qemu-e2k/accel/tcg/cpu-exec.c

791 lines
24 KiB
C
Raw Normal View History

/*
* emulator main execution loop
*
* Copyright (c) 2003-2005 Fabrice Bellard
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "qemu/osdep.h"
#include "qemu-common.h"
#include "qemu/qemu-print.h"
#include "cpu.h"
#include "trace.h"
#include "disas/disas.h"
#include "exec/exec-all.h"
#include "tcg/tcg.h"
#include "qemu/atomic.h"
#include "sysemu/qtest.h"
cpu-exec: Add sleeping algorithm The goal is to sleep qemu whenever the guest clock is in advance compared to the host clock (we use the monotonic clocks). The amount of time to sleep is calculated in the execution loop in cpu_exec. At first, we tried to approximate at each for loop the real time elapsed while searching for a TB (generating or retrieving from cache) and executing it. We would then approximate the virtual time corresponding to the number of virtual instructions executed. The difference between these 2 values would allow us to know if the guest is in advance or delayed. However, the function used for measuring the real time (qemu_clock_get_ns(QEMU_CLOCK_REALTIME)) proved to be very expensive. We had an added overhead of 13% of the total run time. Therefore, we modified the algorithm and only take into account the difference between the 2 clocks at the begining of the cpu_exec function. During the for loop we try to reduce the advance of the guest only by computing the virtual time elapsed and sleeping if necessary. The overhead is thus reduced to 3%. Even though this method still has a noticeable overhead, it no longer is a bottleneck in trying to achieve a better guest frequency for which the guest clock is faster than the host one. As for the the alignement of the 2 clocks, with the first algorithm the guest clock was oscillating between -1 and 1ms compared to the host clock. Using the second algorithm we notice that the guest is 5ms behind the host, which is still acceptable for our use case. The tests where conducted using fio and stress. The host machine in an i5 CPU at 3.10GHz running Debian Jessie (kernel 3.12). The guest machine is an arm versatile-pb built with buildroot. Currently, on our test machine, the lowest icount we can achieve that is suitable for aligning the 2 clocks is 6. However, we observe that the IO tests (using fio) are slower than the cpu tests (using stress). Signed-off-by: Sebastian Tanase <sebastian.tanase@openwide.fr> Tested-by: Camille Bégué <camille.begue@openwide.fr> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-07-25 11:56:31 +02:00
#include "qemu/timer.h"
#include "qemu/rcu.h"
#include "exec/tb-hash.h"
tcg: consolidate TB lookups in tb_lookup__cpu_state This avoids duplicating code. cpu_exec_step will also use the new common function once we integrate parallel_cpus into tb->cflags. Note that in this commit we also fix a race, described by Richard Henderson during review. Think of this scenario with threads A and B: (A) Lookup succeeds for TB in hash without tb_lock (B) Sets the TB's tb->invalid flag (B) Removes the TB from tb_htable (B) Clears all CPU's tb_jmp_cache (A) Store TB into local tb_jmp_cache Given that order of events, (A) will keep executing that invalid TB until another flush of its tb_jmp_cache happens, which in theory might never happen. We can fix this by checking the tb->invalid flag every time we look up a TB from tb_jmp_cache, so that in the above scenario, next time we try to find that TB in tb_jmp_cache, we won't, and will therefore be forced to look it up in tb_htable. Performance-wise, I measured a small improvement when booting debian-arm. Note that inlining pays off: Performance counter stats for 'taskset -c 0 qemu-system-arm \ -machine type=virt -nographic -smp 1 -m 4096 \ -netdev user,id=unet,hostfwd=tcp::2222-:22 \ -device virtio-net-device,netdev=unet \ -drive file=jessie.qcow2,id=myblock,index=0,if=none \ -device virtio-blk-device,drive=myblock \ -kernel kernel.img -append console=ttyAMA0 root=/dev/vda1 \ -name arm,debug-threads=on -smp 1' (10 runs): Before: 18714.917392 task-clock # 0.952 CPUs utilized ( +- 0.95% ) 23,142 context-switches # 0.001 M/sec ( +- 0.50% ) 1 CPU-migrations # 0.000 M/sec 10,558 page-faults # 0.001 M/sec ( +- 0.95% ) 53,957,727,252 cycles # 2.883 GHz ( +- 0.91% ) [83.33%] 24,440,599,852 stalled-cycles-frontend # 45.30% frontend cycles idle ( +- 1.20% ) [83.33%] 16,495,714,424 stalled-cycles-backend # 30.57% backend cycles idle ( +- 0.95% ) [66.66%] 76,267,572,582 instructions # 1.41 insns per cycle # 0.32 stalled cycles per insn ( +- 0.87% ) [83.34%] 12,692,186,323 branches # 678.186 M/sec ( +- 0.92% ) [83.35%] 263,486,879 branch-misses # 2.08% of all branches ( +- 0.73% ) [83.34%] 19.648474449 seconds time elapsed ( +- 0.82% ) After, w/ inline (this patch): 18471.376627 task-clock # 0.955 CPUs utilized ( +- 0.96% ) 23,048 context-switches # 0.001 M/sec ( +- 0.48% ) 1 CPU-migrations # 0.000 M/sec 10,708 page-faults # 0.001 M/sec ( +- 0.81% ) 53,208,990,796 cycles # 2.881 GHz ( +- 0.98% ) [83.34%] 23,941,071,673 stalled-cycles-frontend # 44.99% frontend cycles idle ( +- 0.95% ) [83.34%] 16,161,773,848 stalled-cycles-backend # 30.37% backend cycles idle ( +- 0.76% ) [66.67%] 75,786,269,766 instructions # 1.42 insns per cycle # 0.32 stalled cycles per insn ( +- 1.24% ) [83.34%] 12,573,617,143 branches # 680.708 M/sec ( +- 1.34% ) [83.33%] 260,235,550 branch-misses # 2.07% of all branches ( +- 0.66% ) [83.33%] 19.340502161 seconds time elapsed ( +- 0.56% ) After, w/o inline: 18791.253967 task-clock # 0.954 CPUs utilized ( +- 0.78% ) 23,230 context-switches # 0.001 M/sec ( +- 0.42% ) 1 CPU-migrations # 0.000 M/sec 10,563 page-faults # 0.001 M/sec ( +- 1.27% ) 54,168,674,622 cycles # 2.883 GHz ( +- 0.80% ) [83.34%] 24,244,712,629 stalled-cycles-frontend # 44.76% frontend cycles idle ( +- 1.37% ) [83.33%] 16,288,648,572 stalled-cycles-backend # 30.07% backend cycles idle ( +- 0.95% ) [66.66%] 77,659,755,503 instructions # 1.43 insns per cycle # 0.31 stalled cycles per insn ( +- 0.97% ) [83.34%] 12,922,780,045 branches # 687.702 M/sec ( +- 1.06% ) [83.34%] 261,962,386 branch-misses # 2.03% of all branches ( +- 0.71% ) [83.35%] 19.700174670 seconds time elapsed ( +- 0.56% ) Reviewed-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-07-11 23:33:33 +02:00
#include "exec/tb-lookup.h"
#include "exec/log.h"
tcg: drop global lock during TCG code execution This finally allows TCG to benefit from the iothread introduction: Drop the global mutex while running pure TCG CPU code. Reacquire the lock when entering MMIO or PIO emulation, or when leaving the TCG loop. We have to revert a few optimization for the current TCG threading model, namely kicking the TCG thread in qemu_mutex_lock_iothread and not kicking it in qemu_cpu_kick. We also need to disable RAM block reordering until we have a more efficient locking mechanism at hand. Still, a Linux x86 UP guest and my Musicpal ARM model boot fine here. These numbers demonstrate where we gain something: 20338 jan 20 0 331m 75m 6904 R 99 0.9 0:50.95 qemu-system-arm 20337 jan 20 0 331m 75m 6904 S 20 0.9 0:26.50 qemu-system-arm The guest CPU was fully loaded, but the iothread could still run mostly independent on a second core. Without the patch we don't get beyond 32206 jan 20 0 330m 73m 7036 R 82 0.9 1:06.00 qemu-system-arm 32204 jan 20 0 330m 73m 7036 S 21 0.9 0:17.03 qemu-system-arm We don't benefit significantly, though, when the guest is not fully loading a host CPU. Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> Message-Id: <1439220437-23957-10-git-send-email-fred.konrad@greensocs.com> [FK: Rebase, fix qemu_devices_reset deadlock, rm address_space_* mutex] Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com> [EGC: fixed iothread lock for cpu-exec IRQ handling] Signed-off-by: Emilio G. Cota <cota@braap.org> [AJB: -smp single-threaded fix, clean commit msg, BQL fixes] Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Pranith Kumar <bobby.prani@gmail.com> [PM: target-arm changes] Acked-by: Peter Maydell <peter.maydell@linaro.org>
2017-02-23 19:29:11 +01:00
#include "qemu/main-loop.h"
#if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
#include "hw/i386/apic.h"
#endif
#include "sysemu/cpus.h"
#include "exec/cpu-all.h"
#include "sysemu/cpu-timers.h"
#include "sysemu/replay.h"
cpu-exec: Add sleeping algorithm The goal is to sleep qemu whenever the guest clock is in advance compared to the host clock (we use the monotonic clocks). The amount of time to sleep is calculated in the execution loop in cpu_exec. At first, we tried to approximate at each for loop the real time elapsed while searching for a TB (generating or retrieving from cache) and executing it. We would then approximate the virtual time corresponding to the number of virtual instructions executed. The difference between these 2 values would allow us to know if the guest is in advance or delayed. However, the function used for measuring the real time (qemu_clock_get_ns(QEMU_CLOCK_REALTIME)) proved to be very expensive. We had an added overhead of 13% of the total run time. Therefore, we modified the algorithm and only take into account the difference between the 2 clocks at the begining of the cpu_exec function. During the for loop we try to reduce the advance of the guest only by computing the virtual time elapsed and sleeping if necessary. The overhead is thus reduced to 3%. Even though this method still has a noticeable overhead, it no longer is a bottleneck in trying to achieve a better guest frequency for which the guest clock is faster than the host one. As for the the alignement of the 2 clocks, with the first algorithm the guest clock was oscillating between -1 and 1ms compared to the host clock. Using the second algorithm we notice that the guest is 5ms behind the host, which is still acceptable for our use case. The tests where conducted using fio and stress. The host machine in an i5 CPU at 3.10GHz running Debian Jessie (kernel 3.12). The guest machine is an arm versatile-pb built with buildroot. Currently, on our test machine, the lowest icount we can achieve that is suitable for aligning the 2 clocks is 6. However, we observe that the IO tests (using fio) are slower than the cpu tests (using stress). Signed-off-by: Sebastian Tanase <sebastian.tanase@openwide.fr> Tested-by: Camille Bégué <camille.begue@openwide.fr> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-07-25 11:56:31 +02:00
/* -icount align implementation. */
typedef struct SyncClocks {
int64_t diff_clk;
int64_t last_cpu_icount;
int64_t realtime_clock;
cpu-exec: Add sleeping algorithm The goal is to sleep qemu whenever the guest clock is in advance compared to the host clock (we use the monotonic clocks). The amount of time to sleep is calculated in the execution loop in cpu_exec. At first, we tried to approximate at each for loop the real time elapsed while searching for a TB (generating or retrieving from cache) and executing it. We would then approximate the virtual time corresponding to the number of virtual instructions executed. The difference between these 2 values would allow us to know if the guest is in advance or delayed. However, the function used for measuring the real time (qemu_clock_get_ns(QEMU_CLOCK_REALTIME)) proved to be very expensive. We had an added overhead of 13% of the total run time. Therefore, we modified the algorithm and only take into account the difference between the 2 clocks at the begining of the cpu_exec function. During the for loop we try to reduce the advance of the guest only by computing the virtual time elapsed and sleeping if necessary. The overhead is thus reduced to 3%. Even though this method still has a noticeable overhead, it no longer is a bottleneck in trying to achieve a better guest frequency for which the guest clock is faster than the host one. As for the the alignement of the 2 clocks, with the first algorithm the guest clock was oscillating between -1 and 1ms compared to the host clock. Using the second algorithm we notice that the guest is 5ms behind the host, which is still acceptable for our use case. The tests where conducted using fio and stress. The host machine in an i5 CPU at 3.10GHz running Debian Jessie (kernel 3.12). The guest machine is an arm versatile-pb built with buildroot. Currently, on our test machine, the lowest icount we can achieve that is suitable for aligning the 2 clocks is 6. However, we observe that the IO tests (using fio) are slower than the cpu tests (using stress). Signed-off-by: Sebastian Tanase <sebastian.tanase@openwide.fr> Tested-by: Camille Bégué <camille.begue@openwide.fr> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-07-25 11:56:31 +02:00
} SyncClocks;
#if !defined(CONFIG_USER_ONLY)
/* Allow the guest to have a max 3ms advance.
* The difference between the 2 clocks could therefore
* oscillate around 0.
*/
#define VM_CLOCK_ADVANCE 3000000
#define THRESHOLD_REDUCE 1.5
#define MAX_DELAY_PRINT_RATE 2000000000LL
#define MAX_NB_PRINTS 100
cpu-exec: Add sleeping algorithm The goal is to sleep qemu whenever the guest clock is in advance compared to the host clock (we use the monotonic clocks). The amount of time to sleep is calculated in the execution loop in cpu_exec. At first, we tried to approximate at each for loop the real time elapsed while searching for a TB (generating or retrieving from cache) and executing it. We would then approximate the virtual time corresponding to the number of virtual instructions executed. The difference between these 2 values would allow us to know if the guest is in advance or delayed. However, the function used for measuring the real time (qemu_clock_get_ns(QEMU_CLOCK_REALTIME)) proved to be very expensive. We had an added overhead of 13% of the total run time. Therefore, we modified the algorithm and only take into account the difference between the 2 clocks at the begining of the cpu_exec function. During the for loop we try to reduce the advance of the guest only by computing the virtual time elapsed and sleeping if necessary. The overhead is thus reduced to 3%. Even though this method still has a noticeable overhead, it no longer is a bottleneck in trying to achieve a better guest frequency for which the guest clock is faster than the host one. As for the the alignement of the 2 clocks, with the first algorithm the guest clock was oscillating between -1 and 1ms compared to the host clock. Using the second algorithm we notice that the guest is 5ms behind the host, which is still acceptable for our use case. The tests where conducted using fio and stress. The host machine in an i5 CPU at 3.10GHz running Debian Jessie (kernel 3.12). The guest machine is an arm versatile-pb built with buildroot. Currently, on our test machine, the lowest icount we can achieve that is suitable for aligning the 2 clocks is 6. However, we observe that the IO tests (using fio) are slower than the cpu tests (using stress). Signed-off-by: Sebastian Tanase <sebastian.tanase@openwide.fr> Tested-by: Camille Bégué <camille.begue@openwide.fr> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-07-25 11:56:31 +02:00
static int64_t max_delay;
static int64_t max_advance;
static void align_clocks(SyncClocks *sc, CPUState *cpu)
cpu-exec: Add sleeping algorithm The goal is to sleep qemu whenever the guest clock is in advance compared to the host clock (we use the monotonic clocks). The amount of time to sleep is calculated in the execution loop in cpu_exec. At first, we tried to approximate at each for loop the real time elapsed while searching for a TB (generating or retrieving from cache) and executing it. We would then approximate the virtual time corresponding to the number of virtual instructions executed. The difference between these 2 values would allow us to know if the guest is in advance or delayed. However, the function used for measuring the real time (qemu_clock_get_ns(QEMU_CLOCK_REALTIME)) proved to be very expensive. We had an added overhead of 13% of the total run time. Therefore, we modified the algorithm and only take into account the difference between the 2 clocks at the begining of the cpu_exec function. During the for loop we try to reduce the advance of the guest only by computing the virtual time elapsed and sleeping if necessary. The overhead is thus reduced to 3%. Even though this method still has a noticeable overhead, it no longer is a bottleneck in trying to achieve a better guest frequency for which the guest clock is faster than the host one. As for the the alignement of the 2 clocks, with the first algorithm the guest clock was oscillating between -1 and 1ms compared to the host clock. Using the second algorithm we notice that the guest is 5ms behind the host, which is still acceptable for our use case. The tests where conducted using fio and stress. The host machine in an i5 CPU at 3.10GHz running Debian Jessie (kernel 3.12). The guest machine is an arm versatile-pb built with buildroot. Currently, on our test machine, the lowest icount we can achieve that is suitable for aligning the 2 clocks is 6. However, we observe that the IO tests (using fio) are slower than the cpu tests (using stress). Signed-off-by: Sebastian Tanase <sebastian.tanase@openwide.fr> Tested-by: Camille Bégué <camille.begue@openwide.fr> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-07-25 11:56:31 +02:00
{
int64_t cpu_icount;
if (!icount_align_option) {
return;
}
cpu_icount = cpu->icount_extra + cpu_neg(cpu)->icount_decr.u16.low;
cpu-exec: Add sleeping algorithm The goal is to sleep qemu whenever the guest clock is in advance compared to the host clock (we use the monotonic clocks). The amount of time to sleep is calculated in the execution loop in cpu_exec. At first, we tried to approximate at each for loop the real time elapsed while searching for a TB (generating or retrieving from cache) and executing it. We would then approximate the virtual time corresponding to the number of virtual instructions executed. The difference between these 2 values would allow us to know if the guest is in advance or delayed. However, the function used for measuring the real time (qemu_clock_get_ns(QEMU_CLOCK_REALTIME)) proved to be very expensive. We had an added overhead of 13% of the total run time. Therefore, we modified the algorithm and only take into account the difference between the 2 clocks at the begining of the cpu_exec function. During the for loop we try to reduce the advance of the guest only by computing the virtual time elapsed and sleeping if necessary. The overhead is thus reduced to 3%. Even though this method still has a noticeable overhead, it no longer is a bottleneck in trying to achieve a better guest frequency for which the guest clock is faster than the host one. As for the the alignement of the 2 clocks, with the first algorithm the guest clock was oscillating between -1 and 1ms compared to the host clock. Using the second algorithm we notice that the guest is 5ms behind the host, which is still acceptable for our use case. The tests where conducted using fio and stress. The host machine in an i5 CPU at 3.10GHz running Debian Jessie (kernel 3.12). The guest machine is an arm versatile-pb built with buildroot. Currently, on our test machine, the lowest icount we can achieve that is suitable for aligning the 2 clocks is 6. However, we observe that the IO tests (using fio) are slower than the cpu tests (using stress). Signed-off-by: Sebastian Tanase <sebastian.tanase@openwide.fr> Tested-by: Camille Bégué <camille.begue@openwide.fr> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-07-25 11:56:31 +02:00
