[Qemu-devel] [PULL 06/16] exec: move rcu_read_lock/unlock to address_space_translate callers

[Paolo Bonzini]

Once address_space_translate will be called outside the BQL, the returned
MemoryRegion might disappear as soon as the RCU read-side critical section
ends.  Avoid this by moving the critical section to the callers.

Signed-off-by: Paolo Bonzini <address@hidden>
Message-Id: <address@hidden>
---
 exec.c                | 33 +++++++++++++++++++++++++++++----
 hw/vfio/common.c      |  7 +++++--
 include/exec/memory.h |  4 +++-
 translate-all.c       |  3 +++
 4 files changed, 40 insertions(+), 7 deletions(-)

diff --git a/exec.c b/exec.c
index ae37b98..e19ab22 100644
--- a/exec.c
+++ b/exec.c
@@ -373,6 +373,7 @@ static inline bool memory_access_is_direct(MemoryRegion 
*mr, bool is_write)
     return false;
 }
 
+/* Called from RCU critical section */
 MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
                                       hwaddr *xlat, hwaddr *plen,
                                       bool is_write)
@@ -381,7 +382,6 @@ MemoryRegion *address_space_translate(AddressSpace *as, 
hwaddr addr,
     MemoryRegionSection *section;
     MemoryRegion *mr;
 
-    rcu_read_lock();
     for (;;) {
         AddressSpaceDispatch *d = atomic_rcu_read(&as->dispatch);
         section = address_space_translate_internal(d, addr, &addr, plen, true);
@@ -409,7 +409,6 @@ MemoryRegion *address_space_translate(AddressSpace *as, 
hwaddr addr,
     }
 
     *xlat = addr;
-    rcu_read_unlock();
     return mr;
 }
 
@@ -2329,6 +2328,7 @@ MemTxResult address_space_rw(AddressSpace *as, hwaddr 
addr, MemTxAttrs attrs,
     MemoryRegion *mr;
     MemTxResult result = MEMTX_OK;
 
+    rcu_read_lock();
     while (len > 0) {
         l = len;
         mr = address_space_translate(as, addr, &addr1, &l, is_write);
@@ -2415,6 +2415,7 @@ MemTxResult address_space_rw(AddressSpace *as, hwaddr 
addr, MemTxAttrs attrs,
         buf += l;
         addr += l;
     }
+    rcu_read_unlock();
 
     return result;
 }
@@ -2452,6 +2453,7 @@ static inline void 
cpu_physical_memory_write_rom_internal(AddressSpace *as,
     hwaddr addr1;
     MemoryRegion *mr;
 
+    rcu_read_lock();
     while (len > 0) {
         l = len;
         mr = address_space_translate(as, addr, &addr1, &l, true);
@@ -2477,6 +2479,7 @@ static inline void 
cpu_physical_memory_write_rom_internal(AddressSpace *as,
         buf += l;
         addr += l;
     }
+    rcu_read_unlock();
 }
 
 /* used for ROM loading : can write in RAM and ROM */
@@ -2585,6 +2588,7 @@ bool address_space_access_valid(AddressSpace *as, hwaddr 
addr, int len, bool is_
     MemoryRegion *mr;
     hwaddr l, xlat;
 
+    rcu_read_lock();
     while (len > 0) {
         l = len;
         mr = address_space_translate(as, addr, &xlat, &l, is_write);
@@ -2598,6 +2602,7 @@ bool address_space_access_valid(AddressSpace *as, hwaddr 
addr, int len, bool is_
         len -= l;
         addr += l;
     }
+    rcu_read_unlock();
     return true;
 }
 
@@ -2624,9 +2629,12 @@ void *address_space_map(AddressSpace *as,
     }
 
     l = len;
+    rcu_read_lock();
     mr = address_space_translate(as, addr, &xlat, &l, is_write);
+
     if (!memory_access_is_direct(mr, is_write)) {
         if (atomic_xchg(&bounce.in_use, true)) {
+            rcu_read_unlock();
             return NULL;
         }
         /* Avoid unbounded allocations */
@@ -2642,6 +2650,7 @@ void *address_space_map(AddressSpace *as,
                                bounce.buffer, l);
         }
 
