syscall_emul: [patch 13/22] add system call retry capability

This changeset adds functionality that allows system calls to retry without
affecting thread context state such as the program counter or register values
for the associated thread context (when system calls return with a retry
fault).

This functionality is needed to solve problems with blocking system calls
in multi-process or multi-threaded simulations where information is passed
between processes/threads. Blocking system calls can cause deadlock because
the simulator itself is single threaded. There is only a single thread
servicing the event queue which can cause deadlock if the thread hits a
blocking system call instruction.

To illustrate the problem, consider two processes using the producer/consumer
sharing model. The processes can use file descriptors and the read and write
calls to pass information to one another. If the consumer calls the blocking
read system call before the producer has produced anything, the call will
block the event queue (while executing the system call instruction) and
deadlock the simulation.

The solution implemented in this changeset is to recognize that the system
calls will block and then generate a special retry fault. The fault will
be sent back up through the function call chain until it is exposed to the
cpu model's pipeline where the fault becomes visible. The fault will trigger
the cpu model to replay the instruction at a future tick where the call has
a chance to succeed without actually going into a blocking state.

In subsequent patches, we recognize that a syscall will block by calling a
non-blocking poll (from inside the system call implementation) and checking
for events. When events show up during the poll, it signifies that the call
would not have blocked and the syscall is allowed to proceed (calling an
underlying host system call if necessary). If no events are returned from the
poll, we generate the fault and try the instruction for the thread context
at a distant tick. Note that retrying every tick is not efficient.

As an aside, the simulator has some multi-threading support for the event
queue, but it is not used by default and needs work. Even if the event queue
was completely multi-threaded, meaning that there is a hardware thread on
the host servicing a single simulator thread contexts with a 1:1 mapping
between them, it's still possible to run into deadlock due to the event queue
barriers on quantum boundaries. The solution of replaying at a later tick
is the simplest solution and solves the problem generally.

src/arch/alpha/isa/decoder.isa

@@ -840,7 +840,7 @@ decode OPCODE default Unknown::unknown() {
                     exitSimLoop("halt instruction encountered");
                 }}, IsNonSpeculative);
                 0x83: callsys({{
-                    xc->syscall(R0);
+                    xc->syscall(R0, &fault);
                 }}, IsSerializeAfter, IsNonSpeculative, IsSyscall);
                 // Read uniq reg into ABI return value register (r0)
                 0x9e: rduniq({{ R0 = Runiq; }}, IsIprAccess);

src/arch/arm/faults.cc

@@ -784,7 +784,8 @@ SupervisorCall::invoke(ThreadContext *tc, const StaticInstPtr &inst)
         callNum = tc->readIntReg(INTREG_X8);
     else
         callNum = tc->readIntReg(INTREG_R7);
-    tc->syscall(callNum);
+    Fault fault;
+    tc->syscall(callNum, &fault);
 
     // Advance the PC since that won't happen automatically.
     PCState pc = tc->pcState();

src/arch/mips/isa/decoder.isa

@@ -164,7 +164,7 @@ decode OPCODE_HI default Unknown::unknown() {
                     0x2: movz({{ Rd = (Rt == 0) ? Rs : Rd; }});
                     0x3: movn({{ Rd = (Rt != 0) ? Rs : Rd; }});
                     0x4: decode FullSystemInt {
-                        0: syscall_se({{ xc->syscall(R2); }},
+                        0: syscall_se({{ xc->syscall(R2, &fault); }},
                                 IsSerializeAfter, IsNonSpeculative);
                       default: syscall({{ fault = std::make_shared<SystemCallFault>(); }});
                     }

src/arch/power/isa/decoder.isa

@@ -512,7 +512,7 @@ decode OPCODE default Unknown::unknown() {
         55: stfdu({{ Mem_df = Fs; }});
     }
 