sc->diff_clk += cpu_icount_to_ns(sc->last_cpu_icount - cpu_icount);
sc->last_cpu_icount = cpu_icount;
if (sc->diff_clk > VM_CLOCK_ADVANCE) {
#ifndef _WIN32
struct timespec sleep_delay, rem_delay;
sleep_delay.tv_sec = sc->diff_clk / 1000000000LL;
sleep_delay.tv_nsec = sc->diff_clk % 1000000000LL;
if (nanosleep(&sleep_delay, &rem_delay) < 0) {
sc->diff_clk = rem_delay.tv_sec * 1000000000LL + rem_delay.tv_nsec;
cpu-exec: Add sleeping algorithm The goal is to sleep qemu whenever the guest clock is in advance compared to the host clock (we use the monotonic clocks). The amount of time to sleep is calculated in the execution loop in cpu_exec. At first, we tried to approximate at each for loop the real time elapsed while searching for a TB (generating or retrieving from cache) and executing it. We would then approximate the virtual time corresponding to the number of virtual instructions executed. The difference between these 2 values would allow us to know if the guest is in advance or delayed. However, the function used for measuring the real time (qemu_clock_get_ns(QEMU_CLOCK_REALTIME)) proved to be very expensive. We had an added overhead of 13% of the total run time. Therefore, we modified the algorithm and only take into account the difference between the 2 clocks at the begining of the cpu_exec function. During the for loop we try to reduce the advance of the guest only by computing the virtual time elapsed and sleeping if necessary. The overhead is thus reduced to 3%. Even though this method still has a noticeable overhead, it no longer is a bottleneck in trying to achieve a better guest frequency for which the guest clock is faster than the host one. As for the the alignement of the 2 clocks, with the first algorithm the guest clock was oscillating between -1 and 1ms compared to the host clock. Using the second algorithm we notice that the guest is 5ms behind the host, which is still acceptable for our use case. The tests where conducted using fio and stress. The host machine in an i5 CPU at 3.10GHz running Debian Jessie (kernel 3.12). The guest machine is an arm versatile-pb built with buildroot. Currently, on our test machine, the lowest icount we can achieve that is suitable for aligning the 2 clocks is 6. However, we observe that the IO tests (using fio) are slower than the cpu tests (using stress). Signed-off-by: Sebastian Tanase <sebastian.tanase@openwide.fr> Tested-by: Camille Bégué <camille.begue@openwide.fr> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-07-25 11:56:31 +02:00
} else {
sc->diff_clk = 0;
}
#else
Sleep(sc->diff_clk / SCALE_MS);
sc->diff_clk = 0;
#endif
}
}
static void print_delay(const SyncClocks *sc)
{
static float threshold_delay;
static int64_t last_realtime_clock;
static int nb_prints;
if (icount_align_option &&
sc->realtime_clock - last_realtime_clock >= MAX_DELAY_PRINT_RATE &&
nb_prints < MAX_NB_PRINTS) {
if ((-sc->diff_clk / (float)1000000000LL > threshold_delay) ||
(-sc->diff_clk / (float)1000000000LL <
(threshold_delay - THRESHOLD_REDUCE))) {
threshold_delay = (-sc->diff_clk / 1000000000LL) + 1;
qemu_printf("Warning: The guest is now late by %.1f to %.1f seconds\n",
threshold_delay - 1,
threshold_delay);
nb_prints++;
last_realtime_clock = sc->realtime_clock;
}
}
}
static void init_delay_params(SyncClocks *sc, CPUState *cpu)
cpu-exec: Add sleeping algorithm The goal is to sleep qemu whenever the guest clock is in advance compared to the host clock (we use the monotonic clocks). The amount of time to sleep is calculated in the execution loop in cpu_exec. At first, we tried to approximate at each for loop the real time elapsed while searching for a TB (generating or retrieving from cache) and executing it. We would then approximate the virtual time corresponding to the number of virtual instructions executed. The difference between these 2 values would allow us to know if the guest is in advance or delayed. However, the function used for measuring the real time (qemu_clock_get_ns(QEMU_CLOCK_REALTIME)) proved to be very expensive. We had an added overhead of 13% of the total run time. Therefore, we modified the algorithm and only take into account the difference between the 2 clocks at the begining of the cpu_exec function. During the for loop we try to reduce the advance of the guest only by computing the virtual time elapsed and sleeping if necessary. The overhead is thus reduced to 3%. Even though this method still has a noticeable overhead, it no longer is a bottleneck in trying to achieve a better guest frequency for which the guest clock is faster than the host one. As for the the alignement of the 2 clocks, with the first algorithm the guest clock was oscillating between -1 and 1ms compared to the host clock. Using the second algorithm we notice that the guest is 5ms behind the host, which is still acceptable for our use case. The tests where conducted using fio and stress. The host machine in an i5 CPU at 3.10GHz running Debian Jessie (kernel 3.12). The guest machine is an arm versatile-pb built with buildroot. Currently, on our test machine, the lowest icount we can achieve that is suitable for aligning the 2 clocks is 6. However, we observe that the IO tests (using fio) are slower than the cpu tests (using stress). Signed-off-by: Sebastian Tanase <sebastian.tanase@openwide.fr> Tested-by: Camille Bégué <camille.begue@openwide.fr> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-07-25 11:56:31 +02:00
{
if (!icount_align_option) {
return;
}
sc->realtime_clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT);
sc->diff_clk = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) - sc->realtime_clock;
sc->last_cpu_icount
= cpu->icount_extra + cpu_neg(cpu)->icount_decr.u16.low;
if (sc->diff_clk < max_delay) {
max_delay = sc->diff_clk;
}
if (sc->diff_clk > max_advance) {
max_advance = sc->diff_clk;
}
/* Print every 2s max if the guest is late. We limit the number
of printed messages to NB_PRINT_MAX(currently 100) */
print_delay(sc);
cpu-exec: Add sleeping algorithm The goal is to sleep qemu whenever the guest clock is in advance compared to the host clock (we use the monotonic clocks). The amount of time to sleep is calculated in the execution loop in cpu_exec. At first, we tried to approximate at each for loop the real time elapsed while searching for a TB (generating or retrieving from cache) and executing it. We would then approximate the virtual time corresponding to the number of virtual instructions executed. The difference between these 2 values would allow us to know if the guest is in advance or delayed. However, the function used for measuring the real time (qemu_clock_get_ns(QEMU_CLOCK_REALTIME)) proved to be very expensive. We had an added overhead of 13% of the total run time. Therefore, we modified the algorithm and only take into account the difference between the 2 clocks at the begining of the cpu_exec function. During the for loop we try to reduce the advance of the guest only by computing the virtual time elapsed and sleeping if necessary. The overhead is thus reduced to 3%. Even though this method still has a noticeable overhead, it no longer is a bottleneck in trying to achieve a better guest frequency for which the guest clock is faster than the host one. As for the the alignement of the 2 clocks, with the first algorithm the guest clock was oscillating between -1 and 1ms compared to the host clock. Using the second algorithm we notice that the guest is 5ms behind the host, which is still acceptable for our use case. The tests where conducted using fio and stress. The host machine in an i5 CPU at 3.10GHz running Debian Jessie (kernel 3.12). The guest machine is an arm versatile-pb built with buildroot. Currently, on our test machine, the lowest icount we can achieve that is suitable for aligning the 2 clocks is 6. However, we observe that the IO tests (using fio) are slower than the cpu tests (using stress). Signed-off-by: Sebastian Tanase <sebastian.tanase@openwide.fr> Tested-by: Camille Bégué <camille.begue@openwide.fr> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-07-25 11:56:31 +02:00
}
#else
static void align_clocks(SyncClocks *sc, const CPUState *cpu)
{
}
static void init_delay_params(SyncClocks *sc, const CPUState *cpu)
{
}
#endif /* CONFIG USER ONLY */
/* Execute a TB, and fix up the CPU state afterwards if necessary */
static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, TranslationBlock *itb)
{
CPUArchState *env = cpu->env_ptr;
uintptr_t ret;
TranslationBlock *last_tb;
int tb_exit;
uint8_t *tb_ptr = itb->tc.ptr;
qemu_log_mask_and_addr(CPU_LOG_EXEC, itb->pc,
"Trace %d: %p ["
TARGET_FMT_lx "/" TARGET_FMT_lx "/%#x] %s\n",
cpu->cpu_index, itb->tc.ptr,
itb->cs_base, itb->pc, itb->flags,
lookup_symbol(itb->pc));
#if defined(DEBUG_DISAS)
if (qemu_loglevel_mask(CPU_LOG_TB_CPU)
&& qemu_log_in_addr_range(itb->pc)) {
FILE *logfile = qemu_log_lock();
int flags = 0;
if (qemu_loglevel_mask(CPU_LOG_TB_FPU)) {
flags |= CPU_DUMP_FPU;
}
#if defined(TARGET_I386)
flags |= CPU_DUMP_CCOP;
#endif
log_cpu_state(cpu, flags);
qemu_log_unlock(logfile);
}
#endif /* DEBUG_DISAS */
ret = tcg_qemu_tb_exec(env, tb_ptr);
cpu->can_do_io = 1;
last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
tb_exit = ret & TB_EXIT_MASK;
trace_exec_tb_exit(last_tb, tb_exit);
if (tb_exit > TB_EXIT_IDX1) {
/* We didn't start executing this TB (eg because the instruction
* counter hit zero); we must restore the guest PC to the address
* of the start of the TB.
*/
CPUClass *cc = CPU_GET_CLASS(cpu);
qemu_log_mask_and_addr(CPU_LOG_EXEC, last_tb->pc,
"Stopped execution of TB chain before %p ["
TARGET_FMT_lx "] %s\n",
last_tb->tc.ptr, last_tb->pc,
lookup_symbol(last_tb->pc));
if (cc->synchronize_from_tb) {
cc->synchronize_from_tb(cpu, last_tb);
} else {
assert(cc->set_pc);
cc->set_pc(cpu, last_tb->pc);
}
}
return ret;
}
#ifndef CONFIG_USER_ONLY
/* Execute the code without caching the generated code. An interpreter
could be used if available. */
static void cpu_exec_nocache(CPUState *cpu, int max_cycles,
TranslationBlock *orig_tb, bool ignore_icount)
{
TranslationBlock *tb;
uint32_t cflags = curr_cflags() | CF_NOCACHE;
if (ignore_icount) {
cflags &= ~CF_USE_ICOUNT;
}
/* Should never happen.
We only end up here when an existing TB is too long. */
cflags |= MIN(max_cycles, CF_COUNT_MASK);
tcg: remove tb_lock Use mmap_lock in user-mode to protect TCG state and the page descriptors. In !user-mode, each vCPU has its own TCG state, so no locks needed. Per-page locks are used to protect the page descriptors. Per-TB locks are used in both modes to protect TB jumps. Some notes: - tb_lock is removed from notdirty_mem_write by passing a locked page_collection to tb_invalidate_phys_page_fast. - tcg_tb_lookup/remove/insert/etc have their own internal lock(s), so there is no need to further serialize access to them. - do_tb_flush is run in a safe async context, meaning no other vCPU threads are running. Therefore acquiring mmap_lock there is just to please tools such as thread sanitizer. - Not visible in the diff, but tb_invalidate_phys_page already has an assert_memory_lock. - cpu_io_recompile is !user-only, so no mmap_lock there. - Added mmap_unlock()'s before all siglongjmp's that could be called in user-mode while mmap_lock is held. + Added an assert for !have_mmap_lock() after returning from the longjmp in cpu_exec, just like we do in cpu_exec_step_atomic. Performance numbers before/after: Host: AMD Opteron(tm) Processor 6376 ubuntu 17.04 ppc64 bootup+shutdown time 700 +-+--+----+------+------------+-----------+------------*--+-+ | + + + + + *B | | before ***B*** ** * | |tb lock removal ###D### *** | 600 +-+ *** +-+ | ** # | | *B* #D | | *** * ## | 500 +-+ *** ### +-+ | * *** ### | | *B* # ## | | ** * #D# | 400 +-+ ** ## +-+ | ** ### | | ** ## | | ** # ## | 300 +-+ * B* #D# +-+ | B *** ### | | * ** #### | | * *** ### | 200 +-+ B *B #D# +-+ | #B* * ## # | | #* ## | | + D##D# + + + + | 100 +-+--+----+------+------------+-----------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/HwmBHXe debian jessie aarch64 bootup+shutdown time 90 +-+--+-----+-----+------------+------------+------------+--+-+ | + + + + + + | | before ***B*** B | 80 +tb lock removal ###D### **D +-+ | **### | | **## | 70 +-+ ** # +-+ | ** ## | | ** # | 60 +-+ *B ## +-+ | ** ## | | *** #D | 50 +-+ *** ## +-+ | * ** ### | | **B* ### | 40 +-+ **** # ## +-+ | **** #D# | | ***B** ### | 30 +-+ B***B** #### +-+ | B * * # ### | | B ###D# | 20 +-+ D ##D## +-+ | D# | | + + + + + + | 10 +-+--+-----+-----+------------+------------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/iGpGFtv The gains are high for 4-8 CPUs. Beyond that point, however, unrelated lock contention significantly hurts scalability. Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-05 05:46:31 +02:00
mmap_lock();
tb = tb_gen_code(cpu, orig_tb->pc, orig_tb->cs_base,
orig_tb->flags, cflags);
tb->orig_tb = orig_tb;
tcg: remove tb_lock Use mmap_lock in user-mode to protect TCG state and the page descriptors. In !user-mode, each vCPU has its own TCG state, so no locks needed. Per-page locks are used to protect the page descriptors. Per-TB locks are used in both modes to protect TB jumps. Some notes: - tb_lock is removed from notdirty_mem_write by passing a locked page_collection to tb_invalidate_phys_page_fast. - tcg_tb_lookup/remove/insert/etc have their own internal lock(s), so there is no need to further serialize access to them. - do_tb_flush is run in a safe async context, meaning no other vCPU threads are running. Therefore acquiring mmap_lock there is just to please tools such as thread sanitizer. - Not visible in the diff, but tb_invalidate_phys_page already has an assert_memory_lock. - cpu_io_recompile is !user-only, so no mmap_lock there. - Added mmap_unlock()'s before all siglongjmp's that could be called in user-mode while mmap_lock is held. + Added an assert for !have_mmap_lock() after returning from the longjmp in cpu_exec, just like we do in cpu_exec_step_atomic. Performance numbers before/after: Host: AMD Opteron(tm) Processor 6376 ubuntu 17.04 ppc64 bootup+shutdown time 700 +-+--+----+------+------------+-----------+------------*--+-+ | + + + + + *B | | before ***B*** ** * | |tb lock removal ###D### *** | 600 +-+ *** +-+ | ** # | | *B* #D | | *** * ## | 500 +-+ *** ### +-+ | * *** ### | | *B* # ## | | ** * #D# | 400 +-+ ** ## +-+ | ** ### | | ** ## | | ** # ## | 300 +-+ * B* #D# +-+ | B *** ### | | * ** #### | | * *** ### | 200 +-+ B *B #D# +-+ | #B* * ## # | | #* ## | | + D##D# + + + + | 100 +-+--+----+------+------------+-----------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/HwmBHXe debian jessie aarch64 bootup+shutdown time 90 +-+--+-----+-----+------------+------------+------------+--+-+ | + + + + + + | | before ***B*** B | 80 +tb lock removal ###D### **D +-+ | **### | | **## | 70 +-+ ** # +-+ | ** ## | | ** # | 60 +-+ *B ## +-+ | ** ## | | *** #D | 50 +-+ *** ## +-+ | * ** ### | | **B* ### | 40 +-+ **** # ## +-+ | **** #D# | | ***B** ### | 30 +-+ B***B** #### +-+ | B * * # ### | | B ###D# | 20 +-+ D ##D## +-+ | D# | | + + + + + + | 10 +-+--+-----+-----+------------+------------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/iGpGFtv The gains are high for 4-8 CPUs. Beyond that point, however, unrelated lock contention significantly hurts scalability. Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-05 05:46:31 +02:00
mmap_unlock();
/* execute the generated code */
trace_exec_tb_nocache(tb, tb->pc);
cpu_tb_exec(cpu, tb);
tcg: remove tb_lock Use mmap_lock in user-mode to protect TCG state and the page descriptors. In !user-mode, each vCPU has its own TCG state, so no locks needed. Per-page locks are used to protect the page descriptors. Per-TB locks are used in both modes to protect TB jumps. Some notes: - tb_lock is removed from notdirty_mem_write by passing a locked page_collection to tb_invalidate_phys_page_fast. - tcg_tb_lookup/remove/insert/etc have their own internal lock(s), so there is no need to further serialize access to them. - do_tb_flush is run in a safe async context, meaning no other vCPU threads are running. Therefore acquiring mmap_lock there is just to please tools such as thread sanitizer. - Not visible in the diff, but tb_invalidate_phys_page already has an assert_memory_lock. - cpu_io_recompile is !user-only, so no mmap_lock there. - Added mmap_unlock()'s before all siglongjmp's that could be called in user-mode while mmap_lock is held. + Added an assert for !have_mmap_lock() after returning from the longjmp in cpu_exec, just like we do in cpu_exec_step_atomic. Performance numbers before/after: Host: AMD Opteron(tm) Processor 6376 ubuntu 17.04 ppc64 bootup+shutdown time 700 +-+--+----+------+------------+-----------+------------*--+-+ | + + + + + *B | | before ***B*** ** * | |tb lock removal ###D### *** | 600 +-+ *** +-+ | ** # | | *B* #D | | *** * ## | 500 +-+ *** ### +-+ | * *** ### | | *B* # ## | | ** * #D# | 400 +-+ ** ## +-+ | ** ### | | ** ## | | ** # ## | 300 +-+ * B* #D# +-+ | B *** ### | | * ** #### | | * *** ### | 200 +-+ B *B #D# +-+ | #B* * ## # | | #* ## | | + D##D# + + + + | 100 +-+--+----+------+------------+-----------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/HwmBHXe debian jessie aarch64 bootup+shutdown time 90 +-+--+-----+-----+------------+------------+------------+--+-+ | + + + + + + | | before ***B*** B | 80 +tb lock removal ###D### **D +-+ | **### | | **## | 70 +-+ ** # +-+ | ** ## | | ** # | 60 +-+ *B ## +-+ | ** ## | | *** #D | 50 +-+ *** ## +-+ | * ** ### | | **B* ### | 40 +-+ **** # ## +-+ | **** #D# | | ***B** ### | 30 +-+ B***B** #### +-+ | B * * # ### | | B ###D# | 20 +-+ D ##D## +-+ | D# | | + + + + + + | 10 +-+--+-----+-----+------------+------------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/iGpGFtv The gains are high for 4-8 CPUs. Beyond that point, however, unrelated lock contention significantly hurts scalability. Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-05 05:46:31 +02:00
mmap_lock();
tb_phys_invalidate(tb, -1);
