summaryrefslogtreecommitdiff
path: root/arch/alpha
diff options
context:
space:
mode:
authorSteve Reinhardt <stever@eecs.umich.edu>2003-11-03 20:31:15 -0800
committerSteve Reinhardt <stever@eecs.umich.edu>2003-11-03 20:31:15 -0800
commit02795babaf52cbf6f8c29bbb2aecfc0e60b46b63 (patch)
tree2b449d92d76e9b5bda962d60bb981469fe238f9c /arch/alpha
parent40b9a3878aed4e2400727ca87e1c9976cdecef58 (diff)
downloadgem5-02795babaf52cbf6f8c29bbb2aecfc0e60b46b63.tar.xz
Fix bugs in and expand syscall emulation code.
arch/alpha/fake_syscall.cc: Fix a couple of bugs: - error return codes weren't making it through due to inadvertent cast to unsigned - sigreturn broken in not one but two ways - make all file descriptors look like plain files (not ttys) Added implementations of setuid, getgid, fcntl, and getdirentries from Dave Oehmke --HG-- extra : convert_revision : 53d3f13e1b05e3bde9e68ada3774ca39fa4c0d4c
Diffstat (limited to 'arch/alpha')
-rw-r--r--arch/alpha/fake_syscall.cc296
1 files changed, 222 insertions, 74 deletions
diff --git a/arch/alpha/fake_syscall.cc b/arch/alpha/fake_syscall.cc
index b2e42daaf..43bb0a9cc 100644
--- a/arch/alpha/fake_syscall.cc
+++ b/arch/alpha/fake_syscall.cc
@@ -31,6 +31,7 @@
#include <sys/types.h>
#include <sys/stat.h>
#include <string.h> // for memset()
+#include <dirent.h>
#include "sim/host.hh"
#include "cpu/base_cpu.hh"
@@ -61,11 +62,17 @@ class SyscallDesc {
FuncPtr funcPtr;
int flags;
+ enum Flags {
+ SuppressReturnValue = 1
+ };
+
SyscallDesc(const char *_name, FuncPtr _funcPtr, int _flags = 0)
: name(_name), funcPtr(_funcPtr), flags(_flags)
- {}
+ {
+ }
- int doFunc(int num, Process *proc, ExecContext *xc) {
+ int doFunc(int num, Process *proc, ExecContext *xc)
+ {
return (*funcPtr)(this, num, proc, xc);
}
};
@@ -75,7 +82,8 @@ class BaseBufferArg {
public:
- BaseBufferArg(Addr _addr, int _size) : addr(_addr), size(_size) {
+ BaseBufferArg(Addr _addr, int _size) : addr(_addr), size(_size)
+ {
bufPtr = new uint8_t[size];
// clear out buffer: in case we only partially populate this,
// and then do a copyOut(), we want to make sure we don't
@@ -88,7 +96,8 @@ class BaseBufferArg {
//
// copy data into simulator space (read from target memory)
//
- virtual bool copyIn(FunctionalMemory *mem) {
+ virtual bool copyIn(FunctionalMemory *mem)
+ {
mem->access(Read, addr, bufPtr, size);
return true; // no EFAULT detection for now
}
@@ -96,7 +105,8 @@ class BaseBufferArg {
//
// copy data out of simulator space (write to target memory)
//
- virtual bool copyOut(FunctionalMemory *mem) {
+ virtual bool copyOut(FunctionalMemory *mem)
+ {
mem->access(Write, addr, bufPtr, size);
return true; // no EFAULT detection for now
}
@@ -154,7 +164,7 @@ setArg(ExecContext *xc, int i, IntReg val)
static void
-set_return_value(ExecContext *xc, IntReg return_value)
+set_return_value(ExecContext *xc, int64_t return_value)
{
// check for error condition. Alpha syscall convention is to
// indicate success/failure in reg a3 (r19) and put the
@@ -194,60 +204,27 @@ int
ioctlFunc(SyscallDesc *desc, int callnum, Process *process,
ExecContext *xc)
{
- int fd = process->sim_fd(getArg(xc, 0));
+ int fd = getArg(xc, 0);
unsigned req = getArg(xc, 1);
- switch (req) {
- case OSF::TIOCGETP: {
- // get tty parameters: the main use of this call is by
- // isatty(), which really just wants to see whether it
- // succeeds or returns ENOTTY to determine whether this is
- // a terminal or not. This call is in turn used by the
- // stdio library to determine whether to do line buffering
- // or block buffering on a specific file descriptor.
