exccause == EXCCAUSE_ITLB_MISS || exccause == EXCCAUSE_FETCH_CACHE_ATTRIBUTE) ? 1 : 0; -#if 0 +#ifdef DEBUG_PAGE_FAULT printk("[%s:%d:%08x:%d:%08x:%s%s]\n", current->comm, current->pid, address, exccause, regs->pc, is_write? "w":"", is_exec? "x":""); #endif @@ -219,7 +221,7 @@ bad_page_fault(struct pt_regs *regs, unsigned long address, int sig) /* Are we prepared to handle this kernel fault? */ if ((entry = search_exception_tables(regs->pc)) != NULL) { -#if 1 +#ifdef DEBUG_PAGE_FAULT printk(KERN_DEBUG "%s: Exception at pc=%#010lx (%lx)\n", current->comm, regs->pc, entry->fixup); #endif diff --git a/arch/xtensa/mm/init.c b/arch/xtensa/mm/init.c index 8415c76f11c..b3086f34a8e 100644 --- a/arch/xtensa/mm/init.c +++ b/arch/xtensa/mm/init.c @@ -15,40 +15,24 @@ * Kevin Chea */ -#include -#include -#include #include #include -#include -#include -#include #include #include +#include +#include +#include +#include #include #include #include #include -#include #include #include -#include - -#define DEBUG 0 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); -//static DEFINE_SPINLOCK(tlb_lock); - -/* - * This flag is used to indicate that the page was mapped and modified in - * kernel space, so the cache is probably dirty at that address. - * If cache aliasing is enabled and the page color mismatches, update_mmu_cache - * synchronizes the caches if this bit is set. - */ - -#define PG_cache_clean PG_arch_1 /* References to section boundaries */ @@ -323,228 +307,22 @@ void show_mem(void) printk("%d free pages\n", free); } -/* ------------------------------------------------------------------------- */ - -#if (DCACHE_WAY_SIZE > PAGE_SIZE) - -/* - * With cache aliasing, the page color of the page in kernel space and user - * space might mismatch. We temporarily map the page to a different virtual - * address with the same color and clear the page there. - */ - -void clear_user_page(void *kaddr, unsigned long vaddr, struct page* page) -{ - - /* There shouldn't be any entries for this page. */ - - __flush_invalidate_dcache_page_phys(__pa(page_address(page))); - - if (!PAGE_COLOR_EQ(vaddr, kaddr)) { - unsigned long v, p; - - /* Temporarily map page to DTLB_WAY_DCACHE_ALIAS0. */ - - spin_lock(&tlb_lock); - - p = (unsigned long)pte_val((mk_pte(page,PAGE_KERNEL))); - kaddr = (void*)PAGE_COLOR_MAP0(vaddr); - v = (unsigned long)kaddr | DTLB_WAY_DCACHE_ALIAS0; - __asm__ __volatile__("wdtlb %0,%1; dsync" : :"a" (p), "a" (v)); - - clear_page(kaddr); - - spin_unlock(&tlb_lock); - } else { - clear_page(kaddr); - } - - /* We need to make sure that i$ and d$ are coherent. */ - - clear_bit(PG_cache_clean, &page->flags); -} - -/* - * With cache aliasing, we have to make sure that the page color of the page - * in kernel space matches that of the virtual user address before we read - * the page. If the page color differ, we create a temporary DTLB entry with - * the corrent page color and use this 'temporary' address as the source. - * We then use the same approach as in clear_user_page and copy the data - * to the kernel space and clear the