accel/tcg: Use the alignment test in tlb_fill_align

When we have a tlb miss, defer the alignment check to
the new tlb_fill_align hook.  Move the existing alignment
check so that we only perform it with a tlb hit.

Reviewed-by: Helge Deller <deller@gmx.de>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

accel/tcg/cputlb.c

@@ -1620,6 +1620,7 @@ typedef struct MMULookupLocals {
  * mmu_lookup1: translate one page
  * @cpu: generic cpu state
  * @data: lookup parameters
+ * @memop: memory operation for the access, or 0
  * @mmu_idx: virtual address context
  * @access_type: load/store/code
  * @ra: return address into tcg generated code, or 0
@@ -1629,7 +1630,7 @@ typedef struct MMULookupLocals {
  * tlb_fill_align will longjmp out.  Return true if the softmmu tlb for
  * @mmu_idx may have resized.
  */
-static bool mmu_lookup1(CPUState *cpu, MMULookupPageData *data,
+static bool mmu_lookup1(CPUState *cpu, MMULookupPageData *data, MemOp memop,
                         int mmu_idx, MMUAccessType access_type, uintptr_t ra)
 {
     vaddr addr = data->addr;
@@ -1645,7 +1646,7 @@ static bool mmu_lookup1(CPUState *cpu, MMULookupPageData *data,
         if (!victim_tlb_hit(cpu, mmu_idx, index, access_type,
                             addr & TARGET_PAGE_MASK)) {
             tlb_fill_align(cpu, addr, access_type, mmu_idx,
-                           0, data->size, false, ra);
+                           memop, data->size, false, ra);
             maybe_resized = true;
             index = tlb_index(cpu, mmu_idx, addr);
             entry = tlb_entry(cpu, mmu_idx, addr);
@@ -1657,6 +1658,25 @@ static bool mmu_lookup1(CPUState *cpu, MMULookupPageData *data,
     flags = tlb_addr & (TLB_FLAGS_MASK & ~TLB_FORCE_SLOW);
     flags |= full->slow_flags[access_type];
 
+    if (likely(!maybe_resized)) {
+        /* Alignment has not been checked by tlb_fill_align. */
+        int a_bits = memop_alignment_bits(memop);
+
+        /*
+         * This alignment check differs from the one above, in that this is
+         * based on the atomicity of the operation. The intended use case is
+         * the ARM memory type field of each PTE, where access to pages with
+         * Device memory type require alignment.
+         */
+        if (unlikely(flags & TLB_CHECK_ALIGNED)) {
+            int at_bits = memop_atomicity_bits(memop);
+            a_bits = MAX(a_bits, at_bits);
+        }
+        if (unlikely(addr & ((1 << a_bits) - 1))) {
+            cpu_unaligned_access(cpu, addr, access_type, mmu_idx, ra);
+        }
+    }
+
     data->full = full;
     data->flags = flags;
     /* Compute haddr speculatively; depending on flags it might be invalid. */
@@ -1713,7 +1733,6 @@ static void mmu_watch_or_dirty(CPUState *cpu, MMULookupPageData *data,
 static bool mmu_lookup(CPUState *cpu, vaddr addr, MemOpIdx oi,
                        uintptr_t ra, MMUAccessType type, MMULookupLocals *l)
 {
-    unsigned a_bits;
     bool crosspage;
     int flags;
 
@@ -1722,12 +1741,6 @@ static bool mmu_lookup(CPUState *cpu, vaddr addr, MemOpIdx oi,
 
     tcg_debug_assert(l->mmu_idx < NB_MMU_MODES);
 
-    /* Handle CPU specific unaligned behaviour */
-    a_bits = memop_alignment_bits(l->memop);
-    if (addr & ((1 << a_bits) - 1)) {
-        cpu_unaligned_access(cpu, addr, type, l->mmu_idx, ra);
-    }
-
     l->page[0].addr = addr;
     l->page[0].size = memop_size(l->memop);
     l->page[1].addr = (addr + l->page[0].size - 1) & TARGET_PAGE_MASK;
@@ -1735,7 +1748,7 @@ static bool mmu_lookup(CPUState *cpu, vaddr addr, MemOpIdx oi,
     crosspage = (addr ^ l->page[1].addr) & TARGET_PAGE_MASK;
 
     if (likely(!crosspage)) {
-        mmu_lookup1(cpu, &l->page[0], l->mmu_idx, type, ra);
+        mmu_lookup1(cpu, &l->page[0], l->memop, l->mmu_idx, type, ra);
 
