[Qemu-devel] [RFC PATCH V7 09/19] Drop global lock during TCG code execution

[fred . konrad]

From: KONRAD Frederic <address@hidden>

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.

I'm pretty sure some cases are still broken, definitely SMP (we no
longer perform round-robin scheduling "by chance"). 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.

Note that this patch depends on
http://thread.gmane.org/gmane.comp.emulators.qemu/118657

Changes from Fred Konrad:
  * Rebase on the current HEAD.
  * Fixes a deadlock in qemu_devices_reset().
  * Remove the mutex in address_space_*
---
 cpus.c                    | 20 +++-----------------
 cputlb.c                  |  5 +++++
 target-i386/misc_helper.c | 27 ++++++++++++++++++++++++---
 translate-all.c           |  2 ++
 vl.c                      |  6 ++++++
 5 files changed, 40 insertions(+), 20 deletions(-)

diff --git a/cpus.c b/cpus.c
index 2550be2..154a081 100644
--- a/cpus.c
+++ b/cpus.c
@@ -1232,23 +1232,7 @@ bool qemu_mutex_iothread_locked(void)
 
 void qemu_mutex_lock_iothread(void)
 {
-    atomic_inc(&iothread_requesting_mutex);
-    /* In the simple case there is no need to bump the VCPU thread out of
-     * TCG code execution.
-     */
-    if (!tcg_enabled() || qemu_in_vcpu_thread() ||
-        !first_cpu || !first_cpu->thread) {
-        qemu_mutex_lock(&qemu_global_mutex);
-        atomic_dec(&iothread_requesting_mutex);
-    } else {
-        if (qemu_mutex_trylock(&qemu_global_mutex)) {
-            qemu_cpu_kick_thread(first_cpu);
-            qemu_mutex_lock(&qemu_global_mutex);
-        }
-        atomic_dec(&iothread_requesting_mutex);
-        qemu_cond_broadcast(&qemu_io_proceeded_cond);
-    }
-    iothread_locked = true;
+    qemu_mutex_lock(&qemu_global_mutex);
 }
 
 void qemu_mutex_unlock_iothread(void)
@@ -1469,7 +1453,9 @@ static int tcg_cpu_exec(CPUState *cpu)
         cpu->icount_decr.u16.low = decr;
         cpu->icount_extra = count;
     }
+    qemu_mutex_unlock_iothread();
     ret = cpu_exec(cpu);
+    qemu_mutex_lock_iothread();
 #ifdef CONFIG_PROFILER
     tcg_time += profile_getclock() - ti;
 #endif
diff --git a/cputlb.c b/cputlb.c
index a506086..79fff1c 100644
--- a/cputlb.c
+++ b/cputlb.c
@@ -30,6 +30,9 @@
 #include "exec/ram_addr.h"
 #include "tcg/tcg.h"
 
+void qemu_mutex_lock_iothread(void);
+void qemu_mutex_unlock_iothread(void);
+
 //#define DEBUG_TLB
 //#define DEBUG_TLB_CHECK
 
@@ -125,8 +128,10 @@ void tlb_flush_page(CPUState *cpu, target_ulong addr)
    can be detected */
 void tlb_protect_code(ram_addr_t ram_addr)
 {
+    qemu_mutex_lock_iothread();
     cpu_physical_memory_test_and_clear_dirty(ram_addr, TARGET_PAGE_SIZE,
                                              DIRTY_MEMORY_CODE);
+    qemu_mutex_unlock_iothread();
 }
 
 /* update the TLB so that writes in physical page 'phys_addr' are no longer
diff --git a/target-i386/misc_helper.c b/target-i386/misc_helper.c
index 52c5d65..55f63bf 100644
--- a/target-i386/misc_helper.c
+++ b/target-i386/misc_helper.c
@@ -27,8 +27,10 @@ void helper_outb(CPUX86State *env, uint32_t port, uint32_t 
data)
 #ifdef CONFIG_USER_ONLY
     fprintf(stderr, "outb: port=0x%04x, data=%02x\n", port, data);
 #else
+    qemu_mutex_lock_iothread();
     address_space_stb(&address_space_io, port, data,
                       cpu_get_mem_attrs(env), NULL);
+    qemu_mutex_unlock_iothread();
 #endif
 }
 
