From 3547cdb1526aabd38b1e71dba93d545340fe41f8 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Fri, 3 Aug 2007 15:55:07 -0700 Subject: [XTENSA] add missing system calls Add missing system calls that have been recently added to the kernel for the Xtensa architecture and define __IGNORE macros for system calls that we don't need for Xtensa. Signed-off-by: Chris Zankel --- include/asm-xtensa/syscall.h | 22 +++++++- include/asm-xtensa/unistd.h | 122 +++++++++++++++++++++++++++++++++++++++++-- 2 files changed, 140 insertions(+), 4 deletions(-) (limited to 'include') diff --git a/include/asm-xtensa/syscall.h b/include/asm-xtensa/syscall.h index 6cb0d42f11c8..ea9b5132215e 100644 --- a/include/asm-xtensa/syscall.h +++ b/include/asm-xtensa/syscall.h @@ -1,3 +1,13 @@ +/* + * include/asm-xtensa/syscall.h + * + * 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 - 2007 Tensilica Inc. + */ + struct pt_regs; struct sigaction; asmlinkage long xtensa_execve(char*, char**, char**, struct pt_regs*); @@ -17,4 +27,14 @@ asmlinkage long sys_rt_sigaction(int, const struct sigaction __user *, struct sigaction __user *, size_t); -asmlinkage long xtensa_shmat(int shmid, char __user *shmaddr, int shmflg); +asmlinkage long xtensa_shmat(int, char __user *, int); + +/* Should probably move to linux/syscalls.h */ +struct pollfd; +asmlinkage long sys_pselect6(int n, fd_set __user *inp, fd_set __user *outp, + fd_set __user *exp, struct timespec __user *tsp, void __user *sig); +asmlinkage long sys_ppoll(struct pollfd __user *ufds, unsigned int nfds, + struct timespec __user *tsp, const sigset_t __user *sigmask, + size_t sigsetsize); + + diff --git a/include/asm-xtensa/unistd.h b/include/asm-xtensa/unistd.h index 9bd34024431c..78ad761e6d5c 100644 --- a/include/asm-xtensa/unistd.h +++ b/include/asm-xtensa/unistd.h @@ -577,7 +577,112 @@ __SYSCALL(258, sys_keyctl, 5) #define __NR_available259 259 __SYSCALL(259, sys_ni_syscall, 0) -#define __NR_syscall_count 261 + +#define __NR_readahead 260 +__SYSCALL(260, sys_readahead, 5) +#define __NR_remap_file_pages 261 +__SYSCALL(261, sys_remap_file_pages, 5) +#define __NR_migrate_pages 262 +__SYSCALL(262, sys_migrate_pages, 0) +#define __NR_mbind 263 +__SYSCALL(263, sys_mbind, 6) +#define __NR_get_mempolicy 264 +__SYSCALL(264, sys_get_mempolicy, 5) +#define __NR_set_mempolicy 265 +__SYSCALL(265, sys_set_mempolicy, 3) +#define __NR_unshare 266 +__SYSCALL(266, sys_unshare, 1) +#define __NR_move_pages 267 +__SYSCALL(267, sys_move_pages, 0) +#define __NR_splice 268 +__SYSCALL(268, sys_splice, 0) +#define __NR_tee 269 +__SYSCALL(269, sys_tee, 0) +#define __NR_vmsplice 270 +__SYSCALL(270, sys_vmsplice, 0) +#define __NR_available271 271 +__SYSCALL(271, sys_ni_syscall, 0) + +#define __NR_pselect6 272 +__SYSCALL(272, sys_pselect6, 0) +#define __NR_ppoll 273 +__SYSCALL(273, sys_ppoll, 0) +#define __NR_epoll_pwait 274 +__SYSCALL(274, sys_epoll_pwait, 0) +#define __NR_available275 275 +__SYSCALL(275, sys_ni_syscall, 0) + +#define __NR_inotify_init 276 +__SYSCALL(276, sys_inotify_init, 0) +#define __NR_inotify_add_watch 277 +__SYSCALL(277, sys_inotify_add_watch, 3) +#define __NR_inotify_rm_watch 278 +__SYSCALL(278, sys_inotify_rm_watch, 2) +#define __NR_available279 279 +__SYSCALL(279, sys_ni_syscall, 0) + +#define __NR_getcpu 280 +__SYSCALL(280, sys_getcpu, 0) +#define __NR_kexec_load 281 +__SYSCALL(281, sys_ni_syscall, 0) + +#define __NR_ioprio_set 282 +__SYSCALL(282, sys_ioprio_set, 2) +#define __NR_ioprio_get 283 +__SYSCALL(283, sys_ioprio_get, 3) + +#define __NR_set_robust_list 284 +__SYSCALL(284, sys_set_robust_list, 3) +#define __NR_get_robust_list 285 +__SYSCALL(285, sys_get_robust_list, 3) +#define __NR_reserved286 286 /* sync_file_rangeX */ +__SYSCALL(286, sys_ni_syscall, 3) +#define __NR_available287 287 +__SYSCALL(287, sys_faccessat, 0) + +/* Relative File Operations */ + +#define __NR_openat 288 +__SYSCALL(288, sys_openat, 4) +#define __NR_mkdirat 289 +__SYSCALL(289, sys_mkdirat, 3) +#define __NR_mknodat 290 +__SYSCALL(290, sys_mknodat, 4) +#define __NR_unlinkat 291 +__SYSCALL(291, sys_unlinkat, 3) +#define __NR_renameat 292 +__SYSCALL(292, sys_renameat, 4) +#define __NR_linkat 293 +__SYSCALL(293, sys_linkat, 5) +#define __NR_symlinkat 294 +__SYSCALL(294, sys_symlinkat, 3) +#define __NR_readlinkat 295 +__SYSCALL(295, sys_readlinkat, 4) +#define __NR_utimensat 296 +__SYSCALL(296, sys_utimensat, 0) +#define __NR_fchownat 297 +__SYSCALL(297, sys_fchownat, 5) +#define __NR_futimesat 298 +__SYSCALL(298, sys_futimesat, 4) +#define __NR_fstatat64 299 +__SYSCALL(299, sys_fstatat64, 0) +#define __NR_fchmodat 300 +__SYSCALL(300, sys_fchmodat, 4) +#define __NR_faccessat 301 +__SYSCALL(301, sys_faccessat, 4) +#define __NR_available302 302 +__SYSCALL(302, sys_ni_syscall, 0) +#define __NR_available303 303 +__SYSCALL(303, sys_ni_syscall, 0) + +#define __NR_signalfd 304 +__SYSCALL(304, sys_signalfd, 3) +#define __NR_timerfd 305 +__SYSCALL(305, sys_timerfd, 4) +#define __NR_eventfd 306 +__SYSCALL(306, sys_eventfd, 1) + +#define __NR_syscall_count 307 /* * sysxtensa syscall handler @@ -612,8 +717,19 @@ __SYSCALL(259, sys_ni_syscall, 0) #define __ARCH_WANT_SYS_LLSEEK #define __ARCH_WANT_SYS_RT_SIGACTION #define __ARCH_WANT_SYS_RT_SIGSUSPEND +#define __ARCH_WANT_SYS_GETPGRP -#endif /* __KERNEL__ */ +/* + * Ignore legacy system calls in the checksyscalls.sh script + */ -#endif /* _XTENSA_UNISTD_H */ +#define __IGNORE_fork /* use clone */ +#define __IGNORE_time +#define __IGNORE_alarm /* use setitimer */ +#define __IGNORE_pause +#define __IGNORE_mmap /* use mmap2 */ +#define __IGNORE_vfork /* use clone */ +#define __IGNORE_fadvise64 /* use fadvise64_64 */ +#endif /* __KERNEL__ */ +#endif /* _XTENSA_UNISTD_H */ -- cgit v1.2.3 From cef9287ead2775e4d7eaa280a73ee99d7269c3e4 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Sat, 4 Aug 2007 09:20:43 -0700 Subject: [XTENSA] Add getpgrp system-call to unistd.h Although __ARCH_WANT_SYS_GETPGRP was defined, the actualy entry for the getpgrp system-call was missing. Signed-off-by: Chris Zankel --- include/asm-xtensa/unistd.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'include') diff --git a/include/asm-xtensa/unistd.h b/include/asm-xtensa/unistd.h index 78ad761e6d5c..bed04102407b 100644 --- a/include/asm-xtensa/unistd.h +++ b/include/asm-xtensa/unistd.h @@ -339,8 +339,8 @@ __SYSCALL(148, sys_setpgid, 2) __SYSCALL(149, sys_getpgid, 1) #define __NR_getppid 150 __SYSCALL(150, sys_getppid, 0) -#define __NR_available151 151 -__SYSCALL(151, sys_ni_syscall, 0) +#define __NR_getpgrp 151 +__SYSCALL(151, sys_getpgrp, 0) #define __NR_reserved152 152 /* set_thread_area */ __SYSCALL(152, sys_ni_syscall, 0) -- cgit v1.2.3 From 068732d9b465cafcc56a745c3931f2fc4e58f48f Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Sun, 5 Aug 2007 10:16:54 -0700 Subject: [XTENSA] Remove extraneous include statement We don't need to include processor.h in bug.h. Signed-off-by: Chris Zankel --- include/asm-xtensa/bugs.h | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-) (limited to 'include') diff --git a/include/asm-xtensa/bugs.h b/include/asm-xtensa/bugs.h index c42285320133..69b29d198249 100644 --- a/include/asm-xtensa/bugs.h +++ b/include/asm-xtensa/bugs.h @@ -13,10 +13,6 @@ #ifndef _XTENSA_BUGS_H #define _XTENSA_BUGS_H -#include - -static void __init check_bugs(void) -{ -} +static void check_bugs(void) { } #endif /* _XTENSA_BUGS_H */ -- 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 +++++ include/asm-xtensa/syscall.h | 2 ++ include/asm-xtensa/unistd.h | 2 +- 3 files changed, 8 insertions(+), 1 deletion(-) (limited to 'include') diff --git a/arch/xtensa/kernel/syscall.c b/arch/xtensa/kernel/syscall.c index fe3834bc1dbf..f3e16efcd47a 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); +} + diff --git a/include/asm-xtensa/syscall.h b/include/asm-xtensa/syscall.h index ea9b5132215e..05cebf8f62b1 100644 --- a/include/asm-xtensa/syscall.h +++ b/include/asm-xtensa/syscall.h @@ -28,6 +28,8 @@ asmlinkage long sys_rt_sigaction(int, struct sigaction __user *, size_t); asmlinkage long xtensa_shmat(int, char __user *, int); +asmlinkage long xtensa_fadvise64_64(int, int, + unsigned long long, unsigned long long); /* Should probably move to linux/syscalls.h */ struct pollfd; diff --git a/include/asm-xtensa/unistd.h b/include/asm-xtensa/unistd.h index bed04102407b..92968aabe34e 100644 --- a/include/asm-xtensa/unistd.h +++ b/include/asm-xtensa/unistd.h @@ -151,7 +151,7 @@ __SYSCALL( 61, sys_fcntl64, 3) #define __NR_available62 62 __SYSCALL( 62, sys_ni_syscall, 0) #define __NR_fadvise64_64 63 -__SYSCALL( 63, sys_fadvise64_64, 6) +__SYSCALL( 63, xtensa_fadvise64_64, 6) #define __NR_utime 64 /* glibc 2.3.3 ?? */ __SYSCALL( 64, sys_utime, 2) #define __NR_utimes 65 -- 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 +++++++++++++++++++++++++++++----------------- include/asm-xtensa/timex.h | 4 ++-- 2 files changed, 31 insertions(+), 19 deletions(-) (limited to 'include') diff --git a/arch/xtensa/kernel/time.c b/arch/xtensa/kernel/time.c index 22949be4a5d8..60d29fe0b1bd 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; } diff --git a/include/asm-xtensa/timex.h b/include/asm-xtensa/timex.h