syscall_emul.hh revision 13933:b4382461066d
1/*
2 * Copyright (c) 2012-2013, 2015 ARM Limited
3 * Copyright (c) 2015 Advanced Micro Devices, Inc.
4 * All rights reserved
5 *
6 * The license below extends only to copyright in the software and shall
7 * not be construed as granting a license to any other intellectual
8 * property including but not limited to intellectual property relating
9 * to a hardware implementation of the functionality of the software
10 * licensed hereunder.  You may use the software subject to the license
11 * terms below provided that you ensure that this notice is replicated
12 * unmodified and in its entirety in all distributions of the software,
13 * modified or unmodified, in source code or in binary form.
14 *
15 * Copyright (c) 2003-2005 The Regents of The University of Michigan
16 * All rights reserved.
17 *
18 * Redistribution and use in source and binary forms, with or without
19 * modification, are permitted provided that the following conditions are
20 * met: redistributions of source code must retain the above copyright
21 * notice, this list of conditions and the following disclaimer;
22 * redistributions in binary form must reproduce the above copyright
23 * notice, this list of conditions and the following disclaimer in the
24 * documentation and/or other materials provided with the distribution;
25 * neither the name of the copyright holders nor the names of its
26 * contributors may be used to endorse or promote products derived from
27 * this software without specific prior written permission.
28 *
29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
33 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
34 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
35 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
36 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
37 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
38 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
39 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 *
41 * Authors: Steve Reinhardt
42 *          Kevin Lim
43 */
44
45#ifndef __SIM_SYSCALL_EMUL_HH__
46#define __SIM_SYSCALL_EMUL_HH__
47
48#if (defined(__APPLE__) || defined(__OpenBSD__) ||      \
49     defined(__FreeBSD__) || defined(__CYGWIN__) ||     \
50     defined(__NetBSD__))
51#define NO_STAT64 1
52#else
53#define NO_STAT64 0
54#endif
55
56///
57/// @file syscall_emul.hh
58///
59/// This file defines objects used to emulate syscalls from the target
60/// application on the host machine.
61
62#ifdef __linux__
63#include <sys/eventfd.h>
64#include <sys/statfs.h>
65
66#endif
67
68#ifdef __CYGWIN32__
69#include <sys/fcntl.h>
70
71#endif
72#include <fcntl.h>
73#include <net/if.h>
74#include <poll.h>
75#include <sys/ioctl.h>
76#include <sys/mman.h>
77#include <sys/socket.h>
78#include <sys/stat.h>
79#include <sys/time.h>
80#include <sys/types.h>
81#include <sys/uio.h>
82#include <unistd.h>
83
84#include <cerrno>
85#include <memory>
86#include <string>
87
88#include "arch/generic/tlb.hh"
89#include "arch/utility.hh"
90#include "base/intmath.hh"
91#include "base/loader/object_file.hh"
92#include "base/logging.hh"
93#include "base/trace.hh"
94#include "base/types.hh"
95#include "config/the_isa.hh"
96#include "cpu/base.hh"
97#include "cpu/thread_context.hh"
98#include "mem/page_table.hh"
99#include "params/Process.hh"
100#include "sim/emul_driver.hh"
101#include "sim/futex_map.hh"
102#include "sim/process.hh"
103#include "sim/syscall_debug_macros.hh"
104#include "sim/syscall_desc.hh"
105#include "sim/syscall_emul_buf.hh"
106#include "sim/syscall_return.hh"
107
108#if defined(__APPLE__) && defined(__MACH__) && !defined(CMSG_ALIGN)
109#define CMSG_ALIGN(len) (((len) + sizeof(size_t) - 1) & ~(sizeof(size_t) - 1))
110#endif
111
112//////////////////////////////////////////////////////////////////////
113//
114// The following emulation functions are generic enough that they
115// don't need to be recompiled for different emulated OS's.  They are
116// defined in sim/syscall_emul.cc.
117//
118//////////////////////////////////////////////////////////////////////
119
120void warnUnsupportedOS(std::string syscall_name);
121
122/// Handler for unimplemented syscalls that we haven't thought about.
123SyscallReturn unimplementedFunc(SyscallDesc *desc, int num,
124                                Process *p, ThreadContext *tc);
125
126/// Handler for unimplemented syscalls that we never intend to
127/// implement (signal handling, etc.) and should not affect the correct
128/// behavior of the program.  Print a warning only if the appropriate
129/// trace flag is enabled.  Return success to the target program.
130SyscallReturn ignoreFunc(SyscallDesc *desc, int num,
131                         Process *p, ThreadContext *tc);
132
133// Target fallocateFunc() handler.
134SyscallReturn fallocateFunc(SyscallDesc *desc, int num,
135                            Process *p, ThreadContext *tc);
136
137/// Target exit() handler: terminate current context.
138SyscallReturn exitFunc(SyscallDesc *desc, int num,
139                       Process *p, ThreadContext *tc);
140
141/// Target exit_group() handler: terminate simulation. (exit all threads)
142SyscallReturn exitGroupFunc(SyscallDesc *desc, int num,
143                       Process *p, ThreadContext *tc);
144
145/// Target set_tid_address() handler.
146SyscallReturn setTidAddressFunc(SyscallDesc *desc, int num,
147                                Process *p, ThreadContext *tc);
148
149/// Target getpagesize() handler.
150SyscallReturn getpagesizeFunc(SyscallDesc *desc, int num,
151                              Process *p, ThreadContext *tc);
152
153/// Target brk() handler: set brk address.
154SyscallReturn brkFunc(SyscallDesc *desc, int num,
155                      Process *p, ThreadContext *tc);
156
157/// Target close() handler.
158SyscallReturn closeFunc(SyscallDesc *desc, int num,
159                        Process *p, ThreadContext *tc);
160
161/// Target lseek() handler.
162SyscallReturn lseekFunc(SyscallDesc *desc, int num,
163                        Process *p, ThreadContext *tc);
164
165/// Target _llseek() handler.
166SyscallReturn _llseekFunc(SyscallDesc *desc, int num,
167                          Process *p, ThreadContext *tc);
168
169/// Target munmap() handler.
170SyscallReturn munmapFunc(SyscallDesc *desc, int num,
171                         Process *p, ThreadContext *tc);
172
173/// Target shutdown() handler.
174SyscallReturn shutdownFunc(SyscallDesc *desc, int num,
175                           Process *p, ThreadContext *tc);
176
177/// Target gethostname() handler.
178SyscallReturn gethostnameFunc(SyscallDesc *desc, int num,
179                              Process *p, ThreadContext *tc);
180
181/// Target getcwd() handler.
182SyscallReturn getcwdFunc(SyscallDesc *desc, int num,
183                         Process *p, ThreadContext *tc);
184
185/// Target readlink() handler.
186SyscallReturn readlinkFunc(SyscallDesc *desc, int num,
187                           Process *p, ThreadContext *tc,
188                           int index = 0);
189SyscallReturn readlinkFunc(SyscallDesc *desc, int num,
190                           Process *p, ThreadContext *tc);
191
192/// Target unlink() handler.
193SyscallReturn unlinkHelper(SyscallDesc *desc, int num,
194                           Process *p, ThreadContext *tc,
195                           int index);
196SyscallReturn unlinkFunc(SyscallDesc *desc, int num,
197                         Process *p, ThreadContext *tc);
198
199/// Target link() handler
200SyscallReturn linkFunc(SyscallDesc *desc, int num, Process *p,
201                       ThreadContext *tc);
202
203/// Target symlink() handler.
204SyscallReturn symlinkFunc(SyscallDesc *desc, int num, Process *p,
205                          ThreadContext *tc);
206
207/// Target mkdir() handler.
208SyscallReturn mkdirFunc(SyscallDesc *desc, int num,
209                        Process *p, ThreadContext *tc);
210
211/// Target mknod() handler.
212SyscallReturn mknodFunc(SyscallDesc *desc, int num,
213                        Process *p, ThreadContext *tc);
214
215/// Target chdir() handler.
216SyscallReturn chdirFunc(SyscallDesc *desc, int num,
217                        Process *p, ThreadContext *tc);
218
219// Target rmdir() handler.
220SyscallReturn rmdirFunc(SyscallDesc *desc, int num,
221                        Process *p, ThreadContext *tc);
222
223/// Target rename() handler.
224SyscallReturn renameFunc(SyscallDesc *desc, int num,
225                         Process *p, ThreadContext *tc);
226
227
228/// Target truncate() handler.
229SyscallReturn truncateFunc(SyscallDesc *desc, int num,
230                           Process *p, ThreadContext *tc);
231
232
233/// Target ftruncate() handler.
234SyscallReturn ftruncateFunc(SyscallDesc *desc, int num,
235                            Process *p, ThreadContext *tc);
236
237
238/// Target truncate64() handler.
239SyscallReturn truncate64Func(SyscallDesc *desc, int num,
240                             Process *p, ThreadContext *tc);
241
242/// Target ftruncate64() handler.
243SyscallReturn ftruncate64Func(SyscallDesc *desc, int num,
244                              Process *p, ThreadContext *tc);
245
246
247/// Target umask() handler.
248SyscallReturn umaskFunc(SyscallDesc *desc, int num,
249                        Process *p, ThreadContext *tc);
250
251/// Target gettid() handler.
252SyscallReturn gettidFunc(SyscallDesc *desc, int num,
253                         Process *p, ThreadContext *tc);
254
255/// Target chown() handler.
256SyscallReturn chownFunc(SyscallDesc *desc, int num,
257                        Process *p, ThreadContext *tc);
258
259/// Target setpgid() handler.
260SyscallReturn setpgidFunc(SyscallDesc *desc, int num,
261                          Process *p, ThreadContext *tc);
262
263/// Target fchown() handler.
264SyscallReturn fchownFunc(SyscallDesc *desc, int num,
265                         Process *p, ThreadContext *tc);
266
267/// Target dup() handler.
268SyscallReturn dupFunc(SyscallDesc *desc, int num,
269                      Process *process, ThreadContext *tc);
270
271/// Target dup2() handler.
272SyscallReturn dup2Func(SyscallDesc *desc, int num,
273                       Process *process, ThreadContext *tc);
274
275/// Target fcntl() handler.
276SyscallReturn fcntlFunc(SyscallDesc *desc, int num,
277                        Process *process, ThreadContext *tc);
278
279/// Target fcntl64() handler.
280SyscallReturn fcntl64Func(SyscallDesc *desc, int num,
281                          Process *process, ThreadContext *tc);
282
283/// Target setuid() handler.
284SyscallReturn setuidFunc(SyscallDesc *desc, int num,
285                         Process *p, ThreadContext *tc);
286
287/// Target pipe() handler.
288SyscallReturn pipeFunc(SyscallDesc *desc, int num,
289                       Process *p, ThreadContext *tc);
290
291/// Internal pipe() handler.
292SyscallReturn pipeImpl(SyscallDesc *desc, int num, Process *p,
293                       ThreadContext *tc, bool pseudoPipe);
294
295/// Target getpid() handler.
296SyscallReturn getpidFunc(SyscallDesc *desc, int num,
297                         Process *p, ThreadContext *tc);
298
299// Target getpeername() handler.
300SyscallReturn getpeernameFunc(SyscallDesc *desc, int num,
301                              Process *p, ThreadContext *tc);
302
303// Target bind() handler.
304SyscallReturn bindFunc(SyscallDesc *desc, int num,
305                       Process *p, ThreadContext *tc);
306
307// Target listen() handler.
308SyscallReturn listenFunc(SyscallDesc *desc, int num,
309                         Process *p, ThreadContext *tc);
310
311// Target connect() handler.
312SyscallReturn connectFunc(SyscallDesc *desc, int num,
313                          Process *p, ThreadContext *tc);
314
315#if defined(SYS_getdents)
316// Target getdents() handler.
317SyscallReturn getdentsFunc(SyscallDesc *desc, int num,
318                           Process *p, ThreadContext *tc);
319#endif
320
321#if defined(SYS_getdents64)
322// Target getdents() handler.
323SyscallReturn getdents64Func(SyscallDesc *desc, int num,
324                           Process *p, ThreadContext *tc);
325#endif
326
327// Target sendto() handler.
328SyscallReturn sendtoFunc(SyscallDesc *desc, int num,
329                         Process *p, ThreadContext *tc);
330
331// Target recvfrom() handler.
332SyscallReturn recvfromFunc(SyscallDesc *desc, int num,
333                           Process *p, ThreadContext *tc);
334
335// Target recvmsg() handler.
336SyscallReturn recvmsgFunc(SyscallDesc *desc, int num,
337                          Process *p, ThreadContext *tc);
338
339// Target sendmsg() handler.
340SyscallReturn sendmsgFunc(SyscallDesc *desc, int num,
341                          Process *p, ThreadContext *tc);
342
343// Target getuid() handler.
344SyscallReturn getuidFunc(SyscallDesc *desc, int num,
345                         Process *p, ThreadContext *tc);
346
347/// Target getgid() handler.
348SyscallReturn getgidFunc(SyscallDesc *desc, int num,
349                         Process *p, ThreadContext *tc);
350
351/// Target getppid() handler.
352SyscallReturn getppidFunc(SyscallDesc *desc, int num,
353                          Process *p, ThreadContext *tc);
354
355/// Target geteuid() handler.
356SyscallReturn geteuidFunc(SyscallDesc *desc, int num,
357                          Process *p, ThreadContext *tc);
358
359/// Target getegid() handler.
360SyscallReturn getegidFunc(SyscallDesc *desc, int num,
361                          Process *p, ThreadContext *tc);
362
363/// Target access() handler
364SyscallReturn accessFunc(SyscallDesc *desc, int num,
365                         Process *p, ThreadContext *tc);
366SyscallReturn accessFunc(SyscallDesc *desc, int num,
367                         Process *p, ThreadContext *tc,
368                         int index);
369
370// Target getsockopt() handler.
371SyscallReturn getsockoptFunc(SyscallDesc *desc, int num,
372                             Process *p, ThreadContext *tc);
373
374// Target setsockopt() handler.
375SyscallReturn setsockoptFunc(SyscallDesc *desc, int num,
376                             Process *p, ThreadContext *tc);
377
378// Target getsockname() handler.
