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#if defined(__linux__)
63#include <sys/eventfd.h>
64#include <sys/statfs.h>
65
66#else
67#include <sys/mount.h>
68
69#endif
70
71#ifdef __CYGWIN32__
72#include <sys/fcntl.h>
73
74#endif
75#include <fcntl.h>
76#include <net/if.h>
77#include <poll.h>
78#include <sys/ioctl.h>
79#include <sys/mman.h>
80#include <sys/socket.h>
81#include <sys/stat.h>
82#include <sys/time.h>
83#include <sys/types.h>
84#include <sys/uio.h>
85#include <unistd.h>
86
87#include <cerrno>
88#include <memory>
89#include <string>
90
91#include "arch/generic/tlb.hh"
92#include "arch/utility.hh"
93#include "base/intmath.hh"
94#include "base/loader/object_file.hh"
95#include "base/logging.hh"
96#include "base/trace.hh"
97#include "base/types.hh"
98#include "config/the_isa.hh"
99#include "cpu/base.hh"
100#include "cpu/thread_context.hh"
101#include "mem/page_table.hh"
102#include "params/Process.hh"
103#include "sim/emul_driver.hh"
104#include "sim/futex_map.hh"
105#include "sim/process.hh"
106#include "sim/syscall_debug_macros.hh"
107#include "sim/syscall_desc.hh"
108#include "sim/syscall_emul_buf.hh"
109#include "sim/syscall_return.hh"
110
111#if defined(__APPLE__) && defined(__MACH__) && !defined(CMSG_ALIGN)
112#define CMSG_ALIGN(len) (((len) + sizeof(size_t) - 1) & ~(sizeof(size_t) - 1))
113#endif
114
115//////////////////////////////////////////////////////////////////////
116//
117// The following emulation functions are generic enough that they
118// don't need to be recompiled for different emulated OS's.  They are
119// defined in sim/syscall_emul.cc.
120//
121//////////////////////////////////////////////////////////////////////
122
123void warnUnsupportedOS(std::string syscall_name);
124
125/// Handler for unimplemented syscalls that we haven't thought about.
126SyscallReturn unimplementedFunc(SyscallDesc *desc, int num, ThreadContext *tc);
127
128/// Handler for unimplemented syscalls that we never intend to
129/// implement (signal handling, etc.) and should not affect the correct
130/// behavior of the program.  Print a warning only if the appropriate
131/// trace flag is enabled.  Return success to the target program.
132SyscallReturn ignoreFunc(SyscallDesc *desc, int num, ThreadContext *tc);
133
134// Target fallocateFunc() handler.
135SyscallReturn fallocateFunc(SyscallDesc *desc, int num, ThreadContext *tc);
136
137/// Target exit() handler: terminate current context.
138SyscallReturn exitFunc(SyscallDesc *desc, int num, ThreadContext *tc);
139
140/// Target exit_group() handler: terminate simulation. (exit all threads)
141SyscallReturn exitGroupFunc(SyscallDesc *desc, int num, ThreadContext *tc);
142
143/// Target set_tid_address() handler.
144SyscallReturn setTidAddressFunc(SyscallDesc *desc, int num, ThreadContext *tc);
145
146/// Target getpagesize() handler.
147SyscallReturn getpagesizeFunc(SyscallDesc *desc, int num, ThreadContext *tc);
148
149/// Target brk() handler: set brk address.
150SyscallReturn brkFunc(SyscallDesc *desc, int num, ThreadContext *tc);
151
152/// Target close() handler.
153SyscallReturn closeFunc(SyscallDesc *desc, int num, ThreadContext *tc);
154
155/// Target lseek() handler.
156SyscallReturn lseekFunc(SyscallDesc *desc, int num, ThreadContext *tc);
157
158/// Target _llseek() handler.
159SyscallReturn _llseekFunc(SyscallDesc *desc, int num, ThreadContext *tc);
160
161/// Target munmap() handler.
162SyscallReturn munmapFunc(SyscallDesc *desc, int num, ThreadContext *tc);
163
164/// Target shutdown() handler.
165SyscallReturn shutdownFunc(SyscallDesc *desc, int num, ThreadContext *tc);
166
167/// Target gethostname() handler.
168SyscallReturn gethostnameFunc(SyscallDesc *desc, int num, ThreadContext *tc);
169
170/// Target getcwd() handler.
171SyscallReturn getcwdFunc(SyscallDesc *desc, int num, ThreadContext *tc);
172
173/// Target readlink() handler.
174SyscallReturn readlinkFunc(SyscallDesc *desc, int num, ThreadContext *tc,
175                           int index = 0);
176SyscallReturn readlinkFunc(SyscallDesc *desc, int num, ThreadContext *tc);
177
178/// Target unlink() handler.
179SyscallReturn unlinkHelper(SyscallDesc *desc, int num, ThreadContext *tc,
180                           int index);
181SyscallReturn unlinkFunc(SyscallDesc *desc, int num, ThreadContext *tc);
182
183/// Target link() handler
184SyscallReturn linkFunc(SyscallDesc *desc, int num, ThreadContext *tc);
185
186/// Target symlink() handler.
187SyscallReturn symlinkFunc(SyscallDesc *desc, int num, ThreadContext *tc);
188
189/// Target mkdir() handler.
190SyscallReturn mkdirFunc(SyscallDesc *desc, int num, ThreadContext *tc);
191
192/// Target mknod() handler.
193SyscallReturn mknodFunc(SyscallDesc *desc, int num, ThreadContext *tc);
194
195/// Target chdir() handler.
196SyscallReturn chdirFunc(SyscallDesc *desc, int num, ThreadContext *tc);
197
198// Target rmdir() handler.
199SyscallReturn rmdirFunc(SyscallDesc *desc, int num, ThreadContext *tc);
200
201/// Target rename() handler.
202SyscallReturn renameFunc(SyscallDesc *desc, int num, ThreadContext *tc);
203
204
205/// Target truncate() handler.
206SyscallReturn truncateFunc(SyscallDesc *desc, int num, ThreadContext *tc);
207
208
209/// Target ftruncate() handler.
210SyscallReturn ftruncateFunc(SyscallDesc *desc, int num, ThreadContext *tc);
211
212
213/// Target truncate64() handler.
214SyscallReturn truncate64Func(SyscallDesc *desc, int num, ThreadContext *tc);
215
216/// Target ftruncate64() handler.
217SyscallReturn ftruncate64Func(SyscallDesc *desc, int num, ThreadContext *tc);
218
219
220/// Target umask() handler.
221SyscallReturn umaskFunc(SyscallDesc *desc, int num, ThreadContext *tc);
222
223/// Target gettid() handler.
224SyscallReturn gettidFunc(SyscallDesc *desc, int num, ThreadContext *tc);
225
226/// Target chown() handler.
227SyscallReturn chownFunc(SyscallDesc *desc, int num, ThreadContext *tc);
228
229/// Target getpgrpFunc() handler.
230SyscallReturn getpgrpFunc(SyscallDesc *desc, int num, ThreadContext *tc);
231
232/// Target setpgid() handler.
233SyscallReturn setpgidFunc(SyscallDesc *desc, int num, ThreadContext *tc);
234
235/// Target fchown() handler.
236SyscallReturn fchownFunc(SyscallDesc *desc, int num, ThreadContext *tc);
237
238/// Target dup() handler.
239SyscallReturn dupFunc(SyscallDesc *desc, int num, ThreadContext *tc);
240
241/// Target dup2() handler.
242SyscallReturn dup2Func(SyscallDesc *desc, int num, ThreadContext *tc);
243
244/// Target fcntl() handler.
245SyscallReturn fcntlFunc(SyscallDesc *desc, int num, ThreadContext *tc);
246
247/// Target fcntl64() handler.
248SyscallReturn fcntl64Func(SyscallDesc *desc, int num, ThreadContext *tc);
249
250/// Target setuid() handler.
251SyscallReturn setuidFunc(SyscallDesc *desc, int num, ThreadContext *tc);
252
253/// Target pipe() handler.
254SyscallReturn pipeFunc(SyscallDesc *desc, int num, ThreadContext *tc);
255
256/// Internal pipe() handler.
257SyscallReturn pipeImpl(SyscallDesc *desc, int num, ThreadContext *tc,
258                       bool pseudo_pipe, bool is_pipe2=false);
259
260/// Target pipe() handler.
261SyscallReturn pipe2Func(SyscallDesc *desc, int num, ThreadContext *tc);
262
263/// Target getpid() handler.
264SyscallReturn getpidFunc(SyscallDesc *desc, int num, ThreadContext *tc);
265
266// Target getpeername() handler.
267SyscallReturn getpeernameFunc(SyscallDesc *desc, int num, ThreadContext *tc);
268
269// Target bind() handler.
270SyscallReturn bindFunc(SyscallDesc *desc, int num, ThreadContext *tc);
271
272// Target listen() handler.
273SyscallReturn listenFunc(SyscallDesc *desc, int num, ThreadContext *tc);
274
275// Target connect() handler.
276SyscallReturn connectFunc(SyscallDesc *desc, int num, ThreadContext *tc);
277
278#if defined(SYS_getdents)
279// Target getdents() handler.
280SyscallReturn getdentsFunc(SyscallDesc *desc, int num, ThreadContext *tc);
281#endif
282
283#if defined(SYS_getdents64)
284// Target getdents() handler.
285SyscallReturn getdents64Func(SyscallDesc *desc, int num, ThreadContext *tc);
286#endif
287
288// Target sendto() handler.
289SyscallReturn sendtoFunc(SyscallDesc *desc, int num, ThreadContext *tc);
290
291// Target recvfrom() handler.
292SyscallReturn recvfromFunc(SyscallDesc *desc, int num, ThreadContext *tc);
293
294// Target recvmsg() handler.
295SyscallReturn recvmsgFunc(SyscallDesc *desc, int num, ThreadContext *tc);
296
297// Target sendmsg() handler.
298SyscallReturn sendmsgFunc(SyscallDesc *desc, int num, ThreadContext *tc);
299
300// Target getuid() handler.
301SyscallReturn getuidFunc(SyscallDesc *desc, int num, ThreadContext *tc);
302
303/// Target getgid() handler.
304SyscallReturn getgidFunc(SyscallDesc *desc, int num, ThreadContext *tc);
305
306/// Target getppid() handler.
307SyscallReturn getppidFunc(SyscallDesc *desc, int num, ThreadContext *tc);
308
309/// Target geteuid() handler.
310SyscallReturn geteuidFunc(SyscallDesc *desc, int num, ThreadContext *tc);
311
312/// Target getegid() handler.
313SyscallReturn getegidFunc(SyscallDesc *desc, int num, ThreadContext *tc);
314
315/// Target access() handler
316SyscallReturn accessFunc(SyscallDesc *desc, int num, ThreadContext *tc);
317SyscallReturn accessFunc(SyscallDesc *desc, int num, ThreadContext *tc,
318                         int index);
319
320// Target getsockopt() handler.
321SyscallReturn getsockoptFunc(SyscallDesc *desc, int num, ThreadContext *tc);
322
323// Target setsockopt() handler.
324SyscallReturn setsockoptFunc(SyscallDesc *desc, int num, ThreadContext *tc);
325
326// Target getsockname() handler.
327SyscallReturn getsocknameFunc(SyscallDesc *desc, int num, ThreadContext *tc);
328
329/// Futex system call
330/// Implemented by Daniel Sanchez
331/// Used by printf's in multi-threaded apps
332template <class OS>
333SyscallReturn
334futexFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
335{
336    using namespace std;
337
338    int index = 0;
339    auto process = tc->getProcessPtr();
340
341    Addr uaddr = process->getSyscallArg(tc, index);
342    int op = process->getSyscallArg(tc, index);
343    int val = process->getSyscallArg(tc, index);
344    int timeout M5_VAR_USED = process->getSyscallArg(tc, index);
345    Addr uaddr2 M5_VAR_USED = process->getSyscallArg(tc, index);
346    int val3 = process->getSyscallArg(tc, index);
347
348    /*
349     * Unsupported option that does not affect the correctness of the
350     * application. This is a performance optimization utilized by Linux.
351     */
352    op &= ~OS::TGT_FUTEX_PRIVATE_FLAG;
353    op &= ~OS::TGT_FUTEX_CLOCK_REALTIME_FLAG;
354
355    FutexMap &futex_map = tc->getSystemPtr()->futexMap;
356
357    if (OS::TGT_FUTEX_WAIT == op || OS::TGT_FUTEX_WAIT_BITSET == op) {
358        // Ensure futex system call accessed atomically.
359        BufferArg buf(uaddr, sizeof(int));
360        buf.copyIn(tc->getVirtProxy());
361        int mem_val = *(int*)buf.bufferPtr();
362
363        /*
364         * The value in memory at uaddr is not equal with the expected val
365         * (a different thread must have changed it before the system call was
366         * invoked). In this case, we need to throw an error.
367         */
368        if (val != mem_val)
369            return -OS::TGT_EWOULDBLOCK;
370
371        if (OS::TGT_FUTEX_WAIT) {
372            futex_map.suspend(uaddr, process->tgid(), tc);
373        } else {
374            futex_map.suspend_bitset(uaddr, process->tgid(), tc, val3);
375        }
376
377        return 0;
378    } else if (OS::TGT_FUTEX_WAKE == op) {
379        return futex_map.wakeup(uaddr, process->tgid(), val);
380    } else if (OS::TGT_FUTEX_WAKE_BITSET == op) {
381        return futex_map.wakeup_bitset(uaddr, process->tgid(), val3);
382    } else if (OS::TGT_FUTEX_REQUEUE == op ||
383               OS::TGT_FUTEX_CMP_REQUEUE == op) {
384
385        // Ensure futex system call accessed atomically.
386        BufferArg buf(uaddr, sizeof(int));
387        buf.copyIn(tc->getVirtProxy());
388        int mem_val = *(int*)buf.bufferPtr();
389        /*
390         * For CMP_REQUEUE, the whole operation is only started only if
391         * val3 is still the value of the futex pointed to by uaddr.