index 28c7985a4000..a5fca59fba9e 100644 --- a/include/asm-xtensa/timex.h +++ b/include/asm-xtensa/timex.h @@ -41,10 +41,10 @@ extern unsigned long ccount_per_jiffy; extern unsigned long ccount_nsec; #define CCOUNT_PER_JIFFY ccount_per_jiffy -#define CCOUNT_NSEC ccount_nsec +#define NSEC_PER_CCOUNT ccount_nsec #else #define CCOUNT_PER_JIFFY (CONFIG_XTENSA_CPU_CLOCK*(1000000UL/HZ)) -#define CCOUNT_NSEC (1000000000UL / CONFIG_XTENSA_CPU_CLOCK) +#define NSEC_PER_CCOUNT (1000UL / CONFIG_XTENSA_CPU_CLOCK) #endif -- cgit v1.2.3 From 24a9ab7fa143498802b35c66de10cc3cfdac2c51 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Sun, 5 Aug 2007 11:24:13 -0700 Subject: [XTENSA] Add typecast macro for constants Add macros (__XTENSA_UL and __XTENSA_UL_CONST) for typecasting constants. Signed-off-by: Chris Zankel --- include/asm-xtensa/processor.h | 2 +- include/asm-xtensa/types.h | 9 +++++++++ 2 files changed, 10 insertions(+), 1 deletion(-) (limited to 'include') diff --git a/include/asm-xtensa/processor.h b/include/asm-xtensa/processor.h index 4feb9f7f35a6..35145bcd96eb 100644 --- a/include/asm-xtensa/processor.h +++ b/include/asm-xtensa/processor.h @@ -33,7 +33,7 @@ * the 1 GB requirement applies to the stack as well. */ -#define TASK_SIZE 0x40000000 +#define TASK_SIZE __XTENSA_UL_CONST(0x40000000) /* * General exception cause assigned to debug exceptions. Debug exceptions go diff --git a/include/asm-xtensa/types.h b/include/asm-xtensa/types.h index 9d99a8e9e337..f1e84526f999 100644 --- a/include/asm-xtensa/types.h +++ b/include/asm-xtensa/types.h @@ -11,6 +11,15 @@ #ifndef _XTENSA_TYPES_H #define _XTENSA_TYPES_H + +#ifdef __ASSEMBLY__ +# define __XTENSA_UL(x) (x) +# define __XTENSA_UL_CONST(x) x +#else +# define __XTENSA_UL(x) ((unsigned long)(x)) +# define __XTENSA_UL_CONST(x) x##UL +#endif + #ifndef __ASSEMBLY__ typedef unsigned short umode_t; -- cgit v1.2.3 From 26465f2f4f5a253f22596fc9245a6bb5c0856ee1 Mon Sep 17 00:00:00 2001 From: Chris Zankel Date: Mon, 6 Aug 2007 23:12:24 -0700 Subject: [XTENSA] Use the generic version of get_order Use the generic version of get_order for processor configurations that don't have the 'nsa/nsau' instructions. Signed-off-by: Chris Zankel --- include/asm-xtensa/page.h | 50 +++++++++++++++++++++-------------------------- 1 file changed, 22 insertions(+), 28 deletions(-) (limited to 'include') diff --git a/include/asm-xtensa/page.h b/include/asm-xtensa/page.h index 1213cde75438..2d6ac21136cf 100644 --- a/include/asm-xtensa/page.h +++ b/include/asm-xtensa/page.h @@ -1,11 +1,11 @@ /* - * linux/include/asm-xtensa/page.h + * include/asm-xtensa/page.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 * published by the Free Software Foundation. * - * Copyright (C) 2001 - 2005 Tensilica Inc. + * Copyright (C) 2001 - 2007 Tensilica Inc. */ #ifndef _XTENSA_PAGE_H @@ -14,6 +14,11 @@ #ifdef __KERNEL__ #include +#include + +/* + * Fixed TLB translations in the processor. + */ #define XCHAL_KSEG_CACHED_VADDR 0xd0000000 #define XCHAL_KSEG_BYPASS_VADDR 0xd8000000 @@ -26,13 +31,13 @@ */ #define PAGE_SHIFT 12 -#define PAGE_SIZE (1 << PAGE_SHIFT) +#define PAGE_SIZE (__XTENSA_UL_CONST(1) << PAGE_SHIFT) #define PAGE_MASK (~(PAGE_SIZE-1)) #define PAGE_ALIGN(addr) (((addr)+PAGE_SIZE - 1) & PAGE_MASK) #define PAGE_OFFSET XCHAL_KSEG_CACHED_VADDR -#define MAX_MEM_PFN XCHAL_KSEG_SIZE -#define PGTABLE_START 0x80000000 +#define MAX_MEM_PFN XCHAL_KSEG_SIZE +#define PGTABLE_START 0x80000000 #ifdef __ASSEMBLY__ @@ -58,34 +63,23 @@ typedef struct { unsigned long pgprot; } pgprot_t; /* * Pure 2^n version of get_order + * Use 'nsau' instructions if supported by the processor or the generic version. */ -static inline int get_order(unsigned long size) +#if XCHAL_HAVE_NSA + +static inline __attribute_const__ int get_order(unsigned long size) { - int order; -#ifndef XCHAL_HAVE_NSU - unsigned long x1, x2, x4, x8, x16; - - size = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; - x1 = size & 0xAAAAAAAA; - x2 = size & 0xCCCCCCCC; - x4 = size & 0xF0F0F0F0; - x8 = size & 0xFF00FF00; - x16 = size & 0xFFFF0000; - order = x2 ? 