tcg: remove tb_lock Use mmap_lock in user-mode to protect TCG state and the page descriptors. In !user-mode, each vCPU has its own TCG state, so no locks needed. Per-page locks are used to protect the page descriptors. Per-TB locks are used in both modes to protect TB jumps. Some notes: - tb_lock is removed from notdirty_mem_write by passing a locked page_collection to tb_invalidate_phys_page_fast. - tcg_tb_lookup/remove/insert/etc have their own internal lock(s), so there is no need to further serialize access to them. - do_tb_flush is run in a safe async context, meaning no other vCPU threads are running. Therefore acquiring mmap_lock there is just to please tools such as thread sanitizer. - Not visible in the diff, but tb_invalidate_phys_page already has an assert_memory_lock. - cpu_io_recompile is !user-only, so no mmap_lock there. - Added mmap_unlock()'s before all siglongjmp's that could be called in user-mode while mmap_lock is held. + Added an assert for !have_mmap_lock() after returning from the longjmp in cpu_exec, just like we do in cpu_exec_step_atomic. Performance numbers before/after: Host: AMD Opteron(tm) Processor 6376 ubuntu 17.04 ppc64 bootup+shutdown time 700 +-+--+----+------+------------+-----------+------------*--+-+ | + + + + + *B | | before ***B*** ** * | |tb lock removal ###D### *** | 600 +-+ *** +-+ | ** # | | *B* #D | | *** * ## | 500 +-+ *** ### +-+ | * *** ### | | *B* # ## | | ** * #D# | 400 +-+ ** ## +-+ | ** ### | | ** ## | | ** # ## | 300 +-+ * B* #D# +-+ | B *** ### | | * ** #### | | * *** ### | 200 +-+ B *B #D# +-+ | #B* * ## # | | #* ## | | + D##D# + + + + | 100 +-+--+----+------+------------+-----------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/HwmBHXe debian jessie aarch64 bootup+shutdown time 90 +-+--+-----+-----+------------+------------+------------+--+-+ | + + + + + + | | before ***B*** B | 80 +tb lock removal ###D### **D +-+ | **### | | **## | 70 +-+ ** # +-+ | ** ## | | ** # | 60 +-+ *B ## +-+ | ** ## | | *** #D | 50 +-+ *** ## +-+ | * ** ### | | **B* ### | 40 +-+ **** # ## +-+ | **** #D# | | ***B** ### | 30 +-+ B***B** #### +-+ | B * * # ### | | B ###D# | 20 +-+ D ##D## +-+ | D# | | + + + + + + | 10 +-+--+-----+-----+------------+------------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/iGpGFtv The gains are high for 4-8 CPUs. Beyond that point, however, unrelated lock contention significantly hurts scalability. Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-05 05:46:31 +02:00
mmap_unlock();
tcg: track TBs with per-region BST's This paves the way for enabling scalable parallel generation of TCG code. Instead of tracking TBs with a single binary search tree (BST), use a BST for each TCG region, protecting it with a lock. This is as scalable as it gets, since each TCG thread operates on a separate region. The core of this change is the introduction of struct tcg_region_tree, which contains a pointer to a GTree and an associated lock to serialize accesses to it. We then allocate an array of tcg_region_tree's, adding the appropriate padding to avoid false sharing based on qemu_dcache_linesize. Given a tc_ptr, we first find the corresponding region_tree. This is done by special-casing the first and last regions first, since they might be of size != region.size; otherwise we just divide the offset by region.stride. I was worried about this division (several dozen cycles of latency), but profiling shows that this is not a fast path. Note that region.stride is not required to be a power of two; it is only required to be a multiple of the host's page size. Note that with this design we can also provide consistent snapshots about all region trees at once; for instance, tcg_tb_foreach acquires/releases all region_tree locks before/after iterating over them. For this reason we now drop tb_lock in dump_exec_info(). As an alternative I considered implementing a concurrent BST, but this can be tricky to get right, offers no consistent snapshots of the BST, and performance and scalability-wise I don't think it could ever beat having separate GTrees, given that our workload is insert-mostly (all concurrent BST designs I've seen focus, understandably, on making lookups fast, which comes at the expense of convoluted, non-wait-free insertions/removals). Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-07-26 22:58:05 +02:00
tcg_tb_remove(tb);
}
#endif
void cpu_exec_step_atomic(CPUState *cpu)
{
CPUClass *cc = CPU_GET_CLASS(cpu);
TranslationBlock *tb;
target_ulong cs_base, pc;
uint32_t flags;
uint32_t cflags = 1;
uint32_t cf_mask = cflags & CF_HASH_MASK;
if (sigsetjmp(cpu->jmp_env, 0) == 0) {
start_exclusive();
tb = tb_lookup__cpu_state(cpu, &pc, &cs_base, &flags, cf_mask);
if (tb == NULL) {
mmap_lock();
tb = tb_gen_code(cpu, pc, cs_base, flags, cflags);
mmap_unlock();
}
/* Since we got here, we know that parallel_cpus must be true. */
parallel_cpus = false;
cc->cpu_exec_enter(cpu);
/* execute the generated code */
trace_exec_tb(tb, pc);
cpu_tb_exec(cpu, tb);
cc->cpu_exec_exit(cpu);
} else {
tcg: remove tb_lock Use mmap_lock in user-mode to protect TCG state and the page descriptors. In !user-mode, each vCPU has its own TCG state, so no locks needed. Per-page locks are used to protect the page descriptors. Per-TB locks are used in both modes to protect TB jumps. Some notes: - tb_lock is removed from notdirty_mem_write by passing a locked page_collection to tb_invalidate_phys_page_fast. - tcg_tb_lookup/remove/insert/etc have their own internal lock(s), so there is no need to further serialize access to them. - do_tb_flush is run in a safe async context, meaning no other vCPU threads are running. Therefore acquiring mmap_lock there is just to please tools such as thread sanitizer. - Not visible in the diff, but tb_invalidate_phys_page already has an assert_memory_lock. - cpu_io_recompile is !user-only, so no mmap_lock there. - Added mmap_unlock()'s before all siglongjmp's that could be called in user-mode while mmap_lock is held. + Added an assert for !have_mmap_lock() after returning from the longjmp in cpu_exec, just like we do in cpu_exec_step_atomic. Performance numbers before/after: Host: AMD Opteron(tm) Processor 6376 ubuntu 17.04 ppc64 bootup+shutdown time 700 +-+--+----+------+------------+-----------+------------*--+-+ | + + + + + *B | | before ***B*** ** * | |tb lock removal ###D### *** | 600 +-+ *** +-+ | ** # | | *B* #D | | *** * ## | 500 +-+ *** ### +-+ | * *** ### | | *B* # ## | | ** * #D# | 400 +-+ ** ## +-+ | ** ### | | ** ## | | ** # ## | 300 +-+ * B* #D# +-+ | B *** ### | | * ** #### | | * *** ### | 200 +-+ B *B #D# +-+ | #B* * ## # | | #* ## | | + D##D# + + + + | 100 +-+--+----+------+------------+-----------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/HwmBHXe debian jessie aarch64 bootup+shutdown time 90 +-+--+-----+-----+------------+------------+------------+--+-+ | + + + + + + | | before ***B*** B | 80 +tb lock removal ###D### **D +-+ | **### | | **## | 70 +-+ ** # +-+ | ** ## | | ** # | 60 +-+ *B ## +-+ | ** ## | | *** #D | 50 +-+ *** ## +-+ | * ** ### | | **B* ### | 40 +-+ **** # ## +-+ | **** #D# | | ***B** ### | 30 +-+ B***B** #### +-+ | B * * # ### | | B ###D# | 20 +-+ D ##D## +-+ | D# | | + + + + + + | 10 +-+--+-----+-----+------------+------------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/iGpGFtv The gains are high for 4-8 CPUs. Beyond that point, however, unrelated lock contention significantly hurts scalability. Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-05 05:46:31 +02:00
/*
* The mmap_lock is dropped by tb_gen_code if it runs out of
* memory.
*/
#ifndef CONFIG_SOFTMMU
tcg_debug_assert(!have_mmap_lock());
#endif
if (qemu_mutex_iothread_locked()) {
qemu_mutex_unlock_iothread();
}
assert_no_pages_locked();
qemu_plugin_disable_mem_helpers(cpu);
}
/*
* As we start the exclusive region before codegen we must still
* be in the region if we longjump out of either the codegen or
* the execution.
*/
g_assert(cpu_in_exclusive_context(cpu));
parallel_cpus = true;
end_exclusive();
}
tb hash: track translated blocks with qht Having a fixed-size hash table for keeping track of all translation blocks is suboptimal: some workloads are just too big or too small to get maximum performance from the hash table. The MRU promotion policy helps improve performance when the hash table is a little undersized, but it cannot make up for severely undersized hash tables. Furthermore, frequent MRU promotions result in writes that are a scalability bottleneck. For scalability, lookups should only perform reads, not writes. This is not a big deal for now, but it will become one once MTTCG matures. The appended fixes these issues by using qht as the implementation of the TB hash table. This solution is superior to other alternatives considered, namely: - master: implementation in QEMU before this patchset - xxhash: before this patch, i.e. fixed buckets + xxhash hashing + MRU. - xxhash-rcu: fixed buckets + xxhash + RCU list + MRU. MRU is implemented here by adding an intermediate struct that contains the u32 hash and a pointer to the TB; this allows us, on an MRU promotion, to copy said struct (that is not at the head), and put this new copy at the head. After a grace period, the original non-head struct can be eliminated, and after another grace period, freed. - qht-fixed-nomru: fixed buckets + xxhash + qht without auto-resize + no MRU for lookups; MRU for inserts. The appended solution is the following: - qht-dyn-nomru: dynamic number of buckets + xxhash + qht w/ auto-resize + no MRU for lookups; MRU for inserts. The plots below compare the considered solutions. The Y axis shows the boot time (in seconds) of a debian jessie image with arm-softmmu; the X axis sweeps the number of buckets (or initial number of buckets for qht-autoresize). The plots in PNG format (and with errorbars) can be seen here: http://imgur.com/a/Awgnq Each test runs 5 times, and the entire QEMU process is pinned to a single core for repeatability of results. Host: Intel Xeon E5-2690 28 ++------------+-------------+-------------+-------------+------------++ A***** + + + master **A*** + 27 ++ * xxhash ##B###++ | A******A****** xxhash-rcu $$C$$$ | 26 C$$ A******A****** qht-fixed-nomru*%%D%%%++ D%%$$ A******A******A*qht-dyn-mru A*E****A 25 ++ %%$$ qht-dyn-nomru &&F&&&++ B#####% | 24 ++ #C$$$$$ ++ | B### $ | | ## C$$$$$$ | 23 ++ # C$$$$$$ ++ | B###### C$$$$$$ %%%D 22 ++ %B###### C$$$$$$C$$$$$$C$$$$$$C$$$$$$C$$$$$$C | D%%%%%%B###### @E@@@@@@ %%%D%%%@@@E@@@@@@E 21 E@@@@@@E@@@@@@F&&&@@@E@@@&&&D%%%%%%B######B######B######B######B######B + E@@@ F&&& + E@ + F&&& + + 20 ++------------+-------------+-------------+-------------+------------++ 14 16 18 20 22 24 log2 number of buckets Host: Intel i7-4790K 14.5 ++------------+------------+-------------+------------+------------++ A** + + + master **A*** + 14 ++ ** xxhash ##B###++ 13.5 ++ ** xxhash-rcu $$C$$$++ | qht-fixed-nomru %%D%%% | 13 ++ A****** qht-dyn-mru @@E@@@++ | A*****A******A****** qht-dyn-nomru &&F&&& | 12.5 C$$ A******A******A*****A****** ***A 12 ++ $$ A*** ++ D%%% $$ | 11.5 ++ %% ++ B### %C$$$$$$ | 11 ++ ## D%%%%% C$$$$$ ++ | # % C$$$$$$ | 10.5 F&&&&&&B######D%%%%% C$$$$$$C$$$$$$C$$$$$$C$$$$$C$$$$$$ $$$C 10 E@@@@@@E@@@@@@B#####B######B######E@@@@@@E@@@%%%D%%%%%D%%%###B######B + F&& D%%%%%%B######B######B#####B###@@@D%%% + 9.5 ++------------+------------+-------------+------------+------------++ 14 16 18 20 22 24 log2 number of buckets Note that the original point before this patch series is X=15 for "master"; the little sensitivity to the increased number of buckets is due to the poor hashing function in master. xxhash-rcu has significant overhead due to the constant churn of allocating and deallocating intermediate structs for implementing MRU. An alternative would be do consider failed lookups as "maybe not there", and then acquire the external lock (tb_lock in this case) to really confirm that there was indeed a failed lookup. This, however, would not be enough to implement dynamic resizing--this is more complex: see "Resizable, Scalable, Concurrent Hash Tables via Relativistic Programming" by Triplett, McKenney and Walpole. This solution was discarded due to the very coarse RCU read critical sections that we have in MTTCG; resizing requires waiting for readers after every pointer update, and resizes require many pointer updates, so this would quickly become prohibitive. qht-fixed-nomru shows that MRU promotion is advisable for undersized hash tables. However, qht-dyn-mru shows that MRU promotion is not important if the hash table is properly sized: there is virtually no difference in performance between qht-dyn-nomru and qht-dyn-mru. Before this patch, we're at X=15 on "xxhash"; after this patch, we're at X=15 @ qht-dyn-nomru. This patch thus matches the best performance that we can achieve with optimum sizing of the hash table, while keeping the hash table scalable for readers. The improvement we get before and after this patch for booting debian jessie with arm-softmmu is: - Intel Xeon E5-2690: 10.5% less time - Intel i7-4790K: 5.2% less time We could get this same improvement _for this particular workload_ by statically increasing the size of the hash table. But this would hurt workloads that do not need a large hash table. The dynamic (upward) resizing allows us to start small and enlarge the hash table as needed. A quick note on downsizing: the table is resized back to 2**15 buckets on every tb_flush; this makes sense because it is not guaranteed that the table will reach the same number of TBs later on (e.g. most bootup code is thrown away after boot); it makes sense to grow the hash table as more code blocks are translated. This also avoids the complication of having to build downsizing hysteresis logic into qht. Reviewed-by: Sergey Fedorov <serge.fedorov@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Emilio G. Cota <cota@braap.org> Message-Id: <1465412133-3029-15-git-send-email-cota@braap.org> Signed-off-by: Richard Henderson <rth@twiddle.net>
2016-06-08 20:55:32 +02:00
struct tb_desc {
target_ulong pc;
target_ulong cs_base;
CPUArchState *env;
tb_page_addr_t phys_page1;
uint32_t flags;
uint32_t cf_mask;
uint32_t trace_vcpu_dstate;
tb hash: track translated blocks with qht Having a fixed-size hash table for keeping track of all translation blocks is suboptimal: some workloads are just too big or too small to get maximum performance from the hash table. The MRU promotion policy helps improve performance when the hash table is a little undersized, but it cannot make up for severely undersized hash tables. Furthermore, frequent MRU promotions result in writes that are a scalability bottleneck. For scalability, lookups should only perform reads, not writes. This is not a big deal for now, but it will become one once MTTCG matures. The appended fixes these issues by using qht as the implementation of the TB hash table. This solution is superior to other alternatives considered, namely: - master: implementation in QEMU before this patchset - xxhash: before this patch, i.e. fixed buckets + xxhash hashing + MRU. - xxhash-rcu: fixed buckets + xxhash + RCU list + MRU. MRU is implemented here by adding an intermediate struct that contains the u32 hash and a pointer to the TB; this allows us, on an MRU promotion, to copy said struct (that is not at the head), and put this new copy at the head. After a grace period, the original non-head struct can be eliminated, and after another grace period, freed. - qht-fixed-nomru: fixed buckets + xxhash + qht without auto-resize + no MRU for lookups; MRU for inserts. The appended solution is the following: - qht-dyn-nomru: dynamic number of buckets + xxhash + qht w/ auto-resize + no MRU for lookups; MRU for inserts. The plots below compare the considered solutions. The Y axis shows the boot time (in seconds) of a debian jessie image with arm-softmmu; the X axis sweeps the number of buckets (or initial number of buckets for qht-autoresize). The plots in PNG format (and with errorbars) can be seen here: http://imgur.com/a/Awgnq Each test runs 5 times, and the entire QEMU process is pinned to a single core for repeatability of results. Host: Intel Xeon E5-2690 28 ++------------+-------------+-------------+-------------+------------++ A***** + + + master **A*** + 27 ++ * xxhash ##B###++ | A******A****** xxhash-rcu $$C$$$ | 26 C$$ A******A****** qht-fixed-nomru*%%D%%%++ D%%$$ A******A******A*qht-dyn-mru A*E****A 25 ++ %%$$ qht-dyn-nomru &&F&&&++ B#####% | 24 ++ #C$$$$$ ++ | B### $ | | ## C$$$$$$ | 23 ++ # C$$$$$$ ++ | B###### C$$$$$$ %%%D 22 ++ %B###### C$$$$$$C$$$$$$C$$$$$$C$$$$$$C$$$$$$C | D%%%%%%B###### @E@@@@@@ %%%D%%%@@@E@@@@@@E 21 E@@@@@@E@@@@@@F&&&@@@E@@@&&&D%%%%%%B######B######B######B######B######B + E@@@ F&&& + E@ + F&&& + + 20 ++------------+-------------+-------------+-------------+------------++ 14 16 18 20 22 24 log2 number of buckets Host: Intel i7-4790K 14.5 ++------------+------------+-------------+------------+------------++ A** + + + master **A*** + 14 ++ ** xxhash ##B###++ 13.5 ++ ** xxhash-rcu $$C$$$++ | qht-fixed-nomru %%D%%% | 13 ++ A****** qht-dyn-mru @@E@@@++ | A*****A******A****** qht-dyn-nomru &&F&&& | 12.5 C$$ A******A******A*****A****** ***A 12 ++ $$ A*** ++ D%%% $$ | 11.5 ++ %% ++ B### %C$$$$$$ | 11 ++ ## D%%%%% C$$$$$ ++ | # % C$$$$$$ | 10.5 F&&&&&&B######D%%%%% C$$$$$$C$$$$$$C$$$$$$C$$$$$C$$$$$$ $$$C 10 E@@@@@@E@@@@@@B#####B######B######E@@@@@@E@@@%%%D%%%%%D%%%###B######B + F&& D%%%%%%B######B######B#####B###@@@D%%% + 9.5 ++------------+------------+-------------+------------+------------++ 14 16 18 20 22 24 log2 number of buckets Note that the original point before this patch series is X=15 for "master"; the little sensitivity to the increased number of buckets is due to the poor hashing function in master. xxhash-rcu has significant overhead due to the constant churn of allocating and deallocating intermediate structs for implementing MRU. An alternative would be do consider failed lookups as "maybe not there", and then acquire the external lock (tb_lock in this case) to really confirm that there was indeed a failed lookup. This, however, would not be enough to implement dynamic resizing--this is more complex: see "Resizable, Scalable, Concurrent Hash Tables via Relativistic Programming" by Triplett, McKenney and Walpole. This solution was discarded due to the very coarse RCU read critical sections that we have in MTTCG; resizing requires waiting for readers after every pointer update, and resizes require many pointer updates, so this would quickly become prohibitive. qht-fixed-nomru shows that MRU promotion is advisable for undersized hash tables. However, qht-dyn-mru shows that MRU promotion is not important if the hash table is properly sized: there is virtually no difference in performance between qht-dyn-nomru and qht-dyn-mru. Before this patch, we're at X=15 on "xxhash"; after this patch, we're at X=15 @ qht-dyn-nomru. This patch thus matches the best performance that we can achieve with optimum sizing of the hash table, while keeping the hash table scalable for readers. The improvement we get before and after this patch for booting debian jessie with arm-softmmu is: - Intel Xeon E5-2690: 10.5% less time - Intel i7-4790K: 5.2% less time We could get this same improvement _for this particular workload_ by statically increasing the size of the hash table. But this would hurt workloads that do not need a large hash table. The dynamic (upward) resizing allows us to start small and enlarge the hash table as needed. A quick note on downsizing: the table is resized back to 2**15 buckets on every tb_flush; this makes sense because it is not guaranteed that the table will reach the same number of TBs later on (e.g. most bootup code is thrown away after boot); it makes sense to grow the hash table as more code blocks are translated. This also avoids the complication of having to build downsizing hysteresis logic into qht. Reviewed-by: Sergey Fedorov <serge.fedorov@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Emilio G. Cota <cota@braap.org> Message-Id: <1465412133-3029-15-git-send-email-cota@braap.org> Signed-off-by: Richard Henderson <rth@twiddle.net>
2016-06-08 20:55:32 +02:00
};
static bool tb_lookup_cmp(const void *p, const void *d)