392         */
393        if (OS::TGT_FUTEX_CMP_REQUEUE && val3 != mem_val)
394            return -OS::TGT_EWOULDBLOCK;
395        return futex_map.requeue(uaddr, process->tgid(), val, timeout, uaddr2);
396    } else if (OS::TGT_FUTEX_WAKE_OP == op) {
397        /*
398         * The FUTEX_WAKE_OP operation is equivalent to executing the
399         * following code atomically and totally ordered with respect to
400         * other futex operations on any of the two supplied futex words:
401         *
402         *   int oldval = *(int *) addr2;
403         *   *(int *) addr2 = oldval op oparg;
404         *   futex(addr1, FUTEX_WAKE, val, 0, 0, 0);
405         *   if (oldval cmp cmparg)
406         *        futex(addr2, FUTEX_WAKE, val2, 0, 0, 0);
407         *
408         * (op, oparg, cmp, cmparg are encoded in val3)
409         *
410         * +---+---+-----------+-----------+
411         * |op |cmp|   oparg   |  cmparg   |
412         * +---+---+-----------+-----------+
413         *   4   4       12          12    <== # of bits
414         *
415         * reference: http://man7.org/linux/man-pages/man2/futex.2.html
416         *
417         */
418        // get value from simulated-space
419        BufferArg buf(uaddr2, sizeof(int));
420        buf.copyIn(tc->getVirtProxy());
421        int oldval = *(int*)buf.bufferPtr();
422        int newval = oldval;
423        // extract op, oparg, cmp, cmparg from val3
424        int wake_cmparg =  val3 & 0xfff;
425        int wake_oparg  = (val3 & 0xfff000)   >> 12;
426        int wake_cmp    = (val3 & 0xf000000)  >> 24;
427        int wake_op     = (val3 & 0xf0000000) >> 28;
428        if ((wake_op & OS::TGT_FUTEX_OP_ARG_SHIFT) >> 3 == 1)
429            wake_oparg = (1 << wake_oparg);
430        wake_op &= ~OS::TGT_FUTEX_OP_ARG_SHIFT;
431        // perform operation on the value of the second futex
432        if (wake_op == OS::TGT_FUTEX_OP_SET)
433            newval = wake_oparg;
434        else if (wake_op == OS::TGT_FUTEX_OP_ADD)
435            newval += wake_oparg;
436        else if (wake_op == OS::TGT_FUTEX_OP_OR)
437            newval |= wake_oparg;
438        else if (wake_op == OS::TGT_FUTEX_OP_ANDN)
439            newval &= ~wake_oparg;
440        else if (wake_op == OS::TGT_FUTEX_OP_XOR)
441            newval ^= wake_oparg;
442        // copy updated value back to simulated-space
443        *(int*)buf.bufferPtr() = newval;
444        buf.copyOut(tc->getVirtProxy());
445        // perform the first wake-up
446        int woken1 = futex_map.wakeup(uaddr, process->tgid(), val);
447        int woken2 = 0;
448        // calculate the condition of the second wake-up
449        bool is_wake2 = false;
450        if (wake_cmp == OS::TGT_FUTEX_OP_CMP_EQ)
451            is_wake2 = oldval == wake_cmparg;
452        else if (wake_cmp == OS::TGT_FUTEX_OP_CMP_NE)
453            is_wake2 = oldval != wake_cmparg;
454        else if (wake_cmp == OS::TGT_FUTEX_OP_CMP_LT)
455            is_wake2 = oldval < wake_cmparg;
456        else if (wake_cmp == OS::TGT_FUTEX_OP_CMP_LE)
457            is_wake2 = oldval <= wake_cmparg;
458        else if (wake_cmp == OS::TGT_FUTEX_OP_CMP_GT)
459            is_wake2 = oldval > wake_cmparg;
460        else if (wake_cmp == OS::TGT_FUTEX_OP_CMP_GE)
461            is_wake2 = oldval >= wake_cmparg;
462        // perform the second wake-up
463        if (is_wake2)
464            woken2 = futex_map.wakeup(uaddr2, process->tgid(), timeout);
465
466        return woken1 + woken2;
467    }
468    warn("futex: op %d not implemented; ignoring.", op);
469    return -ENOSYS;
470}
471
472
473/// Pseudo Funcs  - These functions use a different return convension,
474/// returning a second value in a register other than the normal return register
475SyscallReturn pipePseudoFunc(SyscallDesc *desc, int num, ThreadContext *tc);
476
477/// Target getpidPseudo() handler.
478SyscallReturn getpidPseudoFunc(SyscallDesc *desc, int num, ThreadContext *tc);
479
480/// Target getuidPseudo() handler.
481SyscallReturn getuidPseudoFunc(SyscallDesc *desc, int num, ThreadContext *tc);
482
483/// Target getgidPseudo() handler.
484SyscallReturn getgidPseudoFunc(SyscallDesc *desc, int num, ThreadContext *tc);
485
486
487/// A readable name for 1,000,000, for converting microseconds to seconds.
488const int one_million = 1000000;
489/// A readable name for 1,000,000,000, for converting nanoseconds to seconds.
490const int one_billion = 1000000000;
491
492/// Approximate seconds since the epoch (1/1/1970).  About a billion,
493/// by my reckoning.  We want to keep this a constant (not use the
494/// real-world time) to keep simulations repeatable.
495const unsigned seconds_since_epoch = 1000000000;
496
497/// Helper function to convert current elapsed time to seconds and
498/// microseconds.
499template <class T1, class T2>
500void
501getElapsedTimeMicro(T1 &sec, T2 &usec)
502{
503    uint64_t elapsed_usecs = curTick() / SimClock::Int::us;
504    sec = elapsed_usecs / one_million;
505    usec = elapsed_usecs % one_million;
506}
507
508/// Helper function to convert current elapsed time to seconds and
509/// nanoseconds.
510template <class T1, class T2>
511void
512getElapsedTimeNano(T1 &sec, T2 &nsec)
513{
514    uint64_t elapsed_nsecs = curTick() / SimClock::Int::ns;
515    sec = elapsed_nsecs / one_billion;
516    nsec = elapsed_nsecs % one_billion;
517}
518
519//////////////////////////////////////////////////////////////////////
520//
521// The following emulation functions are generic, but need to be
522// templated to account for differences in types, constants, etc.
523//
524//////////////////////////////////////////////////////////////////////
525
526    typedef struct statfs hst_statfs;
527#if NO_STAT64
528    typedef struct stat hst_stat;
529    typedef struct stat hst_stat64;
530#else
531    typedef struct stat hst_stat;
532    typedef struct stat64 hst_stat64;
533#endif
534
535//// Helper function to convert a host stat buffer to a target stat
536//// buffer.  Also copies the target buffer out to the simulated
537//// memory space.  Used by stat(), fstat(), and lstat().
538
539template <typename target_stat, typename host_stat>
540void
541convertStatBuf(target_stat &tgt, host_stat *host, bool fakeTTY = false)
542{
543    using namespace TheISA;
544
545    if (fakeTTY)
546        tgt->st_dev = 0xA;
547    else
548        tgt->st_dev = host->st_dev;
549    tgt->st_dev = TheISA::htog(tgt->st_dev);
550    tgt->st_ino = host->st_ino;
551    tgt->st_ino = TheISA::htog(tgt->st_ino);
552    tgt->st_mode = host->st_mode;
553    if (fakeTTY) {
554        // Claim to be a character device
555        tgt->st_mode &= ~S_IFMT;    // Clear S_IFMT
556        tgt->st_mode |= S_IFCHR;    // Set S_IFCHR
557    }
558    tgt->st_mode = TheISA::htog(tgt->st_mode);
559    tgt->st_nlink = host->st_nlink;
560    tgt->st_nlink = TheISA::htog(tgt->st_nlink);
561    tgt->st_uid = host->st_uid;
562    tgt->st_uid = TheISA::htog(tgt->st_uid);
563    tgt->st_gid = host->st_gid;
564    tgt->st_gid = TheISA::htog(tgt->st_gid);
565    if (fakeTTY)
566        tgt->st_rdev = 0x880d;
567    else
568        tgt->st_rdev = host->st_rdev;
569    tgt->st_rdev = TheISA::htog(tgt->st_rdev);
570    tgt->st_size = host->st_size;
571    tgt->st_size = TheISA::htog(tgt->st_size);
572    tgt->st_atimeX = host->st_atime;
573    tgt->st_atimeX = TheISA::htog(tgt->st_atimeX);
574    tgt->st_mtimeX = host->st_mtime;
575    tgt->st_mtimeX = TheISA::htog(tgt->st_mtimeX);
576    tgt->st_ctimeX = host->st_ctime;
577    tgt->st_ctimeX = TheISA::htog(tgt->st_ctimeX);
578    // Force the block size to be 8KB. This helps to ensure buffered io works
579    // consistently across different hosts.
580    tgt->st_blksize = 0x2000;
581    tgt->st_blksize = TheISA::htog(tgt->st_blksize);
582    tgt->st_blocks = host->st_blocks;
583    tgt->st_blocks = TheISA::htog(tgt->st_blocks);
584}
585
586// Same for stat64
587
588template <typename target_stat, typename host_stat64>
589void
590convertStat64Buf(target_stat &tgt, host_stat64 *host, bool fakeTTY = false)
591{
592    using namespace TheISA;
593
594    convertStatBuf<target_stat, host_stat64>(tgt, host, fakeTTY);
595#if defined(STAT_HAVE_NSEC)
596    tgt->st_atime_nsec = host->st_atime_nsec;
597    tgt->st_atime_nsec = TheISA::htog(tgt->st_atime_nsec);
598    tgt->st_mtime_nsec = host->st_mtime_nsec;
599    tgt->st_mtime_nsec = TheISA::htog(tgt->st_mtime_nsec);
600    tgt->st_ctime_nsec = host->st_ctime_nsec;
601    tgt->st_ctime_nsec = TheISA::htog(tgt->st_ctime_nsec);
602#else
603    tgt->st_atime_nsec = 0;
604    tgt->st_mtime_nsec = 0;
605    tgt->st_ctime_nsec = 0;
606#endif
607}
608
609// Here are a couple of convenience functions
610template<class OS>
611void
612copyOutStatBuf(PortProxy &mem, Addr addr,
613               hst_stat *host, bool fakeTTY = false)
614{
615    typedef TypedBufferArg<typename OS::tgt_stat> tgt_stat_buf;
616    tgt_stat_buf tgt(addr);
617    convertStatBuf<tgt_stat_buf, hst_stat>(tgt, host, fakeTTY);
618    tgt.copyOut(mem);
619}
620
621template<class OS>
622void
623copyOutStat64Buf(PortProxy &mem, Addr addr,
624                 hst_stat64 *host, bool fakeTTY = false)
625{
626    typedef TypedBufferArg<typename OS::tgt_stat64> tgt_stat_buf;
627    tgt_stat_buf tgt(addr);
628    convertStat64Buf<tgt_stat_buf, hst_stat64>(tgt, host, fakeTTY);
629    tgt.copyOut(mem);
630}
631
632template <class OS>
633void
634copyOutStatfsBuf(PortProxy &mem, Addr addr,
635                 hst_statfs *host)
636{
637    TypedBufferArg<typename OS::tgt_statfs> tgt(addr);
638
639    tgt->f_type = TheISA::htog(host->f_type);
640#if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
641    tgt->f_bsize = TheISA::htog(host->f_iosize);
642#else
643    tgt->f_bsize = TheISA::htog(host->f_bsize);
644#endif
645    tgt->f_blocks = TheISA::htog(host->f_blocks);
646    tgt->f_bfree = TheISA::htog(host->f_bfree);
647    tgt->f_bavail = TheISA::htog(host->f_bavail);
648    tgt->f_files = TheISA::htog(host->f_files);
649    tgt->f_ffree = TheISA::htog(host->f_ffree);
650    memcpy(&tgt->f_fsid, &host->f_fsid, sizeof(host->f_fsid));
651#if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
652    tgt->f_namelen = TheISA::htog(host->f_namemax);
653    tgt->f_frsize = TheISA::htog(host->f_bsize);
654#elif defined(__APPLE__)
655    tgt->f_namelen = 0;
656    tgt->f_frsize = 0;
657#else
658    tgt->f_namelen = TheISA::htog(host->f_namelen);
659    tgt->f_frsize = TheISA::htog(host->f_frsize);
660#endif
661#if defined(__linux__)
662    memcpy(&tgt->f_spare, &host->f_spare, sizeof(host->f_spare));
663#else
664    /*
665     * The fields are different sizes per OS. Don't bother with
666     * f_spare or f_reserved on non-Linux for now.
667     */
668    memset(&tgt->f_spare, 0, sizeof(tgt->f_spare));
669#endif
670
671    tgt.copyOut(mem);
672}
673
674/// Target ioctl() handler.  For the most part, programs call ioctl()
675/// only to find out if their stdout is a tty, to determine whether to
676/// do line or block buffering.  We always claim that output fds are
677/// not TTYs to provide repeatable results.