PG_cache_clean bit to synchronize caches - * later. - * - * Note: - * Instead of using another 'way' for the temporary DTLB entry, we could - * probably use the same entry that points to the kernel address (after - * saving the original value and restoring it when we are done). - */ +struct kmem_cache *pgtable_cache __read_mostly; -void copy_user_page(void* to, void* from, unsigned long vaddr, - struct page* to_page) +static void pgd_ctor(void *addr, struct kmem_cache *cache, unsigned long flags) { - /* There shouldn't be any entries for the new page. */ - - __flush_invalidate_dcache_page_phys(__pa(page_address(to_page))); - - spin_lock(&tlb_lock); - - if (!PAGE_COLOR_EQ(vaddr, from)) { - unsigned long v, p, t; - - __asm__ __volatile__ ("pdtlb %1,%2; rdtlb1 %0,%1" - : "=a"(p), "=a"(t) : "a"(from)); - from = (void*)PAGE_COLOR_MAP0(vaddr); - v = (unsigned long)from | DTLB_WAY_DCACHE_ALIAS0; - __asm__ __volatile__ ("wdtlb %0,%1; dsync" ::"a" (p), "a" (v)); - } - - if (!PAGE_COLOR_EQ(vaddr, to)) { - unsigned long v, p; - - p = (unsigned long)pte_val((mk_pte(to_page,PAGE_KERNEL))); - to = (void*)PAGE_COLOR_MAP1(vaddr); - v = (unsigned long)to | DTLB_WAY_DCACHE_ALIAS1; - __asm__ __volatile__ ("wdtlb %0,%1; dsync" ::"a" (p), "a" (v)); - } - copy_page(to, from); - - spin_unlock(&tlb_lock); - - /* We need to make sure that i$ and d$ are coherent. */ - - clear_bit(PG_cache_clean, &to_page->flags); -} - - - -/* - * Any time the kernel writes to a user page cache page, or it is about to - * read from a page cache page this routine is called. - * - * Note: - * The kernel currently only provides one architecture bit in the page - * flags that we use for I$/D$ coherency. Maybe, in future, we can - * use a sepearte bit for deferred dcache aliasing: - * If the page is not mapped yet, we only need to set a flag, - * if mapped, we need to invalidate the page. - */ -// FIXME: we probably need this for WB caches not only for Page Coloring.. - -void flush_dcache_page(struct page *page) -{ - unsigned long addr = __pa(page_address(page)); - struct address_space *mapping = page_mapping(page); - - __flush_invalidate_dcache_page_phys(addr); - - if (!test_bit(PG_cache_clean, &page->flags)) - return; - - /* If this page hasn't been mapped, yet, handle I$/D$ coherency later.*/ -#if 0 - if (mapping && !mapping_mapped(mapping)) - clear_bit(PG_cache_clean, &page->flags); - else -#endif - __invalidate_icache_page_phys(addr); -} - -void flush_cache_range(struct vm_area_struct* vma, unsigned long s, - unsigned long e) -{ - __flush_invalidate_cache_all(); -} - -void flush_cache_page(struct vm_area_struct* vma, unsigned long address, - unsigned long pfn) -{ - struct page *page = pfn_to_page(pfn); - - /* Remove any entry for the old mapping. */ - - if (current->active_mm == vma->vm_mm) { - unsigned long addr = __pa(page_address(page)); - __flush_invalidate_dcache_page_phys(addr); - if ((vma->vm_flags & VM_EXEC) != 0) - __invalidate_icache_page_phys(addr); - } else { - BUG(); - } -} - -#endif /* (DCACHE_WAY_SIZE > PAGE_SIZE) */ - - -pte_t* pte_alloc_one_kernel (struct mm_struct* mm, unsigned long addr) -{ - pte_t* pte = (pte_t*)__get_free_pages(GFP_KERNEL|__GFP_REPEAT, 0); - if (likely(pte)) { - pte_t* ptep = (pte_t*)(pte_val(*pte) + PAGE_OFFSET); - int i; - for (i = 0; i < 1024; i++, ptep++) - pte_clear(mm, addr, ptep); - } - return pte; -} - -struct page* pte_alloc_one(struct mm_struct *mm, unsigned long addr) -{ - struct page *page; - - page = alloc_pages(GFP_KERNEL | __GFP_REPEAT, 0); - - if (likely(page)) { - pte_t* ptep = kmap_atomic(page, KM_USER0); - int i; + pte_t* ptep = (pte_t*)addr; + int i; - for (i = 0; i < 1024; i++, ptep++) - pte_clear(mm, addr, ptep); + for (i = 0; i < 1024; i++, ptep++) + pte_clear(NULL, 0, ptep); - kunmap_atomic(ptep, KM_USER0); - } - return page; } - -/* - * Handle D$/I$ coherency. - * - * Note: - * We only have one architecture bit for the page flags, so we cannot handle - * cache aliasing, yet. - */ - -void -update_mmu_cache(struct vm_area_struct * vma, unsigned long addr, pte_t pte) +void __init pgtable_cache_init(void) { - unsigned long pfn = pte_pfn(pte); - struct page *page; - unsigned long vaddr = addr & PAGE_MASK; - - if (!pfn_valid(pfn)) - return; - - page = pfn_to_page(pfn); - - invalidate_itlb_mapping(addr); - invalidate_dtlb_mapping(addr); - - /* We have a new mapping. Use it. */ - - write_dtlb_entry(pte, dtlb_probe(addr)); - - /* If the processor can execute from this page, synchronize D$/I$. */ - - if ((vma->vm_flags & VM_EXEC) != 0) { - - write_itlb_entry(pte, itlb_probe(addr)); - - /* Synchronize caches, if not clean. */ - - if (!test_and_set_bit(PG_cache_clean, &page->flags)) { - __flush_dcache_page(vaddr); - __invalidate_icache_page(vaddr); - } - } + pgtable_cache = kmem_cache_create("pgd", + PAGE_SIZE, PAGE_SIZE, + SLAB_HWCACHE_ALIGN, + pgd_ctor); } - diff --git a/arch/xtensa/mm/misc.S b/arch/xtensa/mm/misc.S index ae085332c60..e1f880368e3 100644 --- a/arch/xtensa/mm/misc.S +++ b/arch/xtensa/mm/misc.S @@ -7,29 +7,33 @@ * License. See the file "COPYING" in the main directory of this archive * for more details. * - * Copyright (C) 2001 - 2005 Tensilica Inc. + * Copyright (C) 2001 - 2007 Tensilica Inc. * * Chris Zankel */ -/* Note: we might want to implement some of the loops as zero-overhead-loops, - * where applicable and if supported by the processor. - */ #include #include #include #include #include +#include + -/* clear_page (page) */ +/* + * clear_page and clear_user_page are the same for non-cache-aliased configs. + * + * clear_page (unsigned long page) + * a2 + */ ENTRY(clear_page) entry a1, 16 - addi a4, a2, PAGE_SIZE - movi a3, 0 -1: s32i a3, a2, 0 + movi a3, 0 + __loopi a2, a7, PAGE_SIZE, 32 + s32i a3, a2, 0 s32i a3, a2, 4 s32i a3, a2, 8 s32i a3, a2, 12 @@ -37,42 +41,277 @@ ENTRY(clear_page) s32i a3, a2, 20 s32i a3, a2, 24 s32i a3, a2, 28 - addi a2, a2, 32 - blt a2, a4, 1b + __endla a2, a7, 32 retw /* + * copy_page and copy_user_page are the same for non-cache-aliased configs. + * * copy_page (void *to, void *from) - * a2 a3 + * a2 a3 */ ENTRY(copy_page) entry a1, 16 - addi a4, a2, PAGE_SIZE - -1: l32i