379SyscallReturn getsocknameFunc(SyscallDesc *desc, int num,
380                              Process *p, ThreadContext *tc);
381
382/// Futex system call
383/// Implemented by Daniel Sanchez
384/// Used by printf's in multi-threaded apps
385template <class OS>
386SyscallReturn
387futexFunc(SyscallDesc *desc, int callnum, Process *process,
388          ThreadContext *tc)
389{
390    using namespace std;
391
392    int index = 0;
393    Addr uaddr = process->getSyscallArg(tc, index);
394    int op = process->getSyscallArg(tc, index);
395    int val = process->getSyscallArg(tc, index);
396    int timeout M5_VAR_USED = process->getSyscallArg(tc, index);
397    Addr uaddr2 M5_VAR_USED = process->getSyscallArg(tc, index);
398    int val3 = process->getSyscallArg(tc, index);
399
400    /*
401     * Unsupported option that does not affect the correctness of the
402     * application. This is a performance optimization utilized by Linux.
403     */
404    op &= ~OS::TGT_FUTEX_PRIVATE_FLAG;
405    op &= ~OS::TGT_FUTEX_CLOCK_REALTIME_FLAG;
406
407    FutexMap &futex_map = tc->getSystemPtr()->futexMap;
408
409    if (OS::TGT_FUTEX_WAIT == op || OS::TGT_FUTEX_WAIT_BITSET == op) {
410        // Ensure futex system call accessed atomically.
411        BufferArg buf(uaddr, sizeof(int));
412        buf.copyIn(tc->getMemProxy());
413        int mem_val = *(int*)buf.bufferPtr();
414
415        /*
416         * The value in memory at uaddr is not equal with the expected val
417         * (a different thread must have changed it before the system call was
418         * invoked). In this case, we need to throw an error.
419         */
420        if (val != mem_val)
421            return -OS::TGT_EWOULDBLOCK;
422
423        if (OS::TGT_FUTEX_WAIT) {
424            futex_map.suspend(uaddr, process->tgid(), tc);
425        } else {
426            futex_map.suspend_bitset(uaddr, process->tgid(), tc, val3);
427        }
428
429        return 0;
430    } else if (OS::TGT_FUTEX_WAKE == op) {
431        return futex_map.wakeup(uaddr, process->tgid(), val);
432    } else if (OS::TGT_FUTEX_WAKE_BITSET == op) {
433        return futex_map.wakeup_bitset(uaddr, process->tgid(), val3);
434    } else if (OS::TGT_FUTEX_REQUEUE == op ||
435               OS::TGT_FUTEX_CMP_REQUEUE == op) {
436
437        // Ensure futex system call accessed atomically.
438        BufferArg buf(uaddr, sizeof(int));
439        buf.copyIn(tc->getMemProxy());
440        int mem_val = *(int*)buf.bufferPtr();
441        /*
442         * For CMP_REQUEUE, the whole operation is only started only if
443         * val3 is still the value of the futex pointed to by uaddr.
444         */
445        if (OS::TGT_FUTEX_CMP_REQUEUE && val3 != mem_val)
446            return -OS::TGT_EWOULDBLOCK;
447        return futex_map.requeue(uaddr, process->tgid(), val, timeout, uaddr2);
448    } else if (OS::TGT_FUTEX_WAKE_OP == op) {
449        /*
450         * The FUTEX_WAKE_OP operation is equivalent to executing the
451         * following code atomically and totally ordered with respect to
452         * other futex operations on any of the two supplied futex words:
453         *
454         *   int oldval = *(int *) addr2;
455         *   *(int *) addr2 = oldval op oparg;
456         *   futex(addr1, FUTEX_WAKE, val, 0, 0, 0);
457         *   if (oldval cmp cmparg)
458         *        futex(addr2, FUTEX_WAKE, val2, 0, 0, 0);
459         *
460         * (op, oparg, cmp, cmparg are encoded in val3)
461         *
462         * +---+---+-----------+-----------+
463         * |op |cmp|   oparg   |  cmparg   |
464         * +---+---+-----------+-----------+
465         *   4   4       12          12    <== # of bits
466         *
467         * reference: http://man7.org/linux/man-pages/man2/futex.2.html
468         *
469         */
470        // get value from simulated-space
471        BufferArg buf(uaddr2, sizeof(int));
472        buf.copyIn(tc->getMemProxy());
473        int oldval = *(int*)buf.bufferPtr();
474        int newval = oldval;
475        // extract op, oparg, cmp, cmparg from val3
476        int wake_cmparg =  val3 & 0xfff;
477        int wake_oparg  = (val3 & 0xfff000)   >> 12;
478        int wake_cmp    = (val3 & 0xf000000)  >> 24;
479        int wake_op     = (val3 & 0xf0000000) >> 28;
480        if ((wake_op & OS::TGT_FUTEX_OP_ARG_SHIFT) >> 3 == 1)
481            wake_oparg = (1 << wake_oparg);
482        wake_op &= ~OS::TGT_FUTEX_OP_ARG_SHIFT;
483        // perform operation on the value of the second futex
484        if (wake_op == OS::TGT_FUTEX_OP_SET)
485            newval = wake_oparg;
486        else if (wake_op == OS::TGT_FUTEX_OP_ADD)
487            newval += wake_oparg;
488        else if (wake_op == OS::TGT_FUTEX_OP_OR)
489            newval |= wake_oparg;
490        else if (wake_op == OS::TGT_FUTEX_OP_ANDN)
491            newval &= ~wake_oparg;
492        else if (wake_op == OS::TGT_FUTEX_OP_XOR)
493            newval ^= wake_oparg;
494        // copy updated value back to simulated-space
495        *(int*)buf.bufferPtr() = newval;
496        buf.copyOut(tc->getMemProxy());
497        // perform the first wake-up
498        int woken1 = futex_map.wakeup(uaddr, process->tgid(), val);
499        int woken2 = 0;
500        // calculate the condition of the second wake-up
501        bool is_wake2 = false;
502        if (wake_cmp == OS::TGT_FUTEX_OP_CMP_EQ)
503            is_wake2 = oldval == wake_cmparg;
504        else if (wake_cmp == OS::TGT_FUTEX_OP_CMP_NE)
505            is_wake2 = oldval != wake_cmparg;
506        else if (wake_cmp == OS::TGT_FUTEX_OP_CMP_LT)
507            is_wake2 = oldval < wake_cmparg;
508        else if (wake_cmp == OS::TGT_FUTEX_OP_CMP_LE)
509            is_wake2 = oldval <= wake_cmparg;
510        else if (wake_cmp == OS::TGT_FUTEX_OP_CMP_GT)
511            is_wake2 = oldval > wake_cmparg;
512        else if (wake_cmp == OS::TGT_FUTEX_OP_CMP_GE)
513            is_wake2 = oldval >= wake_cmparg;
514        // perform the second wake-up
515        if (is_wake2)
516            woken2 = futex_map.wakeup(uaddr2, process->tgid(), timeout);
517
518        return woken1 + woken2;
519    }
520    warn("futex: op %d not implemented; ignoring.", op);
521    return -ENOSYS;
522}
523
524
525/// Pseudo Funcs  - These functions use a different return convension,
526/// returning a second value in a register other than the normal return register
527SyscallReturn pipePseudoFunc(SyscallDesc *desc, int num,
528                             Process *process, ThreadContext *tc);
529
530/// Target getpidPseudo() handler.
531SyscallReturn getpidPseudoFunc(SyscallDesc *desc, int num,
532                               Process *p, ThreadContext *tc);
533
534/// Target getuidPseudo() handler.
535SyscallReturn getuidPseudoFunc(SyscallDesc *desc, int num,
536                               Process *p, ThreadContext *tc);
537
538/// Target getgidPseudo() handler.
539SyscallReturn getgidPseudoFunc(SyscallDesc *desc, int num,
540                               Process *p, ThreadContext *tc);
541
542
543/// A readable name for 1,000,000, for converting microseconds to seconds.
544const int one_million = 1000000;
545/// A readable name for 1,000,000,000, for converting nanoseconds to seconds.
546const int one_billion = 1000000000;
547
548/// Approximate seconds since the epoch (1/1/1970).  About a billion,
549/// by my reckoning.  We want to keep this a constant (not use the
550/// real-world time) to keep simulations repeatable.
551const unsigned seconds_since_epoch = 1000000000;
552
553/// Helper function to convert current elapsed time to seconds and
554/// microseconds.
555template <class T1, class T2>
556void
557getElapsedTimeMicro(T1 &sec, T2 &usec)
558{
559    uint64_t elapsed_usecs = curTick() / SimClock::Int::us;
560    sec = elapsed_usecs / one_million;
561    usec = elapsed_usecs % one_million;
562}
563
564/// Helper function to convert current elapsed time to seconds and
565/// nanoseconds.
566template <class T1, class T2>
567void
568getElapsedTimeNano(T1 &sec, T2 &nsec)
569{
570    uint64_t elapsed_nsecs = curTick() / SimClock::Int::ns;
571    sec = elapsed_nsecs / one_billion;
572    nsec = elapsed_nsecs % one_billion;
573}
574
575//////////////////////////////////////////////////////////////////////
576//
577// The following emulation functions are generic, but need to be
578// templated to account for differences in types, constants, etc.
579//
580//////////////////////////////////////////////////////////////////////
581
582    typedef struct statfs hst_statfs;
583#if NO_STAT64
584    typedef struct stat hst_stat;
585    typedef struct stat hst_stat64;
586#else
587    typedef struct stat hst_stat;
588    typedef struct stat64 hst_stat64;
589#endif
590
591//// Helper function to convert a host stat buffer to a target stat
592//// buffer.  Also copies the target buffer out to the simulated
593//// memory space.  Used by stat(), fstat(), and lstat().
594
595template <typename target_stat, typename host_stat>
596void
597convertStatBuf(target_stat &tgt, host_stat *host, bool fakeTTY = false)
598{
599    using namespace TheISA;
600
601    if (fakeTTY)
602        tgt->st_dev = 0xA;
603    else
604        tgt->st_dev = host->st_dev;
605    tgt->st_dev = TheISA::htog(tgt->st_dev);
606    tgt->st_ino = host->st_ino;
607    tgt->st_ino = TheISA::htog(tgt->st_ino);
608    tgt->st_mode = host->st_mode;
609    if (fakeTTY) {
610        // Claim to be a character device
611        tgt->st_mode &= ~S_IFMT;    // Clear S_IFMT
612        tgt->st_mode |= S_IFCHR;    // Set S_IFCHR
613    }
614    tgt->st_mode = TheISA::htog(tgt->st_mode);
615    tgt->st_nlink = host->st_nlink;
616    tgt->st_nlink = TheISA::htog(tgt->st_nlink);
617    tgt->st_uid = host->st_uid;
618    tgt->st_uid = TheISA::htog(tgt->st_uid);
619    tgt->st_gid = host->st_gid;
620    tgt->st_gid = TheISA::htog(tgt->st_gid);
621    if (fakeTTY)
622        tgt->st_rdev = 0x880d;
623    else
624        tgt->st_rdev = host->st_rdev;
625    tgt->st_rdev = TheISA::htog(tgt->st_rdev);
626    tgt->st_size = host->st_size;
627    tgt->st_size = TheISA::htog(tgt->st_size);
628    tgt->st_atimeX = host->st_atime;
629    tgt->st_atimeX = TheISA::htog(tgt->st_atimeX);
630    tgt->st_mtimeX = host->st_mtime;
631    tgt->st_mtimeX = TheISA::htog(tgt->st_mtimeX);
632    tgt->st_ctimeX = host->st_ctime;
633    tgt->st_ctimeX = TheISA::htog(tgt->st_ctimeX);
634    // Force the block size to be 8KB. This helps to ensure buffered io works
635    // consistently across different hosts.
636    tgt->st_blksize = 0x2000;
637    tgt->st_blksize = TheISA::htog(tgt->st_blksize);
638    tgt->st_blocks = host->st_blocks;
639    tgt->st_blocks = TheISA::htog(tgt->st_blocks);
640}
641
642// Same for stat64
643
644template <typename target_stat, typename host_stat64>
645void
646convertStat64Buf(target_stat &tgt, host_stat64 *host, bool fakeTTY = false)
647{
648    using namespace TheISA;
649
650    convertStatBuf<target_stat, host_stat64>(tgt, host, fakeTTY);
651#if defined(STAT_HAVE_NSEC)
652    tgt->st_atime_nsec = host->st_atime_nsec;
653    tgt->st_atime_nsec = TheISA::htog(tgt->st_atime_nsec);
654    tgt->st_mtime_nsec = host->st_mtime_nsec;
655    tgt->st_mtime_nsec = TheISA::htog(tgt->st_mtime_nsec);
656    tgt->st_ctime_nsec = host->st_ctime_nsec;
657    tgt->st_ctime_nsec = TheISA::htog(tgt->st_ctime_nsec);
658#else
659    tgt->st_atime_nsec = 0;
660    tgt->st_mtime_nsec = 0;
661    tgt->st_ctime_nsec = 0;
662#endif
663}
664
665// Here are a couple of convenience functions
666template<class OS>
667void
668copyOutStatBuf(SETranslatingPortProxy &mem, Addr addr,
669               hst_stat *host, bool fakeTTY = false)
670{
671    typedef TypedBufferArg<typename OS::tgt_stat> tgt_stat_buf;
672    tgt_stat_buf tgt(addr);
673    convertStatBuf<tgt_stat_buf, hst_stat>(tgt, host, fakeTTY);
674    tgt.copyOut(mem);
675}
676
677template<class OS>
678void
679copyOutStat64Buf(SETranslatingPortProxy &mem, Addr addr,
680                 hst_stat64 *host, bool fakeTTY = false)
681{
682    typedef TypedBufferArg<typename OS::tgt_stat64> tgt_stat_buf;
683    tgt_stat_buf tgt(addr);
684    convertStat64Buf<tgt_stat_buf, hst_stat64>(tgt, host, fakeTTY);
685    tgt.copyOut(mem);
686}
687
688template <class OS>
689void
690copyOutStatfsBuf(SETranslatingPortProxy &mem, Addr addr,
691                 hst_statfs *host)
692{
693    TypedBufferArg<typename OS::tgt_statfs> tgt(addr);
694
695    tgt->f_type = TheISA::htog(host->f_type);
696#if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
697    tgt->f_bsize = TheISA::htog(host->f_iosize);
698#else
699    tgt->f_bsize = TheISA::htog(host->f_bsize);
700#endif
701    tgt->f_blocks = TheISA::htog(host->f_blocks);
702    tgt->f_bfree = TheISA::htog(host->f_bfree);
703    tgt->f_bavail = TheISA::htog(host->f_bavail);
704    tgt->f_files = TheISA::htog(host->f_files);
705    tgt->f_ffree = TheISA::htog(host->f_ffree);
706    memcpy(&tgt->f_fsid, &host->f_fsid, sizeof(host->f_fsid));
707#if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
708    tgt->f_namelen = TheISA::htog(host->f_namemax);
709    tgt->f_frsize = TheISA::htog(host->f_bsize);
710#elif defined(__APPLE__)
711    tgt->f_namelen = 0;
712    tgt->f_frsize = 0;
713#else
714    tgt->f_namelen = TheISA::htog(host->f_namelen);
715    tgt->f_frsize = TheISA::htog(host->f_frsize);
716#endif
717#if defined(__linux__)
718    memcpy(&tgt->f_spare, &host->f_spare, sizeof(host->f_spare));
719#else
720    /*
721     * The fields are different sizes per OS. Don't bother with
722     * f_spare or f_reserved on non-Linux for now.