678template <class OS>
679SyscallReturn
680ioctlFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
681{
682    int index = 0;
683    auto p = tc->getProcessPtr();
684
685    int tgt_fd = p->getSyscallArg(tc, index);
686    unsigned req = p->getSyscallArg(tc, index);
687
688    DPRINTF_SYSCALL(Verbose, "ioctl(%d, 0x%x, ...)\n", tgt_fd, req);
689
690    if (OS::isTtyReq(req))
691        return -ENOTTY;
692
693    auto dfdp = std::dynamic_pointer_cast<DeviceFDEntry>((*p->fds)[tgt_fd]);
694    if (dfdp) {
695        EmulatedDriver *emul_driver = dfdp->getDriver();
696        if (emul_driver)
697            return emul_driver->ioctl(tc, req);
698    }
699
700    auto sfdp = std::dynamic_pointer_cast<SocketFDEntry>((*p->fds)[tgt_fd]);
701    if (sfdp) {
702        int status;
703
704        switch (req) {
705          case SIOCGIFCONF: {
706            Addr conf_addr = p->getSyscallArg(tc, index);
707            BufferArg conf_arg(conf_addr, sizeof(ifconf));
708            conf_arg.copyIn(tc->getVirtProxy());
709
710            ifconf *conf = (ifconf*)conf_arg.bufferPtr();
711            Addr ifc_buf_addr = (Addr)conf->ifc_buf;
712            BufferArg ifc_buf_arg(ifc_buf_addr, conf->ifc_len);
713            ifc_buf_arg.copyIn(tc->getVirtProxy());
714
715            conf->ifc_buf = (char*)ifc_buf_arg.bufferPtr();
716
717            status = ioctl(sfdp->getSimFD(), req, conf_arg.bufferPtr());
718            if (status != -1) {
719                conf->ifc_buf = (char*)ifc_buf_addr;
720                ifc_buf_arg.copyOut(tc->getVirtProxy());
721                conf_arg.copyOut(tc->getVirtProxy());
722            }
723
724            return status;
725          }
726          case SIOCGIFFLAGS:
727#if defined(__linux__)
728          case SIOCGIFINDEX:
729#endif
730          case SIOCGIFNETMASK:
731          case SIOCGIFADDR:
732#if defined(__linux__)
733          case SIOCGIFHWADDR:
734#endif
735          case SIOCGIFMTU: {
736            Addr req_addr = p->getSyscallArg(tc, index);
737            BufferArg req_arg(req_addr, sizeof(ifreq));
738            req_arg.copyIn(tc->getVirtProxy());
739
740            status = ioctl(sfdp->getSimFD(), req, req_arg.bufferPtr());
741            if (status != -1)
742                req_arg.copyOut(tc->getVirtProxy());
743            return status;
744          }
745        }
746    }
747
748    /**
749     * For lack of a better return code, return ENOTTY. Ideally, we should
750     * return something better here, but at least we issue the warning.
751     */
752    warn("Unsupported ioctl call (return ENOTTY): ioctl(%d, 0x%x, ...) @ \n",
753         tgt_fd, req, tc->pcState());
754    return -ENOTTY;
755}
756
757template <class OS>
758SyscallReturn
759openImpl(SyscallDesc *desc, int callnum, ThreadContext *tc, bool isopenat)
760{
761    int index = 0;
762    auto p = tc->getProcessPtr();
763    int tgt_dirfd = -1;
764
765    /**
766     * If using the openat variant, read in the target directory file
767     * descriptor from the simulated process.
768     */
769    if (isopenat)
770        tgt_dirfd = p->getSyscallArg(tc, index);
771
772    /**
773     * Retrieve the simulated process' memory proxy and then read in the path
774     * string from that memory space into the host's working memory space.
775     */
776    std::string path;
777    if (!tc->getVirtProxy().tryReadString(path, p->getSyscallArg(tc, index)))
778        return -EFAULT;
779
780#ifdef __CYGWIN32__
781    int host_flags = O_BINARY;
782#else
783    int host_flags = 0;
784#endif
785    /**
786     * Translate target flags into host flags. Flags exist which are not
787     * ported between architectures which can cause check failures.
788     */
789    int tgt_flags = p->getSyscallArg(tc, index);
790    for (int i = 0; i < OS::NUM_OPEN_FLAGS; i++) {
791        if (tgt_flags & OS::openFlagTable[i].tgtFlag) {
792            tgt_flags &= ~OS::openFlagTable[i].tgtFlag;
793            host_flags |= OS::openFlagTable[i].hostFlag;
794        }
795    }
796    if (tgt_flags) {
797        warn("open%s: cannot decode flags 0x%x",
798             isopenat ? "at" : "", tgt_flags);
799    }
800#ifdef __CYGWIN32__
801    host_flags |= O_BINARY;
802#endif
803
804    int mode = p->getSyscallArg(tc, index);
805
806    /**
807     * If the simulated process called open or openat with AT_FDCWD specified,
808     * take the current working directory value which was passed into the
809     * process class as a Python parameter and append the current path to
810     * create a full path.
811     * Otherwise, openat with a valid target directory file descriptor has
812     * been called. If the path option, which was passed in as a parameter,
813     * is not absolute, retrieve the directory file descriptor's path and
814     * prepend it to the path passed in as a parameter.
815     * In every case, we should have a full path (which is relevant to the
816     * host) to work with after this block has been passed.
817     */
818    std::string redir_path = path;
819    std::string abs_path = path;
820    if (!isopenat || tgt_dirfd == OS::TGT_AT_FDCWD) {
821        abs_path = p->absolutePath(path, true);
822        redir_path = p->checkPathRedirect(path);
823    } else if (!startswith(path, "/")) {
824        std::shared_ptr<FDEntry> fdep = ((*p->fds)[tgt_dirfd]);
825        auto ffdp = std::dynamic_pointer_cast<FileFDEntry>(fdep);
826        if (!ffdp)
827            return -EBADF;
828        abs_path = ffdp->getFileName() + path;
829        redir_path = p->checkPathRedirect(abs_path);
830    }
831
832    /**
833     * Since this is an emulated environment, we create pseudo file
834     * descriptors for device requests that have been registered with
835     * the process class through Python; this allows us to create a file
836     * descriptor for subsequent ioctl or mmap calls.
837     */
838    if (startswith(abs_path, "/dev/")) {
839        std::string filename = abs_path.substr(strlen("/dev/"));
840        EmulatedDriver *drv = p->findDriver(filename);
841        if (drv) {
842            DPRINTF_SYSCALL(Verbose, "open%s: passing call to "
843                            "driver open with path[%s]\n",
844                            isopenat ? "at" : "", abs_path.c_str());
845            return drv->open(tc, mode, host_flags);
846        }
847        /**
848         * Fall through here for pass through to host devices, such
849         * as /dev/zero
850         */
851    }
852
853    /**
854     * We make several attempts resolve a call to open.
855     *
856     * 1) Resolve any path redirection before hand. This will set the path
857     * up with variable 'redir_path' which may contain a modified path or
858     * the original path value. This should already be done in prior code.
859     * 2) Try to handle the access using 'special_paths'. Some special_paths
860     * and files cannot be called on the host and need to be handled as
861     * special cases inside the simulator. These special_paths are handled by
862     * C++ routines to provide output back to userspace.
863     * 3) If the full path that was created above does not match any of the
864     * special cases, pass it through to the open call on the __HOST__ to let
865     * the host open the file on our behalf. Again, the openImpl tries to
866     * USE_THE_HOST_FILESYSTEM_OPEN (with a possible redirection to the
867     * faux-filesystem files). The faux-filesystem is dynamically created
868     * during simulator configuration using Python functions.
869     * 4) If the host cannot open the file, the open attempt failed in "3)".
870     * Return the host's error code back through the system call to the
871     * simulated process. If running a debug trace, also notify the user that
872     * the open call failed.
873     *
874     * Any success will set sim_fd to something other than -1 and skip the
875     * next conditions effectively bypassing them.
876     */
877    int sim_fd = -1;
878    std::string used_path;
879    std::vector<std::string> special_paths =
880            { "/proc/meminfo/", "/system/", "/platform/", "/etc/passwd" };
881    for (auto entry : special_paths) {
882        if (startswith(path, entry)) {
883            sim_fd = OS::openSpecialFile(abs_path, p, tc);
884            used_path = abs_path;
885        }
886    }
887    if (sim_fd == -1) {
888        sim_fd = open(redir_path.c_str(), host_flags, mode);
889        used_path = redir_path;
890    }
891    if (sim_fd == -1) {
892        int local = -errno;
893        DPRINTF_SYSCALL(Verbose, "open%s: failed -> path:%s "
894                        "(inferred from:%s)\n", isopenat ? "at" : "",
895                        used_path.c_str(), path.c_str());
896        return local;
897    }
898
899    /**
900     * The file was opened successfully and needs to be recorded in the
901     * process' file descriptor array so that it can be retrieved later.
902     * The target file descriptor that is chosen will be the lowest unused
903     * file descriptor.
904     * Return the indirect target file descriptor back to the simulated
905     * process to act as a handle for the opened file.
906     */
907    auto ffdp = std::make_shared<FileFDEntry>(sim_fd, host_flags, path, 0);
908    int tgt_fd = p->fds->allocFD(ffdp);
909    DPRINTF_SYSCALL(Verbose, "open%s: sim_fd[%d], target_fd[%d] -> path:%s\n"
910                    "(inferred from:%s)\n", isopenat ? "at" : "",
911                    sim_fd, tgt_fd, used_path.c_str(), path.c_str());
912    return tgt_fd;
913}
914
915/// Target open() handler.
916template <class OS>
917SyscallReturn
918openFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
919{
920    return openImpl<OS>(desc, callnum, tc, false);
921}
922
923/// Target openat() handler.
924template <class OS>
925SyscallReturn
926openatFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
927{
928    return openImpl<OS>(desc, callnum, tc, true);
929}
930
931/// Target unlinkat() handler.
932template <class OS>
933SyscallReturn
934unlinkatFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
935{
936    int index = 0;
937    auto process = tc->getProcessPtr();
938    int dirfd = process->getSyscallArg(tc, index);
939    if (dirfd != OS::TGT_AT_FDCWD)
940        warn("unlinkat: first argument not AT_FDCWD; unlikely to work");
941
942    return unlinkHelper(desc, callnum, tc, 1);
943}
944
945/// Target facessat() handler
946template <class OS>
947SyscallReturn
948faccessatFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
949{
950    int index = 0;
951    auto process = tc->getProcessPtr();
952    int dirfd = process->getSyscallArg(tc, index);
953    if (dirfd != OS::TGT_AT_FDCWD)
954        warn("faccessat: first argument not AT_FDCWD; unlikely to work");
955    return accessFunc(desc, callnum, tc, 1);
956}
957
958/// Target readlinkat() handler
959template <class OS>
960SyscallReturn
961readlinkatFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
962{
963    int index = 0;
964    auto process = tc->getProcessPtr();
965    int dirfd = process->getSyscallArg(tc, index);
966    if (dirfd != OS::TGT_AT_FDCWD)
967        warn("openat: first argument not AT_FDCWD; unlikely to work");
968    return readlinkFunc(desc, callnum, tc, 1);
969}
970
971/// Target renameat() handler.
972template <class OS>
973SyscallReturn
974renameatFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
975{
976    int index = 0;
977    auto process = tc->getProcessPtr();
978
979    int olddirfd = process->getSyscallArg(tc, index);
980    if (olddirfd != OS::TGT_AT_FDCWD)
981        warn("renameat: first argument not AT_FDCWD; unlikely to work");
982
983    std::string old_name;
984
985    if (!tc->getVirtProxy().tryReadString(old_name,
986                                         process->getSyscallArg(tc, index)))
987        return -EFAULT;
988
989    int newdirfd = process->getSyscallArg(tc, index);
990    if (newdirfd != OS::TGT_AT_FDCWD)
991        warn("renameat: third argument not AT_FDCWD; unlikely to work");
992
993    std::string new_name;
994
995    if (!tc->getVirtProxy().tryReadString(new_name,
996                                         process->getSyscallArg(tc, index)))
997        return -EFAULT;
998
999    // Adjust path for cwd and redirection
1000    old_name = process->checkPathRedirect(old_name);
1001    new_name = process->checkPathRedirect(new_name);
1002
1003    int result = rename(old_name.c_str(), new_name.c_str());
1004    return (result == -1) ? -errno : result;
1005}
1006
1007/// Target sysinfo() handler.
1008template <class OS>
1009SyscallReturn
1010sysinfoFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
1011{
1012    int index = 0;
1013    auto process = tc->getProcessPtr();
1014
1015    TypedBufferArg<typename OS::tgt_sysinfo>
1016        sysinfo(process->getSyscallArg(tc, index));
1017
1018    sysinfo->uptime = seconds_since_epoch;
1019    sysinfo->totalram = process->system->memSize();
1020    sysinfo->mem_unit = 1;
1021
1022    sysinfo.copyOut(tc->getVirtProxy());
1023
1024    return 0;
1025}
1026
1027/// Target chmod() handler.
1028template <class OS>
1029SyscallReturn
1030chmodFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
1031{
1032    std::string path;
1033    auto process = tc->getProcessPtr();
1034
1035    int index = 0;
1036    if (!tc->getVirtProxy().tryReadString(path,
1037                process->getSyscallArg(tc, index))) {
1038        return -EFAULT;
1039    }
1040
1041    uint32_t mode = process->getSyscallArg(tc, index);
1042    mode_t hostMode = 0;
1043
1044    // XXX translate mode flags via OS::something???
1045    hostMode = mode;
1046
1047    // Adjust path for cwd and redirection
1048    path = process->checkPathRedirect(path);
1049
1050    // do the chmod
1051    int result = chmod(path.c_str(), hostMode);
1052    if (result < 0)
1053        return -errno;
1054
1055    return 0;
1056}
1057
1058template <class OS>
1059SyscallReturn
1060pollFunc(SyscallDesc *desc, int num, ThreadContext *tc)
1061{
1062    int index = 0;
1063    auto p = tc->getProcessPtr();
1064    Addr fdsPtr = p->getSyscallArg(tc, index);
1065    int nfds = p->getSyscallArg(tc, index);
1066    int tmout = p->getSyscallArg(tc, index);
1067
1068    BufferArg fdsBuf(fdsPtr, sizeof(struct pollfd) * nfds);
1069    fdsBuf.copyIn(tc->getVirtProxy());
1070
1071    /**
1072     * Record the target file descriptors in a local variable. We need to
1073     * replace them with host file descriptors but we need a temporary copy
1074     * for later. Afterwards, replace each target file descriptor in the
1075     * poll_fd array with its host_fd.
