tru64.hh

来自「M5,一个功能强大的多处理器系统模拟器.很多针对处理器架构,性能的研究都使用它作」· HH 代码 · 共 1,211 行 · 第 1/3 页

        TypedBufferArg<T> tgt(addr);

#if defined(__OpenBSD__) || defined(__APPLE__) || defined(__FreeBSD__)
        tgt->f_type = 0;
#else
        tgt->f_type = htog(host->f_type);
#endif
        tgt->f_bsize = htog(host->f_bsize);
        tgt->f_blocks = htog(host->f_blocks);
        tgt->f_bfree = htog(host->f_bfree);
        tgt->f_bavail = htog(host->f_bavail);
        tgt->f_files = htog(host->f_files);
        tgt->f_ffree = htog(host->f_ffree);

        // Is this as string normally?
        memcpy(&tgt->f_fsid, &host->f_fsid, sizeof(host->f_fsid));

        tgt.copyOut(mem);
    }


    /// The target system's hostname.
    static const char *hostname;


    /// Target getdirentries() handler.
    static SyscallReturn
    getdirentriesFunc(SyscallDesc *desc, int callnum, LiveProcess *process,
                      ThreadContext *tc)
    {
        using namespace TheISA;

#ifdef __CYGWIN__
        panic("getdirent not implemented on cygwin!");
#else
        int fd = process->sim_fd(tc->getSyscallArg(0));
        Addr tgt_buf = tc->getSyscallArg(1);
        int tgt_nbytes = tc->getSyscallArg(2);
        Addr tgt_basep = tc->getSyscallArg(3);

        char * const host_buf = new char[tgt_nbytes];

        // just pass basep through uninterpreted.
        TypedBufferArg<int64_t> basep(tgt_basep);
        basep.copyIn(tc->getMemPort());
        long host_basep = (off_t)htog((int64_t)*basep);
        int host_result = getdirentries(fd, host_buf, tgt_nbytes, &host_basep);

        // check for error
        if (host_result < 0) {
            delete [] host_buf;
            return -errno;
        }

        // no error: copy results back to target space
        Addr tgt_buf_ptr = tgt_buf;
        char *host_buf_ptr = host_buf;
        char *host_buf_end = host_buf + host_result;
        while (host_buf_ptr < host_buf_end) {
            global_dirent *host_dp = (global_dirent *)host_buf_ptr;
            int namelen = strlen(host_dp->d_name);

            // Actual size includes padded string rounded up for alignment.
            // Subtract 256 for dummy char array in Tru64::dirent definition.
            // Add 1 to namelen for terminating null char.
            int tgt_bufsize = sizeof(Tru64::dirent) - 256 + roundUp(namelen+1, 8);
            TypedBufferArg<Tru64::dirent> tgt_dp(tgt_buf_ptr, tgt_bufsize);
            tgt_dp->d_ino = host_dp->d_ino;
            tgt_dp->d_reclen = tgt_bufsize;
            tgt_dp->d_namlen = namelen;
            strcpy(tgt_dp->d_name, host_dp->d_name);
            tgt_dp.copyOut(tc->getMemPort());

            tgt_buf_ptr += tgt_bufsize;
            host_buf_ptr += host_dp->d_reclen;
        }

        delete [] host_buf;

        *basep = htog((int64_t)host_basep);
        basep.copyOut(tc->getMemPort());

        return tgt_buf_ptr - tgt_buf;
#endif
    }

    /// Target sigreturn() handler.
    static SyscallReturn
    sigreturnFunc(SyscallDesc *desc, int callnum, LiveProcess *process,
                  ThreadContext *tc)
    {
        using namespace TheISA;

        using TheISA::RegFile;
        TypedBufferArg<Tru64::sigcontext> sc(tc->getSyscallArg(0));

        sc.copyIn(tc->getMemPort());