tb hash: track translated blocks with qht Having a fixed-size hash table for keeping track of all translation blocks is suboptimal: some workloads are just too big or too small to get maximum performance from the hash table. The MRU promotion policy helps improve performance when the hash table is a little undersized, but it cannot make up for severely undersized hash tables. Furthermore, frequent MRU promotions result in writes that are a scalability bottleneck. For scalability, lookups should only perform reads, not writes. This is not a big deal for now, but it will become one once MTTCG matures. The appended fixes these issues by using qht as the implementation of the TB hash table. This solution is superior to other alternatives considered, namely: - master: implementation in QEMU before this patchset - xxhash: before this patch, i.e. fixed buckets + xxhash hashing + MRU. - xxhash-rcu: fixed buckets + xxhash + RCU list + MRU. MRU is implemented here by adding an intermediate struct that contains the u32 hash and a pointer to the TB; this allows us, on an MRU promotion, to copy said struct (that is not at the head), and put this new copy at the head. After a grace period, the original non-head struct can be eliminated, and after another grace period, freed. - qht-fixed-nomru: fixed buckets + xxhash + qht without auto-resize + no MRU for lookups; MRU for inserts. The appended solution is the following: - qht-dyn-nomru: dynamic number of buckets + xxhash + qht w/ auto-resize + no MRU for lookups; MRU for inserts. The plots below compare the considered solutions. The Y axis shows the boot time (in seconds) of a debian jessie image with arm-softmmu; the X axis sweeps the number of buckets (or initial number of buckets for qht-autoresize). The plots in PNG format (and with errorbars) can be seen here: http://imgur.com/a/Awgnq Each test runs 5 times, and the entire QEMU process is pinned to a single core for repeatability of results. Host: Intel Xeon E5-2690 28 ++------------+-------------+-------------+-------------+------------++ A***** + + + master **A*** + 27 ++ * xxhash ##B###++ | A******A****** xxhash-rcu $$C$$$ | 26 C$$ A******A****** qht-fixed-nomru*%%D%%%++ D%%$$ A******A******A*qht-dyn-mru A*E****A 25 ++ %%$$ qht-dyn-nomru &&F&&&++ B#####% | 24 ++ #C$$$$$ ++ | B### $ | | ## C$$$$$$ | 23 ++ # C$$$$$$ ++ | B###### C$$$$$$ %%%D 22 ++ %B###### C$$$$$$C$$$$$$C$$$$$$C$$$$$$C$$$$$$C | D%%%%%%B###### @E@@@@@@ %%%D%%%@@@E@@@@@@E 21 E@@@@@@E@@@@@@F&&&@@@E@@@&&&D%%%%%%B######B######B######B######B######B + E@@@ F&&& + E@ + F&&& + + 20 ++------------+-------------+-------------+-------------+------------++ 14 16 18 20 22 24 log2 number of buckets Host: Intel i7-4790K 14.5 ++------------+------------+-------------+------------+------------++ A** + + + master **A*** + 14 ++ ** xxhash ##B###++ 13.5 ++ ** xxhash-rcu $$C$$$++ | qht-fixed-nomru %%D%%% | 13 ++ A****** qht-dyn-mru @@E@@@++ | A*****A******A****** qht-dyn-nomru &&F&&& | 12.5 C$$ A******A******A*****A****** ***A 12 ++ $$ A*** ++ D%%% $$ | 11.5 ++ %% ++ B### %C$$$$$$ | 11 ++ ## D%%%%% C$$$$$ ++ | # % C$$$$$$ | 10.5 F&&&&&&B######D%%%%% C$$$$$$C$$$$$$C$$$$$$C$$$$$C$$$$$$ $$$C 10 E@@@@@@E@@@@@@B#####B######B######E@@@@@@E@@@%%%D%%%%%D%%%###B######B + F&& D%%%%%%B######B######B#####B###@@@D%%% + 9.5 ++------------+------------+-------------+------------+------------++ 14 16 18 20 22 24 log2 number of buckets Note that the original point before this patch series is X=15 for "master"; the little sensitivity to the increased number of buckets is due to the poor hashing function in master. xxhash-rcu has significant overhead due to the constant churn of allocating and deallocating intermediate structs for implementing MRU. An alternative would be do consider failed lookups as "maybe not there", and then acquire the external lock (tb_lock in this case) to really confirm that there was indeed a failed lookup. This, however, would not be enough to implement dynamic resizing--this is more complex: see "Resizable, Scalable, Concurrent Hash Tables via Relativistic Programming" by Triplett, McKenney and Walpole. This solution was discarded due to the very coarse RCU read critical sections that we have in MTTCG; resizing requires waiting for readers after every pointer update, and resizes require many pointer updates, so this would quickly become prohibitive. qht-fixed-nomru shows that MRU promotion is advisable for undersized hash tables. However, qht-dyn-mru shows that MRU promotion is not important if the hash table is properly sized: there is virtually no difference in performance between qht-dyn-nomru and qht-dyn-mru. Before this patch, we're at X=15 on "xxhash"; after this patch, we're at X=15 @ qht-dyn-nomru. This patch thus matches the best performance that we can achieve with optimum sizing of the hash table, while keeping the hash table scalable for readers. The improvement we get before and after this patch for booting debian jessie with arm-softmmu is: - Intel Xeon E5-2690: 10.5% less time - Intel i7-4790K: 5.2% less time We could get this same improvement _for this particular workload_ by statically increasing the size of the hash table. But this would hurt workloads that do not need a large hash table. The dynamic (upward) resizing allows us to start small and enlarge the hash table as needed. A quick note on downsizing: the table is resized back to 2**15 buckets on every tb_flush; this makes sense because it is not guaranteed that the table will reach the same number of TBs later on (e.g. most bootup code is thrown away after boot); it makes sense to grow the hash table as more code blocks are translated. This also avoids the complication of having to build downsizing hysteresis logic into qht. Reviewed-by: Sergey Fedorov <serge.fedorov@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Emilio G. Cota <cota@braap.org> Message-Id: <1465412133-3029-15-git-send-email-cota@braap.org> Signed-off-by: Richard Henderson <rth@twiddle.net>
2016-06-08 20:55:32 +02:00
{
const TranslationBlock *tb = p;
const struct tb_desc *desc = d;
if (tb->pc == desc->pc &&
tb->page_addr[0] == desc->phys_page1 &&
tb->cs_base == desc->cs_base &&
tb->flags == desc->flags &&
tb->trace_vcpu_dstate == desc->trace_vcpu_dstate &&
(tb_cflags(tb) & (CF_HASH_MASK | CF_INVALID)) == desc->cf_mask) {
tb hash: track translated blocks with qht Having a fixed-size hash table for keeping track of all translation blocks is suboptimal: some workloads are just too big or too small to get maximum performance from the hash table. The MRU promotion policy helps improve performance when the hash table is a little undersized, but it cannot make up for severely undersized hash tables. Furthermore, frequent MRU promotions result in writes that are a scalability bottleneck. For scalability, lookups should only perform reads, not writes. This is not a big deal for now, but it will become one once MTTCG matures. The appended fixes these issues by using qht as the implementation of the TB hash table. This solution is superior to other alternatives considered, namely: - master: implementation in QEMU before this patchset - xxhash: before this patch, i.e. fixed buckets + xxhash hashing + MRU. - xxhash-rcu: fixed buckets + xxhash + RCU list + MRU. MRU is implemented here by adding an intermediate struct that contains the u32 hash and a pointer to the TB; this allows us, on an MRU promotion, to copy said struct (that is not at the head), and put this new copy at the head. After a grace period, the original non-head struct can be eliminated, and after another grace period, freed. - qht-fixed-nomru: fixed buckets + xxhash + qht without auto-resize + no MRU for lookups; MRU for inserts. The appended solution is the following: - qht-dyn-nomru: dynamic number of buckets + xxhash + qht w/ auto-resize + no MRU for lookups; MRU for inserts. The plots below compare the considered solutions. The Y axis shows the boot time (in seconds) of a debian jessie image with arm-softmmu; the X axis sweeps the number of buckets (or initial number of buckets for qht-autoresize). The plots in PNG format (and with errorbars) can be seen here: http://imgur.com/a/Awgnq Each test runs 5 times, and the entire QEMU process is pinned to a single core for repeatability of results. Host: Intel Xeon E5-2690 28 ++------------+-------------+-------------+-------------+------------++ A***** + + + master **A*** + 27 ++ * xxhash ##B###++ | A******A****** xxhash-rcu $$C$$$ | 26 C$$ A******A****** qht-fixed-nomru*%%D%%%++ D%%$$ A******A******A*qht-dyn-mru A*E****A 25 ++ %%$$ qht-dyn-nomru &&F&&&++ B#####% | 24 ++ #C$$$$$ ++ | B### $ | | ## C$$$$$$ | 23 ++ # C$$$$$$ ++ | B###### C$$$$$$ %%%D 22 ++ %B###### C$$$$$$C$$$$$$C$$$$$$C$$$$$$C$$$$$$C | D%%%%%%B###### @E@@@@@@ %%%D%%%@@@E@@@@@@E 21 E@@@@@@E@@@@@@F&&&@@@E@@@&&&D%%%%%%B######B######B######B######B######B + E@@@ F&&& + E@ + F&&& + + 20 ++------------+-------------+-------------+-------------+------------++ 14 16 18 20 22 24 log2 number of buckets Host: Intel i7-4790K 14.5 ++------------+------------+-------------+------------+------------++ A** + + + master **A*** + 14 ++ ** xxhash ##B###++ 13.5 ++ ** xxhash-rcu $$C$$$++ | qht-fixed-nomru %%D%%% | 13 ++ A****** qht-dyn-mru @@E@@@++ | A*****A******A****** qht-dyn-nomru &&F&&& | 12.5 C$$ A******A******A*****A****** ***A 12 ++ $$ A*** ++ D%%% $$ | 11.5 ++ %% ++ B### %C$$$$$$ | 11 ++ ## D%%%%% C$$$$$ ++ | # % C$$$$$$ | 10.5 F&&&&&&B######D%%%%% C$$$$$$C$$$$$$C$$$$$$C$$$$$C$$$$$$ $$$C 10 E@@@@@@E@@@@@@B#####B######B######E@@@@@@E@@@%%%D%%%%%D%%%###B######B + F&& D%%%%%%B######B######B#####B###@@@D%%% + 9.5 ++------------+------------+-------------+------------+------------++ 14 16 18 20 22 24 log2 number of buckets Note that the original point before this patch series is X=15 for "master"; the little sensitivity to the increased number of buckets is due to the poor hashing function in master. xxhash-rcu has significant overhead due to the constant churn of allocating and deallocating intermediate structs for implementing MRU. An alternative would be do consider failed lookups as "maybe not there", and then acquire the external lock (tb_lock in this case) to really confirm that there was indeed a failed lookup. This, however, would not be enough to implement dynamic resizing--this is more complex: see "Resizable, Scalable, Concurrent Hash Tables via Relativistic Programming" by Triplett, McKenney and Walpole. This solution was discarded due to the very coarse RCU read critical sections that we have in MTTCG; resizing requires waiting for readers after every pointer update, and resizes require many pointer updates, so this would quickly become prohibitive. qht-fixed-nomru shows that MRU promotion is advisable for undersized hash tables. However, qht-dyn-mru shows that MRU promotion is not important if the hash table is properly sized: there is virtually no difference in performance between qht-dyn-nomru and qht-dyn-mru. Before this patch, we're at X=15 on "xxhash"; after this patch, we're at X=15 @ qht-dyn-nomru. This patch thus matches the best performance that we can achieve with optimum sizing of the hash table, while keeping the hash table scalable for readers. The improvement we get before and after this patch for booting debian jessie with arm-softmmu is: - Intel Xeon E5-2690: 10.5% less time - Intel i7-4790K: 5.2% less time We could get this same improvement _for this particular workload_ by statically increasing the size of the hash table. But this would hurt workloads that do not need a large hash table. The dynamic (upward) resizing allows us to start small and enlarge the hash table as needed. A quick note on downsizing: the table is resized back to 2**15 buckets on every tb_flush; this makes sense because it is not guaranteed that the table will reach the same number of TBs later on (e.g. most bootup code is thrown away after boot); it makes sense to grow the hash table as more code blocks are translated. This also avoids the complication of having to build downsizing hysteresis logic into qht. Reviewed-by: Sergey Fedorov <serge.fedorov@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Emilio G. Cota <cota@braap.org> Message-Id: <1465412133-3029-15-git-send-email-cota@braap.org> Signed-off-by: Richard Henderson <rth@twiddle.net>
2016-06-08 20:55:32 +02:00
/* check next page if needed */
if (tb->page_addr[1] == -1) {
return true;
} else {
tb_page_addr_t phys_page2;
target_ulong virt_page2;
virt_page2 = (desc->pc & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE;
phys_page2 = get_page_addr_code(desc->env, virt_page2);
if (tb->page_addr[1] == phys_page2) {
return true;
}
}
}
return false;
}
TranslationBlock *tb_htable_lookup(CPUState *cpu, target_ulong pc,
target_ulong cs_base, uint32_t flags,
uint32_t cf_mask)
{
tb hash: track translated blocks with qht Having a fixed-size hash table for keeping track of all translation blocks is suboptimal: some workloads are just too big or too small to get maximum performance from the hash table. The MRU promotion policy helps improve performance when the hash table is a little undersized, but it cannot make up for severely undersized hash tables. Furthermore, frequent MRU promotions result in writes that are a scalability bottleneck. For scalability, lookups should only perform reads, not writes. This is not a big deal for now, but it will become one once MTTCG matures. The appended fixes these issues by using qht as the implementation of the TB hash table. This solution is superior to other alternatives considered, namely: - master: implementation in QEMU before this patchset - xxhash: before this patch, i.e. fixed buckets + xxhash hashing + MRU. - xxhash-rcu: fixed buckets + xxhash + RCU list + MRU. MRU is implemented here by adding an intermediate struct that contains the u32 hash and a pointer to the TB; this allows us, on an MRU promotion, to copy said struct (that is not at the head), and put this new copy at the head. After a grace period, the original non-head struct can be eliminated, and after another grace period, freed. - qht-fixed-nomru: fixed buckets + xxhash + qht without auto-resize + no MRU for lookups; MRU for inserts. The appended solution is the following: - qht-dyn-nomru: dynamic number of buckets + xxhash + qht w/ auto-resize + no MRU for lookups; MRU for inserts. The plots below compare the considered solutions. The Y axis shows the boot time (in seconds) of a debian jessie image with arm-softmmu; the X axis sweeps the number of buckets (or initial number of buckets for qht-autoresize). The plots in PNG format (and with errorbars) can be seen here: http://imgur.com/a/Awgnq Each test runs 5 times, and the entire QEMU process is pinned to a single core for repeatability of results. Host: Intel Xeon E5-2690 28 ++------------+-------------+-------------+-------------+------------++ A***** + + + master **A*** + 27 ++ * xxhash ##B###++ | A******A****** xxhash-rcu $$C$$$ | 26 C$$ A******A****** qht-fixed-nomru*%%D%%%++ D%%$$ A******A******A*qht-dyn-mru A*E****A 25 ++ %%$$ qht-dyn-nomru &&F&&&++ B#####% | 24 ++ #C$$$$$ ++ | B### $ | | ## C$$$$$$ | 23 ++ # C$$$$$$ ++ | B###### C$$$$$$ %%%D 22 ++ %B###### C$$$$$$C$$$$$$C$$$$$$C$$$$$$C$$$$$$C | D%%%%%%B###### @E@@@@@@ %%%D%%%@@@E@@@@@@E 21 E@@@@@@E@@@@@@F&&&@@@E@@@&&&D%%%%%%B######B######B######B######B######B + E@@@ F&&& + E@ + F&&& + + 20 ++------------+-------------+-------------+-------------+------------++ 14 16 18 20 22 24 log2 number of buckets Host: Intel i7-4790K 14.5 ++------------+------------+-------------+------------+------------++ A** + + + master **A*** + 14 ++ ** xxhash ##B###++ 13.5 ++ ** xxhash-rcu $$C$$$++ | qht-fixed-nomru %%D%%% | 13 ++ A****** qht-dyn-mru @@E@@@++ | A*****A******A****** qht-dyn-nomru &&F&&& | 12.5 C$$ A******A******A*****A****** ***A 12 ++ $$ A*** ++ D%%% $$ | 11.5 ++ %% ++ B### %C$$$$$$ | 11 ++ ## D%%%%% C$$$$$ ++ | # % C$$$$$$ | 10.5 F&&&&&&B######D%%%%% C$$$$$$C$$$$$$C$$$$$$C$$$$$C$$$$$$ $$$C 10 E@@@@@@E@@@@@@B#####B######B######E@@@@@@E@@@%%%D%%%%%D%%%###B######B + F&& D%%%%%%B######B######B#####B###@@@D%%% + 9.5 ++------------+------------+-------------+------------+------------++ 14 16 18 20 22 24 log2 number of buckets Note that the original point before this patch series is X=15 for "master"; the little sensitivity to the increased number of buckets is due to the poor hashing function in master. xxhash-rcu has significant overhead due to the constant churn of allocating and deallocating intermediate structs for implementing MRU. An alternative would be do consider failed lookups as "maybe not there", and then acquire the external lock (tb_lock in this case) to really confirm that there was indeed a failed lookup. This, however, would not be enough to implement dynamic resizing--this is more complex: see "Resizable, Scalable, Concurrent Hash Tables via Relativistic Programming" by Triplett, McKenney and Walpole. This solution was discarded due to the very coarse RCU read critical sections that we have in MTTCG; resizing requires waiting for readers after every pointer update, and resizes require many pointer updates, so this would quickly become prohibitive. qht-fixed-nomru shows that MRU promotion is advisable for undersized hash tables. However, qht-dyn-mru shows that MRU promotion is not important if the hash table is properly sized: there is virtually no difference in performance between qht-dyn-nomru and qht-dyn-mru. Before this patch, we're at X=15 on "xxhash"; after this patch, we're at X=15 @ qht-dyn-nomru. This patch thus matches the best performance that we can achieve with optimum sizing of the hash table, while keeping the hash table scalable for readers. The improvement we get before and after this patch for booting debian jessie with arm-softmmu is: - Intel Xeon E5-2690: 10.5% less time - Intel i7-4790K: 5.2% less time We could get this same improvement _for this particular workload_ by statically increasing the size of the hash table. But this would hurt workloads that do not need a large hash table. The dynamic (upward) resizing allows us to start small and enlarge the hash table as needed. A quick note on downsizing: the table is resized back to 2**15 buckets on every tb_flush; this makes sense because it is not guaranteed that the table will reach the same number of TBs later on (e.g. most bootup code is thrown away after boot); it makes sense to grow the hash table as more code blocks are translated. This also avoids the complication of having to build downsizing hysteresis logic into qht. Reviewed-by: Sergey Fedorov <serge.fedorov@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Emilio G. Cota <cota@braap.org> Message-Id: <1465412133-3029-15-git-send-email-cota@braap.org> Signed-off-by: Richard Henderson <rth@twiddle.net>
2016-06-08 20:55:32 +02:00
tb_page_addr_t phys_pc;
struct tb_desc desc;