1076     */
1077    int temp_tgt_fds[nfds];
1078    for (index = 0; index < nfds; index++) {
1079        temp_tgt_fds[index] = ((struct pollfd *)fdsBuf.bufferPtr())[index].fd;
1080        auto tgt_fd = temp_tgt_fds[index];
1081        auto hbfdp = std::dynamic_pointer_cast<HBFDEntry>((*p->fds)[tgt_fd]);
1082        if (!hbfdp)
1083            return -EBADF;
1084        auto host_fd = hbfdp->getSimFD();
1085        ((struct pollfd *)fdsBuf.bufferPtr())[index].fd = host_fd;
1086    }
1087
1088    /**
1089     * We cannot allow an infinite poll to occur or it will inevitably cause
1090     * a deadlock in the gem5 simulator with clone. We must pass in tmout with
1091     * a non-negative value, however it also makes no sense to poll on the
1092     * underlying host for any other time than tmout a zero timeout.
1093     */
1094    int status;
1095    if (tmout < 0) {
1096        status = poll((struct pollfd *)fdsBuf.bufferPtr(), nfds, 0);
1097        if (status == 0) {
1098            /**
1099             * If blocking indefinitely, check the signal list to see if a
1100             * signal would break the poll out of the retry cycle and try
1101             * to return the signal interrupt instead.
1102             */
1103            System *sysh = tc->getSystemPtr();
1104            std::list<BasicSignal>::iterator it;
1105            for (it=sysh->signalList.begin(); it!=sysh->signalList.end(); it++)
1106                if (it->receiver == p)
1107                    return -EINTR;
1108            return SyscallReturn::retry();
1109        }
1110    } else
1111        status = poll((struct pollfd *)fdsBuf.bufferPtr(), nfds, 0);
1112
1113    if (status == -1)
1114        return -errno;
1115
1116    /**
1117     * Replace each host_fd in the returned poll_fd array with its original
1118     * target file descriptor.
1119     */
1120    for (index = 0; index < nfds; index++) {
1121        auto tgt_fd = temp_tgt_fds[index];
1122        ((struct pollfd *)fdsBuf.bufferPtr())[index].fd = tgt_fd;
1123    }
1124
1125    /**
1126     * Copy out the pollfd struct because the host may have updated fields
1127     * in the structure.
1128     */
1129    fdsBuf.copyOut(tc->getVirtProxy());
1130
1131    return status;
1132}
1133
1134/// Target fchmod() handler.
1135template <class OS>
1136SyscallReturn
1137fchmodFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
1138{
1139    int index = 0;
1140    auto p = tc->getProcessPtr();
1141    int tgt_fd = p->getSyscallArg(tc, index);
1142    uint32_t mode = p->getSyscallArg(tc, index);
1143
1144    auto ffdp = std::dynamic_pointer_cast<FileFDEntry>((*p->fds)[tgt_fd]);
1145    if (!ffdp)
1146        return -EBADF;
1147    int sim_fd = ffdp->getSimFD();
1148
1149    mode_t hostMode = mode;
1150
1151    int result = fchmod(sim_fd, hostMode);
1152
1153    return (result < 0) ? -errno : 0;
1154}
1155
1156/// Target mremap() handler.
1157template <class OS>
1158SyscallReturn
1159mremapFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
1160{
1161    int index = 0;
1162    auto process = tc->getProcessPtr();
1163    Addr start = process->getSyscallArg(tc, index);
1164    uint64_t old_length = process->getSyscallArg(tc, index);
1165    uint64_t new_length = process->getSyscallArg(tc, index);
1166    uint64_t flags = process->getSyscallArg(tc, index);
1167    uint64_t provided_address = 0;
1168    bool use_provided_address = flags & OS::TGT_MREMAP_FIXED;
1169
1170    if (use_provided_address)
1171        provided_address = process->getSyscallArg(tc, index);
1172
1173    if ((start % TheISA::PageBytes != 0) ||
1174        (provided_address % TheISA::PageBytes != 0)) {
1175        warn("mremap failing: arguments not page aligned");
1176        return -EINVAL;
1177    }
1178
1179    new_length = roundUp(new_length, TheISA::PageBytes);
1180
1181    if (new_length > old_length) {
1182        std::shared_ptr<MemState> mem_state = process->memState;
1183        Addr mmap_end = mem_state->getMmapEnd();
1184
1185        if ((start + old_length) == mmap_end &&
1186            (!use_provided_address || provided_address == start)) {
1187            // This case cannot occur when growing downward, as
1188            // start is greater than or equal to mmap_end.
1189            uint64_t diff = new_length - old_length;
1190            process->allocateMem(mmap_end, diff);
1191            mem_state->setMmapEnd(mmap_end + diff);
1192            return start;
1193        } else {
1194            if (!use_provided_address && !(flags & OS::TGT_MREMAP_MAYMOVE)) {
1195                warn("can't remap here and MREMAP_MAYMOVE flag not set\n");
1196                return -ENOMEM;
1197            } else {
1198                uint64_t new_start = provided_address;
1199                if (!use_provided_address) {
1200                    new_start = process->mmapGrowsDown() ?
1201                                mmap_end - new_length : mmap_end;
1202                    mmap_end = process->mmapGrowsDown() ?
1203                               new_start : mmap_end + new_length;
1204                    mem_state->setMmapEnd(mmap_end);
1205                }
1206
1207                process->pTable->remap(start, old_length, new_start);
1208                warn("mremapping to new vaddr %08p-%08p, adding %d\n",
1209                     new_start, new_start + new_length,
1210                     new_length - old_length);
1211                // add on the remaining unallocated pages
1212                process->allocateMem(new_start + old_length,
1213                                     new_length - old_length,
1214                                     use_provided_address /* clobber */);
1215                if (use_provided_address &&
1216                    ((new_start + new_length > mem_state->getMmapEnd() &&
1217                      !process->mmapGrowsDown()) ||
1218                    (new_start < mem_state->getMmapEnd() &&
1219                      process->mmapGrowsDown()))) {
1220                    // something fishy going on here, at least notify the user
1221                    // @todo: increase mmap_end?
1222                    warn("mmap region limit exceeded with MREMAP_FIXED\n");
1223                }
1224                warn("returning %08p as start\n", new_start);
1225                return new_start;
1226            }
1227        }
1228    } else {
1229        if (use_provided_address && provided_address != start)
1230            process->pTable->remap(start, new_length, provided_address);
1231        process->pTable->unmap(start + new_length, old_length - new_length);
1232        return use_provided_address ? provided_address : start;
1233    }
1234}
1235
1236/// Target stat() handler.
1237template <class OS>
1238SyscallReturn
1239statFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
1240{
1241    std::string path;
1242    auto process = tc->getProcessPtr();
1243
1244    int index = 0;
1245    if (!tc->getVirtProxy().tryReadString(path,
1246                process->getSyscallArg(tc, index))) {
1247        return -EFAULT;
1248    }
1249    Addr bufPtr = process->getSyscallArg(tc, index);
1250
1251    // Adjust path for cwd and redirection
1252    path = process->checkPathRedirect(path);
1253
1254    struct stat hostBuf;
1255    int result = stat(path.c_str(), &hostBuf);
1256
1257    if (result < 0)
1258        return -errno;
1259
1260    copyOutStatBuf<OS>(tc->getVirtProxy(), bufPtr, &hostBuf);
1261
1262    return 0;
1263}
1264
1265
1266/// Target stat64() handler.
1267template <class OS>
1268SyscallReturn
1269stat64Func(SyscallDesc *desc, int callnum, ThreadContext *tc)
1270{
1271    std::string path;
1272    auto process = tc->getProcessPtr();
1273
1274    int index = 0;
1275    if (!tc->getVirtProxy().tryReadString(path,
1276                process->getSyscallArg(tc, index)))
1277        return -EFAULT;
1278    Addr bufPtr = process->getSyscallArg(tc, index);
1279
1280    // Adjust path for cwd and redirection
1281    path = process->checkPathRedirect(path);
1282
1283#if NO_STAT64
1284    struct stat  hostBuf;
1285    int result = stat(path.c_str(), &hostBuf);
1286#else
1287    struct stat64 hostBuf;
1288    int result = stat64(path.c_str(), &hostBuf);
1289#endif
1290
1291    if (result < 0)
1292        return -errno;
1293
1294    copyOutStat64Buf<OS>(tc->getVirtProxy(), bufPtr, &hostBuf);
1295
1296    return 0;
1297}
1298
1299
1300/// Target fstatat64() handler.
1301template <class OS>
1302SyscallReturn
1303fstatat64Func(SyscallDesc *desc, int callnum, ThreadContext *tc)
1304{
1305    int index = 0;
1306    auto process = tc->getProcessPtr();
1307    int dirfd = process->getSyscallArg(tc, index);
1308    if (dirfd != OS::TGT_AT_FDCWD)
1309        warn("fstatat64: first argument not AT_FDCWD; unlikely to work");
1310
1311    std::string path;
1312    if (!tc->getVirtProxy().tryReadString(path,
1313                process->getSyscallArg(tc, index)))
1314        return -EFAULT;
1315    Addr bufPtr = process->getSyscallArg(tc, index);
1316
1317    // Adjust path for cwd and redirection
1318    path = process->checkPathRedirect(path);
1319
1320#if NO_STAT64
1321    struct stat  hostBuf;
1322    int result = stat(path.c_str(), &hostBuf);
1323#else
1324    struct stat64 hostBuf;
1325    int result = stat64(path.c_str(), &hostBuf);
1326#endif
1327
1328    if (result < 0)
1329        return -errno;
1330
1331    copyOutStat64Buf<OS>(tc->getVirtProxy(), bufPtr, &hostBuf);
1332
1333    return 0;
1334}
1335
1336
1337/// Target fstat64() handler.
1338template <class OS>
1339SyscallReturn
1340fstat64Func(SyscallDesc *desc, int callnum, ThreadContext *tc)
1341{
1342    int index = 0;
1343    auto p = tc->getProcessPtr();
1344    int tgt_fd = p->getSyscallArg(tc, index);
1345    Addr bufPtr = p->getSyscallArg(tc, index);
1346
1347    auto ffdp = std::dynamic_pointer_cast<HBFDEntry>((*p->fds)[tgt_fd]);
1348    if (!ffdp)
1349        return -EBADF;
1350    int sim_fd = ffdp->getSimFD();
1351
1352#if NO_STAT64
1353    struct stat  hostBuf;
1354    int result = fstat(sim_fd, &hostBuf);
1355#else
1356    struct stat64  hostBuf;
1357    int result = fstat64(sim_fd, &hostBuf);
1358#endif
1359
1360    if (result < 0)
1361        return -errno;
1362
1363    copyOutStat64Buf<OS>(tc->getVirtProxy(), bufPtr, &hostBuf, (sim_fd == 1));
1364
1365    return 0;
1366}
1367
1368
1369/// Target lstat() handler.
1370template <class OS>
1371SyscallReturn
1372lstatFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
1373{
1374    std::string path;
1375    auto process = tc->getProcessPtr();
1376
1377    int index = 0;
1378    if (!tc->getVirtProxy().tryReadString(path,
1379                process->getSyscallArg(tc, index))) {
1380        return -EFAULT;
1381    }
1382    Addr bufPtr = process->getSyscallArg(tc, index);
1383
1384    // Adjust path for cwd and redirection
1385    path = process->checkPathRedirect(path);
1386
1387    struct stat hostBuf;
1388    int result = lstat(path.c_str(), &hostBuf);
1389
1390    if (result < 0)
1391        return -errno;
1392
1393    copyOutStatBuf<OS>(tc->getVirtProxy(), bufPtr, &hostBuf);
1394
1395    return 0;
1396}
1397
1398/// Target lstat64() handler.
1399template <class OS>
1400SyscallReturn
1401lstat64Func(SyscallDesc *desc, int callnum, ThreadContext *tc)
1402{
1403    std::string path;
1404    auto process = tc->getProcessPtr();
1405
1406    int index = 0;
1407    if (!tc->getVirtProxy().tryReadString(path,
1408                process->getSyscallArg(tc, index))) {
1409        return -EFAULT;
1410    }
1411    Addr bufPtr = process->getSyscallArg(tc, index);
1412
1413    // Adjust path for cwd and redirection
1414    path = process->checkPathRedirect(path);
1415
1416#if NO_STAT64
1417    struct stat hostBuf;
1418    int result = lstat(path.c_str(), &hostBuf);
1419#else
1420    struct stat64 hostBuf;
1421    int result = lstat64(path.c_str(), &hostBuf);
1422#endif
1423
1424    if (result < 0)
1425        return -errno;
1426
1427    copyOutStat64Buf<OS>(tc->getVirtProxy(), bufPtr, &hostBuf);
1428
1429    return 0;
1430}
1431
1432/// Target fstat() handler.
1433template <class OS>
1434SyscallReturn
1435fstatFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
1436{
1437    int index = 0;
1438    auto p = tc->getProcessPtr();
1439    int tgt_fd = p->getSyscallArg(tc, index);
1440    Addr bufPtr = p->getSyscallArg(tc, index);
1441
1442    DPRINTF_SYSCALL(Verbose, "fstat(%d, ...)\n", tgt_fd);
1443
1444    auto ffdp = std::dynamic_pointer_cast<FileFDEntry>((*p->fds)[tgt_fd]);
1445    if (!ffdp)
1446        return -EBADF;
1447    int sim_fd = ffdp->getSimFD();
1448
1449    struct stat hostBuf;
1450    int result = fstat(sim_fd, &hostBuf);
1451
1452    if (result < 0)
1453        return -errno;
1454
1455    copyOutStatBuf<OS>(tc->getVirtProxy(), bufPtr, &hostBuf, (sim_fd == 1));
1456
1457    return 0;
1458}
1459
1460/// Target statfs() handler.