        // 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.
        tc->setNextPC(htog(sc->sc_pc));

        for (int i = 0; i < 31; ++i) {
            tc->setIntReg(i, htog(sc->sc_regs[i]));
            tc->setFloatRegBits(i, htog(sc->sc_fpregs[i]));
        }

        tc->setMiscRegNoEffect(AlphaISA::MISCREG_FPCR, htog(sc->sc_fpcr));

        return 0;
    }


    //
    // Mach syscalls -- identified by negated syscall numbers
    //

    /// Create a stack region for a thread.
    static SyscallReturn
    stack_createFunc(SyscallDesc *desc, int callnum, LiveProcess *process,
                     ThreadContext *tc)
    {
        using namespace TheISA;

        TypedBufferArg<Tru64::vm_stack> argp(tc->getSyscallArg(0));

        argp.copyIn(tc->getMemPort());

        int stack_size =
            gtoh(argp->rsize) + gtoh(argp->ysize) + gtoh(argp->gsize);

        // if the user chose an address, just let them have it.  Otherwise
        // pick one for them.
        Addr stack_base = gtoh(argp->address);

        if (stack_base == 0) {
            stack_base = process->next_thread_stack_base;
            process->next_thread_stack_base -= stack_size;
        }

        Addr rounded_stack_base = roundDown(stack_base, VMPageSize);
        Addr rounded_stack_size = roundUp(stack_size, VMPageSize);

        DPRINTF(SyscallVerbose,
                "stack_create: allocating stack @ %#x size %#x "
                "(rounded from %#x, %#x)\n",
                rounded_stack_base, rounded_stack_size,
                stack_base, stack_size);

        // map memory
        process->pTable->allocate(rounded_stack_base, rounded_stack_size);

        argp->address = gtoh(rounded_stack_base);
        argp.copyOut(tc->getMemPort());

        return 0;
    }

    /// NXM library version stamp.
    static
    const int NXM_LIB_VERSION = 301003;

    /// This call sets up the interface between the user and kernel
    /// schedulers by creating a shared-memory region.  The shared memory
    /// region has several structs, some global, some per-RAD, some per-VP.
    static SyscallReturn
    nxm_task_initFunc(SyscallDesc *desc, int callnum, LiveProcess *process,
                      ThreadContext *tc)
    {
        using namespace std;
        using namespace TheISA;

        TypedBufferArg<Tru64::nxm_task_attr> attrp(tc->getSyscallArg(0));
        TypedBufferArg<Addr> configptr_ptr(tc->getSyscallArg(1));

        attrp.copyIn(tc->getMemPort());

        if (gtoh(attrp->nxm_version) != NXM_LIB_VERSION) {
            cerr << "nxm_task_init: thread library version mismatch! "
                 << "got " << attrp->nxm_version
                 << ", expected " << NXM_LIB_VERSION << endl;
            abort();
        }

        if (gtoh(attrp->flags) != Tru64::NXM_TASK_INIT_VP) {
            cerr << "nxm_task_init: bad flag value " << attrp->flags
                 << " (expected " << Tru64::NXM_TASK_INIT_VP << ")" << endl;
            abort();
        }

        Addr base_addr = 0x12000; // was 0x3f0000000LL;
        Addr cur_addr = base_addr; // next addresses to use
        // first comes the config_info struct
        Addr config_addr = cur_addr;
        cur_addr += sizeof(Tru64::nxm_config_info);
        // next comes the per-cpu state vector
        Addr slot_state_addr = cur_addr;
        int slot_state_size =
            process->numCpus() * sizeof(Tru64::nxm_slot_state_t);
        cur_addr += slot_state_size;
        // now the per-RAD state struct (we only support one RAD)
        cur_addr = 0x14000;	// bump up addr for alignment
        Addr rad_state_addr = cur_addr;
        int rad_state_size =
            (sizeof(Tru64::nxm_shared)
             + (process->numCpus()-1) * sizeof(Tru64::nxm_sched_state));
        cur_addr += rad_state_size;