723     */
724    memset(&tgt->f_spare, 0, sizeof(tgt->f_spare));
725#endif
726
727    tgt.copyOut(mem);
728}
729
730/// Target ioctl() handler.  For the most part, programs call ioctl()
731/// only to find out if their stdout is a tty, to determine whether to
732/// do line or block buffering.  We always claim that output fds are
733/// not TTYs to provide repeatable results.
734template <class OS>
735SyscallReturn
736ioctlFunc(SyscallDesc *desc, int callnum, Process *p, ThreadContext *tc)
737{
738    int index = 0;
739    int tgt_fd = p->getSyscallArg(tc, index);
740    unsigned req = p->getSyscallArg(tc, index);
741
742    DPRINTF_SYSCALL(Verbose, "ioctl(%d, 0x%x, ...)\n", tgt_fd, req);
743
744    if (OS::isTtyReq(req))
745        return -ENOTTY;
746
747    auto dfdp = std::dynamic_pointer_cast<DeviceFDEntry>((*p->fds)[tgt_fd]);
748    if (dfdp) {
749        EmulatedDriver *emul_driver = dfdp->getDriver();
750        if (emul_driver)
751            return emul_driver->ioctl(p, tc, req);
752    }
753
754    auto sfdp = std::dynamic_pointer_cast<SocketFDEntry>((*p->fds)[tgt_fd]);
755    if (sfdp) {
756        int status;
757
758        switch (req) {
759          case SIOCGIFCONF: {
760            Addr conf_addr = p->getSyscallArg(tc, index);
761            BufferArg conf_arg(conf_addr, sizeof(ifconf));
762            conf_arg.copyIn(tc->getMemProxy());
763
764            ifconf *conf = (ifconf*)conf_arg.bufferPtr();
765            Addr ifc_buf_addr = (Addr)conf->ifc_buf;
766            BufferArg ifc_buf_arg(ifc_buf_addr, conf->ifc_len);
767            ifc_buf_arg.copyIn(tc->getMemProxy());
768
769            conf->ifc_buf = (char*)ifc_buf_arg.bufferPtr();
770
771            status = ioctl(sfdp->getSimFD(), req, conf_arg.bufferPtr());
772            if (status != -1) {
773                conf->ifc_buf = (char*)ifc_buf_addr;
774                ifc_buf_arg.copyOut(tc->getMemProxy());
775                conf_arg.copyOut(tc->getMemProxy());
776            }
777
778            return status;
779          }
780          case SIOCGIFFLAGS:
781#ifdef __linux__
782          case SIOCGIFINDEX:
783#endif
784          case SIOCGIFNETMASK:
785          case SIOCGIFADDR:
786#ifdef __linux__
787          case SIOCGIFHWADDR:
788#endif
789          case SIOCGIFMTU: {
790            Addr req_addr = p->getSyscallArg(tc, index);
791            BufferArg req_arg(req_addr, sizeof(ifreq));
792            req_arg.copyIn(tc->getMemProxy());
793
794            status = ioctl(sfdp->getSimFD(), req, req_arg.bufferPtr());
795            if (status != -1)
796                req_arg.copyOut(tc->getMemProxy());
797            return status;
798          }
799        }
800    }
801
802    /**
803     * For lack of a better return code, return ENOTTY. Ideally, we should
804     * return something better here, but at least we issue the warning.
805     */
806    warn("Unsupported ioctl call (return ENOTTY): ioctl(%d, 0x%x, ...) @ \n",
807         tgt_fd, req, tc->pcState());
808    return -ENOTTY;
809}
810
811template <class OS>
812SyscallReturn
813openImpl(SyscallDesc *desc, int callnum, Process *p, ThreadContext *tc,
814         bool isopenat)
815{
816    int index = 0;
817    int tgt_dirfd = -1;
818
819    /**
820     * If using the openat variant, read in the target directory file
821     * descriptor from the simulated process.
822     */
823    if (isopenat)
824        tgt_dirfd = p->getSyscallArg(tc, index);
825
826    /**
827     * Retrieve the simulated process' memory proxy and then read in the path
828     * string from that memory space into the host's working memory space.
829     */
830    std::string path;
831    if (!tc->getMemProxy().tryReadString(path, p->getSyscallArg(tc, index)))
832        return -EFAULT;
833
834#ifdef __CYGWIN32__
835    int host_flags = O_BINARY;
836#else
837    int host_flags = 0;
838#endif
839    /**
840     * Translate target flags into host flags. Flags exist which are not
841     * ported between architectures which can cause check failures.
842     */
843    int tgt_flags = p->getSyscallArg(tc, index);
844    for (int i = 0; i < OS::NUM_OPEN_FLAGS; i++) {
845        if (tgt_flags & OS::openFlagTable[i].tgtFlag) {
846            tgt_flags &= ~OS::openFlagTable[i].tgtFlag;
847            host_flags |= OS::openFlagTable[i].hostFlag;
848        }
849    }
850    if (tgt_flags) {
851        warn("open%s: cannot decode flags 0x%x",
852             isopenat ? "at" : "", tgt_flags);
853    }
854#ifdef __CYGWIN32__
855    host_flags |= O_BINARY;
856#endif
857
858    int mode = p->getSyscallArg(tc, index);
859
860    /**
861     * If the simulated process called open or openat with AT_FDCWD specified,
862     * take the current working directory value which was passed into the
863     * process class as a Python parameter and append the current path to
864     * create a full path.
865     * Otherwise, openat with a valid target directory file descriptor has
866     * been called. If the path option, which was passed in as a parameter,
867     * is not absolute, retrieve the directory file descriptor's path and
868     * prepend it to the path passed in as a parameter.
869     * In every case, we should have a full path (which is relevant to the
870     * host) to work with after this block has been passed.
871     */
872    std::string redir_path = path;
873    std::string abs_path = path;
874    if (!isopenat || tgt_dirfd == OS::TGT_AT_FDCWD) {
875        abs_path = p->absolutePath(path, true);
876        redir_path = p->checkPathRedirect(path);
877    } else if (!startswith(path, "/")) {
878        std::shared_ptr<FDEntry> fdep = ((*p->fds)[tgt_dirfd]);
879        auto ffdp = std::dynamic_pointer_cast<FileFDEntry>(fdep);
880        if (!ffdp)
881            return -EBADF;
882        abs_path = ffdp->getFileName() + path;
883        redir_path = p->checkPathRedirect(abs_path);
884    }
885
886    /**
887     * Since this is an emulated environment, we create pseudo file
888     * descriptors for device requests that have been registered with
889     * the process class through Python; this allows us to create a file
890     * descriptor for subsequent ioctl or mmap calls.
891     */
892    if (startswith(abs_path, "/dev/")) {
893        std::string filename = abs_path.substr(strlen("/dev/"));
894        EmulatedDriver *drv = p->findDriver(filename);
895        if (drv) {
896            DPRINTF_SYSCALL(Verbose, "open%s: passing call to "
897                            "driver open with path[%s]\n",
898                            isopenat ? "at" : "", abs_path.c_str());
899            return drv->open(p, tc, mode, host_flags);
900        }
901        /**
902         * Fall through here for pass through to host devices, such
903         * as /dev/zero
904         */
905    }
906
907    /**
908     * We make several attempts resolve a call to open.
909     *
910     * 1) Resolve any path redirection before hand. This will set the path
911     * up with variable 'redir_path' which may contain a modified path or
912     * the original path value. This should already be done in prior code.
913     * 2) Try to handle the access using 'special_paths'. Some special_paths
914     * and files cannot be called on the host and need to be handled as
915     * special cases inside the simulator. These special_paths are handled by
916     * C++ routines to provide output back to userspace.
917     * 3) If the full path that was created above does not match any of the
918     * special cases, pass it through to the open call on the __HOST__ to let
919     * the host open the file on our behalf. Again, the openImpl tries to
920     * USE_THE_HOST_FILESYSTEM_OPEN (with a possible redirection to the
921     * faux-filesystem files). The faux-filesystem is dynamically created
922     * during simulator configuration using Python functions.
923     * 4) If the host cannot open the file, the open attempt failed in "3)".
924     * Return the host's error code back through the system call to the
925     * simulated process. If running a debug trace, also notify the user that
926     * the open call failed.
927     *
928     * Any success will set sim_fd to something other than -1 and skip the
929     * next conditions effectively bypassing them.
930     */
931    int sim_fd = -1;
932    std::string used_path;
933    std::vector<std::string> special_paths =
934            { "/proc/meminfo/", "/system/", "/sys/", "/platform/",
935              "/etc/passwd" };
936    for (auto entry : special_paths) {
937        if (startswith(path, entry)) {
938            sim_fd = OS::openSpecialFile(abs_path, p, tc);
939            used_path = abs_path;
940        }
941    }
942    if (sim_fd == -1) {
943        sim_fd = open(redir_path.c_str(), host_flags, mode);
944        used_path = redir_path;
945    }
946    if (sim_fd == -1) {
947        int local = -errno;
948        DPRINTF_SYSCALL(Verbose, "open%s: failed -> path:%s "
949                        "(inferred from:%s)\n", isopenat ? "at" : "",
950                        used_path.c_str(), path.c_str());
951        return local;
952    }
953
954    /**
955     * The file was opened successfully and needs to be recorded in the
956     * process' file descriptor array so that it can be retrieved later.
957     * The target file descriptor that is chosen will be the lowest unused
958     * file descriptor.
959     * Return the indirect target file descriptor back to the simulated
960     * process to act as a handle for the opened file.
961     */
962    auto ffdp = std::make_shared<FileFDEntry>(sim_fd, host_flags, path, 0);
963    int tgt_fd = p->fds->allocFD(ffdp);
964    DPRINTF_SYSCALL(Verbose, "open%s: sim_fd[%d], target_fd[%d] -> path:%s\n"
965                    "(inferred from:%s)\n", isopenat ? "at" : "",
966                    sim_fd, tgt_fd, used_path.c_str(), path.c_str());
967    return tgt_fd;
968}
969
970/// Target open() handler.
971template <class OS>
972SyscallReturn
973openFunc(SyscallDesc *desc, int callnum, Process *process,
974         ThreadContext *tc)
975{
976    return openImpl<OS>(desc, callnum, process, tc, false);
977}
978
979/// Target openat() handler.
980template <class OS>
981SyscallReturn
982openatFunc(SyscallDesc *desc, int callnum, Process *process,
983           ThreadContext *tc)
984{
985    return openImpl<OS>(desc, callnum, process, tc, true);
986}
987
988/// Target unlinkat() handler.
989template <class OS>
990SyscallReturn
991unlinkatFunc(SyscallDesc *desc, int callnum, Process *process,
992             ThreadContext *tc)
993{
994    int index = 0;
995    int dirfd = process->getSyscallArg(tc, index);
996    if (dirfd != OS::TGT_AT_FDCWD)
997        warn("unlinkat: first argument not AT_FDCWD; unlikely to work");
998
999    return unlinkHelper(desc, callnum, process, tc, 1);
1000}
1001
1002/// Target facessat() handler
1003template <class OS>
1004SyscallReturn
1005faccessatFunc(SyscallDesc *desc, int callnum, Process *process,
1006              ThreadContext *tc)
1007{
1008    int index = 0;
1009    int dirfd = process->getSyscallArg(tc, index);
1010    if (dirfd != OS::TGT_AT_FDCWD)
1011        warn("faccessat: first argument not AT_FDCWD; unlikely to work");
1012    return accessFunc(desc, callnum, process, tc, 1);
1013}
1014
1015/// Target readlinkat() handler
1016template <class OS>
1017SyscallReturn
1018readlinkatFunc(SyscallDesc *desc, int callnum, Process *process,
1019               ThreadContext *tc)
1020{
1021    int index = 0;
1022    int dirfd = process->getSyscallArg(tc, index);
1023    if (dirfd != OS::TGT_AT_FDCWD)
1024        warn("openat: first argument not AT_FDCWD; unlikely to work");
1025    return readlinkFunc(desc, callnum, process, tc, 1);
1026}
1027
1028/// Target renameat() handler.
1029template <class OS>
1030SyscallReturn
1031renameatFunc(SyscallDesc *desc, int callnum, Process *process,
1032             ThreadContext *tc)
1033{
1034    int index = 0;
1035
1036    int olddirfd = process->getSyscallArg(tc, index);
1037    if (olddirfd != OS::TGT_AT_FDCWD)
1038        warn("renameat: first argument not AT_FDCWD; unlikely to work");
1039
1040    std::string old_name;
1041
1042    if (!tc->getMemProxy().tryReadString(old_name,
1043                                         process->getSyscallArg(tc, index)))
1044        return -EFAULT;
1045
1046    int newdirfd = process->getSyscallArg(tc, index);
1047    if (newdirfd != OS::TGT_AT_FDCWD)
1048        warn("renameat: third argument not AT_FDCWD; unlikely to work");
1049
1050    std::string new_name;
1051
1052    if (!tc->getMemProxy().tryReadString(new_name,
1053                                         process->getSyscallArg(tc, index)))
1054        return -EFAULT;
1055
1056    // Adjust path for cwd and redirection
1057    old_name = process->checkPathRedirect(old_name);
1058    new_name = process->checkPathRedirect(new_name);
1059
1060    int result = rename(old_name.c_str(), new_name.c_str());
1061    return (result == -1) ? -errno : result;
1062}
1063
1064/// Target sysinfo() handler.