+        rcu_read_unlock();
         *plen = l;
         return bounce.buffer;
     }
@@ -2665,6 +2674,7 @@ void *address_space_map(AddressSpace *as,
     }
 
     memory_region_ref(mr);
+    rcu_read_unlock();
     *plen = done;
     return qemu_ram_ptr_length(raddr + base, plen);
 }
@@ -2728,6 +2738,7 @@ static inline uint32_t 
address_space_ldl_internal(AddressSpace *as, hwaddr addr,
     hwaddr addr1;
     MemTxResult r;
 
+    rcu_read_lock();
     mr = address_space_translate(as, addr, &addr1, &l, false);
     if (l < 4 || !memory_access_is_direct(mr, false)) {
         /* I/O case */
@@ -2762,6 +2773,7 @@ static inline uint32_t 
address_space_ldl_internal(AddressSpace *as, hwaddr addr,
     if (result) {
         *result = r;
     }
+    rcu_read_unlock();
     return val;
 }
 
@@ -2814,6 +2826,7 @@ static inline uint64_t 
address_space_ldq_internal(AddressSpace *as, hwaddr addr,
     hwaddr addr1;
     MemTxResult r;
 
+    rcu_read_lock();
     mr = address_space_translate(as, addr, &addr1, &l,
                                  false);
     if (l < 8 || !memory_access_is_direct(mr, false)) {
@@ -2849,6 +2862,7 @@ static inline uint64_t 
address_space_ldq_internal(AddressSpace *as, hwaddr addr,
     if (result) {
         *result = r;
     }
+    rcu_read_unlock();
     return val;
 }
 
@@ -2921,6 +2935,7 @@ static inline uint32_t 
address_space_lduw_internal(AddressSpace *as,
     hwaddr addr1;
     MemTxResult r;
 
+    rcu_read_lock();
     mr = address_space_translate(as, addr, &addr1, &l,
                                  false);
     if (l < 2 || !memory_access_is_direct(mr, false)) {
@@ -2956,6 +2971,7 @@ static inline uint32_t 
address_space_lduw_internal(AddressSpace *as,
     if (result) {
         *result = r;
     }
+    rcu_read_unlock();
     return val;
 }
 
@@ -3007,6 +3023,7 @@ void address_space_stl_notdirty(AddressSpace *as, hwaddr 
addr, uint32_t val,
     hwaddr addr1;
     MemTxResult r;
 
+    rcu_read_lock();
     mr = address_space_translate(as, addr, &addr1, &l,
                                  true);
     if (l < 4 || !memory_access_is_direct(mr, true)) {
@@ -3029,6 +3046,7 @@ void address_space_stl_notdirty(AddressSpace *as, hwaddr 
addr, uint32_t val,
     if (result) {
         *result = r;
     }
+    rcu_read_unlock();
 }
 
 void stl_phys_notdirty(AddressSpace *as, hwaddr addr, uint32_t val)
@@ -3049,6 +3067,7 @@ static inline void 
address_space_stl_internal(AddressSpace *as,
     hwaddr addr1;
     MemTxResult r;
 
+    rcu_read_lock();
     mr = address_space_translate(as, addr, &addr1, &l,
                                  true);
     if (l < 4 || !memory_access_is_direct(mr, true)) {
@@ -3083,6 +3102,7 @@ static inline void 
address_space_stl_internal(AddressSpace *as,
     if (result) {
         *result = r;
     }
+    rcu_read_unlock();
 }
 
 void address_space_stl(AddressSpace *as, hwaddr addr, uint32_t val,
@@ -3152,6 +3172,7 @@ static inline void 
address_space_stw_internal(AddressSpace *as,
     hwaddr addr1;
     MemTxResult r;
 