        // now initialize a config_info struct and copy it out to user space
        TypedBufferArg<Tru64::nxm_config_info> config(config_addr);

        config->nxm_nslots_per_rad = htog(process->numCpus());
        config->nxm_nrads = htog(1);	// only one RAD in our system!
        config->nxm_slot_state = htog(slot_state_addr);
        config->nxm_rad[0] = htog(rad_state_addr);

        // initialize the slot_state array and copy it out
        TypedBufferArg<Tru64::nxm_slot_state_t> slot_state(slot_state_addr,
                                                           slot_state_size);
        for (int i = 0; i < process->numCpus(); ++i) {
            // CPU 0 is bound to the calling process; all others are available
            // XXX this code should have an endian conversion, but I don't think
            // it works anyway
            slot_state[i] =
                (i == 0) ? Tru64::NXM_SLOT_BOUND : Tru64::NXM_SLOT_AVAIL;
        }

        // same for the per-RAD "shared" struct.  Note that we need to
        // allocate extra bytes for the per-VP array which is embedded at
        // the end.
        TypedBufferArg<Tru64::nxm_shared> rad_state(rad_state_addr,
                                                    rad_state_size);

        rad_state->nxm_callback = attrp->nxm_callback;
        rad_state->nxm_version = attrp->nxm_version;
        rad_state->nxm_uniq_offset = attrp->nxm_uniq_offset;
        for (int i = 0; i < process->numCpus(); ++i) {
            Tru64::nxm_sched_state *ssp = &rad_state->nxm_ss[i];
            ssp->nxm_u.sigmask = htog(0);
            ssp->nxm_u.sig = htog(0);
            ssp->nxm_u.flags = htog(0);
            ssp->nxm_u.cancel_state = htog(0);
            ssp->nxm_u.nxm_ssig = 0;
            ssp->nxm_bits = htog(0);
            ssp->nxm_quantum = attrp->nxm_quantum;
            ssp->nxm_set_quantum = attrp->nxm_quantum;
            ssp->nxm_sysevent = htog(0);

            if (i == 0) {
                uint64_t uniq = tc->readMiscRegNoEffect(AlphaISA::MISCREG_UNIQ);
                ssp->nxm_u.pth_id = htog(uniq + gtoh(attrp->nxm_uniq_offset));
                ssp->nxm_u.nxm_active = htog(uniq | 1);
            }
            else {
                ssp->nxm_u.pth_id = htog(0);
                ssp->nxm_u.nxm_active = htog(0);
            }
        }

        //
        // copy pointer to shared config area out to user
        //
        *configptr_ptr = htog(config_addr);

        // Register this as a valid address range with the process
        base_addr = roundDown(base_addr, VMPageSize);
        int size = cur_addr - base_addr;
        process->pTable->allocate(base_addr, roundUp(size, VMPageSize));

        config.copyOut(tc->getMemPort());
        slot_state.copyOut(tc->getMemPort());
        rad_state.copyOut(tc->getMemPort());
        configptr_ptr.copyOut(tc->getMemPort());

        return 0;
    }

    /// Initialize thread context.
    static void
    init_thread_context(ThreadContext *tc,
                      Tru64::nxm_thread_attr *attrp, uint64_t uniq_val)
    {
        using namespace TheISA;

        tc->clearArchRegs();

        tc->setIntReg(TheISA::ArgumentReg[0], gtoh(attrp->registers.a0));
        tc->setIntReg(27/*t12*/, gtoh(attrp->registers.pc));
        tc->setIntReg(TheISA::StackPointerReg, gtoh(attrp->registers.sp));
        tc->setMiscRegNoEffect(AlphaISA::MISCREG_UNIQ, uniq_val);

        tc->setPC(gtoh(attrp->registers.pc));
        tc->setNextPC(gtoh(attrp->registers.pc) + sizeof(TheISA::MachInst));

        tc->activate();
    }

    /// Create thread.
    static SyscallReturn
    nxm_thread_createFunc(SyscallDesc *desc, int callnum, LiveProcess *process,
                          ThreadContext *tc)
    {
        using namespace std;
        using namespace TheISA;