1065template <class OS>
1066SyscallReturn
1067sysinfoFunc(SyscallDesc *desc, int callnum, Process *process,
1068            ThreadContext *tc)
1069{
1070
1071    int index = 0;
1072    TypedBufferArg<typename OS::tgt_sysinfo>
1073        sysinfo(process->getSyscallArg(tc, index));
1074
1075    sysinfo->uptime = seconds_since_epoch;
1076    sysinfo->totalram = process->system->memSize();
1077    sysinfo->mem_unit = 1;
1078
1079    sysinfo.copyOut(tc->getMemProxy());
1080
1081    return 0;
1082}
1083
1084/// Target chmod() handler.
1085template <class OS>
1086SyscallReturn
1087chmodFunc(SyscallDesc *desc, int callnum, Process *process,
1088          ThreadContext *tc)
1089{
1090    std::string path;
1091
1092    int index = 0;
1093    if (!tc->getMemProxy().tryReadString(path,
1094                process->getSyscallArg(tc, index))) {
1095        return -EFAULT;
1096    }
1097
1098    uint32_t mode = process->getSyscallArg(tc, index);
1099    mode_t hostMode = 0;
1100
1101    // XXX translate mode flags via OS::something???
1102    hostMode = mode;
1103
1104    // Adjust path for cwd and redirection
1105    path = process->checkPathRedirect(path);
1106
1107    // do the chmod
1108    int result = chmod(path.c_str(), hostMode);
1109    if (result < 0)
1110        return -errno;
1111
1112    return 0;
1113}
1114
1115template <class OS>
1116SyscallReturn
1117pollFunc(SyscallDesc *desc, int num, Process *p, ThreadContext *tc)
1118{
1119    int index = 0;
1120    Addr fdsPtr = p->getSyscallArg(tc, index);
1121    int nfds = p->getSyscallArg(tc, index);
1122    int tmout = p->getSyscallArg(tc, index);
1123
1124    BufferArg fdsBuf(fdsPtr, sizeof(struct pollfd) * nfds);
1125    fdsBuf.copyIn(tc->getMemProxy());
1126
1127    /**
1128     * Record the target file descriptors in a local variable. We need to
1129     * replace them with host file descriptors but we need a temporary copy
1130     * for later. Afterwards, replace each target file descriptor in the
1131     * poll_fd array with its host_fd.
1132     */
1133    int temp_tgt_fds[nfds];
1134    for (index = 0; index < nfds; index++) {
1135        temp_tgt_fds[index] = ((struct pollfd *)fdsBuf.bufferPtr())[index].fd;
1136        auto tgt_fd = temp_tgt_fds[index];
1137        auto hbfdp = std::dynamic_pointer_cast<HBFDEntry>((*p->fds)[tgt_fd]);
1138        if (!hbfdp)
1139            return -EBADF;
1140        auto host_fd = hbfdp->getSimFD();
1141        ((struct pollfd *)fdsBuf.bufferPtr())[index].fd = host_fd;
1142    }
1143
1144    /**
1145     * We cannot allow an infinite poll to occur or it will inevitably cause
1146     * a deadlock in the gem5 simulator with clone. We must pass in tmout with
1147     * a non-negative value, however it also makes no sense to poll on the
1148     * underlying host for any other time than tmout a zero timeout.
1149     */
1150    int status;
1151    if (tmout < 0) {
1152        status = poll((struct pollfd *)fdsBuf.bufferPtr(), nfds, 0);
1153        if (status == 0) {
1154            /**
1155             * If blocking indefinitely, check the signal list to see if a
1156             * signal would break the poll out of the retry cycle and try
1157             * to return the signal interrupt instead.
1158             */
1159            System *sysh = tc->getSystemPtr();
1160            std::list<BasicSignal>::iterator it;
1161            for (it=sysh->signalList.begin(); it!=sysh->signalList.end(); it++)
1162                if (it->receiver == p)
1163                    return -EINTR;
1164            return SyscallReturn::retry();
1165        }
1166    } else
1167        status = poll((struct pollfd *)fdsBuf.bufferPtr(), nfds, 0);
1168
1169    if (status == -1)
1170        return -errno;
1171
1172    /**
1173     * Replace each host_fd in the returned poll_fd array with its original
1174     * target file descriptor.
1175     */
1176    for (index = 0; index < nfds; index++) {
1177        auto tgt_fd = temp_tgt_fds[index];
1178        ((struct pollfd *)fdsBuf.bufferPtr())[index].fd = tgt_fd;
1179    }
1180
1181    /**
1182     * Copy out the pollfd struct because the host may have updated fields
1183     * in the structure.
1184     */
1185    fdsBuf.copyOut(tc->getMemProxy());
1186
1187    return status;
1188}
1189
1190/// Target fchmod() handler.
1191template <class OS>
1192SyscallReturn
1193fchmodFunc(SyscallDesc *desc, int callnum, Process *p, ThreadContext *tc)
1194{
1195    int index = 0;
1196    int tgt_fd = p->getSyscallArg(tc, index);
1197    uint32_t mode = p->getSyscallArg(tc, index);
1198
1199    auto ffdp = std::dynamic_pointer_cast<FileFDEntry>((*p->fds)[tgt_fd]);
1200    if (!ffdp)
1201        return -EBADF;
1202    int sim_fd = ffdp->getSimFD();
1203
1204    mode_t hostMode = mode;
1205
1206    int result = fchmod(sim_fd, hostMode);
1207
1208    return (result < 0) ? -errno : 0;
1209}
1210
1211/// Target mremap() handler.
1212template <class OS>
1213SyscallReturn
1214mremapFunc(SyscallDesc *desc, int callnum, Process *process, ThreadContext *tc)
1215{
1216    int index = 0;
1217    Addr start = process->getSyscallArg(tc, index);
1218    uint64_t old_length = process->getSyscallArg(tc, index);
1219    uint64_t new_length = process->getSyscallArg(tc, index);
1220    uint64_t flags = process->getSyscallArg(tc, index);
1221    uint64_t provided_address = 0;
1222    bool use_provided_address = flags & OS::TGT_MREMAP_FIXED;
1223
1224    if (use_provided_address)
1225        provided_address = process->getSyscallArg(tc, index);
1226
1227    if ((start % TheISA::PageBytes != 0) ||
1228        (provided_address % TheISA::PageBytes != 0)) {
1229        warn("mremap failing: arguments not page aligned");
1230        return -EINVAL;
1231    }
1232
1233    new_length = roundUp(new_length, TheISA::PageBytes);
1234
1235    if (new_length > old_length) {
1236        std::shared_ptr<MemState> mem_state = process->memState;
1237        Addr mmap_end = mem_state->getMmapEnd();
1238
1239        if ((start + old_length) == mmap_end &&
1240            (!use_provided_address || provided_address == start)) {
1241            // This case cannot occur when growing downward, as
1242            // start is greater than or equal to mmap_end.
1243            uint64_t diff = new_length - old_length;
1244            process->allocateMem(mmap_end, diff);
1245            mem_state->setMmapEnd(mmap_end + diff);
1246            return start;
1247        } else {
1248            if (!use_provided_address && !(flags & OS::TGT_MREMAP_MAYMOVE)) {
1249                warn("can't remap here and MREMAP_MAYMOVE flag not set\n");
1250                return -ENOMEM;
1251            } else {
1252                uint64_t new_start = provided_address;
1253                if (!use_provided_address) {
1254                    new_start = process->mmapGrowsDown() ?
1255                                mmap_end - new_length : mmap_end;
1256                    mmap_end = process->mmapGrowsDown() ?
1257                               new_start : mmap_end + new_length;
1258                    mem_state->setMmapEnd(mmap_end);
1259                }
1260
1261                process->pTable->remap(start, old_length, new_start);
1262                warn("mremapping to new vaddr %08p-%08p, adding %d\n",
1263                     new_start, new_start + new_length,
1264                     new_length - old_length);
1265                // add on the remaining unallocated pages
1266                process->allocateMem(new_start + old_length,
1267                                     new_length - old_length,
1268                                     use_provided_address /* clobber */);
1269                if (use_provided_address &&
1270                    ((new_start + new_length > mem_state->getMmapEnd() &&
1271                      !process->mmapGrowsDown()) ||
1272                    (new_start < mem_state->getMmapEnd() &&
1273                      process->mmapGrowsDown()))) {
1274                    // something fishy going on here, at least notify the user
1275                    // @todo: increase mmap_end?
1276                    warn("mmap region limit exceeded with MREMAP_FIXED\n");
1277                }
1278                warn("returning %08p as start\n", new_start);
1279                return new_start;
1280            }
1281        }
1282    } else {
1283        if (use_provided_address && provided_address != start)
1284            process->pTable->remap(start, new_length, provided_address);
1285        process->pTable->unmap(start + new_length, old_length - new_length);
1286        return use_provided_address ? provided_address : start;
1287    }
1288}
1289
1290/// Target stat() handler.
1291template <class OS>
1292SyscallReturn
1293statFunc(SyscallDesc *desc, int callnum, Process *process,
1294         ThreadContext *tc)
1295{
1296    std::string path;
1297
1298    int index = 0;
1299    if (!tc->getMemProxy().tryReadString(path,
1300                process->getSyscallArg(tc, index))) {
1301        return -EFAULT;
1302    }
1303    Addr bufPtr = process->getSyscallArg(tc, index);
1304
1305    // Adjust path for cwd and redirection
1306    path = process->checkPathRedirect(path);
1307
1308    struct stat hostBuf;
1309    int result = stat(path.c_str(), &hostBuf);
1310
1311    if (result < 0)
1312        return -errno;
1313
1314    copyOutStatBuf<OS>(tc->getMemProxy(), bufPtr, &hostBuf);
1315
1316    return 0;
1317}
1318
1319
1320/// Target stat64() handler.
1321template <class OS>
1322SyscallReturn
1323stat64Func(SyscallDesc *desc, int callnum, Process *process,
1324           ThreadContext *tc)
1325{
1326    std::string path;
1327
1328    int index = 0;
1329    if (!tc->getMemProxy().tryReadString(path,
1330                process->getSyscallArg(tc, index)))
1331        return -EFAULT;
1332    Addr bufPtr = process->getSyscallArg(tc, index);
1333
1334    // Adjust path for cwd and redirection
1335    path = process->checkPathRedirect(path);
1336
1337#if NO_STAT64
1338    struct stat  hostBuf;
1339    int result = stat(path.c_str(), &hostBuf);
1340#else
1341    struct stat64 hostBuf;
1342    int result = stat64(path.c_str(), &hostBuf);
1343#endif
1344
1345    if (result < 0)
1346        return -errno;
1347
1348    copyOutStat64Buf<OS>(tc->getMemProxy(), bufPtr, &hostBuf);
1349
1350    return 0;
1351}
1352
1353
1354/// Target fstatat64() handler.
1355template <class OS>
1356SyscallReturn
1357fstatat64Func(SyscallDesc *desc, int callnum, Process *process,
1358              ThreadContext *tc)
1359{
1360    int index = 0;
1361    int dirfd = process->getSyscallArg(tc, index);
1362    if (dirfd != OS::TGT_AT_FDCWD)
1363        warn("fstatat64: first argument not AT_FDCWD; unlikely to work");
1364
1365    std::string path;
1366    if (!tc->getMemProxy().tryReadString(path,
1367                process->getSyscallArg(tc, index)))
1368        return -EFAULT;
1369    Addr bufPtr = process->getSyscallArg(tc, index);
1370
1371    // Adjust path for cwd and redirection
1372    path = process->checkPathRedirect(path);
1373
1374#if NO_STAT64
1375    struct stat  hostBuf;
1376    int result = stat(path.c_str(), &hostBuf);
1377#else
1378    struct stat64 hostBuf;
1379    int result = stat64(path.c_str(), &hostBuf);
1380#endif
1381
1382    if (result < 0)
1383        return -errno;
1384
1385    copyOutStat64Buf<OS>(tc->getMemProxy(), bufPtr, &hostBuf);
1386
1387    return 0;
1388}
1389
1390
1391/// Target fstat64() handler.
1392template <class OS>
1393SyscallReturn
1394fstat64Func(SyscallDesc *desc, int callnum, Process *p, ThreadContext *tc)
1395{
1396    int index = 0;
1397    int tgt_fd = p->getSyscallArg(tc, index);
1398    Addr bufPtr = p->getSyscallArg(tc, index);
1399
1400    auto ffdp = std::dynamic_pointer_cast<FileFDEntry>((*p->fds)[tgt_fd]);
1401    if (!ffdp)
1402        return -EBADF;
1403    int sim_fd = ffdp->getSimFD();
1404
1405#if NO_STAT64
1406    struct stat  hostBuf;
1407    int result = fstat(sim_fd, &hostBuf);
1408#else
1409    struct stat64  hostBuf;
1410    int result = fstat64(sim_fd, &hostBuf);
1411#endif
1412
1413    if (result < 0)
1414        return -errno;
1415
1416    copyOutStat64Buf<OS>(tc->getMemProxy(), bufPtr, &hostBuf, (sim_fd == 1));
1417
1418    return 0;
1419}
1420
1421
1422/// Target lstat() handler.
1423template <class OS>
1424SyscallReturn
1425lstatFunc(SyscallDesc *desc, int callnum, Process *process,
1426          ThreadContext *tc)
1427{
1428    std::string path;
1429
1430    int index = 0;
1431    if (!tc->getMemProxy().tryReadString(path,
1432                process->getSyscallArg(tc, index))) {
1433        return -EFAULT;
1434    }
1435    Addr bufPtr = process->getSyscallArg(tc, index);
1436
1437    // Adjust path for cwd and redirection
1438    path = process->checkPathRedirect(path);
1439
1440    struct stat hostBuf;
1441    int result = lstat(path.c_str(), &hostBuf);
1442
1443    if (result < 0)
1444        return -errno;
1445
1446    copyOutStatBuf<OS>(tc->getMemProxy(), bufPtr, &hostBuf);
1447
1448    return 0;
1449}
1450
1451/// Target lstat64() handler.
1452template <class OS>
1453SyscallReturn
1454lstat64Func(SyscallDesc *desc, int callnum, Process *process,
1455            ThreadContext *tc)
1456{
1457    std::string path;
1458
1459    int index = 0;
1460    if (!tc->getMemProxy().tryReadString(path,
1461                process->getSyscallArg(tc, index))) {
1462        return -EFAULT;
1463    }
1464    Addr bufPtr = process->getSyscallArg(tc, index);
1465
1466    // Adjust path for cwd and redirection
1467    path = process->checkPathRedirect(path);
1468
1469#if NO_STAT64
1470    struct stat hostBuf;
1471    int result = lstat(path.c_str(), &hostBuf);
1472#else
1473    struct stat64 hostBuf;
1474    int result = lstat64(path.c_str(), &hostBuf);
1475#endif
1476
1477    if (result < 0)
1478        return -errno;
1479
1480    copyOutStat64Buf<OS>(tc->getMemProxy(), bufPtr, &hostBuf);
1481
1482    return 0;
1483}
1484
1485/// Target fstat() handler.