1461template <class OS>
1462SyscallReturn
1463statfsFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
1464{
1465#if defined(__linux__)
1466    std::string path;
1467    auto process = tc->getProcessPtr();
1468
1469    int index = 0;
1470    if (!tc->getVirtProxy().tryReadString(path,
1471                process->getSyscallArg(tc, index))) {
1472        return -EFAULT;
1473    }
1474    Addr bufPtr = process->getSyscallArg(tc, index);
1475
1476    // Adjust path for cwd and redirection
1477    path = process->checkPathRedirect(path);
1478
1479    struct statfs hostBuf;
1480    int result = statfs(path.c_str(), &hostBuf);
1481
1482    if (result < 0)
1483        return -errno;
1484
1485    copyOutStatfsBuf<OS>(tc->getVirtProxy(), bufPtr, &hostBuf);
1486    return 0;
1487#else
1488    warnUnsupportedOS("statfs");
1489    return -1;
1490#endif
1491}
1492
1493template <class OS>
1494SyscallReturn
1495cloneFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
1496{
1497    int index = 0;
1498
1499    auto p = tc->getProcessPtr();
1500    RegVal flags = p->getSyscallArg(tc, index);
1501    RegVal newStack = p->getSyscallArg(tc, index);
1502    Addr ptidPtr = p->getSyscallArg(tc, index);
1503
1504#if THE_ISA == RISCV_ISA or THE_ISA == ARM_ISA
1505    /**
1506     * Linux sets CLONE_BACKWARDS flag for RISC-V and Arm.
1507     * The flag defines the list of clone() arguments in the following
1508     * order: flags -> newStack -> ptidPtr -> tlsPtr -> ctidPtr
1509     */
1510    Addr tlsPtr = p->getSyscallArg(tc, index);
1511    Addr ctidPtr = p->getSyscallArg(tc, index);
1512#else
1513    Addr ctidPtr = p->getSyscallArg(tc, index);
1514    Addr tlsPtr = p->getSyscallArg(tc, index);
1515#endif
1516
1517    if (((flags & OS::TGT_CLONE_SIGHAND)&& !(flags & OS::TGT_CLONE_VM)) ||
1518        ((flags & OS::TGT_CLONE_THREAD) && !(flags & OS::TGT_CLONE_SIGHAND)) ||
1519        ((flags & OS::TGT_CLONE_FS)     &&  (flags & OS::TGT_CLONE_NEWNS)) ||
1520        ((flags & OS::TGT_CLONE_NEWIPC) &&  (flags & OS::TGT_CLONE_SYSVSEM)) ||
1521        ((flags & OS::TGT_CLONE_NEWPID) &&  (flags & OS::TGT_CLONE_THREAD)) ||
1522        ((flags & OS::TGT_CLONE_VM)     && !(newStack)))
1523        return -EINVAL;
1524
1525    ThreadContext *ctc;
1526    if (!(ctc = p->findFreeContext())) {
1527        DPRINTF_SYSCALL(Verbose, "clone: no spare thread context in system"
1528                        "[cpu %d, thread %d]", tc->cpuId(), tc->threadId());
1529        return -EAGAIN;
1530    }
1531
1532    /**
1533     * Note that ProcessParams is generated by swig and there are no other
1534     * examples of how to create anything but this default constructor. The
1535     * fields are manually initialized instead of passing parameters to the
1536     * constructor.
1537     */
1538    ProcessParams *pp = new ProcessParams();
1539    pp->executable.assign(*(new std::string(p->progName())));
1540    pp->cmd.push_back(*(new std::string(p->progName())));
1541    pp->system = p->system;
1542    pp->cwd.assign(p->tgtCwd);
1543    pp->input.assign("stdin");
1544    pp->output.assign("stdout");
1545    pp->errout.assign("stderr");
1546    pp->uid = p->uid();
1547    pp->euid = p->euid();
1548    pp->gid = p->gid();
1549    pp->egid = p->egid();
1550
1551    /* Find the first free PID that's less than the maximum */
1552    std::set<int> const& pids = p->system->PIDs;
1553    int temp_pid = *pids.begin();
1554    do {
1555        temp_pid++;
1556    } while (pids.find(temp_pid) != pids.end());
1557    if (temp_pid >= System::maxPID)
1558        fatal("temp_pid is too large: %d", temp_pid);
1559
1560    pp->pid = temp_pid;
1561    pp->ppid = (flags & OS::TGT_CLONE_THREAD) ? p->ppid() : p->pid();
1562    pp->useArchPT = p->useArchPT;
1563    pp->kvmInSE = p->kvmInSE;
1564    Process *cp = pp->create();
1565    delete pp;
1566
1567    Process *owner = ctc->getProcessPtr();
1568    ctc->setProcessPtr(cp);
1569    cp->assignThreadContext(ctc->contextId());
1570    owner->revokeThreadContext(ctc->contextId());
1571
1572    if (flags & OS::TGT_CLONE_PARENT_SETTID) {
1573        BufferArg ptidBuf(ptidPtr, sizeof(long));
1574        long *ptid = (long *)ptidBuf.bufferPtr();
1575        *ptid = cp->pid();
1576        ptidBuf.copyOut(tc->getVirtProxy());
1577    }
1578
1579    if (flags & OS::TGT_CLONE_THREAD) {
1580        cp->pTable->shared = true;
1581        cp->useForClone = true;
1582    }
1583    cp->initState();
1584    p->clone(tc, ctc, cp, flags);
1585
1586    if (flags & OS::TGT_CLONE_THREAD) {
1587        delete cp->sigchld;
1588        cp->sigchld = p->sigchld;
1589    } else if (flags & OS::TGT_SIGCHLD) {
1590        *cp->sigchld = true;
1591    }
1592
1593    if (flags & OS::TGT_CLONE_CHILD_SETTID) {
1594        BufferArg ctidBuf(ctidPtr, sizeof(long));
1595        long *ctid = (long *)ctidBuf.bufferPtr();
1596        *ctid = cp->pid();
1597        ctidBuf.copyOut(ctc->getVirtProxy());
1598    }
1599
1600    if (flags & OS::TGT_CLONE_CHILD_CLEARTID)
1601        cp->childClearTID = (uint64_t)ctidPtr;
1602
1603    ctc->clearArchRegs();
1604
1605    OS::archClone(flags, p, cp, tc, ctc, newStack, tlsPtr);
1606
1607    cp->setSyscallReturn(ctc, 0);
1608
1609#if THE_ISA == ALPHA_ISA
1610    ctc->setIntReg(TheISA::SyscallSuccessReg, 0);
1611#elif THE_ISA == SPARC_ISA
1612    tc->setIntReg(TheISA::SyscallPseudoReturnReg, 0);
1613    ctc->setIntReg(TheISA::SyscallPseudoReturnReg, 1);
1614#endif
1615
1616    if (p->kvmInSE) {
1617#if THE_ISA == X86_ISA
1618        ctc->pcState(tc->readIntReg(TheISA::INTREG_RCX));
1619#else
1620        panic("KVM CPU model is not supported for this ISA");
1621#endif
1622    } else {
1623        TheISA::PCState cpc = tc->pcState();
1624        cpc.advance();
1625        ctc->pcState(cpc);
1626    }
1627    ctc->activate();
1628
1629    return cp->pid();
1630}
1631
1632/// Target fstatfs() handler.
1633template <class OS>
1634SyscallReturn
1635fstatfsFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
1636{
1637    int index = 0;
1638    auto p = tc->getProcessPtr();
1639    int tgt_fd = p->getSyscallArg(tc, index);
1640    Addr bufPtr = p->getSyscallArg(tc, index);
1641
1642    auto ffdp = std::dynamic_pointer_cast<FileFDEntry>((*p->fds)[tgt_fd]);
1643    if (!ffdp)
1644        return -EBADF;
1645    int sim_fd = ffdp->getSimFD();
1646
1647    struct statfs hostBuf;
1648    int result = fstatfs(sim_fd, &hostBuf);
1649
1650    if (result < 0)
1651        return -errno;
1652
1653    copyOutStatfsBuf<OS>(tc->getVirtProxy(), bufPtr, &hostBuf);
1654
1655    return 0;
1656}
1657
1658/// Target readv() handler.
1659template <class OS>
1660SyscallReturn
1661readvFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
1662{
1663    int index = 0;
1664    auto p = tc->getProcessPtr();
1665    int tgt_fd = p->getSyscallArg(tc, index);
1666
1667    auto ffdp = std::dynamic_pointer_cast<FileFDEntry>((*p->fds)[tgt_fd]);
1668    if (!ffdp)
1669        return -EBADF;
1670    int sim_fd = ffdp->getSimFD();
1671
1672    PortProxy &prox = tc->getVirtProxy();
1673    uint64_t tiov_base = p->getSyscallArg(tc, index);
1674    size_t count = p->getSyscallArg(tc, index);
1675    typename OS::tgt_iovec tiov[count];
1676    struct iovec hiov[count];
1677    for (size_t i = 0; i < count; ++i) {
1678        prox.readBlob(tiov_base + (i * sizeof(typename OS::tgt_iovec)),
1679                      &tiov[i], sizeof(typename OS::tgt_iovec));
1680        hiov[i].iov_len = TheISA::gtoh(tiov[i].iov_len);
1681        hiov[i].iov_base = new char [hiov[i].iov_len];
1682    }
1683
1684    int result = readv(sim_fd, hiov, count);
1685    int local_errno = errno;
1686
1687    for (size_t i = 0; i < count; ++i) {
1688        if (result != -1) {
1689            prox.writeBlob(TheISA::htog(tiov[i].iov_base),
1690                           hiov[i].iov_base, hiov[i].iov_len);
1691        }
1692        delete [] (char *)hiov[i].iov_base;
1693    }
1694
1695    return (result == -1) ? -local_errno : result;
1696}
1697
1698/// Target writev() handler.
1699template <class OS>
1700SyscallReturn
1701writevFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
1702{
1703    int index = 0;
1704    auto p = tc->getProcessPtr();
1705    int tgt_fd = p->getSyscallArg(tc, index);
1706
1707    auto hbfdp = std::dynamic_pointer_cast<HBFDEntry>((*p->fds)[tgt_fd]);
1708    if (!hbfdp)
1709        return -EBADF;
1710    int sim_fd = hbfdp->getSimFD();
1711
1712    PortProxy &prox = tc->getVirtProxy();
1713    uint64_t tiov_base = p->getSyscallArg(tc, index);
1714    size_t count = p->getSyscallArg(tc, index);
1715    struct iovec hiov[count];
1716    for (size_t i = 0; i < count; ++i) {
1717        typename OS::tgt_iovec tiov;
1718
1719        prox.readBlob(tiov_base + i*sizeof(typename OS::tgt_iovec),
1720                      &tiov, sizeof(typename OS::tgt_iovec));
1721        hiov[i].iov_len = TheISA::gtoh(tiov.iov_len);
1722        hiov[i].iov_base = new char [hiov[i].iov_len];
1723        prox.readBlob(TheISA::gtoh(tiov.iov_base), hiov[i].iov_base,
1724                      hiov[i].iov_len);
1725    }
1726
1727    int result = writev(sim_fd, hiov, count);
1728
1729    for (size_t i = 0; i < count; ++i)
1730        delete [] (char *)hiov[i].iov_base;
1731
1732    return (result == -1) ? -errno : result;
1733}
1734
1735/// Real mmap handler.
1736template <class OS>
1737SyscallReturn
1738mmapImpl(SyscallDesc *desc, int num, ThreadContext *tc, bool is_mmap2)
1739{
1740    int index = 0;
1741    auto p = tc->getProcessPtr();
1742    Addr start = p->getSyscallArg(tc, index);
1743    uint64_t length = p->getSyscallArg(tc, index);
1744    int prot = p->getSyscallArg(tc, index);
1745    int tgt_flags = p->getSyscallArg(tc, index);
1746    int tgt_fd = p->getSyscallArg(tc, index);
1747    int offset = p->getSyscallArg(tc, index);
1748
1749    if (is_mmap2)
1750        offset *= TheISA::PageBytes;
1751
1752    if (start & (TheISA::PageBytes - 1) ||
1753        offset & (TheISA::PageBytes - 1) ||
1754        (tgt_flags & OS::TGT_MAP_PRIVATE &&
1755         tgt_flags & OS::TGT_MAP_SHARED) ||
1756        (!(tgt_flags & OS::TGT_MAP_PRIVATE) &&
1757         !(tgt_flags & OS::TGT_MAP_SHARED)) ||
1758        !length) {
1759        return -EINVAL;
1760    }
1761
1762    if ((prot & PROT_WRITE) && (tgt_flags & OS::TGT_MAP_SHARED)) {
1763        // With shared mmaps, there are two cases to consider:
1764        // 1) anonymous: writes should modify the mapping and this should be
1765        // visible to observers who share the mapping. Currently, it's
1766        // difficult to update the shared mapping because there's no
1767        // structure which maintains information about the which virtual
1768        // memory areas are shared. If that structure existed, it would be
1769        // possible to make the translations point to the same frames.
1770        // 2) file-backed: writes should modify the mapping and the file
1771        // which is backed by the mapping. The shared mapping problem is the
1772        // same as what was mentioned about the anonymous mappings. For
1773        // file-backed mappings, the writes to the file are difficult
1774        // because it requires syncing what the mapping holds with the file
1775        // that resides on the host system. So, any write on a real system
1776        // would cause the change to be propagated to the file mapping at
1777        // some point in the future (the inode is tracked along with the
1778        // mapping). This isn't guaranteed to always happen, but it usually
1779        // works well enough. The guarantee is provided by the msync system
1780        // call. We could force the change through with shared mappings with
1781        // a call to msync, but that again would require more information
1782        // than we currently maintain.
1783        warn("mmap: writing to shared mmap region is currently "
1784             "unsupported. The write succeeds on the target, but it "
1785             "will not be propagated to the host or shared mappings");
1786    }
1787
1788    length = roundUp(length, TheISA::PageBytes);
1789
1790    int sim_fd = -1;
1791    uint8_t *pmap = nullptr;
1792    if (!(tgt_flags & OS::TGT_MAP_ANONYMOUS)) {
1793        std::shared_ptr<FDEntry> fdep = (*p->fds)[tgt_fd];
1794
1795        auto dfdp = std::dynamic_pointer_cast<DeviceFDEntry>(fdep);
1796        if (dfdp) {
1797            EmulatedDriver *emul_driver = dfdp->getDriver();
1798            return emul_driver->mmap(tc, start, length, prot, tgt_flags,
1799                                     tgt_fd, offset);
1800        }
1801
1802        auto ffdp = std::dynamic_pointer_cast<FileFDEntry>(fdep);
1803        if (!ffdp)
1804            return -EBADF;
1805        sim_fd = ffdp->getSimFD();
1806
1807        pmap = (decltype(pmap))mmap(nullptr, length, PROT_READ, MAP_PRIVATE,
1808                                    sim_fd, offset);
1809
1810        if (pmap == (decltype(pmap))-1) {
1811            warn("mmap: failed to map file into host address space");
1812            return -errno;
1813        }
1814    }
1815
1816    // Extend global mmap region if necessary. Note that we ignore the
1817    // start address unless MAP_FIXED is specified.
