cputlb: Move ROM handling from I/O path to TLB path

It does not require going through the whole I/O path
in order to discard a write.
Reviewed-by: NDavid Hildenbrand <david@redhat.com>
Signed-off-by: NRichard Henderson <richard.henderson@linaro.org>

accel/tcg/cputlb.c

@@ -577,7 +577,8 @@ static void tlb_reset_dirty_range_locked(CPUTLBEntry *tlb_entry,
 {
     uintptr_t addr = tlb_entry->addr_write;
 
-    if ((addr & (TLB_INVALID_MASK | TLB_MMIO | TLB_NOTDIRTY)) == 0) {
+    if ((addr & (TLB_INVALID_MASK | TLB_MMIO |
+                 TLB_DISCARD_WRITE | TLB_NOTDIRTY)) == 0) {
         addr &= TARGET_PAGE_MASK;
         addr += tlb_entry->addend;
         if ((addr - start) < length) {
@@ -745,7 +746,6 @@ void tlb_set_page_with_attrs(CPUState *cpu, target_ulong vaddr,
         address |= TLB_MMIO;
         addend = 0;
     } else {
-        /* TLB_MMIO for rom/romd handled below */
         addend = (uintptr_t)memory_region_get_ram_ptr(section->mr) + xlat;
     }
 
@@ -822,16 +822,17 @@ void tlb_set_page_with_attrs(CPUState *cpu, target_ulong vaddr,
 
     tn.addr_write = -1;
     if (prot & PAGE_WRITE) {
-        if ((memory_region_is_ram(section->mr) && section->readonly)
-            || memory_region_is_romd(section->mr)) {
-            /* Write access calls the I/O callback.  */
-            tn.addr_write = address | TLB_MMIO;
-        } else if (memory_region_is_ram(section->mr)
-                   && cpu_physical_memory_is_clean(
-                       memory_region_get_ram_addr(section->mr) + xlat)) {
-            tn.addr_write = address | TLB_NOTDIRTY;
-        } else {
-            tn.addr_write = address;
+        tn.addr_write = address;
+        if (memory_region_is_romd(section->mr)) {
+            /* Use the MMIO path so that the device can switch states. */
+            tn.addr_write |= TLB_MMIO;
+        } else if (memory_region_is_ram(section->mr)) {
+            if (section->readonly) {
+                tn.addr_write |= TLB_DISCARD_WRITE;
+            } else if (cpu_physical_memory_is_clean(
+                        memory_region_get_ram_addr(section->mr) + xlat)) {
+                tn.addr_write |= TLB_NOTDIRTY;
+            }
         }
         if (prot & PAGE_WRITE_INV) {
             tn.addr_write |= TLB_INVALID_MASK;
@@ -904,7 +905,7 @@ static uint64_t io_readx(CPUArchState *env, CPUIOTLBEntry *iotlbentry,
     mr = section->mr;
     mr_offset = (iotlbentry->addr & TARGET_PAGE_MASK) + addr;
     cpu->mem_io_pc = retaddr;
-    if (mr != &io_mem_rom && mr != &io_mem_notdirty && !cpu->can_do_io) {
+    if (mr != &io_mem_notdirty && !cpu->can_do_io) {
         cpu_io_recompile(cpu, retaddr);
     }
 
@@ -945,7 +946,7 @@ static void io_writex(CPUArchState *env, CPUIOTLBEntry *iotlbentry,
     section = iotlb_to_section(cpu, iotlbentry->addr, iotlbentry->attrs);
     mr = section->mr;
     mr_offset = (iotlbentry->addr & TARGET_PAGE_MASK) + addr;
-    if (mr != &io_mem_rom && mr != &io_mem_notdirty && !cpu->can_do_io) {
+    if (mr != &io_mem_notdirty && !cpu->can_do_io) {
         cpu_io_recompile(cpu, retaddr);
     }
     cpu->mem_io_vaddr = addr;
@@ -1125,7 +1126,7 @@ void *probe_access(CPUArchState *env, target_ulong addr, int size,
     }
 
     /* Reject I/O access, or other required slow-path.  */
-    if (tlb_addr & (TLB_NOTDIRTY | TLB_MMIO | TLB_BSWAP)) {
+    if (tlb_addr & (TLB_NOTDIRTY | TLB_MMIO | TLB_BSWAP | TLB_DISCARD_WRITE)) {
         return NULL;
     }
 
@@ -1617,6 +1618,11 @@ store_helper(CPUArchState *env, target_ulong addr, uint64_t val,
             return;
         }
 
+        /* Ignore writes to ROM.  */
+        if (unlikely(tlb_addr & TLB_DISCARD_WRITE)) {
+            return;
+        }
+
         haddr = (void *)((uintptr_t)addr + entry->addend);
 
         /*

exec.c

@@ -88,7 +88,7 @@ static MemoryRegion *system_io;
 AddressSpace address_space_io;
 AddressSpace address_space_memory;
 