1486template <class OS>
1487SyscallReturn
1488fstatFunc(SyscallDesc *desc, int callnum, Process *p, ThreadContext *tc)
1489{
1490    int index = 0;
1491    int tgt_fd = p->getSyscallArg(tc, index);
1492    Addr bufPtr = p->getSyscallArg(tc, index);
1493
1494    DPRINTF_SYSCALL(Verbose, "fstat(%d, ...)\n", tgt_fd);
1495
1496    auto ffdp = std::dynamic_pointer_cast<FileFDEntry>((*p->fds)[tgt_fd]);
1497    if (!ffdp)
1498        return -EBADF;
1499    int sim_fd = ffdp->getSimFD();
1500
1501    struct stat hostBuf;
1502    int result = fstat(sim_fd, &hostBuf);
1503
1504    if (result < 0)
1505        return -errno;
1506
1507    copyOutStatBuf<OS>(tc->getMemProxy(), bufPtr, &hostBuf, (sim_fd == 1));
1508
1509    return 0;
1510}
1511
1512/// Target statfs() handler.
1513template <class OS>
1514SyscallReturn
1515statfsFunc(SyscallDesc *desc, int callnum, Process *process,
1516           ThreadContext *tc)
1517{
1518#ifdef __linux__
1519    std::string path;
1520
1521    int index = 0;
1522    if (!tc->getMemProxy().tryReadString(path,
1523                process->getSyscallArg(tc, index))) {
1524        return -EFAULT;
1525    }
1526    Addr bufPtr = process->getSyscallArg(tc, index);
1527
1528    // Adjust path for cwd and redirection
1529    path = process->checkPathRedirect(path);
1530
1531    struct statfs hostBuf;
1532    int result = statfs(path.c_str(), &hostBuf);
1533
1534    if (result < 0)
1535        return -errno;
1536
1537    copyOutStatfsBuf<OS>(tc->getMemProxy(), bufPtr, &hostBuf);
1538    return 0;
1539#else
1540    warnUnsupportedOS("statfs");
1541    return -1;
1542#endif
1543}
1544
1545template <class OS>
1546SyscallReturn
1547cloneFunc(SyscallDesc *desc, int callnum, Process *p, ThreadContext *tc)
1548{
1549    int index = 0;
1550
1551    RegVal flags = p->getSyscallArg(tc, index);
1552    RegVal newStack = p->getSyscallArg(tc, index);
1553    Addr ptidPtr = p->getSyscallArg(tc, index);
1554
1555#if THE_ISA == RISCV_ISA or THE_ISA == ARM_ISA
1556    /**
1557     * Linux sets CLONE_BACKWARDS flag for RISC-V and Arm.
1558     * The flag defines the list of clone() arguments in the following
1559     * order: flags -> newStack -> ptidPtr -> tlsPtr -> ctidPtr
1560     */
1561    Addr tlsPtr = p->getSyscallArg(tc, index);
1562    Addr ctidPtr = p->getSyscallArg(tc, index);
1563#else
1564    Addr ctidPtr = p->getSyscallArg(tc, index);
1565    Addr tlsPtr = p->getSyscallArg(tc, index);
1566#endif
1567
1568    if (((flags & OS::TGT_CLONE_SIGHAND)&& !(flags & OS::TGT_CLONE_VM)) ||
1569        ((flags & OS::TGT_CLONE_THREAD) && !(flags & OS::TGT_CLONE_SIGHAND)) ||
1570        ((flags & OS::TGT_CLONE_FS)     &&  (flags & OS::TGT_CLONE_NEWNS)) ||
1571        ((flags & OS::TGT_CLONE_NEWIPC) &&  (flags & OS::TGT_CLONE_SYSVSEM)) ||
1572        ((flags & OS::TGT_CLONE_NEWPID) &&  (flags & OS::TGT_CLONE_THREAD)) ||
1573        ((flags & OS::TGT_CLONE_VM)     && !(newStack)))
1574        return -EINVAL;
1575
1576    ThreadContext *ctc;
1577    if (!(ctc = p->findFreeContext())) {
1578        DPRINTF_SYSCALL(Verbose, "clone: no spare thread context in system"
1579                        "[cpu %d, thread %d]", tc->cpuId(), tc->threadId());
1580        return -EAGAIN;
1581    }
1582
1583    /**
1584     * Note that ProcessParams is generated by swig and there are no other
1585     * examples of how to create anything but this default constructor. The
1586     * fields are manually initialized instead of passing parameters to the
1587     * constructor.
1588     */
1589    ProcessParams *pp = new ProcessParams();
1590    pp->executable.assign(*(new std::string(p->progName())));
1591    pp->cmd.push_back(*(new std::string(p->progName())));
1592    pp->system = p->system;
1593    pp->cwd.assign(p->tgtCwd);
1594    pp->input.assign("stdin");
1595    pp->output.assign("stdout");
1596    pp->errout.assign("stderr");
1597    pp->uid = p->uid();
1598    pp->euid = p->euid();
1599    pp->gid = p->gid();
1600    pp->egid = p->egid();
1601
1602    /* Find the first free PID that's less than the maximum */
1603    std::set<int> const& pids = p->system->PIDs;
1604    int temp_pid = *pids.begin();
1605    do {
1606        temp_pid++;
1607    } while (pids.find(temp_pid) != pids.end());
1608    if (temp_pid >= System::maxPID)
1609        fatal("temp_pid is too large: %d", temp_pid);
1610
1611    pp->pid = temp_pid;
1612    pp->ppid = (flags & OS::TGT_CLONE_THREAD) ? p->ppid() : p->pid();
1613    pp->useArchPT = p->useArchPT;
1614    pp->kvmInSE = p->kvmInSE;
1615    Process *cp = pp->create();
1616    delete pp;
1617
1618    Process *owner = ctc->getProcessPtr();
1619    ctc->setProcessPtr(cp);
1620    cp->assignThreadContext(ctc->contextId());
1621    owner->revokeThreadContext(ctc->contextId());
1622
1623    if (flags & OS::TGT_CLONE_PARENT_SETTID) {
1624        BufferArg ptidBuf(ptidPtr, sizeof(long));
1625        long *ptid = (long *)ptidBuf.bufferPtr();
1626        *ptid = cp->pid();
1627        ptidBuf.copyOut(tc->getMemProxy());
1628    }
1629
1630    if (flags & OS::TGT_CLONE_THREAD) {
1631        cp->pTable->shared = true;
1632        cp->useForClone = true;
1633    }
1634    cp->initState();
1635    p->clone(tc, ctc, cp, flags);
1636
1637    if (flags & OS::TGT_CLONE_THREAD) {
1638        delete cp->sigchld;
1639        cp->sigchld = p->sigchld;
1640    } else if (flags & OS::TGT_SIGCHLD) {
1641        *cp->sigchld = true;
1642    }
1643
1644    if (flags & OS::TGT_CLONE_CHILD_SETTID) {
1645        BufferArg ctidBuf(ctidPtr, sizeof(long));
1646        long *ctid = (long *)ctidBuf.bufferPtr();
1647        *ctid = cp->pid();
1648        ctidBuf.copyOut(ctc->getMemProxy());
1649    }
1650
1651    if (flags & OS::TGT_CLONE_CHILD_CLEARTID)
1652        cp->childClearTID = (uint64_t)ctidPtr;
1653
1654    ctc->clearArchRegs();
1655
1656    OS::archClone(flags, p, cp, tc, ctc, newStack, tlsPtr);
1657
1658    cp->setSyscallReturn(ctc, 0);
1659
1660#if THE_ISA == ALPHA_ISA
1661    ctc->setIntReg(TheISA::SyscallSuccessReg, 0);
1662#elif THE_ISA == SPARC_ISA
1663    tc->setIntReg(TheISA::SyscallPseudoReturnReg, 0);
1664    ctc->setIntReg(TheISA::SyscallPseudoReturnReg, 1);
1665#endif
1666
1667    if (p->kvmInSE) {
1668#if THE_ISA == X86_ISA
1669        ctc->pcState(tc->readIntReg(TheISA::INTREG_RCX));
1670#else
1671        panic("KVM CPU model is not supported for this ISA");
1672#endif
1673    } else {
1674        TheISA::PCState cpc = tc->pcState();
1675        cpc.advance();
1676        ctc->pcState(cpc);
1677    }
1678    ctc->activate();
1679
1680    return cp->pid();
1681}
1682
1683/// Target fstatfs() handler.
1684template <class OS>
1685SyscallReturn
1686fstatfsFunc(SyscallDesc *desc, int callnum, Process *p, ThreadContext *tc)
1687{
1688    int index = 0;
1689    int tgt_fd = p->getSyscallArg(tc, index);
1690    Addr bufPtr = p->getSyscallArg(tc, index);
1691
1692    auto ffdp = std::dynamic_pointer_cast<FileFDEntry>((*p->fds)[tgt_fd]);
1693    if (!ffdp)
1694        return -EBADF;
1695    int sim_fd = ffdp->getSimFD();
1696
1697    struct statfs hostBuf;
1698    int result = fstatfs(sim_fd, &hostBuf);
1699
1700    if (result < 0)
1701        return -errno;
1702
1703    copyOutStatfsBuf<OS>(tc->getMemProxy(), bufPtr, &hostBuf);
1704
1705    return 0;
1706}
1707
1708/// Target readv() handler.
1709template <class OS>
1710SyscallReturn
1711readvFunc(SyscallDesc *desc, int callnum, Process *p, ThreadContext *tc)
1712{
1713    int index = 0;
1714    int tgt_fd = p->getSyscallArg(tc, index);
1715
1716    auto ffdp = std::dynamic_pointer_cast<FileFDEntry>((*p->fds)[tgt_fd]);
1717    if (!ffdp)
1718        return -EBADF;
1719    int sim_fd = ffdp->getSimFD();
1720
1721    SETranslatingPortProxy &prox = tc->getMemProxy();
1722    uint64_t tiov_base = p->getSyscallArg(tc, index);
1723    size_t count = p->getSyscallArg(tc, index);
1724    typename OS::tgt_iovec tiov[count];
1725    struct iovec hiov[count];
1726    for (size_t i = 0; i < count; ++i) {
1727        prox.readBlob(tiov_base + (i * sizeof(typename OS::tgt_iovec)),
1728                      (uint8_t*)&tiov[i], sizeof(typename OS::tgt_iovec));
1729        hiov[i].iov_len = TheISA::gtoh(tiov[i].iov_len);
1730        hiov[i].iov_base = new char [hiov[i].iov_len];
1731    }
1732
1733    int result = readv(sim_fd, hiov, count);
1734    int local_errno = errno;
1735
1736    for (size_t i = 0; i < count; ++i) {
1737        if (result != -1) {
1738            prox.writeBlob(TheISA::htog(tiov[i].iov_base),
1739                           (uint8_t*)hiov[i].iov_base, hiov[i].iov_len);
1740        }
1741        delete [] (char *)hiov[i].iov_base;
1742    }
1743
1744    return (result == -1) ? -local_errno : result;
1745}
1746
1747/// Target writev() handler.
1748template <class OS>
1749SyscallReturn
1750writevFunc(SyscallDesc *desc, int callnum, Process *p, ThreadContext *tc)
1751{
1752    int index = 0;
1753    int tgt_fd = p->getSyscallArg(tc, index);
1754
1755    auto hbfdp = std::dynamic_pointer_cast<HBFDEntry>((*p->fds)[tgt_fd]);
1756    if (!hbfdp)
1757        return -EBADF;
1758    int sim_fd = hbfdp->getSimFD();
1759
1760    SETranslatingPortProxy &prox = tc->getMemProxy();
1761    uint64_t tiov_base = p->getSyscallArg(tc, index);
1762    size_t count = p->getSyscallArg(tc, index);
1763    struct iovec hiov[count];
1764    for (size_t i = 0; i < count; ++i) {
1765        typename OS::tgt_iovec tiov;
1766
1767        prox.readBlob(tiov_base + i*sizeof(typename OS::tgt_iovec),
1768                      (uint8_t*)&tiov, sizeof(typename OS::tgt_iovec));
1769        hiov[i].iov_len = TheISA::gtoh(tiov.iov_len);
1770        hiov[i].iov_base = new char [hiov[i].iov_len];
1771        prox.readBlob(TheISA::gtoh(tiov.iov_base), (uint8_t *)hiov[i].iov_base,
1772                      hiov[i].iov_len);
1773    }
1774
1775    int result = writev(sim_fd, hiov, count);
1776
1777    for (size_t i = 0; i < count; ++i)
1778        delete [] (char *)hiov[i].iov_base;
1779
1780    return (result == -1) ? -errno : result;
1781}
1782
1783/// Real mmap handler.
1784template <class OS>
1785SyscallReturn
1786mmapImpl(SyscallDesc *desc, int num, Process *p, ThreadContext *tc,
1787         bool is_mmap2)
1788{
1789    int index = 0;
1790    Addr start = p->getSyscallArg(tc, index);
1791    uint64_t length = p->getSyscallArg(tc, index);
1792    int prot = p->getSyscallArg(tc, index);
1793    int tgt_flags = p->getSyscallArg(tc, index);
1794    int tgt_fd = p->getSyscallArg(tc, index);
1795    int offset = p->getSyscallArg(tc, index);
1796
1797    if (is_mmap2)
1798        offset *= TheISA::PageBytes;
1799
1800    if (start & (TheISA::PageBytes - 1) ||
1801        offset & (TheISA::PageBytes - 1) ||
1802        (tgt_flags & OS::TGT_MAP_PRIVATE &&
1803         tgt_flags & OS::TGT_MAP_SHARED) ||
1804        (!(tgt_flags & OS::TGT_MAP_PRIVATE) &&
1805         !(tgt_flags & OS::TGT_MAP_SHARED)) ||
1806        !length) {
1807        return -EINVAL;
1808    }
1809
1810    if ((prot & PROT_WRITE) && (tgt_flags & OS::TGT_MAP_SHARED)) {
1811        // With shared mmaps, there are two cases to consider:
1812        // 1) anonymous: writes should modify the mapping and this should be
1813        // visible to observers who share the mapping. Currently, it's
1814        // difficult to update the shared mapping because there's no
1815        // structure which maintains information about the which virtual
1816        // memory areas are shared. If that structure existed, it would be
1817        // possible to make the translations point to the same frames.