a5, a3, 0 - l32i a6, a3, 4 - l32i a7, a3, 8 - s32i a5, a2, 0 - s32i a6, a2, 4 - s32i a7, a2, 8 - l32i a5, a3, 12 - l32i a6, a3, 16 - l32i a7, a3, 20 - s32i a5, a2, 12 - s32i a6, a2, 16 - s32i a7, a2, 20 - l32i a5, a3, 24 - l32i a6, a3, 28 - s32i a5, a2, 24 - s32i a6, a2, 28 - addi a2, a2, 32 - addi a3, a3, 32 - blt a2, a4, 1b + __loopi a2, a4, PAGE_SIZE, 32 + + l32i a8, a3, 0 + l32i a9, a3, 4 + s32i a8, a2, 0 + s32i a9, a2, 4 + + l32i a8, a3, 8 + l32i a9, a3, 12 + s32i a8, a2, 8 + s32i a9, a2, 12 + + l32i a8, a3, 16 + l32i a9, a3, 20 + s32i a8, a2, 16 + s32i a9, a2, 20 + + l32i a8, a3, 24 + l32i a9, a3, 28 + s32i a8, a2, 24 + s32i a9, a2, 28 + + addi a2, a2, 32 + addi a3, a3, 32 + + __endl a2, a4 + + retw + +/* + * If we have to deal with cache aliasing, we use temporary memory mappings + * to ensure that the source and destination pages have the same color as + * the virtual address. We use way 0 and 1 for temporary mappings in such cases. + * + * The temporary DTLB entries shouldn't be flushed by interrupts, but are + * flushed by preemptive task switches. Special code in the + * fast_second_level_miss handler re-established the temporary mapping. + * It requires that the PPNs for the destination and source addresses are + * in a6, and a7, respectively. + */ + +/* TLB miss exceptions are treated special in the following region */ + +ENTRY(__tlbtemp_mapping_start) + +#if (DCACHE_WAY_SIZE > PAGE_SIZE) + +/* + * clear_user_page (void *addr, unsigned long vaddr, struct page *page) + * a2 a3 a4 + */ + +ENTRY(clear_user_page) + entry a1, 32 + + /* Mark page dirty and determine alias. */ + + movi a7, (1 << PG_ARCH_1) + l32i a5, a4, PAGE_FLAGS + xor a6, a2, a3 + extui a3, a3, PAGE_SHIFT, DCACHE_ALIAS_ORDER + extui a6, a6, PAGE_SHIFT, DCACHE_ALIAS_ORDER + or a5, a5, a7 + slli a3, a3, PAGE_SHIFT + s32i a5, a4, PAGE_FLAGS + + /* Skip setting up a temporary DTLB if not aliased. */ + + beqz a6, 1f + + /* Invalidate kernel page. */ + + mov a10, a2 + call8 __invalidate_dcache_page + + /* Setup a temporary DTLB with the color of the VPN */ + + movi a4, -PAGE_OFFSET + (PAGE_KERNEL | _PAGE_HW_WRITE) + movi a5, TLBTEMP_BASE_1 # virt + add a6, a2, a4 # ppn + add a2, a5, a3 # add 'color' + + wdtlb a6, a2 + dsync + +1: movi a3, 0 + __loopi a2, a7, PAGE_SIZE, 32 + s32i a3, a2, 0 + s32i a3, a2, 4 + s32i a3, a2, 8 + s32i a3, a2, 12 + s32i a3, a2, 16 + s32i a3, a2, 20 + s32i a3, a2, 24 + s32i a3, a2, 28 + __endla a2, a7, 32 + + bnez a6, 1f + retw + + /* We need to invalidate the temporary idtlb entry, if any. */ + +1: addi a2, a2, -PAGE_SIZE + idtlb a2 + dsync + + retw + +/* + * copy_page_user (void *to, void *from, unsigned long vaddr, struct page *page) + * a2 a3 a4 a5 + */ + +ENTRY(copy_user_page) + + entry a1, 32 + + /* Mark page dirty and determine alias for destination. */ + + movi a8, (1 << PG_ARCH_1) + l32i a9, a5, PAGE_FLAGS + xor a6, a2, a4 + xor a7, a3, a4 + extui a4, a4, PAGE_SHIFT, DCACHE_ALIAS_ORDER + extui a6, a6, PAGE_SHIFT, DCACHE_ALIAS_ORDER + extui a7, a7, PAGE_SHIFT, DCACHE_ALIAS_ORDER + or a9, a9, a8 + slli