@@ -38,8 +40,13 @@ target_ulong helper_inb(CPUX86State *env, uint32_t port)
     fprintf(stderr, "inb: port=0x%04x\n", port);
     return 0;
 #else
-    return address_space_ldub(&address_space_io, port,
+    target_ulong ret;
+
+    qemu_mutex_lock_iothread();
+    ret = address_space_ldub(&address_space_io, port,
                               cpu_get_mem_attrs(env), NULL);
+    qemu_mutex_unlock_iothread();
+    return ret;
 #endif
 }
 
@@ -48,8 +55,10 @@ void helper_outw(CPUX86State *env, uint32_t port, uint32_t 
data)
 #ifdef CONFIG_USER_ONLY
     fprintf(stderr, "outw: port=0x%04x, data=%04x\n", port, data);
 #else
+    qemu_mutex_lock_iothread();
     address_space_stw(&address_space_io, port, data,
                       cpu_get_mem_attrs(env), NULL);
+    qemu_mutex_unlock_iothread();
 #endif
 }
 
@@ -59,8 +68,13 @@ target_ulong helper_inw(CPUX86State *env, uint32_t port)
     fprintf(stderr, "inw: port=0x%04x\n", port);
     return 0;
 #else
-    return address_space_lduw(&address_space_io, port,
+    target_ulong ret;
+
+    qemu_mutex_lock_iothread();
+    ret = address_space_lduw(&address_space_io, port,
                               cpu_get_mem_attrs(env), NULL);
+    qemu_mutex_unlock_iothread();
+    return ret;
 #endif
 }
 
@@ -69,8 +83,10 @@ void helper_outl(CPUX86State *env, uint32_t port, uint32_t 
data)
 #ifdef CONFIG_USER_ONLY
     fprintf(stderr, "outw: port=0x%04x, data=%08x\n", port, data);
 #else
+    qemu_mutex_lock_iothread();
     address_space_stl(&address_space_io, port, data,
                       cpu_get_mem_attrs(env), NULL);
+    qemu_mutex_unlock_iothread();
 #endif
 }
 
@@ -80,8 +96,13 @@ target_ulong helper_inl(CPUX86State *env, uint32_t port)
     fprintf(stderr, "inl: port=0x%04x\n", port);
     return 0;
 #else
-    return address_space_ldl(&address_space_io, port,
+    target_ulong ret;
+
+    qemu_mutex_lock_iothread();
+    ret = address_space_ldl(&address_space_io, port,
                              cpu_get_mem_attrs(env), NULL);
+    qemu_mutex_unlock_iothread();
+    return ret;
 #endif
 }
 
diff --git a/translate-all.c b/translate-all.c
index 046565c..954c67a 100644
--- a/translate-all.c
+++ b/translate-all.c
@@ -1223,6 +1223,7 @@ void tb_invalidate_phys_page_range(tb_page_addr_t start, 
tb_page_addr_t end,
 #endif
 #ifdef TARGET_HAS_PRECISE_SMC
     if (current_tb_modified) {
+        qemu_mutex_unlock_iothread();
         /* we generate a block containing just the instruction
            modifying the memory. It will ensure that it cannot modify
            itself */
@@ -1327,6 +1328,7 @@ static void tb_invalidate_phys_page(tb_page_addr_t addr,
     p->first_tb = NULL;
 #ifdef TARGET_HAS_PRECISE_SMC
     if (current_tb_modified) {
+        qemu_mutex_unlock_iothread();
         /* we generate a block containing just the instruction
            modifying the memory. It will ensure that it cannot modify
            itself */
diff --git a/vl.c b/vl.c
index 3f269dc..922e969 100644
--- a/vl.c
+++ b/vl.c
@@ -1717,10 +1717,16 @@ void qemu_devices_reset(void)
 {
     QEMUResetEntry *re, *nre;
 
+    /*
+     * Some device's reset needs to grab the global_mutex. So just release it
+     * here.
+     */
+    qemu_mutex_unlock_iothread();
     /* reset all devices */
     QTAILQ_FOREACH_SAFE(re, &reset_handlers, entry, nre) {
         re->func(re->opaque);
     }
+    qemu_mutex_lock_iothread();
 }
 
 void qemu_system_reset(bool report)
-- 
1.9.0