1818    if (!(tgt_flags & OS::TGT_MAP_FIXED)) {
1819        std::shared_ptr<MemState> mem_state = p->memState;
1820        Addr mmap_end = mem_state->getMmapEnd();
1821
1822        start = p->mmapGrowsDown() ? mmap_end - length : mmap_end;
1823        mmap_end = p->mmapGrowsDown() ? start : mmap_end + length;
1824
1825        mem_state->setMmapEnd(mmap_end);
1826    }
1827
1828    DPRINTF_SYSCALL(Verbose, " mmap range is 0x%x - 0x%x\n",
1829                    start, start + length - 1);
1830
1831    // We only allow mappings to overwrite existing mappings if
1832    // TGT_MAP_FIXED is set. Otherwise it shouldn't be a problem
1833    // because we ignore the start hint if TGT_MAP_FIXED is not set.
1834    int clobber = tgt_flags & OS::TGT_MAP_FIXED;
1835    if (clobber) {
1836        for (auto tc : p->system->threadContexts) {
1837            // If we might be overwriting old mappings, we need to
1838            // invalidate potentially stale mappings out of the TLBs.
1839            tc->getDTBPtr()->flushAll();
1840            tc->getITBPtr()->flushAll();
1841        }
1842    }
1843
1844    // Allocate physical memory and map it in. If the page table is already
1845    // mapped and clobber is not set, the simulator will issue throw a
1846    // fatal and bail out of the simulation.
1847    p->allocateMem(start, length, clobber);
1848
1849    // Transfer content into target address space.
1850    PortProxy &tp = tc->getVirtProxy();
1851    if (tgt_flags & OS::TGT_MAP_ANONYMOUS) {
1852        // In general, we should zero the mapped area for anonymous mappings,
1853        // with something like:
1854        //     tp.memsetBlob(start, 0, length);
1855        // However, given that we don't support sparse mappings, and
1856        // some applications can map a couple of gigabytes of space
1857        // (intending sparse usage), that can get painfully expensive.
1858        // Fortunately, since we don't properly implement munmap either,
1859        // there's no danger of remapping used memory, so for now all
1860        // newly mapped memory should already be zeroed so we can skip it.
1861    } else {
1862        // It is possible to mmap an area larger than a file, however
1863        // accessing unmapped portions the system triggers a "Bus error"
1864        // on the host. We must know when to stop copying the file from
1865        // the host into the target address space.
1866        struct stat file_stat;
1867        if (fstat(sim_fd, &file_stat) > 0)
1868            fatal("mmap: cannot stat file");
1869
1870        // Copy the portion of the file that is resident. This requires
1871        // checking both the mmap size and the filesize that we are
1872        // trying to mmap into this space; the mmap size also depends
1873        // on the specified offset into the file.
1874        uint64_t size = std::min((uint64_t)file_stat.st_size - offset,
1875                                 length);
1876        tp.writeBlob(start, pmap, size);
1877
1878        // Cleanup the mmap region before exiting this function.
1879        munmap(pmap, length);
1880
1881        // Maintain the symbol table for dynamic executables.
1882        // The loader will call mmap to map the images into its address
1883        // space and we intercept that here. We can verify that we are
1884        // executing inside the loader by checking the program counter value.
1885        // XXX: with multiprogrammed workloads or multi-node configurations,
1886        // this will not work since there is a single global symbol table.
1887        ObjectFile *interpreter = p->getInterpreter();
1888        if (interpreter) {
1889            Addr text_start = interpreter->textBase();
1890            Addr text_end = text_start + interpreter->textSize();
1891
1892            Addr pc = tc->pcState().pc();
1893
1894            if (pc >= text_start && pc < text_end) {
1895                std::shared_ptr<FDEntry> fdep = (*p->fds)[tgt_fd];
1896                auto ffdp = std::dynamic_pointer_cast<FileFDEntry>(fdep);
1897                ObjectFile *lib = createObjectFile(ffdp->getFileName());
1898
1899                if (lib) {
1900                    lib->loadAllSymbols(debugSymbolTable,
1901                                        lib->textBase(), start);
1902                }
1903            }
1904        }
1905
1906        // Note that we do not zero out the remainder of the mapping. This
1907        // is done by a real system, but it probably will not affect
1908        // execution (hopefully).
1909    }
1910
1911    return start;
1912}
1913
1914template <class OS>
1915SyscallReturn
1916pwrite64Func(SyscallDesc *desc, int num, ThreadContext *tc)
1917{
1918    int index = 0;
1919    auto p = tc->getProcessPtr();
1920    int tgt_fd = p->getSyscallArg(tc, index);
1921    Addr bufPtr = p->getSyscallArg(tc, index);
1922    int nbytes = p->getSyscallArg(tc, index);
1923    int offset = p->getSyscallArg(tc, index);
1924
1925    auto ffdp = std::dynamic_pointer_cast<FileFDEntry>((*p->fds)[tgt_fd]);
1926    if (!ffdp)
1927        return -EBADF;
1928    int sim_fd = ffdp->getSimFD();
1929
1930    BufferArg bufArg(bufPtr, nbytes);
1931    bufArg.copyIn(tc->getVirtProxy());
1932
1933    int bytes_written = pwrite(sim_fd, bufArg.bufferPtr(), nbytes, offset);
1934
1935    return (bytes_written == -1) ? -errno : bytes_written;
1936}
1937
1938/// Target mmap() handler.
1939template <class OS>
1940SyscallReturn
1941mmapFunc(SyscallDesc *desc, int num, ThreadContext *tc)
1942{
1943    return mmapImpl<OS>(desc, num, tc, false);
1944}
1945
1946/// Target mmap2() handler.
1947template <class OS>
1948SyscallReturn
1949mmap2Func(SyscallDesc *desc, int num, ThreadContext *tc)
1950{
1951    return mmapImpl<OS>(desc, num, tc, true);
1952}
1953
1954/// Target getrlimit() handler.
1955template <class OS>
1956SyscallReturn
1957getrlimitFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
1958{
1959    int index = 0;
1960    auto process = tc->getProcessPtr();
1961    unsigned resource = process->getSyscallArg(tc, index);
1962    TypedBufferArg<typename OS::rlimit> rlp(process->getSyscallArg(tc, index));
1963
1964    switch (resource) {
1965      case OS::TGT_RLIMIT_STACK:
1966        // max stack size in bytes: make up a number (8MB for now)
1967        rlp->rlim_cur = rlp->rlim_max = 8 * 1024 * 1024;
1968        rlp->rlim_cur = TheISA::htog(rlp->rlim_cur);
1969        rlp->rlim_max = TheISA::htog(rlp->rlim_max);
1970        break;
1971
1972      case OS::TGT_RLIMIT_DATA:
1973        // max data segment size in bytes: make up a number
1974        rlp->rlim_cur = rlp->rlim_max = 256 * 1024 * 1024;
1975        rlp->rlim_cur = TheISA::htog(rlp->rlim_cur);
1976        rlp->rlim_max = TheISA::htog(rlp->rlim_max);
1977        break;
1978
1979      case OS::TGT_RLIMIT_NPROC:
1980        rlp->rlim_cur = rlp->rlim_max = tc->getSystemPtr()->numContexts();
1981        rlp->rlim_cur = TheISA::htog(rlp->rlim_cur);
1982        rlp->rlim_max = TheISA::htog(rlp->rlim_max);
1983        break;
1984
1985      default:
1986        warn("getrlimit: unimplemented resource %d", resource);
1987        return -EINVAL;
1988        break;
1989    }
1990
1991    rlp.copyOut(tc->getVirtProxy());
1992    return 0;
1993}
1994
1995template <class OS>
1996SyscallReturn
1997prlimitFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
1998{
1999    int index = 0;
2000    auto process = tc->getProcessPtr();
2001    if (process->getSyscallArg(tc, index) != 0)
2002    {
2003        warn("prlimit: ignoring rlimits for nonzero pid");
2004        return -EPERM;
2005    }
2006    int resource = process->getSyscallArg(tc, index);
2007    Addr n = process->getSyscallArg(tc, index);
2008    if (n != 0)
2009        warn("prlimit: ignoring new rlimit");
2010    Addr o = process->getSyscallArg(tc, index);
2011    if (o != 0)
2012    {
2013        TypedBufferArg<typename OS::rlimit> rlp(o);
2014        switch (resource) {
2015          case OS::TGT_RLIMIT_STACK:
2016            // max stack size in bytes: make up a number (8MB for now)
2017            rlp->rlim_cur = rlp->rlim_max = 8 * 1024 * 1024;
2018            rlp->rlim_cur = TheISA::htog(rlp->rlim_cur);
2019            rlp->rlim_max = TheISA::htog(rlp->rlim_max);
2020            break;
2021          case OS::TGT_RLIMIT_DATA:
2022            // max data segment size in bytes: make up a number
2023            rlp->rlim_cur = rlp->rlim_max = 256*1024*1024;
2024            rlp->rlim_cur = TheISA::htog(rlp->rlim_cur);
2025            rlp->rlim_max = TheISA::htog(rlp->rlim_max);
2026            break;
2027          default:
2028            warn("prlimit: unimplemented resource %d", resource);
2029            return -EINVAL;
2030            break;
2031        }
2032        rlp.copyOut(tc->getVirtProxy());
2033    }
2034    return 0;
2035}
2036
2037/// Target clock_gettime() function.
2038template <class OS>
2039SyscallReturn
2040clock_gettimeFunc(SyscallDesc *desc, int num, ThreadContext *tc)
2041{
2042    int index = 1;
2043    auto p = tc->getProcessPtr();
2044    //int clk_id = p->getSyscallArg(tc, index);
2045    TypedBufferArg<typename OS::timespec> tp(p->getSyscallArg(tc, index));
2046
2047    getElapsedTimeNano(tp->tv_sec, tp->tv_nsec);
2048    tp->tv_sec += seconds_since_epoch;
2049    tp->tv_sec = TheISA::htog(tp->tv_sec);
2050    tp->tv_nsec = TheISA::htog(tp->tv_nsec);
2051
2052    tp.copyOut(tc->getVirtProxy());
2053
2054    return 0;
2055}
2056
2057/// Target clock_getres() function.
2058template <class OS>
2059SyscallReturn
2060clock_getresFunc(SyscallDesc *desc, int num, ThreadContext *tc)
2061{
2062    int index = 1;
2063    auto p = tc->getProcessPtr();
2064    TypedBufferArg<typename OS::timespec> tp(p->getSyscallArg(tc, index));
2065
2066    // Set resolution at ns, which is what clock_gettime() returns
2067    tp->tv_sec = 0;
2068    tp->tv_nsec = 1;
2069
2070    tp.copyOut(tc->getVirtProxy());
2071
2072    return 0;
2073}
2074
2075/// Target gettimeofday() handler.
2076template <class OS>
2077SyscallReturn
2078gettimeofdayFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
2079{
2080    int index = 0;
2081    auto process = tc->getProcessPtr();
2082    TypedBufferArg<typename OS::timeval> tp(process->getSyscallArg(tc, index));
2083
2084    getElapsedTimeMicro(tp->tv_sec, tp->tv_usec);
2085    tp->tv_sec += seconds_since_epoch;
2086    tp->tv_sec = TheISA::htog(tp->tv_sec);
2087    tp->tv_usec = TheISA::htog(tp->tv_usec);
2088
2089    tp.copyOut(tc->getVirtProxy());
2090
2091    return 0;
2092}
2093
2094
2095/// Target utimes() handler.
2096template <class OS>
2097SyscallReturn
2098utimesFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
2099{
2100    std::string path;
2101    auto process = tc->getProcessPtr();
2102
2103    int index = 0;
2104    if (!tc->getVirtProxy().tryReadString(path,
2105                process->getSyscallArg(tc, index))) {
2106        return -EFAULT;
2107    }
2108
2109    TypedBufferArg<typename OS::timeval [2]>
2110        tp(process->getSyscallArg(tc, index));
2111    tp.copyIn(tc->getVirtProxy());
2112
2113    struct timeval hostTimeval[2];
2114    for (int i = 0; i < 2; ++i) {
2115        hostTimeval[i].tv_sec = TheISA::gtoh((*tp)[i].tv_sec);
2116        hostTimeval[i].tv_usec = TheISA::gtoh((*tp)[i].tv_usec);
2117    }
2118
2119    // Adjust path for cwd and redirection
2120    path = process->checkPathRedirect(path);
2121
2122    int result = utimes(path.c_str(), hostTimeval);
2123
2124    if (result < 0)
2125        return -errno;
2126
2127    return 0;
2128}
2129
2130template <class OS>
2131SyscallReturn
2132execveFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
2133{
2134    desc->setFlags(0);
2135    auto p = tc->getProcessPtr();
2136
2137    int index = 0;
2138    std::string path;
2139    PortProxy & mem_proxy = tc->getVirtProxy();
2140    if (!mem_proxy.tryReadString(path, p->getSyscallArg(tc, index)))
2141        return -EFAULT;
2142
2143    if (access(path.c_str(), F_OK) == -1)
2144        return -EACCES;
2145
2146    auto read_in = [](std::vector<std::string> &vect,
2147                      PortProxy &mem_proxy, Addr mem_loc)
2148    {
2149        for (int inc = 0; ; inc++) {
2150            BufferArg b((mem_loc + sizeof(Addr) * inc), sizeof(Addr));
2151            b.copyIn(mem_proxy);
2152
2153            if (!*(Addr*)b.bufferPtr())
2154                break;
2155
2156            vect.push_back(std::string());
2157            mem_proxy.tryReadString(vect[inc], *(Addr*)b.bufferPtr());
2158        }
2159    };
2160
2161    /**
2162     * Note that ProcessParams is generated by swig and there are no other
2163     * examples of how to create anything but this default constructor. The
2164     * fields are manually initialized instead of passing parameters to the
2165     * constructor.
