2 * the Linux incarnation of OS-dependent routines. See also
3 * $(sbcl_arch)-linux-os.c
5 * This file (along with os.h) exports an OS-independent interface to
6 * the operating system VM facilities. Surprise surprise, this
7 * interface looks a lot like the Mach interface (but simpler in some
8 * places). For some operating systems, a subset of these functions
9 * will have to be emulated.
13 * This software is part of the SBCL system. See the README file for
16 * This software is derived from the CMU CL system, which was
17 * written at Carnegie Mellon University and released into the
18 * public domain. The software is in the public domain and is
19 * provided with absolutely no warranty. See the COPYING and CREDITS
20 * files for more information.
24 #include <sys/param.h>
31 #include "interrupt.h"
35 #include "genesis/static-symbols.h"
36 #include "genesis/fdefn.h"
38 #include <sys/socket.h>
39 #include <sys/utsname.h>
42 #include <sys/types.h>
44 /* #include <sys/sysinfo.h> */
48 #include <linux/version.h>
53 #if defined LISP_FEATURE_GENCGC
54 #include "gencgc-internal.h"
57 #ifdef LISP_FEATURE_X86
58 /* Prototype for personality(2). Done inline here since the header file
59 * for this isn't available on old versions of glibc. */
60 int personality (unsigned long);
63 size_t os_vm_page_size;
65 #if defined(LISP_FEATURE_SB_THREAD) && !defined(LISP_FEATURE_SB_LUTEX)
66 #include <sys/syscall.h>
70 /* values taken from the kernel's linux/futex.h. This header file
71 doesn't exist in userspace, which is our excuse for not grovelling
73 #define FUTEX_WAIT (0)
74 #define FUTEX_WAKE (1)
76 #define FUTEX_REQUEUE (3)
78 #define sys_futex sbcl_sys_futex
79 static inline int sys_futex (void *futex, int op, int val, struct timespec *rel)
81 return syscall (SYS_futex, futex, op, val, rel);
85 futex_wait(int *lock_word, int oldval)
87 int t= sys_futex(lock_word,FUTEX_WAIT,oldval, 0);
92 futex_wake(int *lock_word, int n)
94 return sys_futex(lock_word,FUTEX_WAKE,n,0);
99 int linux_sparc_siginfo_bug = 0;
101 #ifdef LISP_FEATURE_SB_THREAD
105 size_t n = confstr (_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
107 char *buf = alloca (n);
108 confstr (_CS_GNU_LIBPTHREAD_VERSION, buf, n);
109 if (strstr (buf, "NPTL")) {
118 os_init(char *argv[], char *envp[])
120 /* Conduct various version checks: do we have enough mmap(), is
121 * this a sparc running 2.2, can we do threads? */
122 #if defined(LISP_FEATURE_SB_THREAD) && !defined(LISP_FEATURE_SB_LUTEX)
132 major_version = atoi(p);
134 minor_version = atoi(p);
136 patch_version = atoi(p);
137 if (major_version<2) {
138 lose("linux kernel version too old: major version=%d (can't run in version < 2.0.0)\n",
141 if (!(major_version>2 || minor_version >= 4)) {
142 #ifdef LISP_FEATURE_SPARC
143 FSHOW((stderr,"linux kernel %d.%d predates 2.4;\n enabling workarounds for SPARC kernel bugs in signal handling.\n", major_version,minor_version));
144 linux_sparc_siginfo_bug = 1;
147 #ifdef LISP_FEATURE_SB_THREAD
148 #if !defined(LISP_FEATURE_SB_LUTEX)
149 futex_wait(futex,-1);
151 lose("This version of SBCL is compiled with threading support, but your kernel\n"
152 "is too old to support this. Please use a more recent kernel or\n"
153 "a version of SBCL without threading support.\n");
157 lose("This version of SBCL only works correctly with the NPTL threading\n"
158 "library. Please use a newer glibc, use an older SBCL, or stop using\n"
159 "LD_ASSUME_KERNEL\n");
162 os_vm_page_size = getpagesize();
164 /* KLUDGE: Disable memory randomization on new Linux kernels
165 * by setting a personality flag and re-executing. (We need
166 * to re-execute, since the memory maps that can conflict with
167 * the SBCL spaces have already been done at this point).