1818        // 2) file-backed: writes should modify the mapping and the file
1819        // which is backed by the mapping. The shared mapping problem is the
1820        // same as what was mentioned about the anonymous mappings. For
1821        // file-backed mappings, the writes to the file are difficult
1822        // because it requires syncing what the mapping holds with the file
1823        // that resides on the host system. So, any write on a real system
1824        // would cause the change to be propagated to the file mapping at
1825        // some point in the future (the inode is tracked along with the
1826        // mapping). This isn't guaranteed to always happen, but it usually
1827        // works well enough. The guarantee is provided by the msync system
1828        // call. We could force the change through with shared mappings with
1829        // a call to msync, but that again would require more information
1830        // than we currently maintain.
1831        warn("mmap: writing to shared mmap region is currently "
1832             "unsupported. The write succeeds on the target, but it "
1833             "will not be propagated to the host or shared mappings");
1834    }
1835
1836    length = roundUp(length, TheISA::PageBytes);
1837
1838    int sim_fd = -1;
1839    uint8_t *pmap = nullptr;
1840    if (!(tgt_flags & OS::TGT_MAP_ANONYMOUS)) {
1841        std::shared_ptr<FDEntry> fdep = (*p->fds)[tgt_fd];
1842
1843        auto dfdp = std::dynamic_pointer_cast<DeviceFDEntry>(fdep);
1844        if (dfdp) {
1845            EmulatedDriver *emul_driver = dfdp->getDriver();
1846            return emul_driver->mmap(p, tc, start, length, prot,
1847                                     tgt_flags, tgt_fd, offset);
1848        }
1849
1850        auto ffdp = std::dynamic_pointer_cast<FileFDEntry>(fdep);
1851        if (!ffdp)
1852            return -EBADF;
1853        sim_fd = ffdp->getSimFD();
1854
1855        pmap = (decltype(pmap))mmap(nullptr, length, PROT_READ, MAP_PRIVATE,
1856                                    sim_fd, offset);
1857
1858        if (pmap == (decltype(pmap))-1) {
1859            warn("mmap: failed to map file into host address space");
1860            return -errno;
1861        }
1862    }
1863
1864    // Extend global mmap region if necessary. Note that we ignore the
1865    // start address unless MAP_FIXED is specified.
1866    if (!(tgt_flags & OS::TGT_MAP_FIXED)) {
1867        std::shared_ptr<MemState> mem_state = p->memState;
1868        Addr mmap_end = mem_state->getMmapEnd();
1869
1870        start = p->mmapGrowsDown() ? mmap_end - length : mmap_end;
1871        mmap_end = p->mmapGrowsDown() ? start : mmap_end + length;
1872
1873        mem_state->setMmapEnd(mmap_end);
1874    }
1875
1876    DPRINTF_SYSCALL(Verbose, " mmap range is 0x%x - 0x%x\n",
1877                    start, start + length - 1);
1878
1879    // We only allow mappings to overwrite existing mappings if
1880    // TGT_MAP_FIXED is set. Otherwise it shouldn't be a problem
1881    // because we ignore the start hint if TGT_MAP_FIXED is not set.
1882    int clobber = tgt_flags & OS::TGT_MAP_FIXED;
1883    if (clobber) {
1884        for (auto tc : p->system->threadContexts) {
1885            // If we might be overwriting old mappings, we need to
1886            // invalidate potentially stale mappings out of the TLBs.
1887            tc->getDTBPtr()->flushAll();
1888            tc->getITBPtr()->flushAll();
1889        }
1890    }
1891
1892    // Allocate physical memory and map it in. If the page table is already
1893    // mapped and clobber is not set, the simulator will issue throw a
1894    // fatal and bail out of the simulation.
1895    p->allocateMem(start, length, clobber);
1896
1897    // Transfer content into target address space.
1898    SETranslatingPortProxy &tp = tc->getMemProxy();
1899    if (tgt_flags & OS::TGT_MAP_ANONYMOUS) {
1900        // In general, we should zero the mapped area for anonymous mappings,
1901        // with something like:
1902        //     tp.memsetBlob(start, 0, length);
1903        // However, given that we don't support sparse mappings, and
1904        // some applications can map a couple of gigabytes of space
1905        // (intending sparse usage), that can get painfully expensive.
1906        // Fortunately, since we don't properly implement munmap either,
1907        // there's no danger of remapping used memory, so for now all
1908        // newly mapped memory should already be zeroed so we can skip it.
1909    } else {
1910        // It is possible to mmap an area larger than a file, however
1911        // accessing unmapped portions the system triggers a "Bus error"
1912        // on the host. We must know when to stop copying the file from
1913        // the host into the target address space.
1914        struct stat file_stat;
1915        if (fstat(sim_fd, &file_stat) > 0)
1916            fatal("mmap: cannot stat file");
1917
1918        // Copy the portion of the file that is resident. This requires
1919        // checking both the mmap size and the filesize that we are
1920        // trying to mmap into this space; the mmap size also depends
1921        // on the specified offset into the file.
1922        uint64_t size = std::min((uint64_t)file_stat.st_size - offset,
1923                                 length);
1924        tp.writeBlob(start, pmap, size);
1925
1926        // Cleanup the mmap region before exiting this function.
1927        munmap(pmap, length);
1928
1929        // Maintain the symbol table for dynamic executables.
1930        // The loader will call mmap to map the images into its address
1931        // space and we intercept that here. We can verify that we are
1932        // executing inside the loader by checking the program counter value.
1933        // XXX: with multiprogrammed workloads or multi-node configurations,
1934        // this will not work since there is a single global symbol table.
1935        ObjectFile *interpreter = p->getInterpreter();
1936        if (interpreter) {
1937            Addr text_start = interpreter->textBase();
1938            Addr text_end = text_start + interpreter->textSize();
1939
1940            Addr pc = tc->pcState().pc();
1941
1942            if (pc >= text_start && pc < text_end) {
1943                std::shared_ptr<FDEntry> fdep = (*p->fds)[tgt_fd];
1944                auto ffdp = std::dynamic_pointer_cast<FileFDEntry>(fdep);
1945                ObjectFile *lib = createObjectFile(ffdp->getFileName());
1946
1947                if (lib) {
1948                    lib->loadAllSymbols(debugSymbolTable,
1949                                        lib->textBase(), start);
1950                }
1951            }
1952        }
1953
1954        // Note that we do not zero out the remainder of the mapping. This
1955        // is done by a real system, but it probably will not affect
1956        // execution (hopefully).
1957    }
1958
1959    return start;
1960}
1961
1962template <class OS>
1963SyscallReturn
1964pwrite64Func(SyscallDesc *desc, int num, Process *p, ThreadContext *tc)
1965{
1966    int index = 0;
1967    int tgt_fd = p->getSyscallArg(tc, index);
1968    Addr bufPtr = p->getSyscallArg(tc, index);
1969    int nbytes = p->getSyscallArg(tc, index);
1970    int offset = p->getSyscallArg(tc, index);
1971
1972    auto ffdp = std::dynamic_pointer_cast<FileFDEntry>((*p->fds)[tgt_fd]);
1973    if (!ffdp)
1974        return -EBADF;
1975    int sim_fd = ffdp->getSimFD();
1976
1977    BufferArg bufArg(bufPtr, nbytes);
1978    bufArg.copyIn(tc->getMemProxy());
1979
1980    int bytes_written = pwrite(sim_fd, bufArg.bufferPtr(), nbytes, offset);
1981
1982    return (bytes_written == -1) ? -errno : bytes_written;
1983}
1984
1985/// Target mmap() handler.
1986template <class OS>
1987SyscallReturn
1988mmapFunc(SyscallDesc *desc, int num, Process *p, ThreadContext *tc)
1989{
1990    return mmapImpl<OS>(desc, num, p, tc, false);
1991}
1992
1993/// Target mmap2() handler.
1994template <class OS>
1995SyscallReturn
1996mmap2Func(SyscallDesc *desc, int num, Process *p, ThreadContext *tc)
1997{
1998    return mmapImpl<OS>(desc, num, p, tc, true);
1999}
2000
2001/// Target getrlimit() handler.
2002template <class OS>
2003SyscallReturn
2004getrlimitFunc(SyscallDesc *desc, int callnum, Process *process,
2005              ThreadContext *tc)
2006{
2007    int index = 0;
2008    unsigned resource = process->getSyscallArg(tc, index);
2009    TypedBufferArg<typename OS::rlimit> rlp(process->getSyscallArg(tc, index));
2010
2011    switch (resource) {
2012      case OS::TGT_RLIMIT_STACK:
2013        // max stack size in bytes: make up a number (8MB for now)
2014        rlp->rlim_cur = rlp->rlim_max = 8 * 1024 * 1024;
2015        rlp->rlim_cur = TheISA::htog(rlp->rlim_cur);
2016        rlp->rlim_max = TheISA::htog(rlp->rlim_max);
2017        break;
2018
2019      case OS::TGT_RLIMIT_DATA:
2020        // max data segment size in bytes: make up a number
2021        rlp->rlim_cur = rlp->rlim_max = 256 * 1024 * 1024;
2022        rlp->rlim_cur = TheISA::htog(rlp->rlim_cur);
2023        rlp->rlim_max = TheISA::htog(rlp->rlim_max);
2024        break;
2025
2026      default:
2027        warn("getrlimit: unimplemented resource %d", resource);
2028        return -EINVAL;
2029        break;
2030    }
2031
2032    rlp.copyOut(tc->getMemProxy());
2033    return 0;
2034}
2035
2036template <class OS>
2037SyscallReturn
2038prlimitFunc(SyscallDesc *desc, int callnum, Process *process,
2039            ThreadContext *tc)
2040{
2041    int index = 0;
2042    if (process->getSyscallArg(tc, index) != 0)
2043    {
2044        warn("prlimit: ignoring rlimits for nonzero pid");
2045        return -EPERM;
2046    }
2047    int resource = process->getSyscallArg(tc, index);
2048    Addr n = process->getSyscallArg(tc, index);
2049    if (n != 0)
2050        warn("prlimit: ignoring new rlimit");
2051    Addr o = process->getSyscallArg(tc, index);
2052    if (o != 0)
2053    {
2054        TypedBufferArg<typename OS::rlimit> rlp(o);
2055        switch (resource) {
2056          case OS::TGT_RLIMIT_STACK:
2057            // max stack size in bytes: make up a number (8MB for now)
2058            rlp->rlim_cur = rlp->rlim_max = 8 * 1024 * 1024;
2059            rlp->rlim_cur = TheISA::htog(rlp->rlim_cur);
2060            rlp->rlim_max = TheISA::htog(rlp->rlim_max);
2061            break;
2062          case OS::TGT_RLIMIT_DATA:
2063            // max data segment size in bytes: make up a number
2064            rlp->rlim_cur = rlp->rlim_max = 256*1024*1024;
2065            rlp->rlim_cur = TheISA::htog(rlp->rlim_cur);
2066            rlp->rlim_max = TheISA::htog(rlp->rlim_max);
2067            break;
2068          default:
2069            warn("prlimit: unimplemented resource %d", resource);
2070            return -EINVAL;
2071            break;
2072        }
2073        rlp.copyOut(tc->getMemProxy());
2074    }
2075    return 0;
2076}
2077
2078/// Target clock_gettime() function.
2079template <class OS>
2080SyscallReturn
2081clock_gettimeFunc(SyscallDesc *desc, int num, Process *p, ThreadContext *tc)
2082{
2083    int index = 1;
2084    //int clk_id = p->getSyscallArg(tc, index);
2085    TypedBufferArg<typename OS::timespec> tp(p->getSyscallArg(tc, index));
2086
2087    getElapsedTimeNano(tp->tv_sec, tp->tv_nsec);
2088    tp->tv_sec += seconds_since_epoch;
2089    tp->tv_sec = TheISA::htog(tp->tv_sec);
2090    tp->tv_nsec = TheISA::htog(tp->tv_nsec);
2091
2092    tp.copyOut(tc->getMemProxy());
2093
2094    return 0;
2095}
2096
2097/// Target clock_getres() function.
2098template <class OS>
2099SyscallReturn
2100clock_getresFunc(SyscallDesc *desc, int num, Process *p, ThreadContext *tc)
2101{
2102    int index = 1;
2103    TypedBufferArg<typename OS::timespec> tp(p->getSyscallArg(tc, index));
2104
2105    // Set resolution at ns, which is what clock_gettime() returns
2106    tp->tv_sec = 0;
2107    tp->tv_nsec = 1;
2108
2109    tp.copyOut(tc->getMemProxy());
2110
2111    return 0;
2112}
2113
2114/// Target gettimeofday() handler.
2115template <class OS>
2116SyscallReturn
2117gettimeofdayFunc(SyscallDesc *desc, int callnum, Process *process,
2118                 ThreadContext *tc)
2119{
2120    int index = 0;
2121    TypedBufferArg<typename OS::timeval> tp(process->getSyscallArg(tc, index));
2122
2123    getElapsedTimeMicro(tp->tv_sec, tp->tv_usec);
2124    tp->tv_sec += seconds_since_epoch;
2125    tp->tv_sec = TheISA::htog(tp->tv_sec);
2126    tp->tv_usec = TheISA::htog(tp->tv_usec);
2127
2128    tp.copyOut(tc->getMemProxy());
2129
2130    return 0;
2131}
2132
2133
2134/// Target utimes() handler.
