Merge remote-tracking branch 'remotes/rth-gitlab/tags/pull-tcg-20210107' into staging

Build fix for ppc64 centos7.
Reduce the use of scratch registers for tcg/i386.
Use _aligned_malloc for Win32.
Enable split w^x code gen buffers.

# gpg: Signature made Thu 07 Jan 2021 20:06:38 GMT
# gpg:                using RSA key 7A481E78868B4DB6A85A05C064DF38E8AF7E215F
# gpg:                issuer "richard.henderson@linaro.org"
# gpg: Good signature from "Richard Henderson <richard.henderson@linaro.org>" [full]
# Primary key fingerprint: 7A48 1E78 868B 4DB6 A85A  05C0 64DF 38E8 AF7E 215F

* remotes/rth-gitlab/tags/pull-tcg-20210107: (47 commits)
  tcg: Constify TCGLabelQemuLdst.raddr
  tcg: Constify tcg_code_gen_epilogue
  tcg: Remove TCG_TARGET_SUPPORT_MIRROR
  tcg/arm: Support split-wx code generation
  tcg/mips: Support split-wx code generation
  tcg/mips: Do not assert on relocation overflow
  accel/tcg: Add mips support to alloc_code_gen_buffer_splitwx_memfd
  tcg/riscv: Support split-wx code generation
  tcg/riscv: Remove branch-over-branch fallback
  tcg/riscv: Fix branch range checks
  tcg/s390: Support split-wx code generation
  tcg/s390: Use tcg_tbrel_diff
  tcg/sparc: Support split-wx code generation
  tcg/sparc: Use tcg_tbrel_diff
  tcg/ppc: Support split-wx code generation
  tcg/ppc: Use tcg_out_mem_long to reset TCG_REG_TB
  tcg/ppc: Use tcg_tbrel_diff
  tcg: Introduce tcg_tbrel_diff
  tcg/tci: Push const down through bytecode reading
  disas: Push const down through host disassembly
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

1 files changed, 91 insertions, 16 deletions

diff --git a/util/cacheflush.c b/util/cacheflush.c
index 2881832a38..6a20723902 100644
--- a/util/cacheflush.c
+++ b/util/cacheflush.c
@@ -7,12 +7,81 @@
 
 #include "qemu/osdep.h"
 #include "qemu/cacheflush.h"
+#include "qemu/bitops.h"
 
 
 #if defined(__i386__) || defined(__x86_64__) || defined(__s390__)
 
 /* Caches are coherent and do not require flushing; symbol inline. */
 
+#elif defined(__aarch64__)
+
+#ifdef CONFIG_DARWIN
+/* Apple does not expose CTR_EL0, so we must use system interfaces. */
+extern void sys_icache_invalidate(void *start, size_t len);
+extern void sys_dcache_flush(void *start, size_t len);
+void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
+{
+    sys_dcache_flush((void *)rw, len);
+    sys_icache_invalidate((void *)rx, len);
+}
+#else
+
+/*
+ * TODO: unify this with cacheinfo.c.
+ * We want to save the whole contents of CTR_EL0, so that we
+ * have more than the linesize, but also IDC and DIC.
+ */
+static unsigned int save_ctr_el0;
+static void __attribute__((constructor)) init_ctr_el0(void)
+{
+    asm volatile("mrs\t%0, ctr_el0" : "=r"(save_ctr_el0));
+}
+
+/*
+ * This is a copy of gcc's __aarch64_sync_cache_range, modified
+ * to fit this three-operand interface.
+ */
+void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
+{
+    const unsigned CTR_IDC = 1u << 28;
+    const unsigned CTR_DIC = 1u << 29;
+    const unsigned int ctr_el0 = save_ctr_el0;
+    const uintptr_t icache_lsize = 4 << extract32(ctr_el0, 0, 4);
+    const uintptr_t dcache_lsize = 4 << extract32(ctr_el0, 16, 4);
+    uintptr_t p;
+
+    /*
+     * If CTR_EL0.IDC is enabled, Data cache clean to the Point of Unification
+     * is not required for instruction to data coherence.
+     */
+    if (!(ctr_el0 & CTR_IDC)) {
+        /*
+         * Loop over the address range, clearing one cache line at once.
+         * Data cache must be flushed to unification first to make sure
+         * the instruction cache fetches the updated data.
+         */
+        for (p = rw & -dcache_lsize; p < rw + len; p += dcache_lsize) {
+            asm volatile("dc\tcvau, %0" : : "r" (p) : "memory");
+        }
+        asm volatile("dsb\tish" : : : "memory");
+    }
+
+    /*
+     * If CTR_EL0.DIC is enabled, Instruction cache cleaning to the Point
+     * of Unification is not required for instruction to data coherence.
+     */
+    if (!(ctr_el0 & CTR_DIC)) {
+        for (p = rx & -icache_lsize; p < rx + len; p += icache_lsize) {
+            asm volatile("ic\tivau, %0" : : "r"(p) : "memory");
+        }
+        asm volatile ("dsb\tish" : : : "memory");
+    }
+
+    asm volatile("isb" : : : "memory");
+}
+#endif /* CONFIG_DARWIN */
+
 #elif defined(__mips__)
 
