diff options
author | Kevin Lim <ktlim@umich.edu> | 2006-03-04 15:18:40 -0500 |
---|---|---|
committer | Kevin Lim <ktlim@umich.edu> | 2006-03-04 15:18:40 -0500 |
commit | f15e492375e8ecd42a1f0ba7ead68cfeb2b4b673 (patch) | |
tree | 42a35f201054c6edff9e024fd8dc5d9e5d30dba3 /cpu/pc_event.cc | |
parent | 96fd6b5c4039c98a1b536ec184126ad75e7d2539 (diff) | |
download | gem5-f15e492375e8ecd42a1f0ba7ead68cfeb2b4b673.tar.xz |
Steps towards setting up the infrastructure to allow the new CPU model to work in full system mode.
The major change is renaming the old ExecContext to CPUExecContext, and creating two new classes, ExecContext (an abstract class), and ProxyExecContext (a templated class that derives from ExecContext).
Code outside of the CPU continues to use ExecContext as normal (other than not being able to access variables within the XC). The CPU uses the CPUExecContext, or however else it stores its own state. It then creates a ProxyExecContext, templated on the class used to hold its state. This proxy is passed to any code outside of the CPU that needs to access the XC. This allows code outside of the CPU to use the ExecContext interface to access any state needed, without knowledge of how that state is laid out.
Note that these changes will not compile without the accompanying revision to automatically rename the shadow registers.
SConscript:
Include new file, cpu_exec_context.cc.
arch/alpha/alpha_linux_process.cc:
arch/alpha/alpha_memory.cc:
arch/alpha/alpha_tru64_process.cc:
arch/alpha/arguments.cc:
arch/alpha/isa/decoder.isa:
arch/alpha/stacktrace.cc:
arch/alpha/vtophys.cc:
base/remote_gdb.cc:
cpu/intr_control.cc:
Avoid directly accessing objects within the XC.
arch/alpha/ev5.cc:
Avoid directly accessing objects within the XC.
KernelStats have been moved to the BaseCPU instead of the XC.
arch/alpha/isa_traits.hh:
Remove clearIprs(). It wasn't used very often and it did not work well with the proxy ExecContext.
cpu/base.cc:
Place kernel stats within the BaseCPU instead of the ExecContext.
For now comment out the profiling code sampling until its exact location is decided upon.
cpu/base.hh:
Kernel stats are now in the BaseCPU instead of the ExecContext.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
cpu/memtest/memtest.cc:
cpu/memtest/memtest.hh:
Changes to support rename of old ExecContext to CPUExecContext. See changeset for more details.
cpu/exetrace.cc:
Remove unneeded include of exec_context.hh.
cpu/intr_control.hh:
cpu/o3/alpha_cpu_builder.cc:
Remove unneeded include of exec_context.hh
cpu/o3/alpha_cpu.hh:
cpu/o3/alpha_cpu_impl.hh:
cpu/o3/cpu.cc:
cpu/o3/cpu.hh:
cpu/simple/cpu.cc:
cpu/simple/cpu.hh:
Changes to support rename of old ExecContext to CPUExecContext. See changeset for more details.
Also avoid accessing anything directly from the XC.
cpu/pc_event.cc:
Avoid accessing objects directly from the XC.
dev/tsunami_cchip.cc:
Avoid accessing objects directly within the XC>
kern/freebsd/freebsd_system.cc:
kern/linux/linux_system.cc:
kern/linux/linux_threadinfo.hh:
kern/tru64/dump_mbuf.cc:
kern/tru64/tru64.hh:
kern/tru64/tru64_events.cc:
sim/syscall_emul.cc:
sim/syscall_emul.hh:
Avoid accessing objects directly within the XC.
kern/kernel_stats.cc:
kern/kernel_stats.hh:
Kernel stats no longer exist within the XC.
kern/system_events.cc:
Avoid accessing objects directly within the XC. Also kernel stats are now in the BaseCPU.
sim/process.cc:
sim/process.hh:
Avoid accessing regs directly within an ExecContext. Instead use a CPUExecContext to initialize the registers and copy them over.
cpu/cpu_exec_context.cc:
Rename old ExecContext to CPUExecContext. This is used by the old CPU models to store any necessary architectural state. Also include the ProxyExecContext, which is used to access the CPUExecContext's state in code outside of the CPU.
cpu/cpu_exec_context.hh:
Rename old ExecContext to CPUExecContext. This is used by the old CPU models to store any necessary architectural state. Also include the ProxyExecContext, which is used to access the CPUExecContext's state in code outside of the CPU.
Remove kernel stats from the ExecContext.
sim/pseudo_inst.cc:
Kernel stats now live within the CPU.
Avoid accessing objects directly within the XC.
--HG--
rename : cpu/exec_context.cc => cpu/cpu_exec_context.cc
rename : cpu/exec_context.hh => cpu/cpu_exec_context.hh
extra : convert_revision : a75393a8945c80cca225b5e9d9c22a16609efb85
Diffstat (limited to 'cpu/pc_event.cc')
-rw-r--r-- | cpu/pc_event.cc | 7 |
1 files changed, 4 insertions, 3 deletions
diff --git a/cpu/pc_event.cc b/cpu/pc_event.cc index 83fbc3e2d..050bf1a88 100644 --- a/cpu/pc_event.cc +++ b/cpu/pc_event.cc @@ -38,6 +38,7 @@ #include "cpu/pc_event.hh" #include "sim/debug.hh" #include "sim/root.hh" +#include "sim/system.hh" using namespace std; @@ -79,7 +80,7 @@ PCEventQueue::schedule(PCEvent *event) bool PCEventQueue::doService(ExecContext *xc) { - Addr pc = xc->regs.pc & ~0x3; + Addr pc = xc->readPC() & ~0x3; int serviced = 0; range_t range = equal_range(pc); for (iterator i = range.first; i != range.second; ++i) { @@ -87,7 +88,7 @@ PCEventQueue::doService(ExecContext *xc) // another event. This for example, prevents two invocations // of the SkipFuncEvent. Maybe we should have separate PC // event queues for each processor? - if (pc != (xc->regs.pc & ~0x3)) + if (pc != (xc->readPC() & ~0x3)) continue; DPRINTF(PCEvent, "PC based event serviced at %#x: %s\n", @@ -126,7 +127,7 @@ BreakPCEvent::BreakPCEvent(PCEventQueue *q, const std::string &desc, Addr addr, void BreakPCEvent::process(ExecContext *xc) { - StringWrap name(xc->cpu->name() + ".break_event"); + StringWrap name(xc->getCpuPtr()->name() + ".break_event"); DPRINTFN("break event %s triggered\n", descr()); debug_break(); if (remove) |