[PULL 01/16] cputlb: Handle NB_MMU_MODES > TARGET_PAGE_BITS_MIN

[Richard Henderson <richard.henderson@linaro.org>]

In target/arm we will shortly have "too many" mmu_idx.
The current minimum barrier is caused by the way in which
tlb_flush_page_by_mmuidx is coded.

We can remove this limitation by allocating memory for
consumption by the worker.  Let us assume that this is
the unlikely case, as will be the case for the majority
of targets which have so far satisfied the BUILD_BUG_ON,
and only allocate memory when necessary.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 accel/tcg/cputlb.c | 167 +++++++++++++++++++++++++++++++++++----------
 1 file changed, 132 insertions(+), 35 deletions(-)

diff --git a/accel/tcg/cputlb.c b/accel/tcg/cputlb.c
index a991ea2964..02b381cf50 100644
--- a/accel/tcg/cputlb.c
+++ b/accel/tcg/cputlb.c
@@ -449,28 +449,29 @@ static void tlb_flush_page_locked(CPUArchState *env, int midx,
     }
 }
 
-/* As we are going to hijack the bottom bits of the page address for a
- * mmuidx bit mask we need to fail to build if we can't do that
+/**
+ * tlb_flush_page_by_mmuidx_async_0:
+ * @cpu: cpu on which to flush
+ * @addr: page of virtual address to flush
+ * @idxmap: set of mmu_idx to flush
+ *
+ * Helper for tlb_flush_page_by_mmuidx and friends, flush one page
+ * at @addr from the tlbs indicated by @idxmap from @cpu.
  */
-QEMU_BUILD_BUG_ON(NB_MMU_MODES > TARGET_PAGE_BITS_MIN);
-
-static void tlb_flush_page_by_mmuidx_async_work(CPUState *cpu,
-                                                run_on_cpu_data data)
+static void tlb_flush_page_by_mmuidx_async_0(CPUState *cpu,
+                                             target_ulong addr,
+                                             uint16_t idxmap)
 {
     CPUArchState *env = cpu->env_ptr;
-    target_ulong addr_and_mmuidx = (target_ulong) data.target_ptr;
-    target_ulong addr = addr_and_mmuidx & TARGET_PAGE_MASK;
-    unsigned long mmu_idx_bitmap = addr_and_mmuidx & ALL_MMUIDX_BITS;
     int mmu_idx;
 
     assert_cpu_is_self(cpu);
 
-    tlb_debug("page addr:" TARGET_FMT_lx " mmu_map:0x%lx\n",
-              addr, mmu_idx_bitmap);
+    tlb_debug("page addr:" TARGET_FMT_lx " mmu_map:0x%x\n", addr, idxmap);
 
     qemu_spin_lock(&env_tlb(env)->c.lock);
     for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
-        if (test_bit(mmu_idx, &mmu_idx_bitmap)) {
+        if ((idxmap >> mmu_idx) & 1) {
             tlb_flush_page_locked(env, mmu_idx, addr);
         }
     }
@@ -479,22 +480,75 @@ static void tlb_flush_page_by_mmuidx_async_work(CPUState *cpu,
     tb_flush_jmp_cache(cpu, addr);
 }
 
+/**
+ * tlb_flush_page_by_mmuidx_async_1:
+ * @cpu: cpu on which to flush
+ * @data: encoded addr + idxmap
+ *
+ * Helper for tlb_flush_page_by_mmuidx and friends, called through
+ * async_run_on_cpu.  The idxmap parameter is encoded in the page
+ * offset of the target_ptr field.  This limits the set of mmu_idx
+ * that can be passed via this method.
+ */
+static void tlb_flush_page_by_mmuidx_async_1(CPUState *cpu,
+                                             run_on_cpu_data data)
+{
+    target_ulong addr_and_idxmap = (target_ulong) data.target_ptr;
+    target_ulong addr = addr_and_idxmap & TARGET_PAGE_MASK;
+    uint16_t idxmap = addr_and_idxmap & ~TARGET_PAGE_MASK;
+
+    tlb_flush_page_by_mmuidx_async_0(cpu, addr, idxmap);
+}
+
+typedef struct {
+    target_ulong addr;
+    uint16_t idxmap;
+} TLBFlushPageByMMUIdxData;
+
+/**
+ * tlb_flush_page_by_mmuidx_async_2:
+ * @cpu: cpu on which to flush
+ * @data: allocated addr + idxmap
+ *
+ * Helper for tlb_flush_page_by_mmuidx and friends, called through
+ * async_run_on_cpu.  The addr+idxmap parameters are stored in a
+ * TLBFlushPageByMMUIdxData structure that has been allocated
+ * specifically for this helper.  Free the structure when done.
+ */
+static void tlb_flush_page_by_mmuidx_async_2(CPUState *cpu,
+                                             run_on_cpu_data data)
+{
+    TLBFlushPageByMMUIdxData *d = data.host_ptr;
+
+    tlb_flush_page_by_mmuidx_async_0(cpu, d->addr, d->idxmap);
+    g_free(d);
+}
+
 void tlb_flush_page_by_mmuidx(CPUState *cpu, target_ulong addr, uint16_t idxmap)
 {
-    target_ulong addr_and_mmu_idx;
-
     tlb_debug("addr: "TARGET_FMT_lx" mmu_idx:%" PRIx16 "\n", addr, idxmap);
 
     /* This should already be page aligned */
-    addr_and_mmu_idx = addr & TARGET_PAGE_MASK;
-    addr_and_mmu_idx |= idxmap;
+    addr &= TARGET_PAGE_MASK;
 