169 * Since randomization is currently implemented only on x86 kernels,
170 * don't do this trick on other platforms.
172 #ifdef LISP_FEATURE_X86
173 if ((major_version == 2
174 /* Some old kernels will apparently lose unsupported personality flags
176 && ((minor_version == 6 && patch_version >= 11)
177 || (minor_version > 6)))
178 || major_version >= 3)
180 int pers = personality(0xffffffffUL);
181 /* 0x40000 aka. ADDR_NO_RANDOMIZE */
182 if (!(pers & 0x40000)) {
183 int retval = personality(pers | 0x40000);
184 /* Allegedly some Linux kernels (the reported case was
185 * "hardened Linux 2.6.7") won't set the new personality,
186 * but nor will they return -1 for an error. So as a
187 * workaround query the new personality...
189 int newpers = personality(0xffffffffUL);
190 /* ... and don't re-execute if either the setting resulted
191 * in an error or if the value didn't change. Otherwise
192 * this might result in an infinite loop.
194 if (retval != -1 && newpers != pers) {
195 /* Use /proc/self/exe instead of trying to figure out
196 * the executable path from PATH and argv[0], since
197 * that's unreliable. We follow the symlink instead of
198 * executing the file directly in order to prevent top
199 * from displaying the name of the process as "exe". */
200 char runtime[PATH_MAX+1];
201 int i = readlink("/proc/self/exe", runtime, PATH_MAX);
204 execve(runtime, argv, envp);
207 /* Either changing the personality or execve() failed. Either
208 * way we might as well continue, and hope that the random
209 * memory maps are ok this time around.
211 fprintf(stderr, "WARNING: Couldn't re-execute SBCL with the proper personality flags (maybe /proc isn't mounted?). Trying to continue anyway.\n");
214 /* Use SSE detector. Recent versions of Linux enable SSE support
215 * on SSE capable CPUs. */
216 /* FIXME: Are there any old versions that does not support SSE? */
217 fast_bzero_pointer = fast_bzero_detect;
222 #ifdef LISP_FEATURE_ALPHA
223 /* The Alpha is a 64 bit CPU. SBCL is a 32 bit application. Due to all
224 * the places that assume we can get a pointer into a fixnum with no
225 * information loss, we have to make sure it allocates all its ram in the
228 static void * under_2gb_free_pointer=DYNAMIC_1_SPACE_END;
232 os_validate(os_vm_address_t addr, os_vm_size_t len)
234 int flags = MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE;
235 os_vm_address_t actual;
237 #ifdef LISP_FEATURE_ALPHA
239 addr=under_2gb_free_pointer;
242 actual = mmap(addr, len, OS_VM_PROT_ALL, flags, -1, 0);
243 if (actual == MAP_FAILED) {
245 return 0; /* caller should check this */
248 if (addr && (addr!=actual)) {
249 fprintf(stderr, "mmap: wanted %lu bytes at %p, actually mapped at %p\n",
250 (unsigned long) len, addr, actual);
254 #ifdef LISP_FEATURE_ALPHA
256 len=(len+(os_vm_page_size-1))&(~(os_vm_page_size-1));
257 under_2gb_free_pointer+=len;
264 os_invalidate(os_vm_address_t addr, os_vm_size_t len)
266 if (munmap(addr,len) == -1) {
272 os_map(int fd, int offset, os_vm_address_t addr, os_vm_size_t len)
274 os_vm_address_t actual;
276 actual = mmap(addr, len, OS_VM_PROT_ALL, MAP_PRIVATE | MAP_FIXED,
