exec: RCUify AddressSpaceDispatch

Note that even after this patch, most callers of address_space_*
functions must still be under the big QEMU lock, otherwise the memory
region returned by address_space_translate can disappear as soon as
address_space_translate returns.  This will be fixed in the next part
of this series.

Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

diff --git a/cpu-exec.c b/cpu-exec.c
index 98f968df607832b31a3eb081eb9f66681ac736e7..adb939a994496d2f13e6e5d91483759c0afcb8f4 100644 (file)
--- a/cpu-exec.c
+++ b/cpu-exec.c
@@ -26,6 +26,7 @@
 #include "qemu/timer.h"
 #include "exec/address-spaces.h"
 #include "exec/memory-internal.h"
+#include "qemu/rcu.h"
 
 /* -icount align implementation. */
 
@@ -146,8 +147,27 @@ void cpu_resume_from_signal(CPUState *cpu, void *puc)
 
 void cpu_reload_memory_map(CPUState *cpu)
 {
+    AddressSpaceDispatch *d;
+
+    if (qemu_in_vcpu_thread()) {
+        /* Do not let the guest prolong the critical section as much as it
+         * as it desires.
+         *
+         * Currently, this is prevented by the I/O thread's periodinc kicking
+         * of the VCPU thread (iothread_requesting_mutex, qemu_cpu_kick_thread)
+         * but this will go away once TCG's execution moves out of the global
+         * mutex.
+         *
+         * This pair matches cpu_exec's rcu_read_lock()/rcu_read_unlock(), which
+         * only protects cpu->as->dispatch.  Since we reload it below, we can
+         * split the critical section.
+         */
+        rcu_read_unlock();
+        rcu_read_lock();
+    }
+
     /* The CPU and TLB are protected by the iothread lock.  */
-    AddressSpaceDispatch *d = cpu->as->dispatch;
+    d = atomic_rcu_read(&cpu->as->dispatch);
     cpu->memory_dispatch = d;
     tlb_flush(cpu, 1);
 }
@@ -362,6 +382,8 @@ int cpu_exec(CPUArchState *env)
      * an instruction scheduling constraint on modern architectures.  */
     smp_mb();
 
+    rcu_read_lock();
+
     if (unlikely(exit_request)) {
         cpu->exit_request = 1;
     }
@@ -564,6 +586,7 @@ int cpu_exec(CPUArchState *env)
     } /* for(;;) */
 
     cc->cpu_exec_exit(cpu);
+    rcu_read_unlock();
 
     /* fail safe : never use current_cpu outside cpu_exec() */
     current_cpu = NULL;

diff --git a/cpus.c b/cpus.c
index 0cdd1d71560bec11c4bd8eed51762a72289ef9a6..b826fac09cfb1d1885c0009edbb97d34fd4d8c6a 100644 (file)
--- a/cpus.c
+++ b/cpus.c
@@ -1104,7 +1104,7 @@ bool qemu_cpu_is_self(CPUState *cpu)
     return qemu_thread_is_self(cpu->thread);
 }
 
-static bool qemu_in_vcpu_thread(void)
+bool qemu_in_vcpu_thread(void)
 {
     return current_cpu && qemu_cpu_is_self(current_cpu);
 }

diff --git a/cputlb.c b/cputlb.c
index f92db5e18341fa78195b1ca3153249a6dda703d3..38f21511666171b7e18f20b1575f9fe5a11903a2 100644 (file)
--- a/cputlb.c
+++ b/cputlb.c
@@ -243,8 +243,12 @@ static void tlb_add_large_page(CPUArchState *env, target_ulong vaddr,
 }
 
