[Qemu-devel] [PATCH] exec: optimize lduw_phys and stw_phys

[Bernhard Kohl <bernhard.kohl@nsn.com>]

Implementation of the optimized code for these two functions.

This is necessary for virtio which reads and writes VirtQueue index
fields using these functions. The assumption is that this are atomic
operations, which is not the case, if the memcpy() function which is
used in the non optimized code does single byte copying. This happens
for example with an older WindRiver glibc.

Signed-off-by: Bernhard Kohl <bernhard.kohl@nsn.com>
---
  exec.c |   67 
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++------
  1 files changed, 60 insertions(+), 7 deletions(-)

diff --git a/exec.c b/exec.c
index 14d1fd7..fb40398 100644
--- a/exec.c
+++ b/exec.c
@@ -3739,12 +3739,36 @@ uint32_t ldub_phys(target_phys_addr_t addr)
      return val;
  }

-/* XXX: optimize */
+/* warning: addr must be aligned */
  uint32_t lduw_phys(target_phys_addr_t addr)
  {
-    uint16_t val;
-    cpu_physical_memory_read(addr, (uint8_t *)&val, 2);
-    return tswap16(val);
+    int io_index;
+    uint8_t *ptr;
+    uint32_t val;
+    unsigned long pd;
+    PhysPageDesc *p;
+
+    p = phys_page_find(addr >> TARGET_PAGE_BITS);
+    if (!p) {
+        pd = IO_MEM_UNASSIGNED;
+    } else {
+        pd = p->phys_offset;
+    }
+
+    if ((pd & ~TARGET_PAGE_MASK) > IO_MEM_ROM &&
+        !(pd & IO_MEM_ROMD)) {
+        /* I/O case */
+        io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
+        if (p)
+            addr = (addr & ~TARGET_PAGE_MASK) + p->region_offset;
+        val = io_mem_read[io_index][1](io_mem_opaque[io_index], addr);
+    } else {
+        /* RAM case */
+        ptr = qemu_get_ram_ptr(pd & TARGET_PAGE_MASK) +
+            (addr & ~TARGET_PAGE_MASK);
+        val = lduw_p(ptr);
+    }
+    return val;
  }

  /* warning: addr must be aligned. The ram page is not masked as dirty
@@ -3861,11 +3885,40 @@ void stb_phys(target_phys_addr_t addr, uint32_t val)
      cpu_physical_memory_write(addr, &v, 1);
  }

-/* XXX: optimize */
+/* warning: addr must be aligned */
  void stw_phys(target_phys_addr_t addr, uint32_t val)
  {
-    uint16_t v = tswap16(val);
-    cpu_physical_memory_write(addr, (const uint8_t *)&v, 2);
+    int io_index;
+    uint8_t *ptr;
+    unsigned long pd;
+    PhysPageDesc *p;
+
+    p = phys_page_find(addr >> TARGET_PAGE_BITS);
+    if (!p) {
+        pd = IO_MEM_UNASSIGNED;
+    } else {
+        pd = p->phys_offset;
+    }
+
+    if ((pd & ~TARGET_PAGE_MASK) != IO_MEM_RAM) {
+        io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
+        if (p)
+            addr = (addr & ~TARGET_PAGE_MASK) + p->region_offset;
+        io_mem_write[io_index][1](io_mem_opaque[io_index], addr, val);
+    } else {
+        unsigned long addr1;
+        addr1 = (pd & TARGET_PAGE_MASK) + (addr & ~TARGET_PAGE_MASK);
+        /* RAM case */
+        ptr = qemu_get_ram_ptr(addr1);
+        stw_p(ptr, val);
+        if (!cpu_physical_memory_is_dirty(addr1)) {
+            /* invalidate code */
+            tb_invalidate_phys_page_range(addr1, addr1 + 2, 0);
+            /* set dirty bit */
+            phys_ram_dirty[addr1 >> TARGET_PAGE_BITS] |=
+                (0xff & ~CODE_DIRTY_FLAG);
+        }
+    }
  }

  /* XXX: optimize */
-- 
1.6.6.1