2 : 0; - order += (x16 != 0) * 16; - order += (x8 != 0) * 8; - order += (x4 != 0) * 4; - order += (x1 != 0); - - return order; -#else - size = (size - 1) >> PAGE_SHIFT; - asm ("nsau %0, %1" : "=r" (order) : "r" (size)); - return 32 - order; -#endif + int lz; + asm ("nsau %0, %1" : "=r" (lz) : "r" ((size - 1) >> PAGE_SHIFT)); + return 32 - lz; } +#else + +# include + +#endif struct page; extern void clear_page(void *page); -- 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 ++++++--- include/asm-xtensa/pgtable.h | 213 ++++++++++++++++++++----------------------- 2 files changed, 129 insertions(+), 126 deletions(-) (limited to 'include') 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__) */ -- 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 ++++++++++++++++++++++++++++++++++++++++---- include/asm-xtensa/elf.h | 50 ++++++++++++ 2 files changed, 227 insertions(+), 18 deletions(-) (limited to 'include') diff --git a/arch/xtensa/kernel/module.c b/arch/xtensa/kernel/module.c index 2ea1755a0858..ddf14dcf2ad9 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"); } diff --git a/include/asm-xtensa/elf.h b/include/asm-xtensa/elf.h index 1569b53cec91..7083d46766a8 100644 --- a/include/asm-xtensa/elf.h +++ b/include/asm-xtensa/elf.h @@ -20,6 +20,56 @@ #define EM_XTENSA 94 #define EM_XTENSA_OLD 0xABC7 +/* Xtensa relocations defined by the ABIs */ + +#define R_XTENSA_NONE 0 +#define R_XTENSA_32 1 +#define R_XTENSA_RTLD 2 +#define R_XTENSA_GLOB_DAT 3 +#define R_XTENSA_JMP_SLOT 4 +#define R_XTENSA_RELATIVE 5 +#define R_XTENSA_PLT 6 +#define R_XTENSA_OP0 8 +#define R_XTENSA_OP1 9 +#define R_XTENSA_OP2 10 +#define R_XTENSA_ASM_EXPAND 11 +#define R_XTENSA_ASM_SIMPLIFY 12 +#define R_XTENSA_GNU_VTINHERIT 15 +#define R_XTENSA_GNU_VTENTRY 16 +#define R_XTENSA_DIFF8 17 +#define R_XTENSA_DIFF16 18 +#define R_XTENSA_DIFF32 19 +#define R_XTENSA_SLOT0_OP 20 +#define R_XTENSA_SLOT1_OP 21 +#define R_XTENSA_SLOT2_OP 22 +#define R_XTENSA_SLOT3_OP 23 +#define R_XTENSA_SLOT4_OP 24 +#define R_XTENSA_SLOT5_OP 25 +#define R_XTENSA_SLOT6_OP 26 +#define R_XTENSA_SLOT7_OP 27 +#define R_XTENSA_SLOT8_OP 28 +#define R_XTENSA_SLOT9_OP 29 +#define R_XTENSA_SLOT10_OP 30 +#define R_XTENSA_SLOT11_OP 31 +#define R_XTENSA_SLOT12_OP 32 +#define R_XTENSA_SLOT13_OP 33 +#define R_XTENSA_SLOT14_OP 34 +#define R_XTENSA_SLOT0_ALT 35 +#define R_XTENSA_SLOT1_ALT 36 +#define R_XTENSA_SLOT2_ALT 37 +#define R_XTENSA_SLOT3_ALT 38 +#define R_XTENSA_SLOT4_ALT 39 +#define R_XTENSA_SLOT5_ALT 40 +#define R_XTENSA_SLOT6_ALT 41 +#define R_XTENSA_SLOT7_ALT 42 +#define R_XTENSA_SLOT8_ALT 43 +#define R_XTENSA_SLOT9_ALT 44 +#define R_XTENSA_SLOT10_ALT 45 +#define R_XTENSA_SLOT11_ALT 46 +#define R_XTENSA_SLOT12_ALT 47 +#define R_XTENSA_SLOT13_ALT 48 +#define R_XTENSA_SLOT14_ALT 49 + /* ELF register definitions. This is needed for core dump support. */ /* -- 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 ++++++++++++++++++++++++++++++++++----- include/asm-xtensa/cache.h | 9 ++ include/asm-xtensa/cacheflush.h | 81 +++++++---- include/asm-xtensa/io.h | 1 + include/asm-xtensa/page.h | 56 ++++++- include/asm-xtensa/pgalloc.h | 107 ++++---------- include/asm-xtensa/pgtable.h | 32 ++-- include/asm-xtensa/tlb.h | 30 +++- 14 files changed, 816 insertions(+), 414 deletions(-) create mode 100644 arch/xtensa/mm/cache.c (limited to 'include') diff --git a/arch/xtensa/kernel/asm-offsets.c b/arch/xtensa/kernel/asm-offsets.c index d0323cd6a2ea..d5ffe7b6443e 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 65741e3368e7..91a689eca43d 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 a5aed5932d7b..10aec22a8f98 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 000000000000..9a1fa9478ae7 --- /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 16004067add3..45d28f217c03 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 8415c76f11c2..b3086f34a8e7 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 ae085332c607..e1f880368e32 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 /* diff --git a/include/asm-xtensa/cache.h b/include/asm-xtensa/cache.h index 1c4a78f29ae2..3bba2a540cf0 100644 --- a/include/asm-xtensa/cache.h +++ b/include/asm-xtensa/cache.h @@ -19,6 +19,15 @@ #define DCACHE_WAY_SIZE (XCHAL_DCACHE_SIZE/XCHAL_DCACHE_WAYS) #define ICACHE_WAY_SIZE (XCHAL_ICACHE_SIZE/XCHAL_ICACHE_WAYS) +#define DCACHE_WAY_SHIFT (XCHAL_DCACHE_SETWIDTH + XCHAL_DCACHE_LINEWIDTH) +#define ICACHE_WAY_SHIFT (XCHAL_ICACHE_SETWIDTH + XCHAL_ICACHE_LINEWIDTH) + +/* Maximum cache size per way. */ +#if DCACHE_WAY_SIZE >= ICACHE_WAY_SIZE +# define CACHE_WAY_SIZE DCACHE_WAY_SIZE +#else +# define CACHE_WAY_SIZE ICACHE_WAY_SIZE +#endif #endif /* _XTENSA_CACHE_H */ diff --git a/include/asm-xtensa/cacheflush.h b/include/asm-xtensa/cacheflush.h index 22ef901b7845..b773c57e75a5 100644 --- a/include/asm-xtensa/cacheflush.h +++ b/include/asm-xtensa/cacheflush.h @@ -5,7 +5,7 @@ * License. See the file "COPYING" in the main directory of this archive * for more details. * - * (C) 2001 - 2006 Tensilica Inc. + * (C) 2001 - 2007 Tensilica Inc. */ #ifndef _XTENSA_CACHEFLUSH_H @@ -18,10 +18,7 @@ #include /* - * flush and invalidate data cache, invalidate instruction cache: - * - * __flush_invalidate_cache_all() - * __flush_invalidate_cache_range(from,sze) + * Lo-level routines for cache flushing. * * invalidate data or instruction cache: * @@ -40,26 +37,39 @@ * __flush_invalidate_dcache_all() * __flush_invalidate_dcache_page(adr) * __flush_invalidate_dcache_range(from,size) + * + * specials for cache aliasing: + * + * __flush_invalidate_dcache_page_alias(vaddr,paddr) + * __invalidate_icache_page_alias(vaddr,paddr) */ -extern void __flush_invalidate_cache_all(void); -extern void __flush_invalidate_cache_range(unsigned long, unsigned long); -extern void __flush_invalidate_dcache_all(void); +extern void __invalidate_dcache_all(void); extern void __invalidate_icache_all(void); - extern void __invalidate_dcache_page(unsigned long); extern void __invalidate_icache_page(unsigned long); extern void __invalidate_icache_range(unsigned long, unsigned long); extern void __invalidate_dcache_range(unsigned long, unsigned long); + #if XCHAL_DCACHE_IS_WRITEBACK +extern void __flush_invalidate_dcache_all(void); extern void __flush_dcache_page(unsigned long); +extern void __flush_dcache_range(unsigned long, unsigned long); extern void __flush_invalidate_dcache_page(unsigned long); extern void __flush_invalidate_dcache_range(unsigned long, unsigned long); #else -# define __flush_dcache_page(p) do { } while(0) -# define __flush_invalidate_dcache_page(p) do { } while(0) -# define __flush_invalidate_dcache_range(p,s) do { } while(0) +# define __flush_dcache_range(p,s) do { } while(0) +# define __flush_dcache_page(p) do { } while(0) +# define __flush_invalidate_dcache_page(p) __invalidate_dcache_page(p) +# define __flush_invalidate_dcache_range(p,s) __invalidate_dcache_range(p,s) +#endif + +#if (DCACHE_WAY_SIZE > PAGE_SIZE) +extern void __flush_invalidate_dcache_page_alias(unsigned long, unsigned long); +#endif +#if (ICACHE_WAY_SIZE > PAGE_SIZE) +extern void __invalidate_icache_page_alias(unsigned long, unsigned long); #endif /* @@ -71,17 +81,21 @@ extern void __flush_invalidate_dcache_range(unsigned long, unsigned long); * (see also Documentation/cachetlb.txt) */ -#if (DCACHE_WAY_SIZE > PAGE_SIZE) && XCHAL_DCACHE_IS_WRITEBACK +#if (DCACHE_WAY_SIZE > PAGE_SIZE) -#define flush_cache_all() __flush_invalidate_cache_all(); -#define flush_cache_mm(mm) __flush_invalidate_cache_all(); -#define flush_cache_dup_mm(mm) __flush_invalidate_cache_all(); +#define flush_cache_all() \ + do { \ + __flush_invalidate_dcache_all(); \ + __invalidate_icache_all(); \ + } while (0) -#define flush_cache_vmap(start,end) __flush_invalidate_cache_all(); -#define flush_cache_vunmap(start,end) __flush_invalidate_cache_all(); +#define flush_cache_mm(mm) flush_cache_all() +#define flush_cache_dup_mm(mm) flush_cache_mm(mm) -extern void flush_dcache_page(struct page*); +#define flush_cache_vmap(start,end) flush_cache_all() +#define flush_cache_vunmap(start,end) flush_cache_all() +extern void flush_dcache_page(struct page*); extern void flush_cache_range(struct vm_area_struct*, ulong, ulong); extern void flush_cache_page(struct vm_area_struct*, unsigned long, unsigned long); @@ -101,24 +115,39 @@ extern void flush_cache_page(struct vm_area_struct*, unsigned long, unsigned lon #endif +/* Ensure consistency between data and instruction cache. */ #define flush_icache_range(start,end) \ - __invalidate_icache_range(start,(end)-(start)) + do { \ + __flush_dcache_range(start, (end) - (start)); \ + __invalidate_icache_range(start,(end) - (start)); \ + } while (0) /* This is not required, see Documentation/cachetlb.txt */ - -#define flush_icache_page(vma,page) do { } while(0) +#define flush_icache_page(vma,page) do { } while (0) #define flush_dcache_mmap_lock(mapping) do { } while (0) #define flush_dcache_mmap_unlock(mapping) do { } while (0) +#if (DCACHE_WAY_SIZE > PAGE_SIZE) -#define copy_to_user_page(vma, page, vaddr, dst, src, len) \ - memcpy(dst, src, len) +extern void copy_to_user_page(struct vm_area_struct*, struct page*, + unsigned long, void*, const void*, unsigned long); +extern void copy_from_user_page(struct vm_area_struct*, struct page*, + unsigned long, void*, const void*, unsigned long); + +#else + +#define copy_to_user_page(vma, page, vaddr, dst, src, len) \ + do { \ + memcpy(dst, src, len); \ + __flush_dcache_range((unsigned long) dst, len); \ + __invalidate_icache_range((unsigned long) dst, len); \ + } while (0) #define copy_from_user_page(vma, page, vaddr, dst, src, len) \ memcpy(dst, src, len) -#endif /* __KERNEL__ */ +#endif +#endif /* __KERNEL__ */ #endif /* _XTENSA_CACHEFLUSH_H */ - diff --git a/include/asm-xtensa/io.h b/include/asm-xtensa/io.h index 0faa614d9696..47c3616ea9ac 100644 --- a/include/asm-xtensa/io.h +++ b/include/asm-xtensa/io.h @@ -14,6 +14,7 @@ #ifdef __KERNEL__ #include #include +#include #include diff --git a/include/asm-xtensa/page.h b/include/asm-xtensa/page.h index 2d6ac21136cf..55ce2c9749a3 100644 --- a/include/asm-xtensa/page.h +++ b/include/asm-xtensa/page.h @@ -15,6 +15,7 @@ #include #include +#include /* * Fixed TLB translations in the processor. @@ -39,6 +40,53 @@ #define MAX_MEM_PFN XCHAL_KSEG_SIZE #define PGTABLE_START 0x80000000 +/* + * Cache aliasing: + * + * If the cache size for one way is greater than the page size, we have to + * deal with cache aliasing. The cache index is wider than the page size: + * + * | |cache| cache index + * | pfn |off| virtual address + * |xxxx:X|zzz| + * | : | | + * | \ / | | + * |trans.| | + * | / \ | | + * |yyyy:Y|zzz| physical address + * + * When the page number is translated to the physical page address, the lowest + * bit(s) (X) that are part of the cache index are also translated (Y). + * If this translation changes bit(s) (X), the cache index is also afected, + * thus resulting in a different cache line than before. + * The kernel does not provide a mechanism to ensure that the page color + * (represented by this bit) remains the same when allocated or when pages + * are remapped. When user pages are mapped into kernel space, the color of + * the page might also change. + * + * We use the address space VMALLOC_END ... VMALLOC_END + DCACHE_WAY_SIZE * 2 + * to temporarily map a patch so we can match the color. + */ + +#if DCACHE_WAY_SIZE > PAGE_SIZE +# define DCACHE_ALIAS_ORDER (DCACHE_WAY_SHIFT - PAGE_SHIFT) +# define DCACHE_ALIAS_MASK (PAGE_MASK & (DCACHE_WAY_SIZE - 1)) +# define DCACHE_ALIAS(a) (((a) & DCACHE_ALIAS_MASK) >> PAGE_SHIFT) +# define DCACHE_ALIAS_EQ(a,b) ((((a) ^ (b)) & DCACHE_ALIAS_MASK) == 0) +#else +# define DCACHE_ALIAS_ORDER 0 +#endif + +#if ICACHE_WAY_SIZE > PAGE_SIZE +# define ICACHE_ALIAS_ORDER (ICACHE_WAY_SHIFT - PAGE_SHIFT) +# define ICACHE_ALIAS_MASK (PAGE_MASK & (ICACHE_WAY_SIZE - 1)) +# define ICACHE_ALIAS(a) (((a) & ICACHE_ALIAS_MASK) >> PAGE_SHIFT) +# define ICACHE_ALIAS_EQ(a,b) ((((a) ^ (b)) & ICACHE_ALIAS_MASK) == 0) +#else +# define ICACHE_ALIAS_ORDER 0 +#endif + + #ifdef __ASSEMBLY__ #define __pgprot(x) (x) @@ -90,11 +138,11 @@ extern void copy_page(void *to, void *from); * some extra work */ -#if (DCACHE_WAY_SIZE > PAGE_SIZE) -void clear_user_page(void *addr, unsigned long vaddr, struct page* page); -void copy_user_page(void *to,void* from,unsigned long vaddr,struct page* page); +#if DCACHE_WAY_SIZE > PAGE_SIZE +extern void clear_user_page(void*, unsigned long, struct page*); +extern void copy_user_page(void*, void*, unsigned long, struct page*); #else -# define clear_user_page(page,vaddr,pg) clear_page(page) +# define clear_user_page(page, vaddr, pg) clear_page(page) # define copy_user_page(to, from, vaddr, pg) copy_page(to, from) #endif diff --git a/include/asm-xtensa/pgalloc.h b/include/asm-xtensa/pgalloc.h index d56ddf2055e1..3e5b56525102 100644 --- a/include/asm-xtensa/pgalloc.h +++ b/include/asm-xtensa/pgalloc.h @@ -1,11 +1,11 @@ /* - * linux/include/asm-xtensa/pgalloc.h + * include/asm-xtensa/pgalloc.h * * This program is free software; you can redistribute it and/or modify * 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_PGALLOC_H @@ -13,103 +13,54 @@ #ifdef __KERNEL__ -#include #include -#include -#include - - -/* Cache aliasing: - * - * If the cache size for one way is greater than the page size, we have to - * deal with cache aliasing. The cache index is wider than the page size: - * - * |cache | - * |pgnum |page| virtual address - * |xxxxxX|zzzz| - * | | | - * \ / | | - * trans.| | - * / \ | | - * |yyyyyY|zzzz| physical address - * - * When the page number is translated to the physical page address, the lowest - * bit(s) (X) that are also part of the cache index are also translated (Y). - * If this translation changes this bit (X), the cache index is also afected, - * thus resulting in a different cache line than before. - * The kernel does not provide a mechanism to ensure that the page color - * (represented by this bit) remains the same when allocated or when pages - * are remapped. When user pages are mapped into kernel space, the color of - * the page might also change. - * - * We use the address space VMALLOC_END ... VMALLOC_END + DCACHE_WAY_SIZE * 2 - * to temporarily map a