 /* Add a new TLB entry. At most one entry for a given virtual address
-   is permitted. Only a single TARGET_PAGE_SIZE region is mapped, the
-   supplied size is only used by tlb_flush_page.  */
+ * is permitted. Only a single TARGET_PAGE_SIZE region is mapped, the
+ * supplied size is only used by tlb_flush_page.
+ *
+ * Called from TCG-generated code, which is under an RCU read-side
+ * critical section.
+ */
 void tlb_set_page(CPUState *cpu, target_ulong vaddr,
                   hwaddr paddr, int prot,
                   int mmu_idx, target_ulong size)

diff --git a/exec.c b/exec.c
index fe1e60a3b893e5aa28575f277cf56b2af3d00967..76b3b6cfe4c9c824a30f9db3b03e513a99b72deb 100644 (file)
--- a/exec.c
+++ b/exec.c
@@ -115,6 +115,8 @@ struct PhysPageEntry {
 typedef PhysPageEntry Node[P_L2_SIZE];
 
 typedef struct PhysPageMap {
+    struct rcu_head rcu;
+
     unsigned sections_nb;
     unsigned sections_nb_alloc;
     unsigned nodes_nb;
@@ -124,6 +126,8 @@ typedef struct PhysPageMap {
 } PhysPageMap;
 
 struct AddressSpaceDispatch {
+    struct rcu_head rcu;
+
     /* This is a multi-level map on the physical address space.
      * The bottom level has pointers to MemoryRegionSections.
      */
@@ -315,6 +319,7 @@ bool memory_region_is_unassigned(MemoryRegion *mr)
         && mr != &io_mem_watch;
 }
 
+/* Called from RCU critical section */
 static MemoryRegionSection *address_space_lookup_region(AddressSpaceDispatch *d,
                                                         hwaddr addr,
                                                         bool resolve_subpage)
@@ -330,6 +335,7 @@ static MemoryRegionSection *address_space_lookup_region(AddressSpaceDispatch *d,
     return section;
 }
 
+/* Called from RCU critical section */
 static MemoryRegionSection *
 address_space_translate_internal(AddressSpaceDispatch *d, hwaddr addr, hwaddr *xlat,
                                  hwaddr *plen, bool resolve_subpage)
@@ -370,8 +376,10 @@ MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
     MemoryRegion *mr;
     hwaddr len = *plen;
 
+    rcu_read_lock();
     for (;;) {
-        section = address_space_translate_internal(as->dispatch, addr, &addr, plen, true);
+        AddressSpaceDispatch *d = atomic_rcu_read(&as->dispatch);
+        section = address_space_translate_internal(d, addr, &addr, plen, true);
         mr = section->mr;
 
         if (!mr->iommu_ops) {
@@ -397,9 +405,11 @@ MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
 
     *plen = len;
     *xlat = addr;
+    rcu_read_unlock();
     return mr;
 }
 
+/* Called from RCU critical section */
 MemoryRegionSection *
 address_space_translate_for_iotlb(CPUState *cpu, hwaddr addr,
                                   hwaddr *xlat, hwaddr *plen)
@@ -852,6 +862,7 @@ static void cpu_physical_memory_set_dirty_tracking(bool enable)
     in_migration = enable;
 }
 
+/* Called from RCU critical section */
 hwaddr memory_region_section_get_iotlb(CPUState *cpu,
                                        MemoryRegionSection *section,
                                        target_ulong vaddr,
@@ -1964,7 +1975,8 @@ static uint16_t dummy_section(PhysPageMap *map, AddressSpace *as,
 
 MemoryRegion *iotlb_to_region(CPUState *cpu, hwaddr index)
 {
-    MemoryRegionSection *sections = cpu->memory_dispatch->map.sections;
+    AddressSpaceDispatch *d = atomic_rcu_read(&cpu->memory_dispatch);
+    MemoryRegionSection *sections = d->map.sections;
 
     return sections[index & ~TARGET_PAGE_MASK].mr;
 }
@@ -2000,6 +2012,12 @@ static void mem_begin(MemoryListener *listener)
     as->next_dispatch = d;
 }
 