-    17: IntOp::sc({{ xc->syscall(R0); }},
+    17: IntOp::sc({{ xc->syscall(R0, &fault); }},
                   [ IsSyscall, IsNonSpeculative, IsSerializeAfter ]);
 
     format FloatArithOp {

src/arch/riscv/faults.cc

@@ -87,5 +87,6 @@ BreakpointFault::invoke_se(ThreadContext *tc, const StaticInstPtr &inst)
 void
 SyscallFault::invoke_se(ThreadContext *tc, const StaticInstPtr &inst)
 {
-    tc->syscall(tc->readIntReg(SyscallNumReg));
+    Fault *fault = NoFault;
+    tc->syscall(tc->readIntReg(SyscallNumReg), fault);
 }

src/arch/sparc/faults.cc

@@ -811,7 +811,8 @@ TrapInstruction::invoke(ThreadContext *tc, const StaticInstPtr &inst)
     SparcProcess *sp = dynamic_cast<SparcProcess *>(p);
     assert(sp);
 
-    sp->handleTrap(_n, tc);
+    Fault fault;
+    sp->handleTrap(_n, tc, &fault);
 
     // We need to explicitly advance the pc, since that's not done for us
     // on a faulting instruction

src/arch/sparc/linux/process.cc

@@ -65,14 +65,15 @@ Sparc32LinuxProcess::Sparc32LinuxProcess(ProcessParams * params,
     : Sparc32Process(params, objFile)
 {}
 
-void Sparc32LinuxProcess::handleTrap(int trapNum, ThreadContext *tc)
+void Sparc32LinuxProcess::handleTrap(int trapNum, ThreadContext *tc,
+                                     Fault *fault)
 {
     switch (trapNum) {
       case 0x10: //Linux 32 bit syscall trap
-        tc->syscall(tc->readIntReg(1));
+        tc->syscall(tc->readIntReg(1), fault);
         break;
       default:
-        SparcProcess::handleTrap(trapNum, tc);
+        SparcProcess::handleTrap(trapNum, tc, fault);
     }
 }
 
@@ -81,14 +82,15 @@ Sparc64LinuxProcess::Sparc64LinuxProcess(ProcessParams * params,
     : Sparc64Process(params, objFile)
 {}
 
-void Sparc64LinuxProcess::handleTrap(int trapNum, ThreadContext *tc)
+void Sparc64LinuxProcess::handleTrap(int trapNum, ThreadContext *tc,
+                                     Fault *fault)
 {
     switch (trapNum) {
       // case 0x10: // Linux 32 bit syscall trap
       case 0x6d: // Linux 64 bit syscall trap
-        tc->syscall(tc->readIntReg(1));
+        tc->syscall(tc->readIntReg(1), fault);
         break;
       default:
-        SparcProcess::handleTrap(trapNum, tc);
+        SparcProcess::handleTrap(trapNum, tc, fault);
     }
 }

src/arch/sparc/linux/process.hh

@@ -70,7 +70,7 @@ class Sparc32LinuxProcess : public SparcLinuxProcess, public Sparc32Process
         return SparcLinuxProcess::getDesc32(callnum);
     }
 
-    void handleTrap(int trapNum, ThreadContext *tc);
+    void handleTrap(int trapNum, ThreadContext *tc, Fault *fault);
 };
 
 /// A process with emulated 32 bit SPARC/Linux syscalls.
@@ -86,7 +86,7 @@ class Sparc64LinuxProcess : public SparcLinuxProcess, public Sparc64Process
         return SparcLinuxProcess::getDesc(callnum);
     }
 
-    void handleTrap(int trapNum, ThreadContext *tc);
+    void handleTrap(int trapNum, ThreadContext *tc, Fault *fault);
 };
 
 SyscallReturn getresuidFunc(SyscallDesc *desc, int num,

src/arch/sparc/process.cc

@@ -71,7 +71,7 @@ SparcProcess::SparcProcess(ProcessParams * params, ObjectFile *objFile,
 }
 
 void
-SparcProcess::handleTrap(int trapNum, ThreadContext *tc)
+SparcProcess::handleTrap(int trapNum, ThreadContext *tc, Fault *fault)
 {
     PCState pc = tc->pcState();
     switch (trapNum) {

src/arch/sparc/process.hh

@@ -61,7 +61,7 @@ class SparcProcess : public Process
   public:
 