tb hash: hash phys_pc, pc, and flags with xxhash For some workloads such as arm bootup, tb_phys_hash is performance-critical. The is due to the high frequency of accesses to the hash table, originated by (frequent) TLB flushes that wipe out the cpu-private tb_jmp_cache's. More info: https://lists.nongnu.org/archive/html/qemu-devel/2016-03/msg05098.html To dig further into this I modified an arm image booting debian jessie to immediately shut down after boot. Analysis revealed that quite a bit of time is unnecessarily spent in tb_phys_hash: the cause is poor hashing that results in very uneven loading of chains in the hash table's buckets; the longest observed chain had ~550 elements. The appended addresses this with two changes: 1) Use xxhash as the hash table's hash function. xxhash is a fast, high-quality hashing function. 2) Feed the hashing function with not just tb_phys, but also pc and flags. This improves performance over using just tb_phys for hashing, since that resulted in some hash buckets having many TB's, while others getting very few; with these changes, the longest observed chain on a single hash bucket is brought down from ~550 to ~40. Tests show that the other element checked for in tb_find_physical, cs_base, is always a match when tb_phys+pc+flags are a match, so hashing cs_base is wasteful. It could be that this is an ARM-only thing, though. UPDATE: On Tue, Apr 05, 2016 at 08:41:43 -0700, Richard Henderson wrote: > The cs_base field is only used by i386 (in 16-bit modes), and sparc (for a TB > consisting of only a delay slot). > It may well still turn out to be reasonable to ignore cs_base for hashing. BTW, after this change the hash table should not be called "tb_hash_phys" anymore; this is addressed later in this series. This change gives consistent bootup time improvements. I tested two host machines: - Intel Xeon E5-2690: 11.6% less time - Intel i7-4790K: 19.2% less time Increasing the number of hash buckets yields further improvements. However, using a larger, fixed number of buckets can degrade performance for other workloads that do not translate as many blocks (600K+ for debian-jessie arm bootup). This is dealt with later in this series. Reviewed-by: Sergey Fedorov <sergey.fedorov@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Message-Id: <1465412133-3029-8-git-send-email-cota@braap.org> Signed-off-by: Richard Henderson <rth@twiddle.net>
2016-06-08 20:55:25 +02:00
uint32_t h;
tb hash: track translated blocks with qht Having a fixed-size hash table for keeping track of all translation blocks is suboptimal: some workloads are just too big or too small to get maximum performance from the hash table. The MRU promotion policy helps improve performance when the hash table is a little undersized, but it cannot make up for severely undersized hash tables. Furthermore, frequent MRU promotions result in writes that are a scalability bottleneck. For scalability, lookups should only perform reads, not writes. This is not a big deal for now, but it will become one once MTTCG matures. The appended fixes these issues by using qht as the implementation of the TB hash table. This solution is superior to other alternatives considered, namely: - master: implementation in QEMU before this patchset - xxhash: before this patch, i.e. fixed buckets + xxhash hashing + MRU. - xxhash-rcu: fixed buckets + xxhash + RCU list + MRU. MRU is implemented here by adding an intermediate struct that contains the u32 hash and a pointer to the TB; this allows us, on an MRU promotion, to copy said struct (that is not at the head), and put this new copy at the head. After a grace period, the original non-head struct can be eliminated, and after another grace period, freed. - qht-fixed-nomru: fixed buckets + xxhash + qht without auto-resize + no MRU for lookups; MRU for inserts. The appended solution is the following: - qht-dyn-nomru: dynamic number of buckets + xxhash + qht w/ auto-resize + no MRU for lookups; MRU for inserts. The plots below compare the considered solutions. The Y axis shows the boot time (in seconds) of a debian jessie image with arm-softmmu; the X axis sweeps the number of buckets (or initial number of buckets for qht-autoresize). The plots in PNG format (and with errorbars) can be seen here: http://imgur.com/a/Awgnq Each test runs 5 times, and the entire QEMU process is pinned to a single core for repeatability of results. Host: Intel Xeon E5-2690 28 ++------------+-------------+-------------+-------------+------------++ A***** + + + master **A*** + 27 ++ * xxhash ##B###++ | A******A****** xxhash-rcu $$C$$$ | 26 C$$ A******A****** qht-fixed-nomru*%%D%%%++ D%%$$ A******A******A*qht-dyn-mru A*E****A 25 ++ %%$$ qht-dyn-nomru &&F&&&++ B#####% | 24 ++ #C$$$$$ ++ | B### $ | | ## C$$$$$$ | 23 ++ # C$$$$$$ ++ | B###### C$$$$$$ %%%D 22 ++ %B###### C$$$$$$C$$$$$$C$$$$$$C$$$$$$C$$$$$$C | D%%%%%%B###### @E@@@@@@ %%%D%%%@@@E@@@@@@E 21 E@@@@@@E@@@@@@F&&&@@@E@@@&&&D%%%%%%B######B######B######B######B######B + E@@@ F&&& + E@ + F&&& + + 20 ++------------+-------------+-------------+-------------+------------++ 14 16 18 20 22 24 log2 number of buckets Host: Intel i7-4790K 14.5 ++------------+------------+-------------+------------+------------++ A** + + + master **A*** + 14 ++ ** xxhash ##B###++ 13.5 ++ ** xxhash-rcu $$C$$$++ | qht-fixed-nomru %%D%%% | 13 ++ A****** qht-dyn-mru @@E@@@++ | A*****A******A****** qht-dyn-nomru &&F&&& | 12.5 C$$ A******A******A*****A****** ***A 12 ++ $$ A*** ++ D%%% $$ | 11.5 ++ %% ++ B### %C$$$$$$ | 11 ++ ## D%%%%% C$$$$$ ++ | # % C$$$$$$ | 10.5 F&&&&&&B######D%%%%% C$$$$$$C$$$$$$C$$$$$$C$$$$$C$$$$$$ $$$C 10 E@@@@@@E@@@@@@B#####B######B######E@@@@@@E@@@%%%D%%%%%D%%%###B######B + F&& D%%%%%%B######B######B#####B###@@@D%%% + 9.5 ++------------+------------+-------------+------------+------------++ 14 16 18 20 22 24 log2 number of buckets Note that the original point before this patch series is X=15 for "master"; the little sensitivity to the increased number of buckets is due to the poor hashing function in master. xxhash-rcu has significant overhead due to the constant churn of allocating and deallocating intermediate structs for implementing MRU. An alternative would be do consider failed lookups as "maybe not there", and then acquire the external lock (tb_lock in this case) to really confirm that there was indeed a failed lookup. This, however, would not be enough to implement dynamic resizing--this is more complex: see "Resizable, Scalable, Concurrent Hash Tables via Relativistic Programming" by Triplett, McKenney and Walpole. This solution was discarded due to the very coarse RCU read critical sections that we have in MTTCG; resizing requires waiting for readers after every pointer update, and resizes require many pointer updates, so this would quickly become prohibitive. qht-fixed-nomru shows that MRU promotion is advisable for undersized hash tables. However, qht-dyn-mru shows that MRU promotion is not important if the hash table is properly sized: there is virtually no difference in performance between qht-dyn-nomru and qht-dyn-mru. Before this patch, we're at X=15 on "xxhash"; after this patch, we're at X=15 @ qht-dyn-nomru. This patch thus matches the best performance that we can achieve with optimum sizing of the hash table, while keeping the hash table scalable for readers. The improvement we get before and after this patch for booting debian jessie with arm-softmmu is: - Intel Xeon E5-2690: 10.5% less time - Intel i7-4790K: 5.2% less time We could get this same improvement _for this particular workload_ by statically increasing the size of the hash table. But this would hurt workloads that do not need a large hash table. The dynamic (upward) resizing allows us to start small and enlarge the hash table as needed. A quick note on downsizing: the table is resized back to 2**15 buckets on every tb_flush; this makes sense because it is not guaranteed that the table will reach the same number of TBs later on (e.g. most bootup code is thrown away after boot); it makes sense to grow the hash table as more code blocks are translated. This also avoids the complication of having to build downsizing hysteresis logic into qht. Reviewed-by: Sergey Fedorov <serge.fedorov@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Emilio G. Cota <cota@braap.org> Message-Id: <1465412133-3029-15-git-send-email-cota@braap.org> Signed-off-by: Richard Henderson <rth@twiddle.net>
2016-06-08 20:55:32 +02:00
desc.env = (CPUArchState *)cpu->env_ptr;
desc.cs_base = cs_base;
desc.flags = flags;
desc.cf_mask = cf_mask;
desc.trace_vcpu_dstate = *cpu->trace_dstate;
tb hash: track translated blocks with qht Having a fixed-size hash table for keeping track of all translation blocks is suboptimal: some workloads are just too big or too small to get maximum performance from the hash table. The MRU promotion policy helps improve performance when the hash table is a little undersized, but it cannot make up for severely undersized hash tables. Furthermore, frequent MRU promotions result in writes that are a scalability bottleneck. For scalability, lookups should only perform reads, not writes. This is not a big deal for now, but it will become one once MTTCG matures. The appended fixes these issues by using qht as the implementation of the TB hash table. This solution is superior to other alternatives considered, namely: - master: implementation in QEMU before this patchset - xxhash: before this patch, i.e. fixed buckets + xxhash hashing + MRU. - xxhash-rcu: fixed buckets + xxhash + RCU list + MRU. MRU is implemented here by adding an intermediate struct that contains the u32 hash and a pointer to the TB; this allows us, on an MRU promotion, to copy said struct (that is not at the head), and put this new copy at the head. After a grace period, the original non-head struct can be eliminated, and after another grace period, freed. - qht-fixed-nomru: fixed buckets + xxhash + qht without auto-resize + no MRU for lookups; MRU for inserts. The appended solution is the following: - qht-dyn-nomru: dynamic number of buckets + xxhash + qht w/ auto-resize + no MRU for lookups; MRU for inserts. The plots below compare the considered solutions. The Y axis shows the boot time (in seconds) of a debian jessie image with arm-softmmu; the X axis sweeps the number of buckets (or initial number of buckets for qht-autoresize). The plots in PNG format (and with errorbars) can be seen here: http://imgur.com/a/Awgnq Each test runs 5 times, and the entire QEMU process is pinned to a single core for repeatability of results. Host: Intel Xeon E5-2690 28 ++------------+-------------+-------------+-------------+------------++ A***** + + + master **A*** + 27 ++ * xxhash ##B###++ | A******A****** xxhash-rcu $$C$$$ | 26 C$$ A******A****** qht-fixed-nomru*%%D%%%++ D%%$$ A******A******A*qht-dyn-mru A*E****A 25 ++ %%$$ qht-dyn-nomru &&F&&&++ B#####% | 24 ++ #C$$$$$ ++ | B### $ | | ## C$$$$$$ | 23 ++ # C$$$$$$ ++ | B###### C$$$$$$ %%%D 22 ++ %B###### C$$$$$$C$$$$$$C$$$$$$C$$$$$$C$$$$$$C | D%%%%%%B###### @E@@@@@@ %%%D%%%@@@E@@@@@@E 21 E@@@@@@E@@@@@@F&&&@@@E@@@&&&D%%%%%%B######B######B######B######B######B + E@@@ F&&& + E@ + F&&& + + 20 ++------------+-------------+-------------+-------------+------------++ 14 16 18 20 22 24 log2 number of buckets Host: Intel i7-4790K 14.5 ++------------+------------+-------------+------------+------------++ A** + + + master **A*** + 14 ++ ** xxhash ##B###++ 13.5 ++ ** xxhash-rcu $$C$$$++ | qht-fixed-nomru %%D%%% | 13 ++ A****** qht-dyn-mru @@E@@@++ | A*****A******A****** qht-dyn-nomru &&F&&& | 12.5 C$$ A******A******A*****A****** ***A 12 ++ $$ A*** ++ D%%% $$ | 11.5 ++ %% ++ B### %C$$$$$$ | 11 ++ ## D%%%%% C$$$$$ ++ | # % C$$$$$$ | 10.5 F&&&&&&B######D%%%%% C$$$$$$C$$$$$$C$$$$$$C$$$$$C$$$$$$ $$$C 10 E@@@@@@E@@@@@@B#####B######B######E@@@@@@E@@@%%%D%%%%%D%%%###B######B + F&& D%%%%%%B######B######B#####B###@@@D%%% + 9.5 ++------------+------------+-------------+------------+------------++ 14 16 18 20 22 24 log2 number of buckets Note that the original point before this patch series is X=15 for "master"; the little sensitivity to the increased number of buckets is due to the poor hashing function in master. xxhash-rcu has significant overhead due to the constant churn of allocating and deallocating intermediate structs for implementing MRU. An alternative would be do consider failed lookups as "maybe not there", and then acquire the external lock (tb_lock in this case) to really confirm that there was indeed a failed lookup. This, however, would not be enough to implement dynamic resizing--this is more complex: see "Resizable, Scalable, Concurrent Hash Tables via Relativistic Programming" by Triplett, McKenney and Walpole. This solution was discarded due to the very coarse RCU read critical sections that we have in MTTCG; resizing requires waiting for readers after every pointer update, and resizes require many pointer updates, so this would quickly become prohibitive. qht-fixed-nomru shows that MRU promotion is advisable for undersized hash tables. However, qht-dyn-mru shows that MRU promotion is not important if the hash table is properly sized: there is virtually no difference in performance between qht-dyn-nomru and qht-dyn-mru. Before this patch, we're at X=15 on "xxhash"; after this patch, we're at X=15 @ qht-dyn-nomru. This patch thus matches the best performance that we can achieve with optimum sizing of the hash table, while keeping the hash table scalable for readers. The improvement we get before and after this patch for booting debian jessie with arm-softmmu is: - Intel Xeon E5-2690: 10.5% less time - Intel i7-4790K: 5.2% less time We could get this same improvement _for this particular workload_ by statically increasing the size of the hash table. But this would hurt workloads that do not need a large hash table. The dynamic (upward) resizing allows us to start small and enlarge the hash table as needed. A quick note on downsizing: the table is resized back to 2**15 buckets on every tb_flush; this makes sense because it is not guaranteed that the table will reach the same number of TBs later on (e.g. most bootup code is thrown away after boot); it makes sense to grow the hash table as more code blocks are translated. This also avoids the complication of having to build downsizing hysteresis logic into qht. Reviewed-by: Sergey Fedorov <serge.fedorov@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Emilio G. Cota <cota@braap.org> Message-Id: <1465412133-3029-15-git-send-email-cota@braap.org> Signed-off-by: Richard Henderson <rth@twiddle.net>
2016-06-08 20:55:32 +02:00
desc.pc = pc;
phys_pc = get_page_addr_code(desc.env, pc);
if (phys_pc == -1) {
return NULL;
}
tb hash: track translated blocks with qht Having a fixed-size hash table for keeping track of all translation blocks is suboptimal: some workloads are just too big or too small to get maximum performance from the hash table. The MRU promotion policy helps improve performance when the hash table is a little undersized, but it cannot make up for severely undersized hash tables. Furthermore, frequent MRU promotions result in writes that are a scalability bottleneck. For scalability, lookups should only perform reads, not writes. This is not a big deal for now, but it will become one once MTTCG matures. The appended fixes these issues by using qht as the implementation of the TB hash table. This solution is superior to other alternatives considered, namely: - master: implementation in QEMU before this patchset - xxhash: before this patch, i.e. fixed buckets + xxhash hashing + MRU. - xxhash-rcu: fixed buckets + xxhash + RCU list + MRU. MRU is implemented here by adding an intermediate struct that contains the u32 hash and a pointer to the TB; this allows us, on an MRU promotion, to copy said struct (that is not at the head), and put this new copy at the head. After a grace period, the original non-head struct can be eliminated, and after another grace period, freed. - qht-fixed-nomru: fixed buckets + xxhash + qht without auto-resize + no MRU for lookups; MRU for inserts. The appended solution is the following: - qht-dyn-nomru: dynamic number of buckets + xxhash + qht w/ auto-resize + no MRU for lookups; MRU for inserts. The plots below compare the considered solutions. The Y axis shows the boot time (in seconds) of a debian jessie image with arm-softmmu; the X axis sweeps the number of buckets (or initial number of buckets for qht-autoresize). The plots in PNG format (and with errorbars) can be seen here: http://imgur.com/a/Awgnq Each test runs 5 times, and the entire QEMU process is pinned to a single core for repeatability of results. Host: Intel Xeon E5-2690 28 ++------------+-------------+-------------+-------------+------------++ A***** + + + master **A*** + 27 ++ * xxhash ##B###++ | A******A****** xxhash-rcu $$C$$$ | 26 C$$ A******A****** qht-fixed-nomru*%%D%%%++ D%%$$ A******A******A*qht-dyn-mru A*E****A 25 ++ %%$$ qht-dyn-nomru &&F&&&++ B#####% | 24 ++ #C$$$$$ ++ | B### $ | | ## C$$$$$$ | 23 ++ # C$$$$$$ ++ | B###### C$$$$$$ %%%D 22 ++ %B###### C$$$$$$C$$$$$$C$$$$$$C$$$$$$C$$$$$$C | D%%%%%%B###### @E@@@@@@ %%%D%%%@@@E@@@@@@E 21 E@@@@@@E@@@@@@F&&&@@@E@@@&&&D%%%%%%B######B######B######B######B######B + E@@@ F&&& + E@ + F&&& + + 20 ++------------+-------------+-------------+-------------+------------++ 14 16 18 20 22 24 log2 number of buckets Host: Intel i7-4790K 14.5 ++------------+------------+-------------+------------+------------++ A** + + + master **A*** + 14 ++ ** xxhash ##B###++ 13.5 ++ ** xxhash-rcu $$C$$$++ | qht-fixed-nomru %%D%%% | 13 ++ A****** qht-dyn-mru @@E@@@++ | A*****A******A****** qht-dyn-nomru &&F&&& | 12.5 C$$ A******A******A*****A****** ***A 12 ++ $$ A*** ++ D%%% $$ | 11.5 ++ %% ++ B### %C$$$$$$ | 11 ++ ## D%%%%% C$$$$$ ++ | # % C$$$$$$ | 10.5 F&&&&&&B######D%%%%% C$$$$$$C$$$$$$C$$$$$$C$$$$$C$$$$$$ $$$C 10 E@@@@@@E@@@@@@B#####B######B######E@@@@@@E@@@%%%D%%%%%D%%%###B######B + F&& D%%%%%%B######B######B#####B###@@@D%%% + 9.5 ++------------+------------+-------------+------------+------------++ 14 16 18 20 22 24 log2 number of buckets Note that the original point before this patch series is X=15 for "master"; the little sensitivity to the increased number of buckets is due to the poor hashing function in master. xxhash-rcu has significant overhead due to the constant churn of allocating and deallocating intermediate structs for implementing MRU. An alternative would be do consider failed lookups as "maybe not there", and then acquire the external lock (tb_lock in this case) to really confirm that there was indeed a failed lookup. This, however, would not be enough to implement dynamic resizing--this is more complex: see "Resizable, Scalable, Concurrent Hash Tables via Relativistic Programming" by Triplett, McKenney and Walpole. This solution was discarded due to the very coarse RCU read critical sections that we have in MTTCG; resizing requires waiting for readers after every pointer update, and resizes require many pointer updates, so this would quickly become prohibitive. qht-fixed-nomru shows that MRU promotion is advisable for undersized hash tables. However, qht-dyn-mru shows that MRU promotion is not important if the hash table is properly sized: there is virtually no difference in performance between qht-dyn-nomru and qht-dyn-mru. Before this patch, we're at X=15 on "xxhash"; after this patch, we're at X=15 @ qht-dyn-nomru. This patch thus matches the best performance that we can achieve with optimum sizing of the hash table, while keeping the hash table scalable for readers. The improvement we get before and after this patch for booting debian jessie with arm-softmmu is: - Intel Xeon E5-2690: 10.5% less time - Intel i7-4790K: 5.2% less time We could get this same improvement _for this particular workload_ by statically increasing the size of the hash table. But this would hurt workloads that do not need a large hash table. The dynamic (upward) resizing allows us to start small and enlarge the hash table as needed. A quick note on downsizing: the table is resized back to 2**15 buckets on every tb_flush; this makes sense because it is not guaranteed that the table will reach the same number of TBs later on (e.g. most bootup code is thrown away after boot); it makes sense to grow the hash table as more code blocks are translated. This also avoids the complication of having to build downsizing hysteresis logic into qht. Reviewed-by: Sergey Fedorov <serge.fedorov@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Emilio G. Cota <cota@braap.org> Message-Id: <1465412133-3029-15-git-send-email-cota@braap.org> Signed-off-by: Richard Henderson <rth@twiddle.net>