        TypedBufferArg<Tru64::nxm_thread_attr> attrp(tc->getSyscallArg(0));
        TypedBufferArg<uint64_t> kidp(tc->getSyscallArg(1));
        int thread_index = tc->getSyscallArg(2);

        // get attribute args
        attrp.copyIn(tc->getMemPort());

        if (gtoh(attrp->version) != NXM_LIB_VERSION) {
            cerr << "nxm_thread_create: thread library version mismatch! "
                 << "got " << attrp->version
                 << ", expected " << NXM_LIB_VERSION << endl;
            abort();
        }

        if (thread_index < 0 | thread_index > process->numCpus()) {
            cerr << "nxm_thread_create: bad thread index " << thread_index
                 << endl;
            abort();
        }

        // On a real machine, the per-RAD shared structure is in
        // shared memory, so both the user and kernel can get at it.
        // We don't have that luxury, so we just copy it in and then
        // back out again.
        int rad_state_size =
            (sizeof(Tru64::nxm_shared) +
             (process->numCpus()-1) * sizeof(Tru64::nxm_sched_state));

        TypedBufferArg<Tru64::nxm_shared> rad_state(0x14000,
                                                    rad_state_size);
        rad_state.copyIn(tc->getMemPort());

        uint64_t uniq_val = gtoh(attrp->pthid) - gtoh(rad_state->nxm_uniq_offset);

        if (gtoh(attrp->type) == Tru64::NXM_TYPE_MANAGER) {
            // DEC pthreads seems to always create one of these (in
            // addition to N application threads), but we don't use it,
            // so don't bother creating it.

            // This is supposed to be a port number.  Make something up.
            *kidp = htog(99);
            kidp.copyOut(tc->getMemPort());

            return 0;
        } else if (gtoh(attrp->type) == Tru64::NXM_TYPE_VP) {
            // A real "virtual processor" kernel thread.  Need to fork
            // this thread on another CPU.
            Tru64::nxm_sched_state *ssp = &rad_state->nxm_ss[thread_index];

            if (gtoh(ssp->nxm_u.nxm_active) != 0)
                return (int) Tru64::KERN_NOT_RECEIVER;

            ssp->nxm_u.pth_id = attrp->pthid;
            ssp->nxm_u.nxm_active = htog(uniq_val | 1);

            rad_state.copyOut(tc->getMemPort());

            Addr slot_state_addr = 0x12000 + sizeof(Tru64::nxm_config_info);
            int slot_state_size =
                process->numCpus() * sizeof(Tru64::nxm_slot_state_t);

            TypedBufferArg<Tru64::nxm_slot_state_t>
                slot_state(slot_state_addr,
                           slot_state_size);

            slot_state.copyIn(tc->getMemPort());

            if (slot_state[thread_index] != Tru64::NXM_SLOT_AVAIL) {
                cerr << "nxm_thread_createFunc: requested VP slot "
                     << thread_index << " not available!" << endl;
                fatal("");
            }

            // XXX This should have an endian conversion but I think this code
            // doesn't work anyway
            slot_state[thread_index] = Tru64::NXM_SLOT_BOUND;

            slot_state.copyOut(tc->getMemPort());

            // Find a free simulator thread context.
            for (int i = 0; i < process->numCpus(); ++i) {
                ThreadContext *tc = process->threadContexts[i];

                if (tc->status() == ThreadContext::Unallocated) {
                    // inactive context... grab it
                    init_thread_context(tc, attrp, uniq_val);

                    // This is supposed to be a port number, but we'll try
                    // and get away with just sticking the thread index
                    // here.
                    *kidp = htog(thread_index);
                    kidp.copyOut(tc->getMemPort());

                    return 0;
                }
            }