-MemoryRegion io_mem_rom, io_mem_notdirty;
+MemoryRegion io_mem_notdirty;
 static MemoryRegion io_mem_unassigned;
 #endif
 
@@ -192,7 +192,6 @@ typedef struct subpage_t {
 
 #define PHYS_SECTION_UNASSIGNED 0
 #define PHYS_SECTION_NOTDIRTY 1
-#define PHYS_SECTION_ROM 2
 
 static void io_mem_init(void);
 static void memory_map_init(void);
@@ -1475,8 +1474,6 @@ hwaddr memory_region_section_get_iotlb(CPUState *cpu,
         iotlb = memory_region_get_ram_addr(section->mr) + xlat;
         if (!section->readonly) {
             iotlb |= PHYS_SECTION_NOTDIRTY;
-        } else {
-            iotlb |= PHYS_SECTION_ROM;
         }
     } else {
         AddressSpaceDispatch *d;
@@ -3002,38 +2999,6 @@ static uint16_t dummy_section(PhysPageMap *map, FlatView *fv, MemoryRegion *mr)
     return phys_section_add(map, &section);
 }
 
-static void readonly_mem_write(void *opaque, hwaddr addr,
-                               uint64_t val, unsigned size)
-{
-    /* Ignore any write to ROM. */
-}
-
-static bool readonly_mem_accepts(void *opaque, hwaddr addr,
-                                 unsigned size, bool is_write,
-                                 MemTxAttrs attrs)
-{
-    return is_write;
-}
-
-/* This will only be used for writes, because reads are special cased
- * to directly access the underlying host ram.
- */
-static const MemoryRegionOps readonly_mem_ops = {
-    .write = readonly_mem_write,
-    .valid.accepts = readonly_mem_accepts,
-    .endianness = DEVICE_NATIVE_ENDIAN,
-    .valid = {
-        .min_access_size = 1,
-        .max_access_size = 8,
-        .unaligned = false,
-    },
-    .impl = {
-        .min_access_size = 1,
-        .max_access_size = 8,
-        .unaligned = false,
-    },
-};
-
 MemoryRegionSection *iotlb_to_section(CPUState *cpu,
                                       hwaddr index, MemTxAttrs attrs)
 {
@@ -3047,8 +3012,6 @@ MemoryRegionSection *iotlb_to_section(CPUState *cpu,
 
 static void io_mem_init(void)
 {
-    memory_region_init_io(&io_mem_rom, NULL, &readonly_mem_ops,
-                          NULL, NULL, UINT64_MAX);
     memory_region_init_io(&io_mem_unassigned, NULL, &unassigned_mem_ops, NULL,
                           NULL, UINT64_MAX);
 
@@ -3069,8 +3032,6 @@ AddressSpaceDispatch *address_space_dispatch_new(FlatView *fv)
     assert(n == PHYS_SECTION_UNASSIGNED);
     n = dummy_section(&d->map, fv, &io_mem_notdirty);
     assert(n == PHYS_SECTION_NOTDIRTY);
-    n = dummy_section(&d->map, fv, &io_mem_rom);
-    assert(n == PHYS_SECTION_ROM);
 
     d->phys_map  = (PhysPageEntry) { .ptr = PHYS_MAP_NODE_NIL, .skip = 1 };
 

include/exec/cpu-all.h

@@ -337,12 +337,15 @@ CPUArchState *cpu_copy(CPUArchState *env);
 #define TLB_WATCHPOINT      (1 << (TARGET_PAGE_BITS_MIN - 4))
 /* Set if TLB entry requires byte swap.  */
 #define TLB_BSWAP           (1 << (TARGET_PAGE_BITS_MIN - 5))
+/* Set if TLB entry writes ignored.  */
+#define TLB_DISCARD_WRITE   (1 << (TARGET_PAGE_BITS_MIN - 6))
 
 /* Use this mask to check interception with an alignment mask
  * in a TCG backend.
  */
 #define TLB_FLAGS_MASK \
-    (TLB_INVALID_MASK | TLB_NOTDIRTY | TLB_MMIO | TLB_WATCHPOINT | TLB_BSWAP)
+    (TLB_INVALID_MASK | TLB_NOTDIRTY | TLB_MMIO \
+    | TLB_WATCHPOINT | TLB_BSWAP | TLB_DISCARD_WRITE)
 
 /**
  * tlb_hit_page: return true if page aligned @addr is a hit against the

include/exec/cpu-common.h

@@ -100,7 +100,6 @@ void qemu_flush_coalesced_mmio_buffer(void);
 
 void cpu_flush_icache_range(hwaddr start, hwaddr len);
 
-extern struct MemoryRegion io_mem_rom;
 extern struct MemoryRegion io_mem_notdirty;
 
 typedef int (RAMBlockIterFunc)(RAMBlock *rb, void *opaque);