Add a MemTxAttrs field to the IOTLB, and allow target-specific
code to set it via a new tlb_set_page_with_attrs() function;
pass the attributes through to the device when making IO accesses.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Make the CPU iotlb a structure rather than a plain hwaddr;
this will allow us to add transaction attributes to it.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
QEMU system mode page table walks are expensive. Taken by running QEMU
qemu-system-x86_64 system mode on Intel PIN , a TLB miss and walking a
4-level page tables in guest Linux OS takes ~450 X86 instructions on
average.
QEMU system mode TLB is implemented using a directly-mapped hashtable.
This structure suffers from conflict misses. Increasing the
associativity of the TLB may not be the solution to conflict misses as
all the ways may have to be walked in serial.
A victim TLB is a TLB used to hold translations evicted from the
primary TLB upon replacement. The victim TLB lies between the main TLB
and its refill path. Victim TLB is of greater associativity (fully
associative in this patch). It takes longer to lookup the victim TLB,
but its likely better than a full page table walk. The memory
translation path is changed as follows :
Before Victim TLB:
1. Inline TLB lookup
2. Exit code cache on TLB miss.
3. Check for unaligned, IO accesses
4. TLB refill.
5. Do the memory access.
6. Return to code cache.
After Victim TLB:
1. Inline TLB lookup
2. Exit code cache on TLB miss.
3. Check for unaligned, IO accesses
4. Victim TLB lookup.
5. If victim TLB misses, TLB refill
6. Do the memory access.
7. Return to code cache
The advantage is that victim TLB can offer more associativity to a
directly mapped TLB and thus potentially fewer page table walks while
still keeping the time taken to flush within reasonable limits.
However, placing a victim TLB before the refill path increase TLB
refill path as the victim TLB is consulted before the TLB refill. The
performance results demonstrate that the pros outweigh the cons.
some performance results taken on SPECINT2006 train
datasets and kernel boot and qemu configure script on an
Intel(R) Xeon(R) CPU E5620 @ 2.40GHz Linux machine are shown in the
Google Doc link below.
https://docs.google.com/spreadsheets/d/1eiItzekZwNQOal_h-5iJmC4tMDi051m9qidi5_nwvH4/edit?usp=sharing
In summary, victim TLB improves the performance of qemu-system-x86_64 by
11% on average on SPECINT2006, kernelboot and qemu configscript and with
highest improvement of in 26% in 456.hmmer. And victim TLB does not result
in any performance degradation in any of the measured benchmarks. Furthermore,
the implemented victim TLB is architecture independent and is expected to
benefit other architectures in QEMU as well.
Although there are measurement fluctuations, the performance
improvement is very significant and by no means in the range of
noises.
Signed-off-by: Xin Tong <trent.tong@gmail.com>
Message-id: 1407202523-23553-1-git-send-email-trent.tong@gmail.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Most targets were using offsetof(CPUFooState, breakpoints) to determine
how much of CPUFooState to clear on reset. Use the next field after
CPU_COMMON instead, if any, or sizeof(CPUFooState) otherwise.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Implement WFE to yield our timeslice to the next CPU.
This avoids slowdowns in multicore configurations caused
by one core busy-waiting on a spinlock which can't possibly
be unlocked until the other core has an opportunity to run.
This speeds up my test case A15 dual-core boot by a factor
of three (though it is still four or five times slower than
a single-core boot).
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1393339545-22111-1-git-send-email-peter.maydell@linaro.org
Reviewed-by: Richard Henderson <rth@twiddle.net>
Tested-by: Rob Herring <rob.herring@linaro.org>
Since this is only read in cpu_copy() and linux-user has a global
cpu_model, drop the field from generic code.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Prepares for changing cpu_single_step() argument to CPUState.
Acked-by: Michael Walle <michael@walle.cc> (for lm32)
Signed-off-by: Andreas Färber <afaerber@suse.de>
All targets have been converted to allocating space for temporaries
on the stack. No need to allocate space within the CPU_COMMON block.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Move next_cpu from CPU_COMMON to CPUState.