278 if (actual == MAP_FAILED || (addr && (addr != actual))) {
280 lose("unexpected mmap(..) failure\n");
287 os_protect(os_vm_address_t address, os_vm_size_t length, os_vm_prot_t prot)
289 if (mprotect(address, length, prot) == -1) {
290 if (errno == ENOMEM) {
291 lose("An mprotect call failed with ENOMEM. This probably means that the maximum amount\n"
292 "of separate memory mappings was exceeded. To fix the problem, either increase\n"
293 "the maximum with e.g. 'echo 262144 > /proc/sys/vm/max_map_count' or recompile\n"
294 "SBCL with a larger value for GENCGC-PAGE-SIZE in 'src/target/parms.lisp'.");
302 is_valid_lisp_addr(os_vm_address_t addr)
305 size_t ad = (size_t) addr;
307 if ((READ_ONLY_SPACE_START <= ad && ad < READ_ONLY_SPACE_END)
308 || (STATIC_SPACE_START <= ad && ad < STATIC_SPACE_END)
309 #if defined LISP_FEATURE_GENCGC
310 || (DYNAMIC_SPACE_START <= ad && ad < DYNAMIC_SPACE_END)
312 || (DYNAMIC_0_SPACE_START <= ad && ad < DYNAMIC_0_SPACE_END)
313 || (DYNAMIC_1_SPACE_START <= ad && ad < DYNAMIC_1_SPACE_END)
317 for_each_thread(th) {
318 if((size_t)(th->control_stack_start) <= ad
319 && ad < (size_t)(th->control_stack_end))
321 if((size_t)(th->binding_stack_start) <= ad
322 && ad < (size_t)(th->binding_stack_start + BINDING_STACK_SIZE))
329 * any OS-dependent special low-level handling for signals
333 * The GC needs to be hooked into whatever signal is raised for
334 * page fault on this OS.
337 sigsegv_handler(int signal, siginfo_t *info, void* void_context)
339 os_context_t *context = arch_os_get_context(&void_context);
340 os_vm_address_t addr = arch_get_bad_addr(signal, info, context);
342 #ifdef LISP_FEATURE_ALPHA
343 /* Alpha stuff: This is the end of a pseudo-atomic section during
344 which a signal was received. We must deal with the pending
345 interrupt (see also interrupt.c, ../code/interrupt.lisp)
347 (how we got here: when interrupting, we set bit 63 in reg_ALLOC.
348 At the end of the atomic section we tried to write to reg_ALLOC,
349 got a SIGSEGV (there's nothing mapped there) so ended up here. */
351 *os_context_register_addr(context, reg_ALLOC) & (1L<<63)) {
352 *os_context_register_addr(context, reg_ALLOC) -= (1L<<63);
353 interrupt_handle_pending(context);
358 #ifdef LISP_FEATURE_GENCGC
359 if (!gencgc_handle_wp_violation(addr))
361 if (!interrupt_maybe_gc(signal, info, context))
363 if (!handle_guard_page_triggered(context, addr))
364 #ifdef LISP_FEATURE_C_STACK_IS_CONTROL_STACK
365 arrange_return_to_lisp_function(context, SymbolFunction(MEMORY_FAULT_ERROR));
367 interrupt_handle_now(signal, info, context);
372 os_install_interrupt_handlers(void)
374 undoably_install_low_level_interrupt_handler(SIG_MEMORY_FAULT,
376 #ifdef LISP_FEATURE_SB_THREAD
377 undoably_install_low_level_interrupt_handler(SIG_INTERRUPT_THREAD,
378 interrupt_thread_handler);
379 undoably_install_low_level_interrupt_handler(SIG_STOP_FOR_GC,
380 sig_stop_for_gc_handler);
385 os_get_runtime_executable_path()
387 char path[PATH_MAX + 1];
390 size = readlink("/proc/self/exe", path, sizeof(path)-1);
396 return copied_string(path);