Enable vDSO support on arm64.

Signed-off-by: Haibo Xu <haibo.xu@arm.com>
Change-Id: I20103cd6d193431ab7e8120005da1f567b9bc2eb
PiperOrigin-RevId: 244280119

8 files changed, 257 insertions, 11 deletions

diff --git a/vdso/BUILD b/vdso/BUILD
index 3df569233..f95f690eb 100644
--- a/vdso/BUILD
+++ b/vdso/BUILD
@@ -12,6 +12,11 @@ config_setting(
     constraint_values = ["@bazel_tools//platforms:x86_64"],
 )
 
+config_setting(
+    name = "aarch64",
+    constraint_values = ["@bazel_tools//platforms:aarch64"],
+)
+
 genrule(
     name = "vdso",
     srcs = [
@@ -21,7 +26,8 @@ genrule(
         "seqlock.h",
         "syscalls.h",
         "vdso.cc",
-        "vdso.lds",
+        "vdso_amd64.lds",
+        "vdso_arm64.lds",
         "vdso_time.h",
         "vdso_time.cc",
     ],
@@ -49,7 +55,13 @@ genrule(
           "-Wl,-Bsymbolic " +
           "-Wl,-z,max-page-size=4096 " +
           "-Wl,-z,common-page-size=4096 " +
-          "-Wl,-T$(location vdso.lds) " +
+          select(
+              {
+                  ":x86_64": "-Wl,-T$(location vdso_amd64.lds) ",
+                  ":aarch64": "-Wl,-T$(location vdso_arm64.lds) ",
+              },
+              no_match_error = "Unsupported architecture",
+          ) +
           "-o $(location vdso.so) " +
           "$(location vdso.cc) " +
           "$(location vdso_time.cc) " +
diff --git a/vdso/barrier.h b/vdso/barrier.h
index 7866af414..5b6c763f6 100644
--- a/vdso/barrier.h
+++ b/vdso/barrier.h
@@ -21,11 +21,25 @@ namespace vdso {
 inline void barrier(void) { __asm__ __volatile__("" ::: "memory"); }
 
 #if __x86_64__
+
 inline void memory_barrier(void) {
   __asm__ __volatile__("mfence" ::: "memory");
 }
 inline void read_barrier(void) { barrier(); }
 inline void write_barrier(void) { barrier(); }
+
+#elif __aarch64__
+
+inline void memory_barrier(void) {
+  __asm__ __volatile__("dmb ish" ::: "memory");
+}
+inline void read_barrier(void) {
+  __asm__ __volatile__("dmb ishld" ::: "memory");
+}
+inline void write_barrier(void) {
+  __asm__ __volatile__("dmb ishst" ::: "memory");
+}
+
 #else
 #error "unsupported architecture"
 #endif
diff --git a/vdso/cycle_clock.h b/vdso/cycle_clock.h
index dfb5b427d..26d6690c0 100644
--- a/vdso/cycle_clock.h
+++ b/vdso/cycle_clock.h
@@ -33,6 +33,15 @@ static inline uint64_t cycle_clock(void) {
   asm volatile("rdtsc" : "=a"(lo), "=d"(hi));
   return ((uint64_t)hi << 32) | lo;
 }
+
+#elif __aarch64__
+
+static inline uint64_t cycle_clock(void) {
+  uint64_t val;
+  asm volatile("mrs %0, CNTVCT_EL0" : "=r"(val)::"memory");
+  return val;
+}
+
 #else
 #error "unsupported architecture"
 #endif
diff --git a/vdso/syscalls.h b/vdso/syscalls.h
index 0be8a7f9b..90fb424ce 100644
--- a/vdso/syscalls.h
+++ b/vdso/syscalls.h
@@ -26,10 +26,12 @@
 #include <stddef.h>
 #include <sys/types.h>
 
-struct getcpu_cache;
-
 namespace vdso {
 
+#if __x86_64__
+
+struct getcpu_cache;
+
 static inline int sys_clock_gettime(clockid_t clock, struct timespec* ts) {
   int num = __NR_clock_gettime;
   asm volatile("syscall\n"
@@ -49,6 +51,49 @@ static inline int sys_getcpu(unsigned* cpu, unsigned* node,
   return num;
 }
 