a4, a4, PAGE_SHIFT + s32i a9, a5, PAGE_FLAGS + movi a5, -PAGE_OFFSET + (PAGE_KERNEL | _PAGE_HW_WRITE) + + beqz a6, 1f + + /* Invalidate dcache */ + + mov a10, a2 + call8 __invalidate_dcache_page + + /* Setup a temporary DTLB with a matching color. */ + + movi a8, TLBTEMP_BASE_1 # base + add a6, a2, a5 # ppn + add a2, a8, a4 # add 'color' + + wdtlb a6, a2 + dsync + + /* Skip setting up a temporary DTLB for destination if not aliased. */ + +1: beqz a7, 1f + + /* Setup a temporary DTLB with a matching color. */ + + movi a8, TLBTEMP_BASE_2 # base + add a7, a3, a5 # ppn + add a3, a8, a4 + addi a8, a3, 1 # way1 + + wdtlb a7, a8 + dsync + +1: __loopi a2, a4, PAGE_SIZE, 32 + + l32i a8, a3, 0 + l32i a9, a3, 4 + s32i a8, a2, 0 + s32i a9, a2, 4 + + l32i a8, a3, 8 + l32i a9, a3, 12 + s32i a8, a2, 8 + s32i a9, a2, 12 + + l32i a8, a3, 16 + l32i a9, a3, 20 + s32i a8, a2, 16 + s32i a9, a2, 20 + + l32i a8, a3, 24 + l32i a9, a3, 28 + s32i a8, a2, 24 + s32i a9, a2, 28 + + addi a2, a2, 32 + addi a3, a3, 32 + + __endl a2, a4 + + /* We need to invalidate any temporary mapping! */ + + bnez a6, 1f + bnez a7, 2f + retw + +1: addi a2, a2, -PAGE_SIZE + idtlb a2 + dsync + bnez a7, 2f + retw + +2: addi a3, a3, -PAGE_SIZE+1 + idtlb a3 + dsync + + retw + +#endif + +#if (DCACHE_WAY_SIZE > PAGE_SIZE) + +/* + * void __flush_invalidate_dcache_page_alias (addr, phys) + * a2 a3 + */ + +ENTRY(__flush_invalidate_dcache_page_alias) + entry sp, 16 + + movi a7, 0 # required for exception handler + addi a6, a3, (PAGE_KERNEL | _PAGE_HW_WRITE) + mov a4, a2 + wdtlb a6, a2 + dsync + + ___flush_invalidate_dcache_page a2 a3 + + idtlb a4 + dsync + + retw + +#endif + +ENTRY(__tlbtemp_mapping_itlb) + +#if (ICACHE_WAY_SIZE > PAGE_SIZE) + +ENTRY(__invalidate_icache_page_alias) + entry sp, 16 + + addi a6, a3, (PAGE_KERNEL | _PAGE_HW_WRITE) + mov a4, a2 + witlb a6, a2 + isync + + ___invalidate_icache_page a2 a3 + + iitlb a4 + isync retw +#endif + +/* End of special treatment in tlb miss exception */ + +ENTRY(__tlbtemp_mapping_end) + /* * void __invalidate_icache_page(ulong start) */ @@ -121,8 +360,6 @@ ENTRY(__flush_dcache_page) dsync retw - - /* * void __invalidate_icache_range(ulong start, ulong size) */ @@ -168,7 +405,6 @@ ENTRY(__invalidate_dcache_range) ___invalidate_dcache_range a2 a3 a4 - retw /* -- cgit v1.2.3 From 969f865f36b65c2fcd1fb1ba09b4a2c6f071f864 Mon Sep 17 00:00:00 2001 From: Frederik Deweerdt Date: Wed, 22 Aug 2007 10:47:58 -0700 Subject: [patch 2/2] xtensa console.c: remove duplicate #include This patch removes one of the two linux/console.h included in arch/xtensa/platform-iss/console.c Signed-off-by: Frederik Deweerdt Signed-off-by: Andrew Morton Signed-off-by: Chris Zankel --- arch/xtensa/platform-iss/console.c | 1 - 1 file changed, 1 deletion(-) (limited to 'arch') diff --git a/arch/xtensa/platform-iss/console.c b/arch/xtensa/platform-iss/console.c index 2f4f20ffe66..854677d0c3f 100644 --- a/arch/xtensa/platform-iss/console.c +++ b/arch/xtensa/platform-iss/console.c @@ -20,7 +20,6 @@ #include #include #include -#include #include #include -- cgit v1.2.3