2016-06-08 20:55:32 +02:00
desc.phys_page1 = phys_pc & TARGET_PAGE_MASK;
h = tb_hash_func(phys_pc, pc, flags, cf_mask, *cpu->trace_dstate);
return qht_lookup_custom(&tb_ctx.htable, &desc, h, tb_lookup_cmp);
}
void tb_set_jmp_target(TranslationBlock *tb, int n, uintptr_t addr)
{
if (TCG_TARGET_HAS_direct_jump) {
uintptr_t offset = tb->jmp_target_arg[n];
uintptr_t tc_ptr = (uintptr_t)tb->tc.ptr;
tb_target_set_jmp_target(tc_ptr, tc_ptr + offset, addr);
} else {
tb->jmp_target_arg[n] = addr;
}
}
static inline void tb_add_jump(TranslationBlock *tb, int n,
TranslationBlock *tb_next)
{
translate-all: protect TB jumps with a per-destination-TB lock This applies to both user-mode and !user-mode emulation. Instead of relying on a global lock, protect the list of incoming jumps with tb->jmp_lock. This lock also protects tb->cflags, so update all tb->cflags readers outside tb->jmp_lock to use atomic reads via tb_cflags(). In order to find the destination TB (and therefore its jmp_lock) from the origin TB, we introduce tb->jmp_dest[]. I considered not using a linked list of jumps, which simplifies code and makes the struct smaller. However, it unnecessarily increases memory usage, which results in a performance decrease. See for instance these numbers booting+shutting down debian-arm: Time (s) Rel. err (%) Abs. err (s) Rel. slowdown (%) ------------------------------------------------------------------------------ before 20.88 0.74 0.154512 0. after 20.81 0.38 0.079078 -0.33524904 GTree 21.02 0.28 0.058856 0.67049808 GHashTable + xxhash 21.63 1.08 0.233604 3.5919540 Using a hash table or a binary tree to keep track of the jumps doesn't really pay off, not only due to the increased memory usage, but also because most TBs have only 0 or 1 jumps to them. The maximum number of jumps when booting debian-arm that I measured is 35, but as we can see in the histogram below a TB with that many incoming jumps is extremely rare; the average TB has 0.80 incoming jumps. n_jumps: 379208; avg jumps/tb: 0.801099 dist: [0.0,1.0)|▄█▁▁▁▁▁▁▁▁▁▁▁ ▁▁▁▁▁▁ ▁▁▁ ▁▁▁ ▁|[34.0,35.0] Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-03 02:34:06 +02:00
uintptr_t old;
assert(n < ARRAY_SIZE(tb->jmp_list_next));
translate-all: protect TB jumps with a per-destination-TB lock This applies to both user-mode and !user-mode emulation. Instead of relying on a global lock, protect the list of incoming jumps with tb->jmp_lock. This lock also protects tb->cflags, so update all tb->cflags readers outside tb->jmp_lock to use atomic reads via tb_cflags(). In order to find the destination TB (and therefore its jmp_lock) from the origin TB, we introduce tb->jmp_dest[]. I considered not using a linked list of jumps, which simplifies code and makes the struct smaller. However, it unnecessarily increases memory usage, which results in a performance decrease. See for instance these numbers booting+shutting down debian-arm: Time (s) Rel. err (%) Abs. err (s) Rel. slowdown (%) ------------------------------------------------------------------------------ before 20.88 0.74 0.154512 0. after 20.81 0.38 0.079078 -0.33524904 GTree 21.02 0.28 0.058856 0.67049808 GHashTable + xxhash 21.63 1.08 0.233604 3.5919540 Using a hash table or a binary tree to keep track of the jumps doesn't really pay off, not only due to the increased memory usage, but also because most TBs have only 0 or 1 jumps to them. The maximum number of jumps when booting debian-arm that I measured is 35, but as we can see in the histogram below a TB with that many incoming jumps is extremely rare; the average TB has 0.80 incoming jumps. n_jumps: 379208; avg jumps/tb: 0.801099 dist: [0.0,1.0)|▄█▁▁▁▁▁▁▁▁▁▁▁ ▁▁▁▁▁▁ ▁▁▁ ▁▁▁ ▁|[34.0,35.0] Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-03 02:34:06 +02:00
qemu_spin_lock(&tb_next->jmp_lock);
/* make sure the destination TB is valid */
if (tb_next->cflags & CF_INVALID) {
goto out_unlock_next;
}
/* Atomically claim the jump destination slot only if it was NULL */
old = qatomic_cmpxchg(&tb->jmp_dest[n], (uintptr_t)NULL,
(uintptr_t)tb_next);
translate-all: protect TB jumps with a per-destination-TB lock This applies to both user-mode and !user-mode emulation. Instead of relying on a global lock, protect the list of incoming jumps with tb->jmp_lock. This lock also protects tb->cflags, so update all tb->cflags readers outside tb->jmp_lock to use atomic reads via tb_cflags(). In order to find the destination TB (and therefore its jmp_lock) from the origin TB, we introduce tb->jmp_dest[]. I considered not using a linked list of jumps, which simplifies code and makes the struct smaller. However, it unnecessarily increases memory usage, which results in a performance decrease. See for instance these numbers booting+shutting down debian-arm: Time (s) Rel. err (%) Abs. err (s) Rel. slowdown (%) ------------------------------------------------------------------------------ before 20.88 0.74 0.154512 0. after 20.81 0.38 0.079078 -0.33524904 GTree 21.02 0.28 0.058856 0.67049808 GHashTable + xxhash 21.63 1.08 0.233604 3.5919540 Using a hash table or a binary tree to keep track of the jumps doesn't really pay off, not only due to the increased memory usage, but also because most TBs have only 0 or 1 jumps to them. The maximum number of jumps when booting debian-arm that I measured is 35, but as we can see in the histogram below a TB with that many incoming jumps is extremely rare; the average TB has 0.80 incoming jumps. n_jumps: 379208; avg jumps/tb: 0.801099 dist: [0.0,1.0)|▄█▁▁▁▁▁▁▁▁▁▁▁ ▁▁▁▁▁▁ ▁▁▁ ▁▁▁ ▁|[34.0,35.0] Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-03 02:34:06 +02:00
if (old) {
goto out_unlock_next;
}
translate-all: protect TB jumps with a per-destination-TB lock This applies to both user-mode and !user-mode emulation. Instead of relying on a global lock, protect the list of incoming jumps with tb->jmp_lock. This lock also protects tb->cflags, so update all tb->cflags readers outside tb->jmp_lock to use atomic reads via tb_cflags(). In order to find the destination TB (and therefore its jmp_lock) from the origin TB, we introduce tb->jmp_dest[]. I considered not using a linked list of jumps, which simplifies code and makes the struct smaller. However, it unnecessarily increases memory usage, which results in a performance decrease. See for instance these numbers booting+shutting down debian-arm: Time (s) Rel. err (%) Abs. err (s) Rel. slowdown (%) ------------------------------------------------------------------------------ before 20.88 0.74 0.154512 0. after 20.81 0.38 0.079078 -0.33524904 GTree 21.02 0.28 0.058856 0.67049808 GHashTable + xxhash 21.63 1.08 0.233604 3.5919540 Using a hash table or a binary tree to keep track of the jumps doesn't really pay off, not only due to the increased memory usage, but also because most TBs have only 0 or 1 jumps to them. The maximum number of jumps when booting debian-arm that I measured is 35, but as we can see in the histogram below a TB with that many incoming jumps is extremely rare; the average TB has 0.80 incoming jumps. n_jumps: 379208; avg jumps/tb: 0.801099 dist: [0.0,1.0)|▄█▁▁▁▁▁▁▁▁▁▁▁ ▁▁▁▁▁▁ ▁▁▁ ▁▁▁ ▁|[34.0,35.0] Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-03 02:34:06 +02:00
/* patch the native jump address */
tb_set_jmp_target(tb, n, (uintptr_t)tb_next->tc.ptr);
/* add in TB jmp list */
tb->jmp_list_next[n] = tb_next->jmp_list_head;
tb_next->jmp_list_head = (uintptr_t)tb | n;
qemu_spin_unlock(&tb_next->jmp_lock);
qemu_log_mask_and_addr(CPU_LOG_EXEC, tb->pc,
"Linking TBs %p [" TARGET_FMT_lx
"] index %d -> %p [" TARGET_FMT_lx "]\n",
tb->tc.ptr, tb->pc, n,
tb_next->tc.ptr, tb_next->pc);
translate-all: protect TB jumps with a per-destination-TB lock This applies to both user-mode and !user-mode emulation. Instead of relying on a global lock, protect the list of incoming jumps with tb->jmp_lock. This lock also protects tb->cflags, so update all tb->cflags readers outside tb->jmp_lock to use atomic reads via tb_cflags(). In order to find the destination TB (and therefore its jmp_lock) from the origin TB, we introduce tb->jmp_dest[]. I considered not using a linked list of jumps, which simplifies code and makes the struct smaller. However, it unnecessarily increases memory usage, which results in a performance decrease. See for instance these numbers booting+shutting down debian-arm: Time (s) Rel. err (%) Abs. err (s) Rel. slowdown (%) ------------------------------------------------------------------------------ before 20.88 0.74 0.154512 0. after 20.81 0.38 0.079078 -0.33524904 GTree 21.02 0.28 0.058856 0.67049808 GHashTable + xxhash 21.63 1.08 0.233604 3.5919540 Using a hash table or a binary tree to keep track of the jumps doesn't really pay off, not only due to the increased memory usage, but also because most TBs have only 0 or 1 jumps to them. The maximum number of jumps when booting debian-arm that I measured is 35, but as we can see in the histogram below a TB with that many incoming jumps is extremely rare; the average TB has 0.80 incoming jumps. n_jumps: 379208; avg jumps/tb: 0.801099 dist: [0.0,1.0)|▄█▁▁▁▁▁▁▁▁▁▁▁ ▁▁▁▁▁▁ ▁▁▁ ▁▁▁ ▁|[34.0,35.0] Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-03 02:34:06 +02:00
return;
translate-all: protect TB jumps with a per-destination-TB lock This applies to both user-mode and !user-mode emulation. Instead of relying on a global lock, protect the list of incoming jumps with tb->jmp_lock. This lock also protects tb->cflags, so update all tb->cflags readers outside tb->jmp_lock to use atomic reads via tb_cflags(). In order to find the destination TB (and therefore its jmp_lock) from the origin TB, we introduce tb->jmp_dest[]. I considered not using a linked list of jumps, which simplifies code and makes the struct smaller. However, it unnecessarily increases memory usage, which results in a performance decrease. See for instance these numbers booting+shutting down debian-arm: Time (s) Rel. err (%) Abs. err (s) Rel. slowdown (%) ------------------------------------------------------------------------------ before 20.88 0.74 0.154512 0. after 20.81 0.38 0.079078 -0.33524904 GTree 21.02 0.28 0.058856 0.67049808 GHashTable + xxhash 21.63 1.08 0.233604 3.5919540 Using a hash table or a binary tree to keep track of the jumps doesn't really pay off, not only due to the increased memory usage, but also because most TBs have only 0 or 1 jumps to them. The maximum number of jumps when booting debian-arm that I measured is 35, but as we can see in the histogram below a TB with that many incoming jumps is extremely rare; the average TB has 0.80 incoming jumps. n_jumps: 379208; avg jumps/tb: 0.801099 dist: [0.0,1.0)|▄█▁▁▁▁▁▁▁▁▁▁▁ ▁▁▁▁▁▁ ▁▁▁ ▁▁▁ ▁|[34.0,35.0] Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-03 02:34:06 +02:00
out_unlock_next:
qemu_spin_unlock(&tb_next->jmp_lock);
return;
}
static inline TranslationBlock *tb_find(CPUState *cpu,
TranslationBlock *last_tb,
int tb_exit, uint32_t cf_mask)
{
TranslationBlock *tb;
target_ulong cs_base, pc;
uint32_t flags;
tb = tb_lookup__cpu_state(cpu, &pc, &cs_base, &flags, cf_mask);
tcg: consolidate TB lookups in tb_lookup__cpu_state This avoids duplicating code. cpu_exec_step will also use the new common function once we integrate parallel_cpus into tb->cflags. Note that in this commit we also fix a race, described by Richard Henderson during review. Think of this scenario with threads A and B: (A) Lookup succeeds for TB in hash without tb_lock (B) Sets the TB's tb->invalid flag (B) Removes the TB from tb_htable (B) Clears all CPU's tb_jmp_cache (A) Store TB into local tb_jmp_cache Given that order of events, (A) will keep executing that invalid TB until another flush of its tb_jmp_cache happens, which in theory might never happen. We can fix this by checking the tb->invalid flag every time we look up a TB from tb_jmp_cache, so that in the above scenario, next time we try to find that TB in tb_jmp_cache, we won't, and will therefore be forced to look it up in tb_htable. Performance-wise, I measured a small improvement when booting debian-arm. Note that inlining pays off: Performance counter stats for 'taskset -c 0 qemu-system-arm \ -machine type=virt -nographic -smp 1 -m 4096 \ -netdev user,id=unet,hostfwd=tcp::2222-:22 \ -device virtio-net-device,netdev=unet \ -drive file=jessie.qcow2,id=myblock,index=0,if=none \ -device virtio-blk-device,drive=myblock \ -kernel kernel.img -append console=ttyAMA0 root=/dev/vda1 \ -name arm,debug-threads=on -smp 1' (10 runs): Before: 18714.917392 task-clock # 0.952 CPUs utilized ( +- 0.95% ) 23,142 context-switches # 0.001 M/sec ( +- 0.50% ) 1 CPU-migrations # 0.000 M/sec 10,558 page-faults # 0.001 M/sec ( +- 0.95% ) 53,957,727,252 cycles # 2.883 GHz ( +- 0.91% ) [83.33%] 24,440,599,852 stalled-cycles-frontend # 45.30% frontend cycles idle ( +- 1.20% ) [83.33%] 16,495,714,424 stalled-cycles-backend # 30.57% backend cycles idle ( +- 0.95% ) [66.66%] 76,267,572,582 instructions # 1.41 insns per cycle # 0.32 stalled cycles per insn ( +- 0.87% ) [83.34%] 12,692,186,323 branches # 678.186 M/sec ( +- 0.92% ) [83.35%] 263,486,879 branch-misses # 2.08% of all branches ( +- 0.73% ) [83.34%] 19.648474449 seconds time elapsed ( +- 0.82% ) After, w/ inline (this patch): 18471.376627 task-clock # 0.955 CPUs utilized ( +- 0.96% ) 23,048 context-switches # 0.001 M/sec ( +- 0.48% ) 1 CPU-migrations # 0.000 M/sec 10,708 page-faults # 0.001 M/sec ( +- 0.81% ) 53,208,990,796 cycles # 2.881 GHz ( +- 0.98% ) [83.34%] 23,941,071,673 stalled-cycles-frontend # 44.99% frontend cycles idle ( +- 0.95% ) [83.34%] 16,161,773,848 stalled-cycles-backend # 30.37% backend cycles idle ( +- 0.76% ) [66.67%] 75,786,269,766 instructions # 1.42 insns per cycle # 0.32 stalled cycles per insn ( +- 1.24% ) [83.34%] 12,573,617,143 branches # 680.708 M/sec ( +- 1.34% ) [83.33%] 260,235,550 branch-misses # 2.07% of all branches ( +- 0.66% ) [83.33%] 19.340502161 seconds time elapsed ( +- 0.56% ) After, w/o inline: 18791.253967 task-clock # 0.954 CPUs utilized ( +- 0.78% ) 23,230 context-switches # 0.001 M/sec ( +- 0.42% ) 1 CPU-migrations # 0.000 M/sec 10,563 page-faults # 0.001 M/sec ( +- 1.27% ) 54,168,674,622 cycles # 2.883 GHz ( +- 0.80% ) [83.34%] 24,244,712,629 stalled-cycles-frontend # 44.76% frontend cycles idle ( +- 1.37% ) [83.33%] 16,288,648,572 stalled-cycles-backend # 30.07% backend cycles idle ( +- 0.95% ) [66.66%] 77,659,755,503 instructions # 1.43 insns per cycle # 0.31 stalled cycles per insn ( +- 0.97% ) [83.34%] 12,922,780,045 branches # 687.702 M/sec ( +- 1.06% ) [83.34%] 261,962,386 branch-misses # 2.03% of all branches ( +- 0.71% ) [83.35%] 19.700174670 seconds time elapsed ( +- 0.56% ) Reviewed-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-07-11 23:33:33 +02:00
if (tb == NULL) {
mmap_lock();
tb = tb_gen_code(cpu, pc, cs_base, flags, cf_mask);
tcg: consolidate TB lookups in tb_lookup__cpu_state This avoids duplicating code. cpu_exec_step will also use the new common function once we integrate parallel_cpus into tb->cflags. Note that in this commit we also fix a race, described by Richard Henderson during review. Think of this scenario with threads A and B: (A) Lookup succeeds for TB in hash without tb_lock (B) Sets the TB's tb->invalid flag (B) Removes the TB from tb_htable (B) Clears all CPU's tb_jmp_cache (A) Store TB into local tb_jmp_cache Given that order of events, (A) will keep executing that invalid TB until another flush of its tb_jmp_cache happens, which in theory might never happen. We can fix this by checking the tb->invalid flag every time we look up a TB from tb_jmp_cache, so that in the above scenario, next time we try to find that TB in tb_jmp_cache, we won't, and will therefore be forced to look it up in tb_htable. Performance-wise, I measured a small improvement when booting debian-arm. Note that inlining pays off: Performance counter stats for 'taskset -c 0 qemu-system-arm \ -machine type=virt -nographic -smp 1 -m 4096 \ -netdev user,id=unet,hostfwd=tcp::2222-:22 \ -device virtio-net-device,netdev=unet \ -drive file=jessie.qcow2,id=myblock,index=0,if=none \ -device virtio-blk-device,drive=myblock \ -kernel kernel.img -append console=ttyAMA0 root=/dev/vda1 \ -name arm,debug-threads=on -smp 1' (10 runs): Before: 18714.917392 task-clock # 0.952 CPUs utilized ( +- 0.95% ) 23,142 context-switches # 0.001 M/sec ( +- 0.50% ) 1 CPU-migrations # 0.000 M/sec 10,558 page-faults # 0.001 M/sec ( +- 0.95% ) 53,957,727,252 cycles # 2.883 GHz ( +- 0.91% ) [83.33%] 24,440,599,852 stalled-cycles-frontend # 45.30% frontend cycles idle ( +- 1.20% ) [83.33%] 16,495,714,424 stalled-cycles-backend # 30.57% backend cycles idle ( +- 0.95% ) [66.66%] 76,267,572,582 instructions # 1.41 insns per cycle # 0.32 stalled cycles per insn ( +- 0.87% ) [83.34%] 12,692,186,323 branches # 678.186 M/sec ( +- 0.92% ) [83.35%] 263,486,879 branch-misses # 2.08% of all branches ( +- 0.73% ) [83.34%] 19.648474449 seconds time elapsed ( +- 0.82% ) After, w/ inline (this patch): 18471.376627 task-clock # 0.955 CPUs utilized ( +- 0.96% ) 23,048 context-switches # 0.001 M/sec ( +- 0.48% ) 1 CPU-migrations # 0.000 M/sec 10,708 page-faults # 0.001 M/sec ( +- 0.81% ) 53,208,990,796 cycles # 2.881 GHz ( +- 0.98% ) [83.34%] 23,941,071,673 stalled-cycles-frontend # 44.99% frontend cycles idle ( +- 0.95% ) [83.34%] 16,161,773,848 stalled-cycles-backend # 30.37% backend cycles idle ( +- 0.76% ) [66.67%] 75,786,269,766 instructions # 1.42 insns per cycle # 0.32 stalled cycles per insn ( +- 1.24% ) [83.34%] 12,573,617,143 branches # 680.708 M/sec ( +- 1.34% ) [83.33%] 260,235,550 branch-misses # 2.07% of all branches ( +- 0.66% ) [83.33%] 19.340502161 seconds time elapsed ( +- 0.56% ) After, w/o inline: 18791.253967 task-clock # 0.954 CPUs utilized ( +- 0.78% ) 23,230 context-switches # 0.001 M/sec ( +- 0.42% ) 1 CPU-migrations # 0.000 M/sec 10,563 page-faults # 0.001 M/sec ( +- 1.27% ) 54,168,674,622 cycles # 2.883 GHz ( +- 0.80% ) [83.34%] 24,244,712,629 stalled-cycles-frontend # 44.76% frontend cycles idle ( +- 1.37% ) [83.33%] 16,288,648,572 stalled-cycles-backend # 30.07% backend cycles idle ( +- 0.95% ) [66.66%] 77,659,755,503 instructions # 1.43 insns per cycle # 0.31 stalled cycles per insn ( +- 0.97% ) [83.34%] 12,922,780,045 branches # 687.702 M/sec ( +- 1.06% ) [83.34%] 261,962,386 branch-misses # 2.03% of all branches ( +- 0.71% ) [83.35%] 19.700174670 seconds time elapsed ( +- 0.56% ) Reviewed-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-07-11 23:33:33 +02:00
mmap_unlock();
/* We add the TB in the virtual pc hash table for the fast lookup */
qatomic_set(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)], tb);
}
#ifndef CONFIG_USER_ONLY
/* We don't take care of direct jumps when address mapping changes in
* system emulation. So it's not safe to make a direct jump to a TB
* spanning two pages because the mapping for the second page can change.