2135template <class OS>
2136SyscallReturn
2137utimesFunc(SyscallDesc *desc, int callnum, Process *process,
2138           ThreadContext *tc)
2139{
2140    std::string path;
2141
2142    int index = 0;
2143    if (!tc->getMemProxy().tryReadString(path,
2144                process->getSyscallArg(tc, index))) {
2145        return -EFAULT;
2146    }
2147
2148    TypedBufferArg<typename OS::timeval [2]>
2149        tp(process->getSyscallArg(tc, index));
2150    tp.copyIn(tc->getMemProxy());
2151
2152    struct timeval hostTimeval[2];
2153    for (int i = 0; i < 2; ++i) {
2154        hostTimeval[i].tv_sec = TheISA::gtoh((*tp)[i].tv_sec);
2155        hostTimeval[i].tv_usec = TheISA::gtoh((*tp)[i].tv_usec);
2156    }
2157
2158    // Adjust path for cwd and redirection
2159    path = process->checkPathRedirect(path);
2160
2161    int result = utimes(path.c_str(), hostTimeval);
2162
2163    if (result < 0)
2164        return -errno;
2165
2166    return 0;
2167}
2168
2169template <class OS>
2170SyscallReturn
2171execveFunc(SyscallDesc *desc, int callnum, Process *p, ThreadContext *tc)
2172{
2173    desc->setFlags(0);
2174
2175    int index = 0;
2176    std::string path;
2177    SETranslatingPortProxy & mem_proxy = tc->getMemProxy();
2178    if (!mem_proxy.tryReadString(path, p->getSyscallArg(tc, index)))
2179        return -EFAULT;
2180
2181    if (access(path.c_str(), F_OK) == -1)
2182        return -EACCES;
2183
2184    auto read_in = [](std::vector<std::string> & vect,
2185                      SETranslatingPortProxy & mem_proxy,
2186                      Addr mem_loc)
2187    {
2188        for (int inc = 0; ; inc++) {
2189            BufferArg b((mem_loc + sizeof(Addr) * inc), sizeof(Addr));
2190            b.copyIn(mem_proxy);
2191
2192            if (!*(Addr*)b.bufferPtr())
2193                break;
2194
2195            vect.push_back(std::string());
2196            mem_proxy.tryReadString(vect[inc], *(Addr*)b.bufferPtr());
2197        }
2198    };
2199
2200    /**
2201     * Note that ProcessParams is generated by swig and there are no other
2202     * examples of how to create anything but this default constructor. The
2203     * fields are manually initialized instead of passing parameters to the
2204     * constructor.
2205     */
2206    ProcessParams *pp = new ProcessParams();
2207    pp->executable = path;
2208    Addr argv_mem_loc = p->getSyscallArg(tc, index);
2209    read_in(pp->cmd, mem_proxy, argv_mem_loc);
2210    Addr envp_mem_loc = p->getSyscallArg(tc, index);
2211    read_in(pp->env, mem_proxy, envp_mem_loc);
2212    pp->uid = p->uid();
2213    pp->egid = p->egid();
2214    pp->euid = p->euid();
2215    pp->gid = p->gid();
2216    pp->ppid = p->ppid();
2217    pp->pid = p->pid();
2218    pp->input.assign("cin");
2219    pp->output.assign("cout");
2220    pp->errout.assign("cerr");
2221    pp->cwd.assign(p->tgtCwd);
2222    pp->system = p->system;
2223    /**
2224     * Prevent process object creation with identical PIDs (which will trip
2225     * a fatal check in Process constructor). The execve call is supposed to
2226     * take over the currently executing process' identity but replace
2227     * whatever it is doing with a new process image. Instead of hijacking
2228     * the process object in the simulator, we create a new process object
2229     * and bind to the previous process' thread below (hijacking the thread).
2230     */
2231    p->system->PIDs.erase(p->pid());
2232    Process *new_p = pp->create();
2233    delete pp;
2234
2235    /**
2236     * Work through the file descriptor array and close any files marked
2237     * close-on-exec.
2238     */
2239    new_p->fds = p->fds;
2240    for (int i = 0; i < new_p->fds->getSize(); i++) {
2241        std::shared_ptr<FDEntry> fdep = (*new_p->fds)[i];
2242        if (fdep && fdep->getCOE())
2243            new_p->fds->closeFDEntry(i);
2244    }
2245
2246    *new_p->sigchld = true;
2247
2248    delete p;
2249    tc->clearArchRegs();
2250    tc->setProcessPtr(new_p);
2251    new_p->assignThreadContext(tc->contextId());
2252    new_p->initState();
2253    tc->activate();
2254    TheISA::PCState pcState = tc->pcState();
2255    tc->setNPC(pcState.instAddr());
2256
2257    desc->setFlags(SyscallDesc::SuppressReturnValue);
2258    return 0;
2259}
2260
2261/// Target getrusage() function.
2262template <class OS>
2263SyscallReturn
2264getrusageFunc(SyscallDesc *desc, int callnum, Process *process,
2265              ThreadContext *tc)
2266{
2267    int index = 0;
2268    int who = process->getSyscallArg(tc, index); // THREAD, SELF, or CHILDREN
2269    TypedBufferArg<typename OS::rusage> rup(process->getSyscallArg(tc, index));
2270
2271    rup->ru_utime.tv_sec = 0;
2272    rup->ru_utime.tv_usec = 0;
2273    rup->ru_stime.tv_sec = 0;
2274    rup->ru_stime.tv_usec = 0;
2275    rup->ru_maxrss = 0;
2276    rup->ru_ixrss = 0;
2277    rup->ru_idrss = 0;
2278    rup->ru_isrss = 0;
2279    rup->ru_minflt = 0;
2280    rup->ru_majflt = 0;
2281    rup->ru_nswap = 0;
2282    rup->ru_inblock = 0;
2283    rup->ru_oublock = 0;
2284    rup->ru_msgsnd = 0;
2285    rup->ru_msgrcv = 0;
2286    rup->ru_nsignals = 0;
2287    rup->ru_nvcsw = 0;
2288    rup->ru_nivcsw = 0;
2289
2290    switch (who) {
2291      case OS::TGT_RUSAGE_SELF:
2292        getElapsedTimeMicro(rup->ru_utime.tv_sec, rup->ru_utime.tv_usec);
2293        rup->ru_utime.tv_sec = TheISA::htog(rup->ru_utime.tv_sec);
2294        rup->ru_utime.tv_usec = TheISA::htog(rup->ru_utime.tv_usec);
2295        break;
2296
2297      case OS::TGT_RUSAGE_CHILDREN:
2298        // do nothing.  We have no child processes, so they take no time.
2299        break;
2300
2301      default:
2302        // don't really handle THREAD or CHILDREN, but just warn and
2303        // plow ahead
2304        warn("getrusage() only supports RUSAGE_SELF.  Parameter %d ignored.",
2305             who);
2306    }
2307
2308    rup.copyOut(tc->getMemProxy());
2309
2310    return 0;
2311}
2312
2313/// Target times() function.
2314template <class OS>
2315SyscallReturn
2316timesFunc(SyscallDesc *desc, int callnum, Process *process,
2317          ThreadContext *tc)
2318{
2319    int index = 0;
2320    TypedBufferArg<typename OS::tms> bufp(process->getSyscallArg(tc, index));
2321
2322    // Fill in the time structure (in clocks)
2323    int64_t clocks = curTick() * OS::M5_SC_CLK_TCK / SimClock::Int::s;
2324    bufp->tms_utime = clocks;
2325    bufp->tms_stime = 0;
2326    bufp->tms_cutime = 0;
2327    bufp->tms_cstime = 0;
2328
2329    // Convert to host endianness
2330    bufp->tms_utime = TheISA::htog(bufp->tms_utime);
2331
2332    // Write back
2333    bufp.copyOut(tc->getMemProxy());
2334
2335    // Return clock ticks since system boot
2336    return clocks;
2337}
2338
2339/// Target time() function.
2340template <class OS>
2341SyscallReturn
2342timeFunc(SyscallDesc *desc, int callnum, Process *process, ThreadContext *tc)
2343{
2344    typename OS::time_t sec, usec;
2345    getElapsedTimeMicro(sec, usec);
2346    sec += seconds_since_epoch;
2347
2348    int index = 0;
2349    Addr taddr = (Addr)process->getSyscallArg(tc, index);
2350    if (taddr != 0) {
2351        typename OS::time_t t = sec;
2352        t = TheISA::htog(t);
2353        SETranslatingPortProxy &p = tc->getMemProxy();
2354        p.writeBlob(taddr, (uint8_t*)&t, (int)sizeof(typename OS::time_t));
2355    }
2356    return sec;
2357}
2358
2359template <class OS>
2360SyscallReturn
2361tgkillFunc(SyscallDesc *desc, int num, Process *process, ThreadContext *tc)
2362{
2363    int index = 0;
2364    int tgid = process->getSyscallArg(tc, index);
2365    int tid = process->getSyscallArg(tc, index);
2366    int sig = process->getSyscallArg(tc, index);
2367
2368    /**
2369     * This system call is intended to allow killing a specific thread
2370     * within an arbitrary thread group if sanctioned with permission checks.
2371     * It's usually true that threads share the termination signal as pointed
2372     * out by the pthread_kill man page and this seems to be the intended
2373     * usage. Due to this being an emulated environment, assume the following:
2374     * Threads are allowed to call tgkill because the EUID for all threads
2375     * should be the same. There is no signal handling mechanism for kernel
2376     * registration of signal handlers since signals are poorly supported in
2377     * emulation mode. Since signal handlers cannot be registered, all
2378     * threads within in a thread group must share the termination signal.
2379     * We never exhaust PIDs so there's no chance of finding the wrong one
2380     * due to PID rollover.
2381     */
2382
2383    System *sys = tc->getSystemPtr();
2384    Process *tgt_proc = nullptr;
2385    for (int i = 0; i < sys->numContexts(); i++) {
2386        Process *temp = sys->threadContexts[i]->getProcessPtr();
2387        if (temp->pid() == tid) {
2388            tgt_proc = temp;
2389            break;
2390        }
2391    }
2392
2393    if (sig != 0 || sig != OS::TGT_SIGABRT)
2394        return -EINVAL;
2395
2396    if (tgt_proc == nullptr)
2397        return -ESRCH;
2398
2399    if (tgid != -1 && tgt_proc->tgid() != tgid)
2400        return -ESRCH;
2401
2402    if (sig == OS::TGT_SIGABRT)
2403        exitGroupFunc(desc, 252, process, tc);
2404
2405    return 0;
2406}
2407
2408template <class OS>
2409SyscallReturn
2410socketFunc(SyscallDesc *desc, int num, Process *p, ThreadContext *tc)
2411{
2412    int index = 0;
2413    int domain = p->getSyscallArg(tc, index);
2414    int type = p->getSyscallArg(tc, index);
2415    int prot = p->getSyscallArg(tc, index);
2416
2417    int sim_fd = socket(domain, type, prot);
2418    if (sim_fd == -1)
2419        return -errno;
2420
2421    auto sfdp = std::make_shared<SocketFDEntry>(sim_fd, domain, type, prot);
2422    int tgt_fd = p->fds->allocFD(sfdp);
2423
2424    return tgt_fd;
2425}
2426
2427template <class OS>
2428SyscallReturn
2429socketpairFunc(SyscallDesc *desc, int num, Process *p, ThreadContext *tc)
2430{
2431    int index = 0;
2432    int domain = p->getSyscallArg(tc, index);
2433    int type = p->getSyscallArg(tc, index);
2434    int prot = p->getSyscallArg(tc, index);
2435    Addr svPtr = p->getSyscallArg(tc, index);
2436
2437    BufferArg svBuf((Addr)svPtr, 2 * sizeof(int));
2438    int status = socketpair(domain, type, prot, (int *)svBuf.bufferPtr());
2439    if (status == -1)
2440        return -errno;
2441
2442    int *fds = (int *)svBuf.bufferPtr();
2443
2444    auto sfdp1 = std::make_shared<SocketFDEntry>(fds[0], domain, type, prot);
2445    fds[0] = p->fds->allocFD(sfdp1);
2446    auto sfdp2 = std::make_shared<SocketFDEntry>(fds[1], domain, type, prot);
2447    fds[1] = p->fds->allocFD(sfdp2);
2448    svBuf.copyOut(tc->getMemProxy());
2449
2450    return status;
2451}
2452
2453template <class OS>
2454SyscallReturn
2455selectFunc(SyscallDesc *desc, int callnum, Process *p, ThreadContext *tc)
2456{
2457    int retval;
2458
2459    int index = 0;
2460    int nfds_t = p->getSyscallArg(tc, index);
2461    Addr fds_read_ptr = p->getSyscallArg(tc, index);
2462    Addr fds_writ_ptr = p->getSyscallArg(tc, index);
2463    Addr fds_excp_ptr = p->getSyscallArg(tc, index);
2464    Addr time_val_ptr = p->getSyscallArg(tc, index);
2465
2466    TypedBufferArg<typename OS::fd_set> rd_t(fds_read_ptr);
2467    TypedBufferArg<typename OS::fd_set> wr_t(fds_writ_ptr);
2468    TypedBufferArg<typename OS::fd_set> ex_t(fds_excp_ptr);
2469    TypedBufferArg<typename OS::timeval> tp(time_val_ptr);
2470
2471    /**
2472     * Host fields. Notice that these use the definitions from the system
2473     * headers instead of the gem5 headers and libraries. If the host and
2474     * target have different header file definitions, this will not work.
2475     */
2476    fd_set rd_h;
2477    FD_ZERO(&rd_h);
2478    fd_set wr_h;
2479    FD_ZERO(&wr_h);
2480    fd_set ex_h;
2481    FD_ZERO(&ex_h);
2482
2483    /**
2484     * Copy in the fd_set from the target.
2485     */
2486    if (fds_read_ptr)
2487        rd_t.copyIn(tc->getMemProxy());
2488    if (fds_writ_ptr)
2489        wr_t.copyIn(tc->getMemProxy());
2490    if (fds_excp_ptr)
2491        ex_t.copyIn(tc->getMemProxy());
2492
2493    /**
2494     * We need to translate the target file descriptor set into a host file
2495     * descriptor set. This involves both our internal process fd array
2496     * and the fd_set defined in Linux header files. The nfds field also
2497     * needs to be updated as it will be only target specific after
2498     * retrieving it from the target; the nfds value is expected to be the
2499     * highest file descriptor that needs to be checked, so we need to extend
2500     * it out for nfds_h when we do the update.
2501     */
2502    int nfds_h = 0;
2503    std::map<int, int> trans_map;
2504    auto try_add_host_set = [&](fd_set *tgt_set_entry,
2505                                fd_set *hst_set_entry,
2506                                int iter) -> bool
2507    {
2508        /**
2509         * By this point, we know that we are looking at a valid file
2510         * descriptor set on the target. We need to check if the target file
2511         * descriptor value passed in as iter is part of the set.
2512         */
2513        if (FD_ISSET(iter, tgt_set_entry)) {
2514            /**
2515             * We know that the target file descriptor belongs to the set,
2516             * but we do not yet know if the file descriptor is valid or
2517             * that we have a host mapping. Check that now.
