Age | Commit message (Collapse) | Author |
|
When each load or store is sent to the LSQ, we check whether it will cross a
cache line boundary and, if so, split it in two. This creates two TLB
translations and two memory requests. Care has to be taken if the first
packet of a split load is sent but the second blocks the cache. Similarly,
for a store, if the first packet cannot be sent, we must store the second
one somewhere to retry later.
This modifies the LSQSenderState class to record both packets in a split
load or store.
Finally, a new const variable, HasUnalignedMemAcc, is added to each ISA
to indicate whether unaligned memory accesses are allowed. This is used
throughout the changed code so that compiler can optimise away code dealing
with split requests for ISAs that don't need them.
|
|
|
|
|
|
|
|
1) Move alpha-specific code out of page_table.cc:serialize().
2) Begin serializing M5_pid and unserializing it, but adding an function to do optional paramIn so that old checkpoints don't need to be fixed up.
3) Fix up alpha startup code so that the unserialized M5_pid value is properly written to DTB_IPR_ASN.
4) Fix the memory unserialize that I forgot somehow in the last changeset.
5) Add in an agg_se.py to handle aggregated checkpoints. --bench foo-bar plus positional arguments foo bar are the only changes in usage from se.py.
Note this aggregation stuff has only been tested for Alpha and nothing else, though it should take a very minimal amount of work to get it to work with another ISA.
|
|
|
|
|
|
|
|
|
|
In Linux, the set_thread_area system call stores the address of the thread
local storage area into a field of the current thread_info structure. Later,
to access that value, the program uses the rdhwr instruction to read a
"hardware register" with index 29. The 64 bit MIPS manual, volume II, says
that index 29 is reserved for a future ABI extension and should cause a
"Reserved Instruction Exception". In Linux (and potentially other ISAs) that
exception is trapped and emulated to return the value stored by
set_thread_area as if that were actually stored by a physical register.
The tp_value address (as named in the Linux kernel) is ironically stored as a
control register so that it goes with a particular ThreadContext. Syscall
emulation will use that to emulate storing to the OS's thread info structure,
and rdhwr will emulate faulting and returning that value from software by
returning the value itself, as if it was in hardware. In other words, we fake
faking the register in SE mode. In an FS mode implementation it should
work as specified in the manual.
|
|
The MIPS ISA object expects to be constructed with a CPU pointer it uses to
look at other thread contexts and allow them to be manipulated with control
registers. Unfortunately, that differs from all the other ISA classes and
would complicate their implementation.
This change makes the event constructor use a CPU pointer pulled out of the
thread context passed to setMiscReg instead.
|
|
These are all after the existing ones, suggesting they were added after the
original list was created.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
--HG--
rename : src/arch/sparc/interrupts.hh => src/arch/arm/interrupts.hh
rename : src/arch/sparc/kernel_stats.hh => src/arch/arm/kernel_stats.hh
rename : src/arch/sparc/stacktrace.cc => src/arch/arm/stacktrace.cc
rename : src/arch/sparc/system.cc => src/arch/arm/system.cc
rename : src/arch/sparc/system.hh => src/arch/arm/system.hh
rename : src/dev/sparc/T1000.py => src/dev/arm/Versatile.py
rename : src/dev/sparc/t1000.cc => src/dev/arm/versatile.cc
rename : src/dev/sparc/t1000.hh => src/dev/arm/versatile.hh
|
|
|
|
|
|
|
|
The last commit was somehow empty. This was what was supposed to go in it.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
These registers can be accessed directly, or through MISCREG_SPSR which will
act as whichever SPSR is appropriate for the current mode.
|
|
|
|
Some of the micro-ops weren't casting 1 to ULL before shifting,
which can cause problems. On the perl makerand input this
caused some values to be negative that shouldn't have been.
The casts are done as ULL(1) instead of 1ULL to match others
in the m5 code base.
|
|
|
|
|
|
|
|
Implement some fault classes using the curriously recurring template pattern,
similar to SPARCs.
|
|
The PC indexes in the various register sets was defined in the section for
unaliased registers which was throwing off the indexing. This moves those
where they belong. Also, to make detecting accesses to the PC easier and
because it's in the same place in all modes, the intRegForceUser function
now passes it through as index 15.
|
|
Unfortunately my implementation of the movd instruction had two bugs.
In one case, when moving a 32-bit value into an xmm register, the
lower half of the xmm register was not zero extended.
The other case is that xmm was used instead of xmmlm as the source
for a register move. My test case didn't notice this at first
as it moved xmm0 to eax, which both have the same register
number.
|
|
This double cast led to rounding errors which caused
some benchmarks to get the wrong values, most notably lucas
which failed spectacularly due to CVTTSD2SI returning an
off-by-one value. equake was also broken.
|
|
|
|
|
|
|
|
|
|
|
|
Specifically, get rid of the big switch statement so more cases can be
handled. Enumerating all the possible settings doesn't scale well. Also do
some minor style clean up.
|
|
In the isa_parser, we need to check case statements.
|
|
This allows those bits to be renamed while allowing the other fields to
control the behavior of the processor.
|
|
|
|
That constant is a carry over from Alpha and doesn't do anything in ARM.
|
|
|