- TypedBufferArg<OSF::sgttyb> buf(getArg(xc, 2));
-
- if (fd < 0) {
- // bad file descriptor
- return -EBADF;
- } else if (0 <= fd < 3) {
- // stdin/stdout/stderr: make it look like a terminal
- // so we get line buffering & not block buffering
- buf->sg_ispeed = 0xf;
- buf->sg_ospeed = 0xf;
- buf->sg_erase = 0x7f;
- buf->sg_kill = 0x15;
- buf->sg_flags = 0x18;
- buf.copyOut(xc->mem);
- return 0;
- } else {
- // any other file descriptor: assume it's a file or
- // pipe and not a terminal
- return -ENOTTY;
- }
- break;
- }
+ DPRINTFR(SyscallVerbose, "ioctl(%d, 0x%x, ...)\n", fd, req);
+ if (fd < 0 || process->sim_fd(fd) < 0) {
+ // doesn't map to any simulator fd: not a valid target fd
+ return -EBADF;
+ }
+
+ switch (req) {
case OSF::TIOCISATTY:
- if (fd < 0) {
- // bad file descriptor
- return -EBADF;
- } else if (0 <= fd < 3) {
- // stdin/stdout/stderr: make it look like a terminal
- // so we get line buffering & not block buffering
- return 0;
- } else {
- // any other file descriptor: assume it's a file or
- // pipe and not a terminal
- return -ENOTTY;
- }
- break;
+ case OSF::TIOCGETP:
+ case OSF::TIOCSETP:
+ case OSF::TIOCSETN:
+ case OSF::TIOCSETC:
+ case OSF::TIOCGETC:
+ return -ENOTTY;
default:
- cerr << "Unsupported ioctl call: ioctl("
- << fd << ", " << req << ", ...)" << endl;
- abort();
- break;
+ fatal("Unsupported ioctl call: ioctl(%d, 0x%x, ...)\n", fd, req);
}
}
@@ -421,6 +398,8 @@ fstatFunc(SyscallDesc *desc, int callnum, Process *process,
{
int fd = process->sim_fd(getArg(xc, 0));
+ DPRINTFR(SyscallVerbose, "fstat(%d, ...)\n", fd);
+
if (fd < 0)
return -EBADF;
@@ -586,17 +565,179 @@ getpidFunc(SyscallDesc *desc, int callnum, Process *process,
// Make up a PID. There's no interprocess communication in
// fake_syscall mode, so there's no way for a process to know it's
// not getting a unique value.
+
+ // This is one of the funky syscalls that has two return values,
+ // with the second one (parent PID) going in r20.
+ xc->regs.intRegFile[20] = 99;
return 100;
}
+
int
getuidFunc(SyscallDesc *desc, int callnum, Process *process,
ExecContext *xc)
{
- // Make up a UID.
+ // Make up a UID and EUID... it shouldn't matter, and we want the
+ // simulation to be deterministic.
+
+ // EUID goes in r20.
+ xc->regs.intRegFile[20] = 100; // EUID
+ return 100; // UID
+}
+
+
+int
+getgidFunc(SyscallDesc *desc, int callnum, Process *process,
+ ExecContext *xc)
+{
+ // Get current group ID. EGID goes in r20.
+ xc->regs.intRegFile[20] = 100;
return 100;
}
+
+int
+setuidFunc(SyscallDesc *desc, int callnum, Process *process,
+ ExecContext *xc)
+{
+ // can't fathom why a benchmark would call this.