2518             */
2519            auto hbfdp = std::dynamic_pointer_cast<HBFDEntry>((*p->fds)[iter]);
2520            if (!hbfdp)
2521                return true;
2522            auto sim_fd = hbfdp->getSimFD();
2523
2524            /**
2525             * Add the sim_fd to tgt_fd translation into trans_map for use
2526             * later when we need to zero the target fd_set structures and
2527             * then update them with hits returned from the host select call.
2528             */
2529            trans_map[sim_fd] = iter;
2530
2531            /**
2532             * We know that the host file descriptor exists so now we check
2533             * if we need to update the max count for nfds_h before passing
2534             * the duplicated structure into the host.
2535             */
2536            nfds_h = std::max(nfds_h - 1, sim_fd + 1);
2537
2538            /**
2539             * Add the host file descriptor to the set that we are going to
2540             * pass into the host.
2541             */
2542            FD_SET(sim_fd, hst_set_entry);
2543        }
2544        return false;
2545    };
2546
2547    for (int i = 0; i < nfds_t; i++) {
2548        if (fds_read_ptr) {
2549            bool ebadf = try_add_host_set((fd_set*)&*rd_t, &rd_h, i);
2550            if (ebadf) return -EBADF;
2551        }
2552        if (fds_writ_ptr) {
2553            bool ebadf = try_add_host_set((fd_set*)&*wr_t, &wr_h, i);
2554            if (ebadf) return -EBADF;
2555        }
2556        if (fds_excp_ptr) {
2557            bool ebadf = try_add_host_set((fd_set*)&*ex_t, &ex_h, i);
2558            if (ebadf) return -EBADF;
2559        }
2560    }
2561
2562    if (time_val_ptr) {
2563        /**
2564         * It might be possible to decrement the timeval based on some
2565         * derivation of wall clock determined from elapsed simulator ticks
2566         * but that seems like overkill. Rather, we just set the timeval with
2567         * zero timeout. (There is no reason to block during the simulation
2568         * as it only decreases simulator performance.)
2569         */
2570        tp->tv_sec = 0;
2571        tp->tv_usec = 0;
2572
2573        retval = select(nfds_h,
2574                        fds_read_ptr ? &rd_h : nullptr,
2575                        fds_writ_ptr ? &wr_h : nullptr,
2576                        fds_excp_ptr ? &ex_h : nullptr,
2577                        (timeval*)&*tp);
2578    } else {
2579        /**
2580         * If the timeval pointer is null, setup a new timeval structure to
2581         * pass into the host select call. Unfortunately, we will need to
2582         * manually check the return value and throw a retry fault if the
2583         * return value is zero. Allowing the system call to block will
2584         * likely deadlock the event queue.
2585         */
2586        struct timeval tv = { 0, 0 };
2587
2588        retval = select(nfds_h,
2589                        fds_read_ptr ? &rd_h : nullptr,
2590                        fds_writ_ptr ? &wr_h : nullptr,
2591                        fds_excp_ptr ? &ex_h : nullptr,
2592                        &tv);
2593
2594        if (retval == 0) {
2595            /**
2596             * If blocking indefinitely, check the signal list to see if a
2597             * signal would break the poll out of the retry cycle and try to
2598             * return the signal interrupt instead.
2599             */
2600            for (auto sig : tc->getSystemPtr()->signalList)
2601                if (sig.receiver == p)
2602                    return -EINTR;
2603            return SyscallReturn::retry();
2604        }
2605    }
2606
2607    if (retval == -1)
2608        return -errno;
2609
2610    FD_ZERO((fd_set*)&*rd_t);
2611    FD_ZERO((fd_set*)&*wr_t);
2612    FD_ZERO((fd_set*)&*ex_t);
2613
2614    /**
2615     * We need to translate the host file descriptor set into a target file
2616     * descriptor set. This involves both our internal process fd array
2617     * and the fd_set defined in header files.
2618     */
2619    for (int i = 0; i < nfds_h; i++) {
2620        if (fds_read_ptr) {
2621            if (FD_ISSET(i, &rd_h))
2622                FD_SET(trans_map[i], (fd_set*)&*rd_t);
2623        }
2624
2625        if (fds_writ_ptr) {
2626            if (FD_ISSET(i, &wr_h))
2627                FD_SET(trans_map[i], (fd_set*)&*wr_t);
2628        }
2629
2630        if (fds_excp_ptr) {
2631            if (FD_ISSET(i, &ex_h))
2632                FD_SET(trans_map[i], (fd_set*)&*ex_t);
2633        }
2634    }
2635
2636    if (fds_read_ptr)
2637        rd_t.copyOut(tc->getMemProxy());
2638    if (fds_writ_ptr)
2639        wr_t.copyOut(tc->getMemProxy());
2640    if (fds_excp_ptr)
2641        ex_t.copyOut(tc->getMemProxy());
2642    if (time_val_ptr)
2643        tp.copyOut(tc->getMemProxy());
2644
2645    return retval;
2646}
2647
2648template <class OS>
2649SyscallReturn
2650readFunc(SyscallDesc *desc, int num, Process *p, ThreadContext *tc)
2651{
2652    int index = 0;
2653    int tgt_fd = p->getSyscallArg(tc, index);
2654    Addr buf_ptr = p->getSyscallArg(tc, index);
2655    int nbytes = p->getSyscallArg(tc, index);
2656
2657    auto hbfdp = std::dynamic_pointer_cast<HBFDEntry>((*p->fds)[tgt_fd]);
2658    if (!hbfdp)
2659        return -EBADF;
2660    int sim_fd = hbfdp->getSimFD();
2661
2662    struct pollfd pfd;
2663    pfd.fd = sim_fd;
2664    pfd.events = POLLIN | POLLPRI;
2665    if ((poll(&pfd, 1, 0) == 0)
2666        && !(hbfdp->getFlags() & OS::TGT_O_NONBLOCK))
2667        return SyscallReturn::retry();
2668
2669    BufferArg buf_arg(buf_ptr, nbytes);
2670    int bytes_read = read(sim_fd, buf_arg.bufferPtr(), nbytes);
2671
2672    if (bytes_read > 0)
2673        buf_arg.copyOut(tc->getMemProxy());
2674
2675    return (bytes_read == -1) ? -errno : bytes_read;
2676}
2677
2678template <class OS>
2679SyscallReturn
2680writeFunc(SyscallDesc *desc, int num, Process *p, ThreadContext *tc)
2681{
2682    int index = 0;
2683    int tgt_fd = p->getSyscallArg(tc, index);
2684    Addr buf_ptr = p->getSyscallArg(tc, index);
2685    int nbytes = p->getSyscallArg(tc, index);
2686
2687    auto hbfdp = std::dynamic_pointer_cast<HBFDEntry>((*p->fds)[tgt_fd]);
2688    if (!hbfdp)
2689        return -EBADF;
2690    int sim_fd = hbfdp->getSimFD();
2691
2692    BufferArg buf_arg(buf_ptr, nbytes);
2693    buf_arg.copyIn(tc->getMemProxy());
2694
2695    struct pollfd pfd;
2696    pfd.fd = sim_fd;
2697    pfd.events = POLLOUT;
2698
2699    /**
2700     * We don't want to poll on /dev/random. The kernel will not enable the
2701     * file descriptor for writing unless the entropy in the system falls
2702     * below write_wakeup_threshold. This is not guaranteed to happen
2703     * depending on host settings.
2704     */
2705    auto ffdp = std::dynamic_pointer_cast<FileFDEntry>(hbfdp);
2706    if (ffdp && (ffdp->getFileName() != "/dev/random")) {
2707        if (!poll(&pfd, 1, 0) && !(ffdp->getFlags() & OS::TGT_O_NONBLOCK))
2708            return SyscallReturn::retry();
2709    }
2710
2711    int bytes_written = write(sim_fd, buf_arg.bufferPtr(), nbytes);
2712
2713    if (bytes_written != -1)
2714        fsync(sim_fd);
2715
2716    return (bytes_written == -1) ? -errno : bytes_written;
2717}
2718
2719template <class OS>
2720SyscallReturn
2721wait4Func(SyscallDesc *desc, int num, Process *p, ThreadContext *tc)
2722{
2723    int index = 0;
2724    pid_t pid = p->getSyscallArg(tc, index);
2725    Addr statPtr = p->getSyscallArg(tc, index);
2726    int options = p->getSyscallArg(tc, index);
2727    Addr rusagePtr = p->getSyscallArg(tc, index);
2728
2729    if (rusagePtr)
2730        DPRINTF_SYSCALL(Verbose, "wait4: rusage pointer provided %lx, however "
2731                 "functionality not supported. Ignoring rusage pointer.\n",
2732                 rusagePtr);
2733
2734    /**
2735     * Currently, wait4 is only implemented so that it will wait for children
2736     * exit conditions which are denoted by a SIGCHLD signals posted into the
2737     * system signal list. We return no additional information via any of the
2738     * parameters supplied to wait4. If nothing is found in the system signal
2739     * list, we will wait indefinitely for SIGCHLD to post by retrying the
2740     * call.
2741     */
2742    System *sysh = tc->getSystemPtr();
2743    std::list<BasicSignal>::iterator iter;
2744    for (iter=sysh->signalList.begin(); iter!=sysh->signalList.end(); iter++) {
2745        if (iter->receiver == p) {
2746            if (pid < -1) {
2747                if ((iter->sender->pgid() == -pid)
2748                    && (iter->signalValue == OS::TGT_SIGCHLD))
2749                    goto success;
2750            } else if (pid == -1) {
2751                if (iter->signalValue == OS::TGT_SIGCHLD)
2752                    goto success;
2753            } else if (pid == 0) {
2754                if ((iter->sender->pgid() == p->pgid())
2755                    && (iter->signalValue == OS::TGT_SIGCHLD))
2756                    goto success;
2757            } else {
2758                if ((iter->sender->pid() == pid)
2759                    && (iter->signalValue == OS::TGT_SIGCHLD))
2760                    goto success;
2761            }
2762        }
2763    }
2764
2765    return (options & OS::TGT_WNOHANG) ? 0 : SyscallReturn::retry();
2766
2767success:
2768    // Set status to EXITED for WIFEXITED evaluations.
2769    const int EXITED = 0;
2770    BufferArg statusBuf(statPtr, sizeof(int));
2771    *(int *)statusBuf.bufferPtr() = EXITED;
2772    statusBuf.copyOut(tc->getMemProxy());
2773
2774    // Return the child PID.
2775    pid_t retval = iter->sender->pid();
2776    sysh->signalList.erase(iter);
2777    return retval;
2778}
2779
2780template <class OS>
2781SyscallReturn
2782acceptFunc(SyscallDesc *desc, int num, Process *p, ThreadContext *tc)
2783{
2784    struct sockaddr sa;
2785    socklen_t addrLen;
2786    int host_fd;
2787    int index = 0;
2788    int tgt_fd = p->getSyscallArg(tc, index);
2789    Addr addrPtr = p->getSyscallArg(tc, index);
2790    Addr lenPtr = p->getSyscallArg(tc, index);
2791
2792    BufferArg *lenBufPtr = nullptr;
2793    BufferArg *addrBufPtr = nullptr;
2794
2795    auto sfdp = std::dynamic_pointer_cast<SocketFDEntry>((*p->fds)[tgt_fd]);
2796    if (!sfdp)
2797        return -EBADF;
2798    int sim_fd = sfdp->getSimFD();
2799
2800    /**
2801     * We poll the socket file descriptor first to guarantee that we do not
2802     * block on our accept call. The socket can be opened without the
2803     * non-blocking flag (it blocks). This will cause deadlocks between
2804     * communicating processes.
2805     */
2806    struct pollfd pfd;
2807    pfd.fd = sim_fd;
2808    pfd.events = POLLIN | POLLPRI;
2809    if ((poll(&pfd, 1, 0) == 0)
2810        && !(sfdp->getFlags() & OS::TGT_O_NONBLOCK))
2811        return SyscallReturn::retry();
2812
2813    if (lenPtr) {
2814        lenBufPtr = new BufferArg(lenPtr, sizeof(socklen_t));
2815        lenBufPtr->copyIn(tc->getMemProxy());
2816        memcpy(&addrLen, (socklen_t *)lenBufPtr->bufferPtr(),
2817               sizeof(socklen_t));
2818    }
2819
2820    if (addrPtr) {
2821        addrBufPtr = new BufferArg(addrPtr, sizeof(struct sockaddr));
2822        addrBufPtr->copyIn(tc->getMemProxy());
2823        memcpy(&sa, (struct sockaddr *)addrBufPtr->bufferPtr(),
2824               sizeof(struct sockaddr));
2825    }
2826
2827    host_fd = accept(sim_fd, &sa, &addrLen);
2828
2829    if (host_fd == -1)
2830        return -errno;
2831
2832    if (addrPtr) {
2833        memcpy(addrBufPtr->bufferPtr(), &sa, sizeof(sa));
2834        addrBufPtr->copyOut(tc->getMemProxy());
2835        delete(addrBufPtr);
2836    }
2837
2838    if (lenPtr) {
2839        *(socklen_t *)lenBufPtr->bufferPtr() = addrLen;
2840        lenBufPtr->copyOut(tc->getMemProxy());
2841        delete(lenBufPtr);
2842    }
2843
2844    auto afdp = std::make_shared<SocketFDEntry>(host_fd, sfdp->_domain,
2845                                                sfdp->_type, sfdp->_protocol);
2846    return p->fds->allocFD(afdp);
2847}
2848
2849/// Target eventfd() function.
2850template <class OS>
2851SyscallReturn
2852eventfdFunc(SyscallDesc *desc, int num, Process *p, ThreadContext *tc)
2853{
2854#ifdef __linux__
2855    int index = 0;
2856    unsigned initval = p->getSyscallArg(tc, index);
2857    int in_flags = p->getSyscallArg(tc, index);
2858
2859    int sim_fd = eventfd(initval, in_flags);
2860    if (sim_fd == -1)
2861        return -errno;
2862
2863    bool cloexec = in_flags & OS::TGT_O_CLOEXEC;
2864
2865    int flags = cloexec ? OS::TGT_O_CLOEXEC : 0;
2866    flags |= (in_flags & OS::TGT_O_NONBLOCK) ? OS::TGT_O_NONBLOCK : 0;
2867
2868    auto hbfdp = std::make_shared<HBFDEntry>(flags, sim_fd, cloexec);
2869    int tgt_fd = p->fds->allocFD(hbfdp);
2870    return tgt_fd;
2871#else
2872    warnUnsupportedOS("eventfd");
2873    return -1;
2874#endif
2875}
2876
2877#endif // __SIM_SYSCALL_EMUL_HH__
2878