+    rcu_read_lock();
     mr = address_space_translate(as, addr, &addr1, &l, true);
     if (l < 2 || !memory_access_is_direct(mr, true)) {
 #if defined(TARGET_WORDS_BIGENDIAN)
@@ -3185,6 +3206,7 @@ static inline void 
address_space_stw_internal(AddressSpace *as,
     if (result) {
         *result = r;
     }
+    rcu_read_unlock();
 }
 
 void address_space_stw(AddressSpace *as, hwaddr addr, uint32_t val,
@@ -3322,12 +3344,15 @@ bool cpu_physical_memory_is_io(hwaddr phys_addr)
 {
     MemoryRegion*mr;
     hwaddr l = 1;
+    bool res;
 
+    rcu_read_lock();
     mr = address_space_translate(&address_space_memory,
                                  phys_addr, &phys_addr, &l, false);
 
-    return !(memory_region_is_ram(mr) ||
-             memory_region_is_romd(mr));
+    res = !(memory_region_is_ram(mr) || memory_region_is_romd(mr));
+    rcu_read_unlock();
+    return res;
 }
 
 void qemu_ram_foreach_block(RAMBlockIterFunc func, void *opaque)
diff --git a/hw/vfio/common.c b/hw/vfio/common.c
index b012620..b1045da 100644
--- a/hw/vfio/common.c
+++ b/hw/vfio/common.c
@@ -270,13 +270,14 @@ static void vfio_iommu_map_notify(Notifier *n, void *data)
      * this IOMMU to its immediate target.  We need to translate
      * it the rest of the way through to memory.
      */
+    rcu_read_lock();
     mr = address_space_translate(&address_space_memory,
                                  iotlb->translated_addr,
                                  &xlat, &len, iotlb->perm & IOMMU_WO);
     if (!memory_region_is_ram(mr)) {
         error_report("iommu map to non memory area %"HWADDR_PRIx"",
                      xlat);
-        return;
+        goto out;
     }
     /*
      * Translation truncates length to the IOMMU page size,
@@ -284,7 +285,7 @@ static void vfio_iommu_map_notify(Notifier *n, void *data)
      */
     if (len & iotlb->addr_mask) {
         error_report("iommu has granularity incompatible with target AS");
-        return;
+        goto out;
     }
 
     if ((iotlb->perm & IOMMU_RW) != IOMMU_NONE) {
@@ -307,6 +308,8 @@ static void vfio_iommu_map_notify(Notifier *n, void *data)
                          iotlb->addr_mask + 1, ret);
         }
     }
+out:
+    rcu_read_unlock();
 }
 
 static void vfio_listener_region_add(MemoryListener *listener,
diff --git a/include/exec/memory.h b/include/exec/memory.h
index 0ccfd3b..b61c84f 100644
--- a/include/exec/memory.h
+++ b/include/exec/memory.h
@@ -1233,7 +1233,9 @@ void address_space_stq(AddressSpace *as, hwaddr addr, 
uint64_t val,
 #endif
 
 /* address_space_translate: translate an address range into an address space
- * into a MemoryRegion and an address range into that section
+ * into a MemoryRegion and an address range into that section.  Should be
+ * called from an RCU critical section, to avoid that the last reference
+ * to the returned region disappears after address_space_translate returns.
  *
  * @as: #AddressSpace to be accessed
  * @addr: address within that address space
diff --git a/translate-all.c b/translate-all.c
index 65a76c5..536008f 100644
--- a/translate-all.c
+++ b/translate-all.c
@@ -1416,14 +1416,17 @@ void tb_invalidate_phys_addr(AddressSpace *as, hwaddr 
addr)
     MemoryRegion *mr;
     hwaddr l = 1;
 
+    rcu_read_lock();
     mr = address_space_translate(as, addr, &addr, &l, false);
     if (!(memory_region_is_ram(mr)
           || memory_region_is_romd(mr))) {
+        rcu_read_unlock();
         return;
     }
     ram_addr = (memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK)
         + addr;
     tb_invalidate_phys_page_range(ram_addr, ram_addr + 1, 0);
+    rcu_read_unlock();
 }
 #endif /* !defined(CONFIG_USER_ONLY) */
 
-- 
2.3.5