patch so we can match the color. - */ - -#if (DCACHE_WAY_SIZE > PAGE_SIZE) -# define PAGE_COLOR_MASK (PAGE_MASK & (DCACHE_WAY_SIZE-1)) -# define PAGE_COLOR(a) \ - (((unsigned long)(a)&PAGE_COLOR_MASK) >> PAGE_SHIFT) -# define PAGE_COLOR_EQ(a,b) \ - ((((unsigned long)(a) ^ (unsigned long)(b)) & PAGE_COLOR_MASK) == 0) -# define PAGE_COLOR_MAP0(v) \ - (VMALLOC_END + ((unsigned long)(v) & PAGE_COLOR_MASK)) -# define PAGE_COLOR_MAP1(v) \ - (VMALLOC_END + ((unsigned long)(v) & PAGE_COLOR_MASK) + DCACHE_WAY_SIZE) -#endif /* * Allocating and freeing a pmd is trivial: the 1-entry pmd is * inside the pgd, so has no extra memory associated with it. */ -#define pgd_free(pgd) free_page((unsigned long)(pgd)) - -#if (DCACHE_WAY_SIZE > PAGE_SIZE) && XCHAL_DCACHE_IS_WRITEBACK +#define pmd_populate_kernel(mm, pmdp, ptep) \ + (pmd_val(*(pmdp)) = ((unsigned long)ptep)) +#define pmd_populate(mm, pmdp, page) \ + (pmd_val(*(pmdp)) = ((unsigned long)page_to_virt(page))) -static inline void -pmd_populate_kernel(struct mm_struct *mm, pmd_t *pmdp, pte_t *pte) +static inline pgd_t* +pgd_alloc(struct mm_struct *mm) { - pmd_val(*(pmdp)) = (unsigned long)(pte); - __asm__ __volatile__ ("memw; dhwb %0, 0; dsync" :: "a" (pmdp)); + return (pgd_t*) __get_free_pages(GFP_KERNEL | __GFP_ZERO, PGD_ORDER); } -static inline void -pmd_populate(struct mm_struct *mm, pmd_t *pmdp, struct page *page) +static inline void pgd_free(pgd_t *pgd) { - pmd_val(*(pmdp)) = (unsigned long)page_to_virt(page); - __asm__ __volatile__ ("memw; dhwb %0, 0; dsync" :: "a" (pmdp)); + free_page((unsigned long)pgd); } +/* Use a slab cache for the pte pages (see also sparc64 implementation) */ +extern struct kmem_cache *pgtable_cache; -#else - -# define pmd_populate_kernel(mm, pmdp, pte) \ - (pmd_val(*(pmdp)) = (unsigned long)(pte)) -# define pmd_populate(mm, pmdp, page) \ - (pmd_val(*(pmdp)) = (unsigned long)page_to_virt(page)) - -#endif - -static inline pgd_t* -pgd_alloc(struct mm_struct *mm) +static inline pte_t *pte_alloc_one_kernel(struct mm_struct *mm, + unsigned long address) { - pgd_t *pgd; - - pgd = (pgd_t *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, PGD_ORDER); - - if (likely(pgd != NULL)) - __flush_dcache_page((unsigned long)pgd); + return kmem_cache_alloc(pgtable_cache, GFP_KERNEL|__GFP_REPEAT); +} - return pgd; +static inline struct page *pte_alloc_one(struct mm_struct *mm, + unsigned long addr) +{ + return virt_to_page(pte_alloc_one_kernel(mm, addr)); } -extern pte_t* pte_alloc_one_kernel(struct mm_struct* mm, unsigned long addr); -extern struct page* pte_alloc_one(struct mm_struct* mm, unsigned long addr); +static inline void pte_free_kernel(pte_t *pte) +{ + kmem_cache_free(pgtable_cache, pte); +} -#define pte_free_kernel(pte) free_page((unsigned long)pte) -#define pte_free(pte) __free_page(pte) +static inline void pte_free(struct page *page) +{ + kmem_cache_free(pgtable_cache, page_address(page)); +} #endif /* __KERNEL__ */ #endif /* _XTENSA_PGALLOC_H */ diff --git a/include/asm-xtensa/pgtable.h b/include/asm-xtensa/pgtable.h index 667a6c46b5a1..c0fcc1c9660c 100644 --- a/include/asm-xtensa/pgtable.h +++ b/include/asm-xtensa/pgtable.h @@ -1,5 +1,5 @@ /* - * linux/include/asm-xtensa/pgtable.h + * 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 version 2 as @@ -60,16 +60,20 @@ #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 +/* + * Virtual memory area. We keep a distance to other memory regions to be * on the safe side. We also use this area for cache aliasing. */ -// FIXME: virtual memory area must be configuration-dependent - #define VMALLOC_START 0xC0000000 -#define VMALLOC_END 0xC7FF0000 +#define VMALLOC_END 0xC6FEFFFF +#define TLBTEMP_BASE_1 0xC6FF0000 +#define TLBTEMP_BASE_2 0xC6FF8000 +#define MODULE_START 0xC7000000 +#define MODULE_END 0xC7FFFFFF -/* Xtensa Linux config PTE layout (when present): +/* + * Xtensa Linux config PTE layout (when present): * 31-12: PPN * 11-6: Software * 5-4: RING @@ -126,12 +130,13 @@ #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) +#define PAGE_KERNEL __pgprot(_PAGE_PRESENT | _PAGE_HW_WRITE) +#define PAGE_KERNEL_EXEC __pgprot(_PAGE_PRESENT|_PAGE_HW_WRITE|_PAGE_HW_EXEC) #if (DCACHE_WAY_SIZE > PAGE_SIZE) -# define _PAGE_DIRECTORY (_PAGE_VALID | _PAGE_ACCESSED | _PAGE_HW_WRITE) +# define _PAGE_DIRECTORY (_PAGE_VALID | _PAGE_ACCESSED) #else -# define _PAGE_DIRECTORY (_PAGE_VALID|_PAGE_ACCESSED|_PAGE_HW_WRITE|_PAGE_CA_WB) +# define _PAGE_DIRECTORY (_PAGE_VALID | _PAGE_ACCESSED | _PAGE_CA_WB) #endif #else /* no mmu */ @@ -244,6 +249,10 @@ 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__ ("dhwb %0, 0" :: "a" (ptep)); +#endif + } struct mm_struct; @@ -383,13 +392,12 @@ extern void update_mmu_cache(struct vm_area_struct * vma, * remap a physical page `pfn' of size `size' with page protection `prot' * into virtual address `from' */ + #define io_remap_pfn_range(vma,from,pfn,size,prot) \ remap_pfn_range(vma, from, pfn, size, prot) -/* No page table caches to init */ - -#define pgtable_cache_init() do { } while (0) +extern void pgtable_cache_init(void); typedef pte_t *pte_addr_t; diff --git a/include/asm-xtensa/tlb.h b/include/asm-xtensa/tlb.h index 4562b2dcfbc0..4830232017af 100644 --- a/include/asm-xtensa/tlb.h +++ b/include/asm-xtensa/tlb.h @@ -11,14 +11,36 @@ #ifndef _XTENSA_TLB_H #define _XTENSA_TLB_H -#define tlb_start_vma(tlb,vma) do { } while (0) -#define tlb_end_vma(tlb,vma) do { } while (0) -#define __tlb_remove_tlb_entry(tlb,pte,addr) do { } while (0) +#include +#include + +#if (DCACHE_WAY_SIZE <= PAGE_SIZE) + +/* Note, read http://lkml.org/lkml/2004/1/15/6 */ + +# define tlb_start_vma(tlb,vma) do { } while (0) +# define tlb_end_vma(tlb,vma) do { } while (0) + +#else +# define tlb_start_vma(tlb, vma) \ + do { \ + if (!tlb->fullmm) \ + flush_cache_range(vma, vma->vm_start, vma->vm_end); \ + } while(0) + +# define tlb_end_vma(tlb, vma) \ + do { \ + if (!tlb->fullmm) \ + flush_tlb_range(vma, vma->vm_start, vma->vm_end); \ + } while(0) + +#endif + +#define __tlb_remove_tlb_entry(tlb,pte,addr) do { } while (0) #define tlb_flush(tlb) flush_tlb_mm((tlb)->mm) #include -#include #define __pte_free_tlb(tlb,pte) pte_free(pte) -- cgit v1.2.3 From ebb2a97b2e7422176d52f4f33e3ee400653875b4 Mon Sep 17 00:00:00 2001 From: Alan Cox Date: Wed, 22 Aug 2007 10:54:43 -0700 Subject: [patch 1/2] Xtensa: enable arbitary tty speed setting ioctls Adding the defines/constants activates the existing code in the tty layer and allows arbitary tty speeds to be requested on this platform Signed-off-by: Alan Cox Signed-off-by: Andrew Morton Signed-off-by: Chris Zankel --- include/asm-xtensa/ioctls.h | 4 ++++ include/asm-xtensa/termbits.h | 5 ++++- include/asm-xtensa/termios.h | 6 ++++-- 3 files changed, 12 insertions(+), 3 deletions(-) (limited to 'include') diff --git a/include/asm-xtensa/ioctls.h b/include/asm-xtensa/ioctls.h index 39e6f23921bb..0ffa942954b9 100644 --- a/include/asm-xtensa/ioctls.h +++ b/include/asm-xtensa/ioctls.h @@ -91,6 +91,10 @@ #define TIOCSBRK _IO('T', 39) /* BSD compatibility */ #define TIOCCBRK _IO('T', 40) /* BSD compatibility */ #define TIOCGSID _IOR('T', 41, pid_t) /* Return the session ID of FD*/ +#define TCGETS2 _IOR('T', 42, struct termios2) +#define TCSETS2 _IOW('T', 43, struct termios2) +#define TCSETSW2 _IOW('T', 44, struct termios2) +#define TCSETSF2 _IOW('T', 45, struct termios2) #define TIOCGPTN _IOR('T',0x30, unsigned int) /* Get Pty Number (of pty-mux device) */ #define TIOCSPTLCK _IOW('T',0x31, int) /* Lock/unlock Pty */ diff --git a/include/asm-xtensa/termbits.h b/include/asm-xtensa/termbits.h index 9972c25ec86f..85aa6a3c0b6e 100644 --- a/include/asm-xtensa/termbits.h +++ b/include/asm-xtensa/termbits.h @@ -157,6 +157,7 @@ struct ktermios { #define HUPCL 0002000 #define CLOCAL 0004000 #define CBAUDEX 0010000 +#define BOTHER 0010000 #define B57600 0010001 #define B115200 0010002 #define B230400 0010003 @@ -172,10 +173,12 @@ struct ktermios { #define B3000000 0010015 #define B3500000 0010016 #define B4000000 0010017 -#define CIBAUD 002003600000 /* input baud rate (not used) */ +#define CIBAUD 002003600000 /* input baud rate */ #define CMSPAR 010000000000 /* mark or space (stick) parity */ #define CRTSCTS 020000000000 /* flow control */ +#define IBSHIFT 16 /* Shift from CBAUD to CIBAUD */ + /* c_lflag bits */ #define ISIG 0000001 diff --git a/include/asm-xtensa/termios.h b/include/asm-xtensa/termios.h index f14b42c8dac0..4673f42f88a7 100644 --- a/include/asm-xtensa/termios.h +++ b/include/asm-xtensa/termios.h @@ -95,8 +95,10 @@ struct termio { copy_to_user((termio)->c_cc, (termios)->c_cc, NCC); \ }) -#define user_termios_to_kernel_termios(k, u) copy_from_user(k, u, sizeof(struct termios)) -#define kernel_termios_to_user_termios(u, k) copy_to_user(u, k, sizeof(struct termios)) +#define user_termios_to_kernel_termios(k, u) copy_from_user(k, u, sizeof(struct termios2)) +#define kernel_termios_to_user_termios(u, k) copy_to_user(u, k, sizeof(struct termios2)) +#define user_termios_to_kernel_termios_1(k, u) copy_from_user(k, u, sizeof(struct termios)) +#define kernel_termios_to_user_termios_1(u, k) copy_to_user(u, k, sizeof(struct termios)) #endif /* __KERNEL__ */ -- cgit v1.2.3