+ warn("Ignoring call to setuid(%d)\n", getArg(xc, 0));
+ return 0;
+}
+
+
+int
+fcntlFunc(SyscallDesc *desc, int callnum, Process *process,
+ ExecContext *xc)
+{
+ int fd = getArg(xc,0);
+
+ if (fd < 0 || process->sim_fd(fd) < 0)
+ return -EBADF;
+
+ int cmd = getArg(xc,1);
+ switch (cmd) {
+ case 0: // F_DUPFD
+ // if we really wanted to support this, we'd need to do it
+ // in the target fd space.
+ warn("fcntl(%d, F_DUPFD) not supported, error returned\n", fd);
+ return -EMFILE;
+
+ case 1: // F_GETFD (get fd flags)
+ case 2: // F_SETFD (set fd flags)
+ case 3: // F_GETFL (get file flags)
+ case 4: // F_SETFL (set file flags)
+ // not sure if this is totally valid, but we'll pass it through
+ // to the underlying OS
+ warn("fcntl(%d, %d) passed through to host\n", fd, cmd);
+ return fcntl(process->sim_fd(fd), cmd);
+ // return 0;
+
+ case 7: // F_GETLK (get lock)
+ case 8: // F_SETLK (set lock)
+ case 9: // F_SETLKW (set lock and wait)
+ // don't mess with file locking... just act like it's OK
+ warn("File lock call (fcntl(%d, %d)) ignored.\n", fd, cmd);
+ return 0;
+
+ default:
+ warn("Unknown fcntl command %d\n", cmd);
+ return 0;
+ }
+}
+
+
+///
+/// @TODO this was stolen from SimpleScalar and needs to be rewritten.
+///
+
+/* returns size of DIRENT structure */
+#define OSF_DIRENT_SZ(STR) \
+ (sizeof(uint32_t) + 2*sizeof(uint16_t) + (((strlen(STR) + 1) + 3)/4)*4)
+ /* was: (sizeof(word_t) + 2*sizeof(half_t) + strlen(STR) + 1) */
+
+struct osf_dirent
+{
+ uint32_t d_ino; /* file number of entry */
+ uint16_t d_reclen; /* length of this record */
+ uint16_t d_namlen; /* length of string in d_name */
+ char d_name[256]; /* DUMMY NAME LENGTH */
+ /* the real maximum length is */
+ /* returned by pathconf() */
+ /* At this time, this MUST */
+ /* be 256 -- the kernel */
+ /* requires it */
+};
+
+
+int
+getdirentriesFunc(SyscallDesc *desc, int callnum, Process *process,
+ ExecContext *xc)
+{
+ int i, cnt, osf_cnt;
+ struct dirent *p;
+ int32_t fd = process->sim_fd(getArg(xc,0));
+ Addr osf_buf = getArg(xc,1);
+ char *buf;
+ int32_t osf_nbytes = getArg(xc,2);
+ Addr osf_pbase = getArg(xc,3);
+ Addr osf_base;
+ long base = 0;
+
+ /* number of entries in simulated memory */
+ if (!osf_nbytes)
+ warn("attempting to get 0 directory entries...");
+
+ /* allocate local memory, whatever fits */
+ buf = (char*)calloc(1, osf_nbytes);
+ if (!buf)
+ fatal("out of virtual memory");
+
+ /* get directory entries */
+ int64_t result = getdirentries((int)fd, buf, (size_t)osf_nbytes, &base);
+
+ /* check for an error condition */
+ if (result != (int64_t) -1) {
+
+ /* anything to copy back? */
+ if (result > 0)
+ {
+ /* copy all possible results to simulated space */
+ for (i=0, cnt=0, osf_cnt=0, p=(struct dirent *)buf;
+ cnt < result && p->d_reclen > 0;