2166     */
2167    ProcessParams *pp = new ProcessParams();
2168    pp->executable = path;
2169    Addr argv_mem_loc = p->getSyscallArg(tc, index);
2170    read_in(pp->cmd, mem_proxy, argv_mem_loc);
2171    Addr envp_mem_loc = p->getSyscallArg(tc, index);
2172    read_in(pp->env, mem_proxy, envp_mem_loc);
2173    pp->uid = p->uid();
2174    pp->egid = p->egid();
2175    pp->euid = p->euid();
2176    pp->gid = p->gid();
2177    pp->ppid = p->ppid();
2178    pp->pid = p->pid();
2179    pp->input.assign("cin");
2180    pp->output.assign("cout");
2181    pp->errout.assign("cerr");
2182    pp->cwd.assign(p->tgtCwd);
2183    pp->system = p->system;
2184    /**
2185     * Prevent process object creation with identical PIDs (which will trip
2186     * a fatal check in Process constructor). The execve call is supposed to
2187     * take over the currently executing process' identity but replace
2188     * whatever it is doing with a new process image. Instead of hijacking
2189     * the process object in the simulator, we create a new process object
2190     * and bind to the previous process' thread below (hijacking the thread).
2191     */
2192    p->system->PIDs.erase(p->pid());
2193    Process *new_p = pp->create();
2194    delete pp;
2195
2196    /**
2197     * Work through the file descriptor array and close any files marked
2198     * close-on-exec.
2199     */
2200    new_p->fds = p->fds;
2201    for (int i = 0; i < new_p->fds->getSize(); i++) {
2202        std::shared_ptr<FDEntry> fdep = (*new_p->fds)[i];
2203        if (fdep && fdep->getCOE())
2204            new_p->fds->closeFDEntry(i);
2205    }
2206
2207    *new_p->sigchld = true;
2208
2209    delete p;
2210    tc->clearArchRegs();
2211    tc->setProcessPtr(new_p);
2212    new_p->assignThreadContext(tc->contextId());
2213    new_p->initState();
2214    tc->activate();
2215    TheISA::PCState pcState = tc->pcState();
2216    tc->setNPC(pcState.instAddr());
2217
2218    desc->setFlags(SyscallDesc::SuppressReturnValue);
2219    return 0;
2220}
2221
2222/// Target getrusage() function.
2223template <class OS>
2224SyscallReturn
2225getrusageFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
2226{
2227    int index = 0;
2228    auto process = tc->getProcessPtr();
2229    int who = process->getSyscallArg(tc, index); // THREAD, SELF, or CHILDREN
2230    TypedBufferArg<typename OS::rusage> rup(process->getSyscallArg(tc, index));
2231
2232    rup->ru_utime.tv_sec = 0;
2233    rup->ru_utime.tv_usec = 0;
2234    rup->ru_stime.tv_sec = 0;
2235    rup->ru_stime.tv_usec = 0;
2236    rup->ru_maxrss = 0;
2237    rup->ru_ixrss = 0;
2238    rup->ru_idrss = 0;
2239    rup->ru_isrss = 0;
2240    rup->ru_minflt = 0;
2241    rup->ru_majflt = 0;
2242    rup->ru_nswap = 0;
2243    rup->ru_inblock = 0;
2244    rup->ru_oublock = 0;
2245    rup->ru_msgsnd = 0;
2246    rup->ru_msgrcv = 0;
2247    rup->ru_nsignals = 0;
2248    rup->ru_nvcsw = 0;
2249    rup->ru_nivcsw = 0;
2250
2251    switch (who) {
2252      case OS::TGT_RUSAGE_SELF:
2253        getElapsedTimeMicro(rup->ru_utime.tv_sec, rup->ru_utime.tv_usec);
2254        rup->ru_utime.tv_sec = TheISA::htog(rup->ru_utime.tv_sec);
2255        rup->ru_utime.tv_usec = TheISA::htog(rup->ru_utime.tv_usec);
2256        break;
2257
2258      case OS::TGT_RUSAGE_CHILDREN:
2259        // do nothing.  We have no child processes, so they take no time.
2260        break;
2261
2262      default:
2263        // don't really handle THREAD or CHILDREN, but just warn and
2264        // plow ahead
2265        warn("getrusage() only supports RUSAGE_SELF.  Parameter %d ignored.",
2266             who);
2267    }
2268
2269    rup.copyOut(tc->getVirtProxy());
2270
2271    return 0;
2272}
2273
2274/// Target times() function.
2275template <class OS>
2276SyscallReturn
2277timesFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
2278{
2279    int index = 0;
2280    auto process = tc->getProcessPtr();
2281    TypedBufferArg<typename OS::tms> bufp(process->getSyscallArg(tc, index));
2282
2283    // Fill in the time structure (in clocks)
2284    int64_t clocks = curTick() * OS::M5_SC_CLK_TCK / SimClock::Int::s;
2285    bufp->tms_utime = clocks;
2286    bufp->tms_stime = 0;
2287    bufp->tms_cutime = 0;
2288    bufp->tms_cstime = 0;
2289
2290    // Convert to host endianness
2291    bufp->tms_utime = TheISA::htog(bufp->tms_utime);
2292
2293    // Write back
2294    bufp.copyOut(tc->getVirtProxy());
2295
2296    // Return clock ticks since system boot
2297    return clocks;
2298}
2299
2300/// Target time() function.
2301template <class OS>
2302SyscallReturn
2303timeFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
2304{
2305    typename OS::time_t sec, usec;
2306    getElapsedTimeMicro(sec, usec);
2307    sec += seconds_since_epoch;
2308
2309    int index = 0;
2310    auto process = tc->getProcessPtr();
2311    Addr taddr = (Addr)process->getSyscallArg(tc, index);
2312    if (taddr != 0) {
2313        typename OS::time_t t = sec;
2314        t = TheISA::htog(t);
2315        PortProxy &p = tc->getVirtProxy();
2316        p.writeBlob(taddr, &t, (int)sizeof(typename OS::time_t));
2317    }
2318    return sec;
2319}
2320
2321template <class OS>
2322SyscallReturn
2323tgkillFunc(SyscallDesc *desc, int num, ThreadContext *tc)
2324{
2325    int index = 0;
2326    auto process = tc->getProcessPtr();
2327    int tgid = process->getSyscallArg(tc, index);
2328    int tid = process->getSyscallArg(tc, index);
2329    int sig = process->getSyscallArg(tc, index);
2330
2331    /**
2332     * This system call is intended to allow killing a specific thread
2333     * within an arbitrary thread group if sanctioned with permission checks.
2334     * It's usually true that threads share the termination signal as pointed
2335     * out by the pthread_kill man page and this seems to be the intended
2336     * usage. Due to this being an emulated environment, assume the following:
2337     * Threads are allowed to call tgkill because the EUID for all threads
2338     * should be the same. There is no signal handling mechanism for kernel
2339     * registration of signal handlers since signals are poorly supported in
2340     * emulation mode. Since signal handlers cannot be registered, all
2341     * threads within in a thread group must share the termination signal.
2342     * We never exhaust PIDs so there's no chance of finding the wrong one
2343     * due to PID rollover.
2344     */
2345
2346    System *sys = tc->getSystemPtr();
2347    Process *tgt_proc = nullptr;
2348    for (int i = 0; i < sys->numContexts(); i++) {
2349        Process *temp = sys->threadContexts[i]->getProcessPtr();
2350        if (temp->pid() == tid) {
2351            tgt_proc = temp;
2352            break;
2353        }
2354    }
2355
2356    if (sig != 0 || sig != OS::TGT_SIGABRT)
2357        return -EINVAL;
2358
2359    if (tgt_proc == nullptr)
2360        return -ESRCH;
2361
2362    if (tgid != -1 && tgt_proc->tgid() != tgid)
2363        return -ESRCH;
2364
2365    if (sig == OS::TGT_SIGABRT)
2366        exitGroupFunc(desc, 252, tc);
2367
2368    return 0;
2369}
2370
2371template <class OS>
2372SyscallReturn
2373socketFunc(SyscallDesc *desc, int num, ThreadContext *tc)
2374{
2375    int index = 0;
2376    auto p = tc->getProcessPtr();
2377    int domain = p->getSyscallArg(tc, index);
2378    int type = p->getSyscallArg(tc, index);
2379    int prot = p->getSyscallArg(tc, index);
2380
2381    int sim_fd = socket(domain, type, prot);
2382    if (sim_fd == -1)
2383        return -errno;
2384
2385    auto sfdp = std::make_shared<SocketFDEntry>(sim_fd, domain, type, prot);
2386    int tgt_fd = p->fds->allocFD(sfdp);
2387
2388    return tgt_fd;
2389}
2390
2391template <class OS>
2392SyscallReturn
2393socketpairFunc(SyscallDesc *desc, int num, ThreadContext *tc)
2394{
2395    int index = 0;
2396    auto p = tc->getProcessPtr();
2397    int domain = p->getSyscallArg(tc, index);
2398    int type = p->getSyscallArg(tc, index);
2399    int prot = p->getSyscallArg(tc, index);
2400    Addr svPtr = p->getSyscallArg(tc, index);
2401
2402    BufferArg svBuf((Addr)svPtr, 2 * sizeof(int));
2403    int status = socketpair(domain, type, prot, (int *)svBuf.bufferPtr());
2404    if (status == -1)
2405        return -errno;
2406
2407    int *fds = (int *)svBuf.bufferPtr();
2408
2409    auto sfdp1 = std::make_shared<SocketFDEntry>(fds[0], domain, type, prot);
2410    fds[0] = p->fds->allocFD(sfdp1);
2411    auto sfdp2 = std::make_shared<SocketFDEntry>(fds[1], domain, type, prot);
2412    fds[1] = p->fds->allocFD(sfdp2);
2413    svBuf.copyOut(tc->getVirtProxy());
2414
2415    return status;
2416}
2417
2418template <class OS>
2419SyscallReturn
2420selectFunc(SyscallDesc *desc, int callnum, ThreadContext *tc)
2421{
2422    int retval;
2423
2424    int index = 0;
2425    auto p = tc->getProcessPtr();
2426    int nfds_t = p->getSyscallArg(tc, index);
2427    Addr fds_read_ptr = p->getSyscallArg(tc, index);
2428    Addr fds_writ_ptr = p->getSyscallArg(tc, index);
2429    Addr fds_excp_ptr = p->getSyscallArg(tc, index);
2430    Addr time_val_ptr = p->getSyscallArg(tc, index);
2431
2432    TypedBufferArg<typename OS::fd_set> rd_t(fds_read_ptr);
2433    TypedBufferArg<typename OS::fd_set> wr_t(fds_writ_ptr);
2434    TypedBufferArg<typename OS::fd_set> ex_t(fds_excp_ptr);
2435    TypedBufferArg<typename OS::timeval> tp(time_val_ptr);
2436
2437    /**
2438     * Host fields. Notice that these use the definitions from the system
2439     * headers instead of the gem5 headers and libraries. If the host and
2440     * target have different header file definitions, this will not work.
2441     */
2442    fd_set rd_h;
2443    FD_ZERO(&rd_h);
2444    fd_set wr_h;
2445    FD_ZERO(&wr_h);
2446    fd_set ex_h;
2447    FD_ZERO(&ex_h);
2448
2449    /**
2450     * Copy in the fd_set from the target.
2451     */
2452    if (fds_read_ptr)
2453        rd_t.copyIn(tc->getVirtProxy());
2454    if (fds_writ_ptr)
2455        wr_t.copyIn(tc->getVirtProxy());
2456    if (fds_excp_ptr)
2457        ex_t.copyIn(tc->getVirtProxy());
2458
2459    /**
2460     * We need to translate the target file descriptor set into a host file
2461     * descriptor set. This involves both our internal process fd array
2462     * and the fd_set defined in Linux header files. The nfds field also
2463     * needs to be updated as it will be only target specific after
2464     * retrieving it from the target; the nfds value is expected to be the
2465     * highest file descriptor that needs to be checked, so we need to extend
2466     * it out for nfds_h when we do the update.
2467     */
2468    int nfds_h = 0;
2469    std::map<int, int> trans_map;
2470    auto try_add_host_set = [&](fd_set *tgt_set_entry,
2471                                fd_set *hst_set_entry,
2472                                int iter) -> bool
2473    {
2474        /**
2475         * By this point, we know that we are looking at a valid file
2476         * descriptor set on the target. We need to check if the target file
2477         * descriptor value passed in as iter is part of the set.
2478         */
2479        if (FD_ISSET(iter, tgt_set_entry)) {
2480            /**
2481             * We know that the target file descriptor belongs to the set,
2482             * but we do not yet know if the file descriptor is valid or
2483             * that we have a host mapping. Check that now.
2484             */
2485            auto hbfdp = std::dynamic_pointer_cast<HBFDEntry>((*p->fds)[iter]);
2486            if (!hbfdp)
2487                return true;
2488            auto sim_fd = hbfdp->getSimFD();
2489
2490            /**
2491             * Add the sim_fd to tgt_fd translation into trans_map for use
2492             * later when we need to zero the target fd_set structures and
2493             * then update them with hits returned from the host select call.
2494             */
2495            trans_map[sim_fd] = iter;
2496
2497            /**
2498             * We know that the host file descriptor exists so now we check
2499             * if we need to update the max count for nfds_h before passing
2500             * the duplicated structure into the host.
2501             */
2502            nfds_h = std::max(nfds_h - 1, sim_fd + 1);
2503
2504            /**
2505             * Add the host file descriptor to the set that we are going to
2506             * pass into the host.