*/
if (tb->page_addr[1] != -1) {
last_tb = NULL;
}
#endif
/* See if we can patch the calling TB. */
if (last_tb) {
translate-all: protect TB jumps with a per-destination-TB lock This applies to both user-mode and !user-mode emulation. Instead of relying on a global lock, protect the list of incoming jumps with tb->jmp_lock. This lock also protects tb->cflags, so update all tb->cflags readers outside tb->jmp_lock to use atomic reads via tb_cflags(). In order to find the destination TB (and therefore its jmp_lock) from the origin TB, we introduce tb->jmp_dest[]. I considered not using a linked list of jumps, which simplifies code and makes the struct smaller. However, it unnecessarily increases memory usage, which results in a performance decrease. See for instance these numbers booting+shutting down debian-arm: Time (s) Rel. err (%) Abs. err (s) Rel. slowdown (%) ------------------------------------------------------------------------------ before 20.88 0.74 0.154512 0. after 20.81 0.38 0.079078 -0.33524904 GTree 21.02 0.28 0.058856 0.67049808 GHashTable + xxhash 21.63 1.08 0.233604 3.5919540 Using a hash table or a binary tree to keep track of the jumps doesn't really pay off, not only due to the increased memory usage, but also because most TBs have only 0 or 1 jumps to them. The maximum number of jumps when booting debian-arm that I measured is 35, but as we can see in the histogram below a TB with that many incoming jumps is extremely rare; the average TB has 0.80 incoming jumps. n_jumps: 379208; avg jumps/tb: 0.801099 dist: [0.0,1.0)|▄█▁▁▁▁▁▁▁▁▁▁▁ ▁▁▁▁▁▁ ▁▁▁ ▁▁▁ ▁|[34.0,35.0] Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-03 02:34:06 +02:00
tb_add_jump(last_tb, tb_exit, tb);
}
return tb;
}
static inline bool cpu_handle_halt(CPUState *cpu)
{
if (cpu->halted) {
#if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
if ((cpu->interrupt_request & CPU_INTERRUPT_POLL)
&& replay_interrupt()) {
X86CPU *x86_cpu = X86_CPU(cpu);
tcg: drop global lock during TCG code execution This finally allows TCG to benefit from the iothread introduction: Drop the global mutex while running pure TCG CPU code. Reacquire the lock when entering MMIO or PIO emulation, or when leaving the TCG loop. We have to revert a few optimization for the current TCG threading model, namely kicking the TCG thread in qemu_mutex_lock_iothread and not kicking it in qemu_cpu_kick. We also need to disable RAM block reordering until we have a more efficient locking mechanism at hand. Still, a Linux x86 UP guest and my Musicpal ARM model boot fine here. These numbers demonstrate where we gain something: 20338 jan 20 0 331m 75m 6904 R 99 0.9 0:50.95 qemu-system-arm 20337 jan 20 0 331m 75m 6904 S 20 0.9 0:26.50 qemu-system-arm The guest CPU was fully loaded, but the iothread could still run mostly independent on a second core. Without the patch we don't get beyond 32206 jan 20 0 330m 73m 7036 R 82 0.9 1:06.00 qemu-system-arm 32204 jan 20 0 330m 73m 7036 S 21 0.9 0:17.03 qemu-system-arm We don't benefit significantly, though, when the guest is not fully loading a host CPU. Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> Message-Id: <1439220437-23957-10-git-send-email-fred.konrad@greensocs.com> [FK: Rebase, fix qemu_devices_reset deadlock, rm address_space_* mutex] Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com> [EGC: fixed iothread lock for cpu-exec IRQ handling] Signed-off-by: Emilio G. Cota <cota@braap.org> [AJB: -smp single-threaded fix, clean commit msg, BQL fixes] Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Pranith Kumar <bobby.prani@gmail.com> [PM: target-arm changes] Acked-by: Peter Maydell <peter.maydell@linaro.org>
2017-02-23 19:29:11 +01:00
qemu_mutex_lock_iothread();
apic_poll_irq(x86_cpu->apic_state);
cpu_reset_interrupt(cpu, CPU_INTERRUPT_POLL);
tcg: drop global lock during TCG code execution This finally allows TCG to benefit from the iothread introduction: Drop the global mutex while running pure TCG CPU code. Reacquire the lock when entering MMIO or PIO emulation, or when leaving the TCG loop. We have to revert a few optimization for the current TCG threading model, namely kicking the TCG thread in qemu_mutex_lock_iothread and not kicking it in qemu_cpu_kick. We also need to disable RAM block reordering until we have a more efficient locking mechanism at hand. Still, a Linux x86 UP guest and my Musicpal ARM model boot fine here. These numbers demonstrate where we gain something: 20338 jan 20 0 331m 75m 6904 R 99 0.9 0:50.95 qemu-system-arm 20337 jan 20 0 331m 75m 6904 S 20 0.9 0:26.50 qemu-system-arm The guest CPU was fully loaded, but the iothread could still run mostly independent on a second core. Without the patch we don't get beyond 32206 jan 20 0 330m 73m 7036 R 82 0.9 1:06.00 qemu-system-arm 32204 jan 20 0 330m 73m 7036 S 21 0.9 0:17.03 qemu-system-arm We don't benefit significantly, though, when the guest is not fully loading a host CPU. Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> Message-Id: <1439220437-23957-10-git-send-email-fred.konrad@greensocs.com> [FK: Rebase, fix qemu_devices_reset deadlock, rm address_space_* mutex] Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com> [EGC: fixed iothread lock for cpu-exec IRQ handling] Signed-off-by: Emilio G. Cota <cota@braap.org> [AJB: -smp single-threaded fix, clean commit msg, BQL fixes] Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Pranith Kumar <bobby.prani@gmail.com> [PM: target-arm changes] Acked-by: Peter Maydell <peter.maydell@linaro.org>
2017-02-23 19:29:11 +01:00
qemu_mutex_unlock_iothread();
}
#endif
if (!cpu_has_work(cpu)) {
return true;
}
cpu->halted = 0;
}
return false;
}
static inline void cpu_handle_debug_exception(CPUState *cpu)
{
CPUClass *cc = CPU_GET_CLASS(cpu);
CPUWatchpoint *wp;
if (!cpu->watchpoint_hit) {
QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
wp->flags &= ~BP_WATCHPOINT_HIT;
}
}
cc->debug_excp_handler(cpu);
}
static inline bool cpu_handle_exception(CPUState *cpu, int *ret)
{
if (cpu->exception_index < 0) {
#ifndef CONFIG_USER_ONLY
if (replay_has_exception()
&& cpu_neg(cpu)->icount_decr.u16.low + cpu->icount_extra == 0) {
/* try to cause an exception pending in the log */
cpu_exec_nocache(cpu, 1, tb_find(cpu, NULL, 0, curr_cflags()), true);
}
#endif
if (cpu->exception_index < 0) {
return false;
}
}
if (cpu->exception_index >= EXCP_INTERRUPT) {
/* exit request from the cpu execution loop */
*ret = cpu->exception_index;
if (*ret == EXCP_DEBUG) {
cpu_handle_debug_exception(cpu);
}
cpu->exception_index = -1;
return true;
} else {
#if defined(CONFIG_USER_ONLY)
/* if user mode only, we simulate a fake exception
which will be handled outside the cpu execution
loop */
#if defined(TARGET_I386)
CPUClass *cc = CPU_GET_CLASS(cpu);
cc->do_interrupt(cpu);
#endif
*ret = cpu->exception_index;
cpu->exception_index = -1;
return true;
#else
if (replay_exception()) {
CPUClass *cc = CPU_GET_CLASS(cpu);
qemu_mutex_lock_iothread();
cc->do_interrupt(cpu);
qemu_mutex_unlock_iothread();
cpu->exception_index = -1;
if (unlikely(cpu->singlestep_enabled)) {
/*
* After processing the exception, ensure an EXCP_DEBUG is
* raised when single-stepping so that GDB doesn't miss the
* next instruction.
*/
*ret = EXCP_DEBUG;
cpu_handle_debug_exception(cpu);
return true;
}
} else if (!replay_has_interrupt()) {
/* give a chance to iothread in replay mode */
*ret = EXCP_INTERRUPT;
return true;
}
#endif
}
return false;
}
static inline bool cpu_handle_interrupt(CPUState *cpu,
TranslationBlock **last_tb)
{
CPUClass *cc = CPU_GET_CLASS(cpu);
/* Clear the interrupt flag now since we're processing
* cpu->interrupt_request and cpu->exit_request.
* Ensure zeroing happens before reading cpu->exit_request or
* cpu->interrupt_request (see also smp_wmb in cpu_exit())
*/
qatomic_mb_set(&cpu_neg(cpu)->icount_decr.u16.high, 0);
if (unlikely(qatomic_read(&cpu->interrupt_request))) {
tcg: drop global lock during TCG code execution This finally allows TCG to benefit from the iothread introduction: Drop the global mutex while running pure TCG CPU code. Reacquire the lock when entering MMIO or PIO emulation, or when leaving the TCG loop. We have to revert a few optimization for the current TCG threading model, namely kicking the TCG thread in qemu_mutex_lock_iothread and not kicking it in qemu_cpu_kick. We also need to disable RAM block reordering until we have a more efficient locking mechanism at hand. Still, a Linux x86 UP guest and my Musicpal ARM model boot fine here. These numbers demonstrate where we gain something: 20338 jan 20 0 331m 75m 6904 R 99 0.9 0:50.95 qemu-system-arm 20337 jan 20 0 331m 75m 6904 S 20 0.9 0:26.50 qemu-system-arm The guest CPU was fully loaded, but the iothread could still run mostly independent on a second core. Without the patch we don't get beyond 32206 jan 20 0 330m 73m 7036 R 82 0.9 1:06.00 qemu-system-arm 32204 jan 20 0 330m 73m 7036 S 21 0.9 0:17.03 qemu-system-arm We don't benefit significantly, though, when the guest is not fully loading a host CPU. Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> Message-Id: <1439220437-23957-10-git-send-email-fred.konrad@greensocs.com> [FK: Rebase, fix qemu_devices_reset deadlock, rm address_space_* mutex] Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com> [EGC: fixed iothread lock for cpu-exec IRQ handling] Signed-off-by: Emilio G. Cota <cota@braap.org> [AJB: -smp single-threaded fix, clean commit msg, BQL fixes] Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Pranith Kumar <bobby.prani@gmail.com> [PM: target-arm changes] Acked-by: Peter Maydell <peter.maydell@linaro.org>
2017-02-23 19:29:11 +01:00
int interrupt_request;
qemu_mutex_lock_iothread();
interrupt_request = cpu->interrupt_request;
if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
/* Mask out external interrupts for this step. */
interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
}
if (interrupt_request & CPU_INTERRUPT_DEBUG) {
cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
cpu->exception_index = EXCP_DEBUG;
tcg: drop global lock during TCG code execution This finally allows TCG to benefit from the iothread introduction: Drop the global mutex while running pure TCG CPU code. Reacquire the lock when entering MMIO or PIO emulation, or when leaving the TCG loop. We have to revert a few optimization for the current TCG threading model, namely kicking the TCG thread in qemu_mutex_lock_iothread and not kicking it in qemu_cpu_kick. We also need to disable RAM block reordering until we have a more efficient locking mechanism at hand. Still, a Linux x86 UP guest and my Musicpal ARM model boot fine here. These numbers demonstrate where we gain something: 20338 jan 20 0 331m 75m 6904 R 99 0.9 0:50.95 qemu-system-arm 20337 jan 20 0 331m 75m 6904 S 20 0.9 0:26.50 qemu-system-arm The guest CPU was fully loaded, but the iothread could still run mostly independent on a second core. Without the patch we don't get beyond 32206 jan 20 0 330m 73m 7036 R 82 0.9 1:06.00 qemu-system-arm 32204 jan 20 0 330m 73m 7036 S 21 0.9 0:17.03 qemu-system-arm We don't benefit significantly, though, when the guest is not fully loading a host CPU. Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> Message-Id: <1439220437-23957-10-git-send-email-fred.konrad@greensocs.com> [FK: Rebase, fix qemu_devices_reset deadlock, rm address_space_* mutex] Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com> [EGC: fixed iothread lock for cpu-exec IRQ handling] Signed-off-by: Emilio G. Cota <cota@braap.org> [AJB: -smp single-threaded fix, clean commit msg, BQL fixes] Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Pranith Kumar <bobby.prani@gmail.com> [PM: target-arm changes] Acked-by: Peter Maydell <peter.maydell@linaro.org>
2017-02-23 19:29:11 +01:00
qemu_mutex_unlock_iothread();
return true;
}
if (replay_mode == REPLAY_MODE_PLAY && !replay_has_interrupt()) {
/* Do nothing */
} else if (interrupt_request & CPU_INTERRUPT_HALT) {
replay_interrupt();
cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
cpu->halted = 1;
cpu->exception_index = EXCP_HLT;
tcg: drop global lock during TCG code execution This finally allows TCG to benefit from the iothread introduction: Drop the global mutex while running pure TCG CPU code. Reacquire the lock when entering MMIO or PIO emulation, or when leaving the TCG loop. We have to revert a few optimization for the current TCG threading model, namely kicking the TCG thread in qemu_mutex_lock_iothread and not kicking it in qemu_cpu_kick. We also need to disable RAM block reordering until we have a more efficient locking mechanism at hand. Still, a Linux x86 UP guest and my Musicpal ARM model boot fine here. These numbers demonstrate where we gain something: 20338 jan 20 0 331m 75m 6904 R 99 0.9 0:50.95 qemu-system-arm 20337 jan 20 0 331m 75m 6904 S 20 0.9 0:26.50 qemu-system-arm The guest CPU was fully loaded, but the iothread could still run mostly independent on a second core. Without the patch we don't get beyond 32206 jan 20 0 330m 73m 7036 R 82 0.9 1:06.00 qemu-system-arm 32204 jan 20 0 330m 73m 7036 S 21 0.9 0:17.03 qemu-system-arm We don't benefit significantly, though, when the guest is not fully loading a host CPU. Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> Message-Id: <1439220437-23957-10-git-send-email-fred.konrad@greensocs.com> [FK: Rebase, fix qemu_devices_reset deadlock, rm address_space_* mutex] Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com> [EGC: fixed iothread lock for cpu-exec IRQ handling] Signed-off-by: Emilio G. Cota <cota@braap.org> [AJB: -smp single-threaded fix, clean commit msg, BQL fixes] Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Pranith Kumar <bobby.prani@gmail.com> [PM: target-arm changes] Acked-by: Peter Maydell <peter.maydell@linaro.org>
2017-02-23 19:29:11 +01:00
qemu_mutex_unlock_iothread();
return true;
}
#if defined(TARGET_I386)
else if (interrupt_request & CPU_INTERRUPT_INIT) {
X86CPU *x86_cpu = X86_CPU(cpu);
CPUArchState *env = &x86_cpu->env;
replay_interrupt();
cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0, 0);
do_cpu_init(x86_cpu);
cpu->exception_index = EXCP_HALTED;
tcg: drop global lock during TCG code execution This finally allows TCG to benefit from the iothread introduction: Drop the global mutex while running pure TCG CPU code. Reacquire the lock when entering MMIO or PIO emulation, or when leaving the TCG loop. We have to revert a few optimization for the current TCG threading model, namely kicking the TCG thread in qemu_mutex_lock_iothread and not kicking it in qemu_cpu_kick. We also need to disable RAM block reordering until we have a more efficient locking mechanism at hand. Still, a Linux x86 UP guest and my Musicpal ARM model boot fine here. These numbers demonstrate where we gain something: 20338 jan 20 0 331m 75m 6904 R 99 0.9 0:50.95 qemu-system-arm 20337 jan 20 0 331m 75m 6904 S 20 0.9 0:26.50 qemu-system-arm The guest CPU was fully loaded, but the iothread could still run mostly independent on a second core. Without the patch we don't get beyond 32206 jan 20 0 330m 73m 7036 R 82 0.9 1:06.00 qemu-system-arm 32204 jan 20 0 330m 73m 7036 S 21 0.9 0:17.03 qemu-system-arm We don't benefit significantly, though, when the guest is not fully loading a host CPU. Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> Message-Id: <1439220437-23957-10-git-send-email-fred.konrad@greensocs.com> [FK: Rebase, fix qemu_devices_reset deadlock, rm address_space_* mutex] Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com> [EGC: fixed iothread lock for cpu-exec IRQ handling] Signed-off-by: Emilio G. Cota <cota@braap.org> [AJB: -smp single-threaded fix, clean commit msg, BQL fixes] Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Pranith Kumar <bobby.prani@gmail.com> [PM: target-arm changes] Acked-by: Peter Maydell <peter.maydell@linaro.org>
2017-02-23 19:29:11 +01:00
qemu_mutex_unlock_iothread();
return true;
}
#else
else if (interrupt_request & CPU_INTERRUPT_RESET) {
replay_interrupt();
cpu_reset(cpu);
tcg: drop global lock during TCG code execution This finally allows TCG to benefit from the iothread introduction: Drop the global mutex while running pure TCG CPU code. Reacquire the lock when entering MMIO or PIO emulation, or when leaving the TCG loop. We have to revert a few optimization for the current TCG threading model, namely kicking the TCG thread in qemu_mutex_lock_iothread and not kicking it in qemu_cpu_kick. We also need to disable RAM block reordering until we have a more efficient locking mechanism at hand. Still, a Linux x86 UP guest and my Musicpal ARM model boot fine here. These numbers demonstrate where we gain something: 20338 jan 20 0 331m 75m 6904 R 99 0.9 0:50.95 qemu-system-arm 20337 jan 20 0 331m 75m 6904 S 20 0.9 0:26.50 qemu-system-arm The guest CPU was fully loaded, but the iothread could still run mostly independent on a second core. Without the patch we don't get beyond 32206 jan 20 0 330m 73m 7036 R 82 0.9 1:06.00 qemu-system-arm 32204 jan 20 0 330m 73m 7036 S 21 0.9 0:17.03 qemu-system-arm We don't benefit significantly, though, when the guest is not fully loading a host CPU. Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> Message-Id: <1439220437-23957-10-git-send-email-fred.konrad@greensocs.com> [FK: Rebase, fix qemu_devices_reset deadlock, rm address_space_* mutex] Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com> [EGC: fixed iothread lock for cpu-exec IRQ handling] Signed-off-by: Emilio G. Cota <cota@braap.org> [AJB: -smp single-threaded fix, clean commit msg, BQL fixes] Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Pranith Kumar <bobby.prani@gmail.com> [PM: target-arm changes] Acked-by: Peter Maydell <peter.maydell@linaro.org>
2017-02-23 19:29:11 +01:00
qemu_mutex_unlock_iothread();
return true;
}
#endif
/* The target hook has 3 exit conditions:
False when the interrupt isn't processed,
True when it is, and we should restart on a new TB,
and via longjmp via cpu_loop_exit. */
else {
if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
replay_interrupt();
/*
* After processing the interrupt, ensure an EXCP_DEBUG is
* raised when single-stepping so that GDB doesn't miss the
* next instruction.