+ i++, cnt += p->d_reclen, p=(struct dirent *)(buf+cnt))
+ {
+ struct osf_dirent osf_dirent;
+
+ osf_dirent.d_ino = p->d_ino;
+ osf_dirent.d_namlen = strlen(p->d_name);
+ strcpy(osf_dirent.d_name, p->d_name);
+ osf_dirent.d_reclen = OSF_DIRENT_SZ(p->d_name);
+
+ xc->mem->access(Write, osf_buf + osf_cnt,
+ &osf_dirent, OSF_DIRENT_SZ(p->d_name));
+
+ osf_cnt += OSF_DIRENT_SZ(p->d_name);
+ }
+
+ if (osf_pbase != 0)
+ {
+ osf_base = (Addr)base;
+
+ xc->mem->access(Write, osf_pbase, &osf_base, sizeof(osf_base));
+ }
+
+ /* update V0 to indicate translated read length */
+ result = osf_cnt;
+ }
+ }
+
+ free(buf);
+
+ return result;
+}
+
+
int
getrlimitFunc(SyscallDesc *desc, int callnum, Process *process,
ExecContext *xc)
@@ -704,9 +845,11 @@ sigreturnFunc(SyscallDesc *desc, int callnum, Process *process,
sc.copyIn(xc->mem);
- // restore state from sigcontext structure
- regs->pc = sc->sc_pc;
- regs->npc = regs->pc + sizeof(MachInst);
+ // Restore state from sigcontext structure.
+ // Note that we'll advance PC <- NPC before the end of the cycle,
+ // so we need to restore the desired PC into NPC.
+ // The current regs->pc will get clobbered.
+ regs->npc = sc->sc_pc;
for (int i = 0; i < 31; ++i) {
regs->intRegFile[i] = sc->sc_regs[i];
@@ -807,7 +950,7 @@ exitFunc(SyscallDesc *desc, int callnum, Process *process,
SyscallDesc syscallDescs[] = {
- /* 0 */ SyscallDesc("syscall (#0)", indirectSyscallFunc),
+ /* 0 */ SyscallDesc("syscall (#0)", indirectSyscallFunc, SyscallDesc::SuppressReturnValue),
/* 1 */ SyscallDesc("exit", exitFunc),
/* 2 */ SyscallDesc("fork", unimplementedFunc),
/* 3 */ SyscallDesc("read", readFunc),
@@ -830,7 +973,7 @@ SyscallDesc syscallDescs[] = {
/* 20 */ SyscallDesc("getpid", getpidFunc),
/* 21 */ SyscallDesc("mount", unimplementedFunc),
/* 22 */ SyscallDesc("unmount", unimplementedFunc),
- /* 23 */ SyscallDesc("setuid", unimplementedFunc),
+ /* 23 */ SyscallDesc("setuid", setuidFunc),
/* 24 */ SyscallDesc("getuid", getuidFunc),
/* 25 */ SyscallDesc("exec_with_loader", unimplementedFunc),
/* 26 */ SyscallDesc("ptrace", unimplementedFunc),
@@ -854,7 +997,7 @@ SyscallDesc syscallDescs[] = {
/* 44 */ SyscallDesc("profil", unimplementedFunc),
/* 45 */ SyscallDesc("open", openFunc),
/* 46 */ SyscallDesc("obsolete osigaction", unimplementedFunc),
- /* 47 */ SyscallDesc("getgid", unimplementedFunc),
+ /* 47 */ SyscallDesc("getgid", getgidFunc),
/* 48 */ SyscallDesc("sigprocmask", ignoreFunc),
/* 49 */ SyscallDesc("getlogin", unimplementedFunc),
/* 50 */ SyscallDesc("setlogin", unimplementedFunc),
@@ -899,7 +1042,7 @@ SyscallDesc syscallDescs[] = {
/* 89 */ SyscallDesc("getdtablesize", unimplementedFunc),
/* 90 */ SyscallDesc("dup2", unimplementedFunc),
/* 91 */ SyscallDesc("pre_F64_fstat", unimplementedFunc),