2507             */
2508            FD_SET(sim_fd, hst_set_entry);
2509        }
2510        return false;
2511    };
2512
2513    for (int i = 0; i < nfds_t; i++) {
2514        if (fds_read_ptr) {
2515            bool ebadf = try_add_host_set((fd_set*)&*rd_t, &rd_h, i);
2516            if (ebadf) return -EBADF;
2517        }
2518        if (fds_writ_ptr) {
2519            bool ebadf = try_add_host_set((fd_set*)&*wr_t, &wr_h, i);
2520            if (ebadf) return -EBADF;
2521        }
2522        if (fds_excp_ptr) {
2523            bool ebadf = try_add_host_set((fd_set*)&*ex_t, &ex_h, i);
2524            if (ebadf) return -EBADF;
2525        }
2526    }
2527
2528    if (time_val_ptr) {
2529        /**
2530         * It might be possible to decrement the timeval based on some
2531         * derivation of wall clock determined from elapsed simulator ticks
2532         * but that seems like overkill. Rather, we just set the timeval with
2533         * zero timeout. (There is no reason to block during the simulation
2534         * as it only decreases simulator performance.)
2535         */
2536        tp->tv_sec = 0;
2537        tp->tv_usec = 0;
2538
2539        retval = select(nfds_h,
2540                        fds_read_ptr ? &rd_h : nullptr,
2541                        fds_writ_ptr ? &wr_h : nullptr,
2542                        fds_excp_ptr ? &ex_h : nullptr,
2543                        (timeval*)&*tp);
2544    } else {
2545        /**
2546         * If the timeval pointer is null, setup a new timeval structure to
2547         * pass into the host select call. Unfortunately, we will need to
2548         * manually check the return value and throw a retry fault if the
2549         * return value is zero. Allowing the system call to block will
2550         * likely deadlock the event queue.
2551         */
2552        struct timeval tv = { 0, 0 };
2553
2554        retval = select(nfds_h,
2555                        fds_read_ptr ? &rd_h : nullptr,
2556                        fds_writ_ptr ? &wr_h : nullptr,
2557                        fds_excp_ptr ? &ex_h : nullptr,
2558                        &tv);
2559
2560        if (retval == 0) {
2561            /**
2562             * If blocking indefinitely, check the signal list to see if a
2563             * signal would break the poll out of the retry cycle and try to
2564             * return the signal interrupt instead.
2565             */
2566            for (auto sig : tc->getSystemPtr()->signalList)
2567                if (sig.receiver == p)
2568                    return -EINTR;
2569            return SyscallReturn::retry();
2570        }
2571    }
2572
2573    if (retval == -1)
2574        return -errno;
2575
2576    FD_ZERO((fd_set*)&*rd_t);
2577    FD_ZERO((fd_set*)&*wr_t);
2578    FD_ZERO((fd_set*)&*ex_t);
2579
2580    /**
2581     * We need to translate the host file descriptor set into a target file
2582     * descriptor set. This involves both our internal process fd array
2583     * and the fd_set defined in header files.
2584     */
2585    for (int i = 0; i < nfds_h; i++) {
2586        if (fds_read_ptr) {
2587            if (FD_ISSET(i, &rd_h))
2588                FD_SET(trans_map[i], (fd_set*)&*rd_t);
2589        }
2590
2591        if (fds_writ_ptr) {
2592            if (FD_ISSET(i, &wr_h))
2593                FD_SET(trans_map[i], (fd_set*)&*wr_t);
2594        }
2595
2596        if (fds_excp_ptr) {
2597            if (FD_ISSET(i, &ex_h))
2598                FD_SET(trans_map[i], (fd_set*)&*ex_t);
2599        }
2600    }
2601
2602    if (fds_read_ptr)
2603        rd_t.copyOut(tc->getVirtProxy());
2604    if (fds_writ_ptr)
2605        wr_t.copyOut(tc->getVirtProxy());
2606    if (fds_excp_ptr)
2607        ex_t.copyOut(tc->getVirtProxy());
2608    if (time_val_ptr)
2609        tp.copyOut(tc->getVirtProxy());
2610
2611    return retval;
2612}
2613
2614template <class OS>
2615SyscallReturn
2616readFunc(SyscallDesc *desc, int num, ThreadContext *tc)
2617{
2618    int index = 0;
2619    auto p = tc->getProcessPtr();
2620    int tgt_fd = p->getSyscallArg(tc, index);
2621    Addr buf_ptr = p->getSyscallArg(tc, index);
2622    int nbytes = p->getSyscallArg(tc, index);
2623
2624    auto hbfdp = std::dynamic_pointer_cast<HBFDEntry>((*p->fds)[tgt_fd]);
2625    if (!hbfdp)
2626        return -EBADF;
2627    int sim_fd = hbfdp->getSimFD();
2628
2629    struct pollfd pfd;
2630    pfd.fd = sim_fd;
2631    pfd.events = POLLIN | POLLPRI;
2632    if ((poll(&pfd, 1, 0) == 0)
2633        && !(hbfdp->getFlags() & OS::TGT_O_NONBLOCK))
2634        return SyscallReturn::retry();
2635
2636    BufferArg buf_arg(buf_ptr, nbytes);
2637    int bytes_read = read(sim_fd, buf_arg.bufferPtr(), nbytes);
2638
2639    if (bytes_read > 0)
2640        buf_arg.copyOut(tc->getVirtProxy());
2641
2642    return (bytes_read == -1) ? -errno : bytes_read;
2643}
2644
2645template <class OS>
2646SyscallReturn
2647writeFunc(SyscallDesc *desc, int num, ThreadContext *tc)
2648{
2649    int index = 0;
2650    auto p = tc->getProcessPtr();
2651    int tgt_fd = p->getSyscallArg(tc, index);
2652    Addr buf_ptr = p->getSyscallArg(tc, index);
2653    int nbytes = p->getSyscallArg(tc, index);
2654
2655    auto hbfdp = std::dynamic_pointer_cast<HBFDEntry>((*p->fds)[tgt_fd]);
2656    if (!hbfdp)
2657        return -EBADF;
2658    int sim_fd = hbfdp->getSimFD();
2659
2660    BufferArg buf_arg(buf_ptr, nbytes);
2661    buf_arg.copyIn(tc->getVirtProxy());
2662
2663    struct pollfd pfd;
2664    pfd.fd = sim_fd;
2665    pfd.events = POLLOUT;
2666
2667    /**
2668     * We don't want to poll on /dev/random. The kernel will not enable the
2669     * file descriptor for writing unless the entropy in the system falls
2670     * below write_wakeup_threshold. This is not guaranteed to happen
2671     * depending on host settings.
2672     */
2673    auto ffdp = std::dynamic_pointer_cast<FileFDEntry>(hbfdp);
2674    if (ffdp && (ffdp->getFileName() != "/dev/random")) {
2675        if (!poll(&pfd, 1, 0) && !(ffdp->getFlags() & OS::TGT_O_NONBLOCK))
2676            return SyscallReturn::retry();
2677    }
2678
2679    int bytes_written = write(sim_fd, buf_arg.bufferPtr(), nbytes);
2680
2681    if (bytes_written != -1)
2682        fsync(sim_fd);
2683
2684    return (bytes_written == -1) ? -errno : bytes_written;
2685}
2686
2687template <class OS>
2688SyscallReturn
2689wait4Func(SyscallDesc *desc, int num, ThreadContext *tc)
2690{
2691    int index = 0;
2692    auto p = tc->getProcessPtr();
2693    pid_t pid = p->getSyscallArg(tc, index);
2694    Addr statPtr = p->getSyscallArg(tc, index);
2695    int options = p->getSyscallArg(tc, index);
2696    Addr rusagePtr = p->getSyscallArg(tc, index);
2697
2698    if (rusagePtr)
2699        DPRINTF_SYSCALL(Verbose, "wait4: rusage pointer provided %lx, however "
2700                 "functionality not supported. Ignoring rusage pointer.\n",
2701                 rusagePtr);
2702
2703    /**
2704     * Currently, wait4 is only implemented so that it will wait for children
2705     * exit conditions which are denoted by a SIGCHLD signals posted into the
2706     * system signal list. We return no additional information via any of the
2707     * parameters supplied to wait4. If nothing is found in the system signal
2708     * list, we will wait indefinitely for SIGCHLD to post by retrying the
2709     * call.
2710     */
2711    System *sysh = tc->getSystemPtr();
2712    std::list<BasicSignal>::iterator iter;
2713    for (iter=sysh->signalList.begin(); iter!=sysh->signalList.end(); iter++) {
2714        if (iter->receiver == p) {
2715            if (pid < -1) {
2716                if ((iter->sender->pgid() == -pid)
2717                    && (iter->signalValue == OS::TGT_SIGCHLD))
2718                    goto success;
2719            } else if (pid == -1) {
2720                if (iter->signalValue == OS::TGT_SIGCHLD)
2721                    goto success;
2722            } else if (pid == 0) {
2723                if ((iter->sender->pgid() == p->pgid())
2724                    && (iter->signalValue == OS::TGT_SIGCHLD))
2725                    goto success;
2726            } else {
2727                if ((iter->sender->pid() == pid)
2728                    && (iter->signalValue == OS::TGT_SIGCHLD))
2729                    goto success;
2730            }
2731        }
2732    }
2733
2734    return (options & OS::TGT_WNOHANG) ? 0 : SyscallReturn::retry();
2735
2736success:
2737    // Set status to EXITED for WIFEXITED evaluations.
2738    const int EXITED = 0;
2739    BufferArg statusBuf(statPtr, sizeof(int));
2740    *(int *)statusBuf.bufferPtr() = EXITED;
2741    statusBuf.copyOut(tc->getVirtProxy());
2742
2743    // Return the child PID.
2744    pid_t retval = iter->sender->pid();
2745    sysh->signalList.erase(iter);
2746    return retval;
2747}
2748
2749template <class OS>
2750SyscallReturn
2751acceptFunc(SyscallDesc *desc, int num, ThreadContext *tc)
2752{
2753    struct sockaddr sa;
2754    socklen_t addrLen;
2755    int host_fd;
2756    int index = 0;
2757    auto p = tc->getProcessPtr();
2758    int tgt_fd = p->getSyscallArg(tc, index);
2759    Addr addrPtr = p->getSyscallArg(tc, index);
2760    Addr lenPtr = p->getSyscallArg(tc, index);
2761
2762    BufferArg *lenBufPtr = nullptr;
2763    BufferArg *addrBufPtr = nullptr;
2764
2765    auto sfdp = std::dynamic_pointer_cast<SocketFDEntry>((*p->fds)[tgt_fd]);
2766    if (!sfdp)
2767        return -EBADF;
2768    int sim_fd = sfdp->getSimFD();
2769
2770    /**
2771     * We poll the socket file descriptor first to guarantee that we do not
2772     * block on our accept call. The socket can be opened without the
2773     * non-blocking flag (it blocks). This will cause deadlocks between
2774     * communicating processes.
2775     */
2776    struct pollfd pfd;
2777    pfd.fd = sim_fd;
2778    pfd.events = POLLIN | POLLPRI;
2779    if ((poll(&pfd, 1, 0) == 0)
2780        && !(sfdp->getFlags() & OS::TGT_O_NONBLOCK))
2781        return SyscallReturn::retry();
2782
2783    if (lenPtr) {
2784        lenBufPtr = new BufferArg(lenPtr, sizeof(socklen_t));
2785        lenBufPtr->copyIn(tc->getVirtProxy());
2786        memcpy(&addrLen, (socklen_t *)lenBufPtr->bufferPtr(),
2787               sizeof(socklen_t));
2788    }
2789
2790    if (addrPtr) {
2791        addrBufPtr = new BufferArg(addrPtr, sizeof(struct sockaddr));
2792        addrBufPtr->copyIn(tc->getVirtProxy());
2793        memcpy(&sa, (struct sockaddr *)addrBufPtr->bufferPtr(),
2794               sizeof(struct sockaddr));
2795    }
2796
2797    host_fd = accept(sim_fd, &sa, &addrLen);
2798
2799    if (host_fd == -1)
2800        return -errno;
2801
2802    if (addrPtr) {
2803        memcpy(addrBufPtr->bufferPtr(), &sa, sizeof(sa));
2804        addrBufPtr->copyOut(tc->getVirtProxy());
2805        delete(addrBufPtr);
2806    }
2807
2808    if (lenPtr) {
2809        *(socklen_t *)lenBufPtr->bufferPtr() = addrLen;
2810        lenBufPtr->copyOut(tc->getVirtProxy());
2811        delete(lenBufPtr);
2812    }
2813
2814    auto afdp = std::make_shared<SocketFDEntry>(host_fd, sfdp->_domain,
2815                                                sfdp->_type, sfdp->_protocol);
2816    return p->fds->allocFD(afdp);
2817}
2818
2819/// Target eventfd() function.
2820template <class OS>
2821SyscallReturn
2822eventfdFunc(SyscallDesc *desc, int num, ThreadContext *tc)
2823{
2824#if defined(__linux__)
2825    int index = 0;
2826    auto p = tc->getProcessPtr();
2827    unsigned initval = p->getSyscallArg(tc, index);
2828    int in_flags = p->getSyscallArg(tc, index);
2829
2830    int sim_fd = eventfd(initval, in_flags);
2831    if (sim_fd == -1)
2832        return -errno;
2833
2834    bool cloexec = in_flags & OS::TGT_O_CLOEXEC;
2835
2836    int flags = cloexec ? OS::TGT_O_CLOEXEC : 0;
2837    flags |= (in_flags & OS::TGT_O_NONBLOCK) ? OS::TGT_O_NONBLOCK : 0;
2838
2839    auto hbfdp = std::make_shared<HBFDEntry>(flags, sim_fd, cloexec);
2840    int tgt_fd = p->fds->allocFD(hbfdp);
2841    return tgt_fd;
2842#else
2843    warnUnsupportedOS("eventfd");
2844    return -1;
2845#endif
2846}
2847
2848#endif // __SIM_SYSCALL_EMUL_HH__
2849