*/
cpu->exception_index =
(cpu->singlestep_enabled ? EXCP_DEBUG : -1);
*last_tb = NULL;
}
/* The target hook may have updated the 'cpu->interrupt_request';
* reload the 'interrupt_request' value */
interrupt_request = cpu->interrupt_request;
}
if (interrupt_request & CPU_INTERRUPT_EXITTB) {
cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
/* ensure that no TB jump will be modified as
the program flow was changed */
*last_tb = NULL;
}
tcg: drop global lock during TCG code execution This finally allows TCG to benefit from the iothread introduction: Drop the global mutex while running pure TCG CPU code. Reacquire the lock when entering MMIO or PIO emulation, or when leaving the TCG loop. We have to revert a few optimization for the current TCG threading model, namely kicking the TCG thread in qemu_mutex_lock_iothread and not kicking it in qemu_cpu_kick. We also need to disable RAM block reordering until we have a more efficient locking mechanism at hand. Still, a Linux x86 UP guest and my Musicpal ARM model boot fine here. These numbers demonstrate where we gain something: 20338 jan 20 0 331m 75m 6904 R 99 0.9 0:50.95 qemu-system-arm 20337 jan 20 0 331m 75m 6904 S 20 0.9 0:26.50 qemu-system-arm The guest CPU was fully loaded, but the iothread could still run mostly independent on a second core. Without the patch we don't get beyond 32206 jan 20 0 330m 73m 7036 R 82 0.9 1:06.00 qemu-system-arm 32204 jan 20 0 330m 73m 7036 S 21 0.9 0:17.03 qemu-system-arm We don't benefit significantly, though, when the guest is not fully loading a host CPU. Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> Message-Id: <1439220437-23957-10-git-send-email-fred.konrad@greensocs.com> [FK: Rebase, fix qemu_devices_reset deadlock, rm address_space_* mutex] Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com> [EGC: fixed iothread lock for cpu-exec IRQ handling] Signed-off-by: Emilio G. Cota <cota@braap.org> [AJB: -smp single-threaded fix, clean commit msg, BQL fixes] Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Pranith Kumar <bobby.prani@gmail.com> [PM: target-arm changes] Acked-by: Peter Maydell <peter.maydell@linaro.org>
2017-02-23 19:29:11 +01:00
/* If we exit via cpu_loop_exit/longjmp it is reset in cpu_exec */
qemu_mutex_unlock_iothread();
}
tcg: drop global lock during TCG code execution This finally allows TCG to benefit from the iothread introduction: Drop the global mutex while running pure TCG CPU code. Reacquire the lock when entering MMIO or PIO emulation, or when leaving the TCG loop. We have to revert a few optimization for the current TCG threading model, namely kicking the TCG thread in qemu_mutex_lock_iothread and not kicking it in qemu_cpu_kick. We also need to disable RAM block reordering until we have a more efficient locking mechanism at hand. Still, a Linux x86 UP guest and my Musicpal ARM model boot fine here. These numbers demonstrate where we gain something: 20338 jan 20 0 331m 75m 6904 R 99 0.9 0:50.95 qemu-system-arm 20337 jan 20 0 331m 75m 6904 S 20 0.9 0:26.50 qemu-system-arm The guest CPU was fully loaded, but the iothread could still run mostly independent on a second core. Without the patch we don't get beyond 32206 jan 20 0 330m 73m 7036 R 82 0.9 1:06.00 qemu-system-arm 32204 jan 20 0 330m 73m 7036 S 21 0.9 0:17.03 qemu-system-arm We don't benefit significantly, though, when the guest is not fully loading a host CPU. Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> Message-Id: <1439220437-23957-10-git-send-email-fred.konrad@greensocs.com> [FK: Rebase, fix qemu_devices_reset deadlock, rm address_space_* mutex] Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com> [EGC: fixed iothread lock for cpu-exec IRQ handling] Signed-off-by: Emilio G. Cota <cota@braap.org> [AJB: -smp single-threaded fix, clean commit msg, BQL fixes] Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Pranith Kumar <bobby.prani@gmail.com> [PM: target-arm changes] Acked-by: Peter Maydell <peter.maydell@linaro.org>
2017-02-23 19:29:11 +01:00
/* Finally, check if we need to exit to the main loop. */
if (unlikely(qatomic_read(&cpu->exit_request))
|| (icount_enabled()
&& cpu_neg(cpu)->icount_decr.u16.low + cpu->icount_extra == 0)) {
qatomic_set(&cpu->exit_request, 0);
if (cpu->exception_index == -1) {
cpu->exception_index = EXCP_INTERRUPT;
}
return true;
}
return false;
}
static inline void cpu_loop_exec_tb(CPUState *cpu, TranslationBlock *tb,
TranslationBlock **last_tb, int *tb_exit)
{
uintptr_t ret;
int32_t insns_left;
trace_exec_tb(tb, tb->pc);
ret = cpu_tb_exec(cpu, tb);
tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
*tb_exit = ret & TB_EXIT_MASK;
if (*tb_exit != TB_EXIT_REQUESTED) {
*last_tb = tb;
return;
}
*last_tb = NULL;
insns_left = qatomic_read(&cpu_neg(cpu)->icount_decr.u32);
if (insns_left < 0) {
/* Something asked us to stop executing chained TBs; just
* continue round the main loop. Whatever requested the exit
* will also have set something else (eg exit_request or
* interrupt_request) which will be handled by
* cpu_handle_interrupt. cpu_handle_interrupt will also
* clear cpu->icount_decr.u16.high.
*/
return;
}
/* Instruction counter expired. */
assert(icount_enabled());
#ifndef CONFIG_USER_ONLY
/* Ensure global icount has gone forward */
cpu_update_icount(cpu);
/* Refill decrementer and continue execution. */
insns_left = MIN(0xffff, cpu->icount_budget);
cpu_neg(cpu)->icount_decr.u16.low = insns_left;
cpu->icount_extra = cpu->icount_budget - insns_left;
if (!cpu->icount_extra) {
/* Execute any remaining instructions, then let the main loop
* handle the next event.
*/
if (insns_left > 0) {
cpu_exec_nocache(cpu, insns_left, tb, false);
}
}
#endif
}
/* main execution loop */
int cpu_exec(CPUState *cpu)
{
CPUClass *cc = CPU_GET_CLASS(cpu);
int ret;
SyncClocks sc = { 0 };
cpu-exec: Add sleeping algorithm The goal is to sleep qemu whenever the guest clock is in advance compared to the host clock (we use the monotonic clocks). The amount of time to sleep is calculated in the execution loop in cpu_exec. At first, we tried to approximate at each for loop the real time elapsed while searching for a TB (generating or retrieving from cache) and executing it. We would then approximate the virtual time corresponding to the number of virtual instructions executed. The difference between these 2 values would allow us to know if the guest is in advance or delayed. However, the function used for measuring the real time (qemu_clock_get_ns(QEMU_CLOCK_REALTIME)) proved to be very expensive. We had an added overhead of 13% of the total run time. Therefore, we modified the algorithm and only take into account the difference between the 2 clocks at the begining of the cpu_exec function. During the for loop we try to reduce the advance of the guest only by computing the virtual time elapsed and sleeping if necessary. The overhead is thus reduced to 3%. Even though this method still has a noticeable overhead, it no longer is a bottleneck in trying to achieve a better guest frequency for which the guest clock is faster than the host one. As for the the alignement of the 2 clocks, with the first algorithm the guest clock was oscillating between -1 and 1ms compared to the host clock. Using the second algorithm we notice that the guest is 5ms behind the host, which is still acceptable for our use case. The tests where conducted using fio and stress. The host machine in an i5 CPU at 3.10GHz running Debian Jessie (kernel 3.12). The guest machine is an arm versatile-pb built with buildroot. Currently, on our test machine, the lowest icount we can achieve that is suitable for aligning the 2 clocks is 6. However, we observe that the IO tests (using fio) are slower than the cpu tests (using stress). Signed-off-by: Sebastian Tanase <sebastian.tanase@openwide.fr> Tested-by: Camille Bégué <camille.begue@openwide.fr> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-07-25 11:56:31 +02:00
/* replay_interrupt may need current_cpu */
current_cpu = cpu;
if (cpu_handle_halt(cpu)) {
return EXCP_HALTED;
}
rcu_read_lock();
cc->cpu_exec_enter(cpu);
cpu-exec: Add sleeping algorithm The goal is to sleep qemu whenever the guest clock is in advance compared to the host clock (we use the monotonic clocks). The amount of time to sleep is calculated in the execution loop in cpu_exec. At first, we tried to approximate at each for loop the real time elapsed while searching for a TB (generating or retrieving from cache) and executing it. We would then approximate the virtual time corresponding to the number of virtual instructions executed. The difference between these 2 values would allow us to know if the guest is in advance or delayed. However, the function used for measuring the real time (qemu_clock_get_ns(QEMU_CLOCK_REALTIME)) proved to be very expensive. We had an added overhead of 13% of the total run time. Therefore, we modified the algorithm and only take into account the difference between the 2 clocks at the begining of the cpu_exec function. During the for loop we try to reduce the advance of the guest only by computing the virtual time elapsed and sleeping if necessary. The overhead is thus reduced to 3%. Even though this method still has a noticeable overhead, it no longer is a bottleneck in trying to achieve a better guest frequency for which the guest clock is faster than the host one. As for the the alignement of the 2 clocks, with the first algorithm the guest clock was oscillating between -1 and 1ms compared to the host clock. Using the second algorithm we notice that the guest is 5ms behind the host, which is still acceptable for our use case. The tests where conducted using fio and stress. The host machine in an i5 CPU at 3.10GHz running Debian Jessie (kernel 3.12). The guest machine is an arm versatile-pb built with buildroot. Currently, on our test machine, the lowest icount we can achieve that is suitable for aligning the 2 clocks is 6. However, we observe that the IO tests (using fio) are slower than the cpu tests (using stress). Signed-off-by: Sebastian Tanase <sebastian.tanase@openwide.fr> Tested-by: Camille Bégué <camille.begue@openwide.fr> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-07-25 11:56:31 +02:00
/* Calculate difference between guest clock and host clock.
* This delay includes the delay of the last cycle, so
* what we have to do is sleep until it is 0. As for the
* advance/delay we gain here, we try to fix it next time.
*/
init_delay_params(&sc, cpu);
/* prepare setjmp context for exception handling */
if (sigsetjmp(cpu->jmp_env, 0) != 0) {
#if defined(__clang__) || !QEMU_GNUC_PREREQ(4, 6)
/* Some compilers wrongly smash all local variables after
* siglongjmp. There were bug reports for gcc 4.5.0 and clang.
* Reload essential local variables here for those compilers.
* Newer versions of gcc would complain about this code (-Wclobbered). */
cpu = current_cpu;
cc = CPU_GET_CLASS(cpu);
#else /* buggy compiler */
/* Assert that the compiler does not smash local variables. */
g_assert(cpu == current_cpu);
g_assert(cc == CPU_GET_CLASS(cpu));
#endif /* buggy compiler */
tcg: remove tb_lock Use mmap_lock in user-mode to protect TCG state and the page descriptors. In !user-mode, each vCPU has its own TCG state, so no locks needed. Per-page locks are used to protect the page descriptors. Per-TB locks are used in both modes to protect TB jumps. Some notes: - tb_lock is removed from notdirty_mem_write by passing a locked page_collection to tb_invalidate_phys_page_fast. - tcg_tb_lookup/remove/insert/etc have their own internal lock(s), so there is no need to further serialize access to them. - do_tb_flush is run in a safe async context, meaning no other vCPU threads are running. Therefore acquiring mmap_lock there is just to please tools such as thread sanitizer. - Not visible in the diff, but tb_invalidate_phys_page already has an assert_memory_lock. - cpu_io_recompile is !user-only, so no mmap_lock there. - Added mmap_unlock()'s before all siglongjmp's that could be called in user-mode while mmap_lock is held. + Added an assert for !have_mmap_lock() after returning from the longjmp in cpu_exec, just like we do in cpu_exec_step_atomic. Performance numbers before/after: Host: AMD Opteron(tm) Processor 6376 ubuntu 17.04 ppc64 bootup+shutdown time 700 +-+--+----+------+------------+-----------+------------*--+-+ | + + + + + *B | | before ***B*** ** * | |tb lock removal ###D### *** | 600 +-+ *** +-+ | ** # | | *B* #D | | *** * ## | 500 +-+ *** ### +-+ | * *** ### | | *B* # ## | | ** * #D# | 400 +-+ ** ## +-+ | ** ### | | ** ## | | ** # ## | 300 +-+ * B* #D# +-+ | B *** ### | | * ** #### | | * *** ### | 200 +-+ B *B #D# +-+ | #B* * ## # | | #* ## | | + D##D# + + + + | 100 +-+--+----+------+------------+-----------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/HwmBHXe debian jessie aarch64 bootup+shutdown time 90 +-+--+-----+-----+------------+------------+------------+--+-+ | + + + + + + | | before ***B*** B | 80 +tb lock removal ###D### **D +-+ | **### | | **## | 70 +-+ ** # +-+ | ** ## | | ** # | 60 +-+ *B ## +-+ | ** ## | | *** #D | 50 +-+ *** ## +-+ | * ** ### | | **B* ### | 40 +-+ **** # ## +-+ | **** #D# | | ***B** ### | 30 +-+ B***B** #### +-+ | B * * # ### | | B ###D# | 20 +-+ D ##D## +-+ | D# | | + + + + + + | 10 +-+--+-----+-----+------------+------------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/iGpGFtv The gains are high for 4-8 CPUs. Beyond that point, however, unrelated lock contention significantly hurts scalability. Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-05 05:46:31 +02:00
#ifndef CONFIG_SOFTMMU
tcg_debug_assert(!have_mmap_lock());
#endif
tcg: drop global lock during TCG code execution This finally allows TCG to benefit from the iothread introduction: Drop the global mutex while running pure TCG CPU code. Reacquire the lock when entering MMIO or PIO emulation, or when leaving the TCG loop. We have to revert a few optimization for the current TCG threading model, namely kicking the TCG thread in qemu_mutex_lock_iothread and not kicking it in qemu_cpu_kick. We also need to disable RAM block reordering until we have a more efficient locking mechanism at hand. Still, a Linux x86 UP guest and my Musicpal ARM model boot fine here. These numbers demonstrate where we gain something: 20338 jan 20 0 331m 75m 6904 R 99 0.9 0:50.95 qemu-system-arm 20337 jan 20 0 331m 75m 6904 S 20 0.9 0:26.50 qemu-system-arm The guest CPU was fully loaded, but the iothread could still run mostly independent on a second core. Without the patch we don't get beyond 32206 jan 20 0 330m 73m 7036 R 82 0.9 1:06.00 qemu-system-arm 32204 jan 20 0 330m 73m 7036 S 21 0.9 0:17.03 qemu-system-arm We don't benefit significantly, though, when the guest is not fully loading a host CPU. Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> Message-Id: <1439220437-23957-10-git-send-email-fred.konrad@greensocs.com> [FK: Rebase, fix qemu_devices_reset deadlock, rm address_space_* mutex] Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com> [EGC: fixed iothread lock for cpu-exec IRQ handling] Signed-off-by: Emilio G. Cota <cota@braap.org> [AJB: -smp single-threaded fix, clean commit msg, BQL fixes] Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Pranith Kumar <bobby.prani@gmail.com> [PM: target-arm changes] Acked-by: Peter Maydell <peter.maydell@linaro.org>
2017-02-23 19:29:11 +01:00
if (qemu_mutex_iothread_locked()) {
qemu_mutex_unlock_iothread();
}
qemu_plugin_disable_mem_helpers(cpu);
assert_no_pages_locked();
}
/* if an exception is pending, we execute it here */
while (!cpu_handle_exception(cpu, &ret)) {
TranslationBlock *last_tb = NULL;
int tb_exit = 0;
while (!cpu_handle_interrupt(cpu, &last_tb)) {
uint32_t cflags = cpu->cflags_next_tb;
TranslationBlock *tb;
/* When requested, use an exact setting for cflags for the next
execution. This is used for icount, precise smc, and stop-
after-access watchpoints. Since this request should never
have CF_INVALID set, -1 is a convenient invalid value that
does not require tcg headers for cpu_common_reset. */
if (cflags == -1) {
cflags = curr_cflags();
} else {
cpu->cflags_next_tb = -1;
}
tb = tb_find(cpu, last_tb, tb_exit, cflags);
cpu_loop_exec_tb(cpu, tb, &last_tb, &tb_exit);
/* Try to align the host and virtual clocks
if the guest is in advance */
align_clocks(&sc, cpu);
}
}
cc->cpu_exec_exit(cpu);
rcu_read_unlock();
return ret;
}
#ifndef CONFIG_USER_ONLY
void dump_drift_info(void)
{
if (!icount_enabled()) {
return;
}
qemu_printf("Host - Guest clock %"PRIi64" ms\n",
(cpu_get_clock() - cpu_get_icount()) / SCALE_MS);
if (icount_align_option) {
qemu_printf("Max guest delay %"PRIi64" ms\n",
-max_delay / SCALE_MS);
qemu_printf("Max guest advance %"PRIi64" ms\n",
max_advance / SCALE_MS);
} else {
qemu_printf("Max guest delay NA\n");
qemu_printf("Max guest advance NA\n");
}
}
#endif /* !CONFIG_USER_ONLY */