- /* 92 */ SyscallDesc("fcntl", unimplementedFunc),
+ /* 92 */ SyscallDesc("fcntl", fcntlFunc),
/* 93 */ SyscallDesc("select", unimplementedFunc),
/* 94 */ SyscallDesc("poll", unimplementedFunc),
/* 95 */ SyscallDesc("fsync", unimplementedFunc),
@@ -910,7 +1053,7 @@ SyscallDesc syscallDescs[] = {
/* 100 */ SyscallDesc("getpriority", unimplementedFunc),
/* 101 */ SyscallDesc("old_send", unimplementedFunc),
/* 102 */ SyscallDesc("old_recv", unimplementedFunc),
- /* 103 */ SyscallDesc("sigreturn", sigreturnFunc),
+ /* 103 */ SyscallDesc("sigreturn", sigreturnFunc, SyscallDesc::SuppressReturnValue),
/* 104 */ SyscallDesc("bind", unimplementedFunc),
/* 105 */ SyscallDesc("setsockopt", unimplementedFunc),
/* 106 */ SyscallDesc("listen", unimplementedFunc),
@@ -966,7 +1109,7 @@ SyscallDesc syscallDescs[] = {
/* 156 */ SyscallDesc("sigaction", ignoreFunc),
/* 157 */ SyscallDesc("sigwaitprim", unimplementedFunc),
/* 158 */ SyscallDesc("nfssvc", unimplementedFunc),
- /* 159 */ SyscallDesc("getdirentries", unimplementedFunc),
+ /* 159 */ SyscallDesc("getdirentries", getdirentriesFunc),
/* 160 */ SyscallDesc("pre_F64_statfs", unimplementedFunc),
/* 161 */ SyscallDesc("pre_F64_fstatfs", unimplementedFunc),
/* 162 */ SyscallDesc("unknown #162", unimplementedFunc),
@@ -1688,9 +1831,8 @@ const int Min_Syscall_Desc = -Max_Mach_Syscall_Desc;
// helper function for invoking syscalls
//
static
-int
-doSyscall(int callnum, Process *process,
- ExecContext *xc)
+void
+doSyscall(int callnum, Process *process, ExecContext *xc)
{
if (callnum < Min_Syscall_Desc || callnum > Max_Syscall_Desc) {
cerr << "Syscall " << callnum << " out of range" << endl;
@@ -1700,10 +1842,16 @@ doSyscall(int callnum, Process *process,
SyscallDesc *desc =
(callnum < 0) ? &machSyscallDescs[-callnum] : &syscallDescs[callnum];
- DCOUT(SyscallVerbose) << xc->cpu->name() << ": syscall " << desc->name
- << " called @ " << curTick << endl;
+ DPRINTFR(SyscallVerbose, "%s: syscall %s called\n",
+ xc->cpu->name(), desc->name);
+
+ int retval = desc->doFunc(callnum, process, xc);
- return desc->doFunc(callnum, process, xc);
+ DPRINTFR(SyscallVerbose, "%s: syscall %s returns %d\n",
+ xc->cpu->name(), desc->name, retval);
+
+ if (!((desc->flags & SyscallDesc::SuppressReturnValue) && retval == 0))
+ set_return_value(xc, retval);
}
//
@@ -1718,7 +1866,9 @@ indirectSyscallFunc(SyscallDesc *desc, int callnum, Process *process,
for (int i = 0; i < 5; ++i)
setArg(xc, i, getArg(xc, i+1));
- return doSyscall(new_callnum, process, xc);
+ doSyscall(new_callnum, process, xc);
+
+ return 0;
}
@@ -1727,7 +1877,5 @@ fake_syscall(Process *process, ExecContext *xc)
{
int64_t callnum = xc->regs.intRegFile[ReturnValueReg];
- int retval = doSyscall(callnum, process, xc);
-
- set_return_value(xc, retval);
+ doSyscall(callnum, process, xc);
}