Move first_cpu variable to qom/cpu.h.
gdbstub needs to use CPUState::env_ptr for now.
cpu_copy() no longer needs to save and restore cpu_next.
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
[AF: Rebased, simplified cpu_copy()]
Signed-off-by: Andreas Färber <afaerber@suse.de>
While not normally needed for *-user, it can safely be used there since
always based on uint64_t, to avoid ifdeffery.
To avoid accidental uses, move the guards from exec/hwaddr.h to its
inclusion sites. No need for them in include/hw/.
Prepares for hwaddr use in qom/cpu.h.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Rather than a hand-coded version of the same thing.
Reviewed-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: liguang <lig.fnst@cn.fujitsu.com>
Signed-off-by: Richard Henderson <rth@twiddle.net>
The alignment is a characteristic of the ABI, not the CPU.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@gmail.com>
Previously, this was done for target_long/ulong, and propagated to
abi_long/ulong via a typedef. But target_long/ulong should not
have any specific alignment, it is never used to access guest
memory.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@gmail.com>
The alignment is a characteristic of the ABI, not the CPU.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@gmail.com>
The alignment is a characteristic of the ABI, not the CPU.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@gmail.com>
Both fields are used in VMState, thus need to be moved together.
Explicitly zero them on reset since they were located before
breakpoints.
Pass PowerPCCPU to kvmppc_handle_halt().
Signed-off-by: Andreas Färber <afaerber@suse.de>
The setjmp() function doesn't specify whether signal masks are saved and
restored; on Linux they are not, but on BSD (including MacOSX) they are.
We want to have consistent behaviour across platforms, so we should
always use "don't save/restore signal mask" (this is also generally
going to be faster). This also works around a bug in MacOSX where the
signal-restoration on longjmp() affects the signal mask for a completely
different thread, not just the mask for the thread which did the longjmp.
The most visible effect of this was that ctrl-C was ignored on MacOSX
because the CPU thread did a longjmp which resulted in its signal mask
being applied to every thread, so that all threads had SIGINT and SIGTERM
blocked.
The POSIX-sanctioned portable way to do a jump without affecting signal
masks is to siglongjmp() to a sigjmp_buf which was created by calling
sigsetjmp() with a zero savemask parameter, so change all uses of
setjmp()/longjmp() accordingly. [Technically POSIX allows sigsetjmp(buf, 0)
to save the signal mask; however the following siglongjmp() must not
restore the signal mask, so the pair can be effectively considered as
"sigjmp/longjmp which don't touch the mask".]
For Windows we provide a trivial sigsetjmp/siglongjmp in terms of
setjmp/longjmp -- this is OK because no user will ever pass a non-zero
savemask.
The setjmp() uses in tests/tcg/test-i386.c and tests/tcg/linux-test.c
are left untouched because these are self-contained singlethreaded
test programs intended to be run under QEMU's Linux emulation, so they
have neither the portability nor the multithreading issues to deal with.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Tested-by: Stefan Weil <sw@weilnetz.de>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Explictly NULL it on CPU reset since it was located before breakpoints.
Change vapic_report_tpr_access() argument to CPUState. This also
resolves the use of void* for cpu.h independence.
Change vAPIC patch_instruction() argument to X86CPU.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Commit c64ca8140e (cpu: Move
queued_work_{first,last} to CPUState) moved the qemu_work_item fields
away. Clean up the now unused prototype.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Note that target-alpha accesses this field from TCG, now using a
negative offset. Therefore the field is placed last in CPUState.
Pass PowerPCCPU to [kvm]ppc_fixup_cpu() to facilitate this change.
Move common parts of mips cpu_state_reset() to mips_cpu_reset().
Acked-by: Richard Henderson <rth@twiddle.net> (for alpha)
[AF: Rebased onto ppc CPU subclasses and openpic changes]
Signed-off-by: Andreas Färber <afaerber@suse.de>
To facilitate the field movements, pass MIPSCPU to malta_mips_config();
avoid that for mips_cpu_map_tc() since callers only access MIPS Thread
Contexts, inside TCG helpers.
Signed-off-by: Andreas Färber <afaerber@suse.de>