+static void address_space_dispatch_free(AddressSpaceDispatch *d)
+{
+    phys_sections_free(&d->map);
+    g_free(d);
+}
+
 static void mem_commit(MemoryListener *listener)
 {
     AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
@@ -2008,11 +2026,9 @@ static void mem_commit(MemoryListener *listener)
 
     phys_page_compact_all(next, next->map.nodes_nb);
 
-    as->dispatch = next;
-
+    atomic_rcu_set(&as->dispatch, next);
     if (cur) {
-        phys_sections_free(&cur->map);
-        g_free(cur);
+        call_rcu(cur, address_space_dispatch_free, rcu);
     }
 }
 
@@ -2071,8 +2087,10 @@ void address_space_destroy_dispatch(AddressSpace *as)
 {
     AddressSpaceDispatch *d = as->dispatch;
 
-    g_free(d);
-    as->dispatch = NULL;
+    atomic_rcu_set(&as->dispatch, NULL);
+    if (d) {
+        call_rcu(d, address_space_dispatch_free, rcu);
+    }
 }
 
 static void memory_map_init(void)

diff --git a/hw/i386/intel_iommu.c b/hw/i386/intel_iommu.c
index 0a4282adf300007533cd47de15e84b1934b9ea7d..7da70ff3492e89f7d3c44b690f728278a953803a 100644 (file)
--- a/hw/i386/intel_iommu.c
+++ b/hw/i386/intel_iommu.c
@@ -745,6 +745,9 @@ static inline bool vtd_is_interrupt_addr(hwaddr addr)
 
 /* Map dev to context-entry then do a paging-structures walk to do a iommu
  * translation.
+ *
+ * Called from RCU critical section.
+ *
  * @bus_num: The bus number
  * @devfn: The devfn, which is the  combined of device and function number
  * @is_write: The access is a write operation

diff --git a/hw/pci-host/apb.c b/hw/pci-host/apb.c
index f573875baf9e36511ce106af21bbe880afa362d0..832b6c72483784b9cff47c10fe54c4ffe913f408 100644 (file)
--- a/hw/pci-host/apb.c
+++ b/hw/pci-host/apb.c
@@ -205,6 +205,7 @@ static AddressSpace *pbm_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
     return &is->iommu_as;
 }
 
+/* Called from RCU critical section */
 static IOMMUTLBEntry pbm_translate_iommu(MemoryRegion *iommu, hwaddr addr,
                                          bool is_write)
 {

diff --git a/hw/ppc/spapr_iommu.c b/hw/ppc/spapr_iommu.c
index da474740c056ef21c50396da252ec7e776d3d750..ba003da39efb996309e3975c1d1d29f3f9cf44dd 100644 (file)
--- a/hw/ppc/spapr_iommu.c
+++ b/hw/ppc/spapr_iommu.c
@@ -59,6 +59,7 @@ static sPAPRTCETable *spapr_tce_find_by_liobn(uint32_t liobn)
     return NULL;
 }
 
+/* Called from RCU critical section */
 static IOMMUTLBEntry spapr_tce_translate_iommu(MemoryRegion *iommu, hwaddr addr,
                                                bool is_write)
 {

diff --git a/include/exec/exec-all.h b/include/exec/exec-all.h
index bb3fd37dc653e2a1c13504cf2fe5d37e678fb567..8eb0db3910e8611b339af66778b9c3c40d85921f 100644 (file)
--- a/include/exec/exec-all.h
+++ b/include/exec/exec-all.h
@@ -96,6 +96,7 @@ void tb_invalidate_phys_page_range(tb_page_addr_t start, tb_page_addr_t end,
 void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t end,
                               int is_cpu_write_access);
 #if !defined(CONFIG_USER_ONLY)
+bool qemu_in_vcpu_thread(void);
 void cpu_reload_memory_map(CPUState *cpu);
 void tcg_cpu_address_space_init(CPUState *cpu, AddressSpace *as);
 /* cputlb.c */