 #ifdef __OpenBSD__
@@ -21,29 +90,32 @@
 #include <sys/cachectl.h>
 #endif
 
-void flush_icache_range(uintptr_t start, uintptr_t stop)
+void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
 {
-    cacheflush((void *)start, stop - start, ICACHE);
+    if (rx != rw) {
+        cacheflush((void *)rw, len, DCACHE);
+    }
+    cacheflush((void *)rx, len, ICACHE);
 }
 
 #elif defined(__powerpc__)
 
-void flush_icache_range(uintptr_t start, uintptr_t stop)
+void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
 {
-    uintptr_t p, start1, stop1;
+    uintptr_t p, b, e;
     size_t dsize = qemu_dcache_linesize;
     size_t isize = qemu_icache_linesize;
 
-    start1 = start & ~(dsize - 1);
-    stop1 = (stop + dsize - 1) & ~(dsize - 1);
-    for (p = start1; p < stop1; p += dsize) {
+    b = rw & ~(dsize - 1);
+    e = (rw + len + dsize - 1) & ~(dsize - 1);
+    for (p = b; p < e; p += dsize) {
         asm volatile ("dcbst 0,%0" : : "r"(p) : "memory");
     }
     asm volatile ("sync" : : : "memory");
 
-    start &= start & ~(isize - 1);
-    stop1 = (stop + isize - 1) & ~(isize - 1);
-    for (p = start1; p < stop1; p += isize) {
+    b = rx & ~(isize - 1);
+    e = (rx + len + isize - 1) & ~(isize - 1);
+    for (p = b; p < e; p += isize) {
         asm volatile ("icbi 0,%0" : : "r"(p) : "memory");
     }
     asm volatile ("sync" : : : "memory");
@@ -52,20 +124,23 @@ void flush_icache_range(uintptr_t start, uintptr_t stop)
 
 #elif defined(__sparc__)
 
-void flush_icache_range(uintptr_t start, uintptr_t stop)
+void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
 {
-    uintptr_t p;
-
-    for (p = start & -8; p < ((stop + 7) & -8); p += 8) {
+    /* No additional data flush to the RW virtual address required. */
+    uintptr_t p, end = (rx + len + 7) & -8;
+    for (p = rx & -8; p < end; p += 8) {
         __asm__ __volatile__("flush\t%0" : : "r" (p));
     }
 }
 
 #else
 
-void flush_icache_range(uintptr_t start, uintptr_t stop)
+void flush_idcache_range(uintptr_t rx, uintptr_t rw, size_t len)
 {
-    __builtin___clear_cache((char *)start, (char *)stop);
+    if (rw != rx) {
+        __builtin___clear_cache((char *)rw, (char *)rw + len);
+    }
+    __builtin___clear_cache((char *)rx, (char *)rx + len);
 }
 
 #endif