         flags = l->page[0].flags;
         if (unlikely(flags & (TLB_WATCHPOINT | TLB_NOTDIRTY))) {
@@ -1754,8 +1767,8 @@ static bool mmu_lookup(CPUState *cpu, vaddr addr, MemOpIdx oi,
          * Lookup both pages, recognizing exceptions from either.  If the
          * second lookup potentially resized, refresh first CPUTLBEntryFull.
          */
-        mmu_lookup1(cpu, &l->page[0], l->mmu_idx, type, ra);
+        mmu_lookup1(cpu, &l->page[0], l->memop, l->mmu_idx, type, ra);
-        if (mmu_lookup1(cpu, &l->page[1], l->mmu_idx, type, ra)) {
+        if (mmu_lookup1(cpu, &l->page[1], 0, l->mmu_idx, type, ra)) {
             uintptr_t index = tlb_index(cpu, l->mmu_idx, addr);
             l->page[0].full = &cpu->neg.tlb.d[l->mmu_idx].fulltlb[index];
         }
@@ -1774,19 +1787,6 @@ static bool mmu_lookup(CPUState *cpu, vaddr addr, MemOpIdx oi,
         tcg_debug_assert((flags & TLB_BSWAP) == 0);
     }
 
-    /*
-     * This alignment check differs from the one above, in that this is
-     * based on the atomicity of the operation. The intended use case is
-     * the ARM memory type field of each PTE, where access to pages with
-     * Device memory type require alignment.
-     */
-    if (unlikely(flags & TLB_CHECK_ALIGNED)) {
-        a_bits = memop_atomicity_bits(l->memop);
-        if (addr & ((1 << a_bits) - 1)) {
-            cpu_unaligned_access(cpu, addr, type, l->mmu_idx, ra);
-        }
-    }
-
     return crosspage;
 }
 
@@ -1799,34 +1799,18 @@ static void *atomic_mmu_lookup(CPUState *cpu, vaddr addr, MemOpIdx oi,
 {
     uintptr_t mmu_idx = get_mmuidx(oi);
     MemOp mop = get_memop(oi);
-    int a_bits = memop_alignment_bits(mop);
     uintptr_t index;
     CPUTLBEntry *tlbe;
     vaddr tlb_addr;
     void *hostaddr;
     CPUTLBEntryFull *full;
+    bool did_tlb_fill = false;
 
     tcg_debug_assert(mmu_idx < NB_MMU_MODES);
 
     /* Adjust the given return address.  */
     retaddr -= GETPC_ADJ;
 
-    /* Enforce guest required alignment.  */
-    if (unlikely(a_bits > 0 && (addr & ((1 << a_bits) - 1)))) {
-        /* ??? Maybe indicate atomic op to cpu_unaligned_access */
-        cpu_unaligned_access(cpu, addr, MMU_DATA_STORE,
-                             mmu_idx, retaddr);
-    }
-
-    /* Enforce qemu required alignment.  */
-    if (unlikely(addr & (size - 1))) {
-        /* We get here if guest alignment was not requested,
-           or was not enforced by cpu_unaligned_access above.
-           We might widen the access and emulate, but for now
-           mark an exception and exit the cpu loop.  */
-        goto stop_the_world;
-    }
-
     index = tlb_index(cpu, mmu_idx, addr);
     tlbe = tlb_entry(cpu, mmu_idx, addr);
 
@@ -1836,7 +1820,8 @@ static void *atomic_mmu_lookup(CPUState *cpu, vaddr addr, MemOpIdx oi,
         if (!victim_tlb_hit(cpu, mmu_idx, index, MMU_DATA_STORE,
                             addr & TARGET_PAGE_MASK)) {
             tlb_fill_align(cpu, addr, MMU_DATA_STORE, mmu_idx,
-                           0, size, false, retaddr);
+                           mop, size, false, retaddr);
+            did_tlb_fill = true;
             index = tlb_index(cpu, mmu_idx, addr);
             tlbe = tlb_entry(cpu, mmu_idx, addr);
         }
@@ -1859,6 +1844,23 @@ static void *atomic_mmu_lookup(CPUState *cpu, vaddr addr, MemOpIdx oi,
          */
         g_assert_not_reached();
     }
+
+    /* Enforce guest required alignment, if not handled by tlb_fill_align. */
+    if (!did_tlb_fill && (addr & ((1 << memop_alignment_bits(mop)) - 1))) {
+        cpu_unaligned_access(cpu, addr, MMU_DATA_STORE, mmu_idx, retaddr);
+    }
+
+    /* Enforce qemu required alignment.  */
+    if (unlikely(addr & (size - 1))) {
+        /*
+         * We get here if guest alignment was not requested, or was not
+         * enforced by cpu_unaligned_access or tlb_fill_align above.
+         * We might widen the access and emulate, but for now
+         * mark an exception and exit the cpu loop.
+         */
+        goto stop_the_world;
+    }
+
     /* Collect tlb flags for read. */
     tlb_addr |= tlbe->addr_read;