     // Handles traps which request services from the operating system
-    virtual void handleTrap(int trapNum, ThreadContext *tc);
+    virtual void handleTrap(int trapNum, ThreadContext *tc, Fault *fault);
 
     Addr readFillStart() { return fillStart; }
     Addr readSpillStart() { return spillStart; }

src/arch/x86/isa/decoder/one_byte_opcodes.isa

@@ -400,8 +400,9 @@
                         // will sign extend it, and there's no easy way to
                         // specify only checking the first byte.
                         0xffffffffffffff80:
-                            SyscallInst::int80('xc->syscall(Rax)',
-                                 IsSyscall, IsNonSpeculative, IsSerializeAfter);
+                            SyscallInst::int80('xc->syscall(Rax, &fault)',
+                                               IsSyscall, IsNonSpeculative,
+                                               IsSerializeAfter);
                     }
 
                     default: Inst::INT(Ib);

src/arch/x86/isa/decoder/two_byte_opcodes.isa

@@ -235,8 +235,9 @@
                 }
             }
             0x05: decode FullSystemInt {
-                0: SyscallInst::syscall('xc->syscall(Rax)',
-                        IsSyscall, IsNonSpeculative, IsSerializeAfter);
+                0: SyscallInst::syscall('xc->syscall(Rax, &fault)',
+                                        IsSyscall, IsNonSpeculative,
+                                        IsSerializeAfter);
                 default: decode MODE_MODE {
                     0x0: decode MODE_SUBMODE {
                         0x0: Inst::SYSCALL_64();
@@ -422,8 +423,9 @@
             0x2: Inst::RDMSR();
             0x3: rdpmc();
             0x4: decode FullSystemInt {
-                0: SyscallInst::sysenter('xc->syscall(Rax)',
-                        IsSyscall, IsNonSpeculative, IsSerializeAfter);
+                0: SyscallInst::sysenter('xc->syscall(Rax, &fault)',
+                                         IsSyscall, IsNonSpeculative,
+                                         IsSerializeAfter);
                 default: sysenter();
             }
             0x5: sysexit();

src/arch/x86/process.cc

@@ -134,7 +134,7 @@ X86_64Process::X86_64Process(ProcessParams *params, ObjectFile *objFile,
 }
 
 void
-I386Process::syscall(int64_t callnum, ThreadContext *tc)
+I386Process::syscall(int64_t callnum, ThreadContext *tc, Fault *fault)
 {
     TheISA::PCState pc = tc->pcState();
     Addr eip = pc.pc();
@@ -143,7 +143,7 @@ I386Process::syscall(int64_t callnum, ThreadContext *tc)
         pc.npc(vsyscallPage.base + vsyscallPage.vsysexitOffset);
         tc->pcState(pc);
     }
-    X86Process::syscall(callnum, tc);
+    X86Process::syscall(callnum, tc, fault);
 }
 
 

src/arch/x86/process.hh

@@ -130,7 +130,7 @@ namespace X86ISA
         void argsInit(int intSize, int pageSize);
         void initState();
 
-        void syscall(int64_t callnum, ThreadContext *tc);
+        void syscall(int64_t callnum, ThreadContext *tc, Fault *fault);
         X86ISA::IntReg getSyscallArg(ThreadContext *tc, int &i);
         X86ISA::IntReg getSyscallArg(ThreadContext *tc, int &i, int width);
         void setSyscallArg(ThreadContext *tc, int i, X86ISA::IntReg val);

src/arch/x86/pseudo_inst.cc

@@ -49,7 +49,9 @@ m5Syscall(ThreadContext *tc)
 {
     DPRINTF(PseudoInst, "PseudoInst::m5Syscall()\n");
 
-    tc->syscall(tc->readIntReg(INTREG_RAX));
+    Fault fault;
+    tc->syscall(tc->readIntReg(INTREG_RAX), &fault);
+
     MiscReg rflags = tc->readMiscReg(MISCREG_RFLAGS);
     rflags &= ~(1 << 16);
     tc->setMiscReg(MISCREG_RFLAGS, rflags);