-    if (!qemu_cpu_is_self(cpu)) {
-        async_run_on_cpu(cpu, tlb_flush_page_by_mmuidx_async_work,
-                         RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
+    if (qemu_cpu_is_self(cpu)) {
+        tlb_flush_page_by_mmuidx_async_0(cpu, addr, idxmap);
+    } else if (idxmap < TARGET_PAGE_SIZE) {
+        /*
+         * Most targets have only a few mmu_idx.  In the case where
+         * we can stuff idxmap into the low TARGET_PAGE_BITS, avoid
+         * allocating memory for this operation.
+         */
+        async_run_on_cpu(cpu, tlb_flush_page_by_mmuidx_async_1,
+                         RUN_ON_CPU_TARGET_PTR(addr | idxmap));
     } else {
-        tlb_flush_page_by_mmuidx_async_work(
-            cpu, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
+        TLBFlushPageByMMUIdxData *d = g_new(TLBFlushPageByMMUIdxData, 1);
+
+        /* Otherwise allocate a structure, freed by the worker.  */
+        d->addr = addr;
+        d->idxmap = idxmap;
+        async_run_on_cpu(cpu, tlb_flush_page_by_mmuidx_async_2,
+                         RUN_ON_CPU_HOST_PTR(d));
     }
 }
 
@@ -506,17 +560,36 @@ void tlb_flush_page(CPUState *cpu, target_ulong addr)
 void tlb_flush_page_by_mmuidx_all_cpus(CPUState *src_cpu, target_ulong addr,
                                        uint16_t idxmap)
 {
-    const run_on_cpu_func fn = tlb_flush_page_by_mmuidx_async_work;
-    target_ulong addr_and_mmu_idx;
-
     tlb_debug("addr: "TARGET_FMT_lx" mmu_idx:%"PRIx16"\n", addr, idxmap);
 
     /* This should already be page aligned */
-    addr_and_mmu_idx = addr & TARGET_PAGE_MASK;
-    addr_and_mmu_idx |= idxmap;
+    addr &= TARGET_PAGE_MASK;
 
-    flush_all_helper(src_cpu, fn, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
-    fn(src_cpu, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
+    /*
+     * Allocate memory to hold addr+idxmap only when needed.
+     * See tlb_flush_page_by_mmuidx for details.
+     */
+    if (idxmap < TARGET_PAGE_SIZE) {
+        flush_all_helper(src_cpu, tlb_flush_page_by_mmuidx_async_1,
+                         RUN_ON_CPU_TARGET_PTR(addr | idxmap));
+    } else {
+        CPUState *dst_cpu;
+
+        /* Allocate a separate data block for each destination cpu.  */
+        CPU_FOREACH(dst_cpu) {
+            if (dst_cpu != src_cpu) {
+                TLBFlushPageByMMUIdxData *d
+                    = g_new(TLBFlushPageByMMUIdxData, 1);
+
+                d->addr = addr;
+                d->idxmap = idxmap;
+                async_run_on_cpu(dst_cpu, tlb_flush_page_by_mmuidx_async_2,
+                                 RUN_ON_CPU_HOST_PTR(d));
+            }
+        }
+    }
+
+    tlb_flush_page_by_mmuidx_async_0(src_cpu, addr, idxmap);
 }
 
 void tlb_flush_page_all_cpus(CPUState *src, target_ulong addr)
@@ -528,17 +601,41 @@ void tlb_flush_page_by_mmuidx_all_cpus_synced(CPUState *src_cpu,
                                               target_ulong addr,
                                               uint16_t idxmap)
 {
-    const run_on_cpu_func fn = tlb_flush_page_by_mmuidx_async_work;
-    target_ulong addr_and_mmu_idx;
-
     tlb_debug("addr: "TARGET_FMT_lx" mmu_idx:%"PRIx16"\n", addr, idxmap);
 
     /* This should already be page aligned */
-    addr_and_mmu_idx = addr & TARGET_PAGE_MASK;
-    addr_and_mmu_idx |= idxmap;
+    addr &= TARGET_PAGE_MASK;
 
-    flush_all_helper(src_cpu, fn, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
-    async_safe_run_on_cpu(src_cpu, fn, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
+    /*
+     * Allocate memory to hold addr+idxmap only when needed.
+     * See tlb_flush_page_by_mmuidx for details.
+     */
+    if (idxmap < TARGET_PAGE_SIZE) {
+        flush_all_helper(src_cpu, tlb_flush_page_by_mmuidx_async_1,
+                         RUN_ON_CPU_TARGET_PTR(addr | idxmap));
+        async_safe_run_on_cpu(src_cpu, tlb_flush_page_by_mmuidx_async_1,
+                              RUN_ON_CPU_TARGET_PTR(addr | idxmap));
+    } else {
+        CPUState *dst_cpu;
+        TLBFlushPageByMMUIdxData *d;
+
+        /* Allocate a separate data block for each destination cpu.  */
+        CPU_FOREACH(dst_cpu) {
+            if (dst_cpu != src_cpu) {
+                d = g_new(TLBFlushPageByMMUIdxData, 1);
+                d->addr = addr;
+                d->idxmap = idxmap;
+                async_run_on_cpu(dst_cpu, tlb_flush_page_by_mmuidx_async_2,
+                                 RUN_ON_CPU_HOST_PTR(d));
+            }
+        }
+
+        d = g_new(TLBFlushPageByMMUIdxData, 1);
+        d->addr = addr;
+        d->idxmap = idxmap;
+        async_safe_run_on_cpu(src_cpu, tlb_flush_page_by_mmuidx_async_2,
+                              RUN_ON_CPU_HOST_PTR(d));
+    }
 }
 
 void tlb_flush_page_all_cpus_synced(CPUState *src, target_ulong addr)
-- 
2.20.1