+#elif __aarch64__
+
+static inline int sys_rt_sigreturn(void) {
+  int num = __NR_rt_sigreturn;
+
+  asm volatile(
+      "mov x8, %0\n"
+      "svc #0    \n"
+      : "+r"(num)
+      :
+      :);
+  return num;
+}
+
+static inline int sys_clock_gettime(clockid_t _clkid, struct timespec *_ts) {
+  register struct timespec *ts asm("x1") = _ts;
+  register clockid_t clkid asm("x0") = _clkid;
+  register long ret asm("x0");
+  register long nr asm("x8") = __NR_clock_gettime;
+
+  asm volatile("svc #0\n"
+               : "=r"(ret)
+               : "r"(clkid), "r"(ts), "r"(nr)
+               : "memory");
+  return ret;
+}
+
+static inline int sys_clock_getres(clockid_t _clkid, struct timespec *_ts) {
+  register struct timespec *ts asm("x1") = _ts;
+  register clockid_t clkid asm("x0") = _clkid;
+  register long ret asm("x0");
+  register long nr asm("x8") = __NR_clock_getres;
+
+  asm volatile("svc #0\n"
+               : "=r"(ret)
+               : "r"(clkid), "r"(ts), "r"(nr)
+               : "memory");
+  return ret;
+}
+
+#else
+#error "unsupported architecture"
+#endif
 }  // namespace vdso
 
 #endif  // VDSO_SYSCALLS_H_
diff --git a/vdso/vdso.cc b/vdso/vdso.cc
index f30dc26a2..550729035 100644
--- a/vdso/vdso.cc
+++ b/vdso/vdso.cc
@@ -23,9 +23,9 @@
 #include "vdso/vdso_time.h"
 
 namespace vdso {
+namespace {
 
-// __vdso_clock_gettime() implements clock_gettime()
-extern "C" int __vdso_clock_gettime(clockid_t clock, struct timespec* ts) {
+int __common_clock_gettime(clockid_t clock, struct timespec* ts) {
   int ret;
 
   switch (clock) {
@@ -44,11 +44,8 @@ extern "C" int __vdso_clock_gettime(clockid_t clock, struct timespec* ts) {
 
   return ret;
 }
-extern "C" int clock_gettime(clockid_t clock, struct timespec* ts)
-    __attribute__((weak, alias("__vdso_clock_gettime")));
 
-// __vdso_gettimeofday() implements gettimeofday()
-extern "C" int __vdso_gettimeofday(struct timeval* tv, struct timezone* tz) {
+int __common_gettimeofday(struct timeval* tv, struct timezone* tz) {
   if (tv) {
     struct timespec ts;
     int ret = ClockRealtime(&ts);
@@ -68,6 +65,21 @@ extern "C" int __vdso_gettimeofday(struct timeval* tv, struct timezone* tz) {
 
   return 0;
 }
+}  // namespace
+
+#if __x86_64__
+
+// __vdso_clock_gettime() implements clock_gettime()
+extern "C" int __vdso_clock_gettime(clockid_t clock, struct timespec* ts) {
+  return __common_clock_gettime(clock, ts);
+}
+extern "C" int clock_gettime(clockid_t clock, struct timespec* ts)
+    __attribute__((weak, alias("__vdso_clock_gettime")));
+
+// __vdso_gettimeofday() implements gettimeofday()
+extern "C" int __vdso_gettimeofday(struct timeval* tv, struct timezone* tz) {
+  return __common_gettimeofday(tv, tz);
+}
 extern "C" int gettimeofday(struct timeval* tv, struct timezone* tz)
     __attribute__((weak, alias("__vdso_gettimeofday")));
 
@@ -92,4 +104,45 @@ extern "C" long getcpu(unsigned* cpu, unsigned* node,
                        struct getcpu_cache* cache)
     __attribute__((weak, alias("__vdso_getcpu")));
 
+#elif __aarch64__
+
+// __kernel_clock_gettime() implements clock_gettime()
+extern "C" int __kernel_clock_gettime(clockid_t clock, struct timespec* ts) {
+  return __common_clock_gettime(clock, ts);
+}
+
+// __kernel_gettimeofday() implements gettimeofday()
+extern "C" int __kernel_gettimeofday(struct timeval* tv, struct timezone* tz) {
+  return __common_gettimeofday(tv, tz);
+}
+
+// __kernel_clock_getres() implements clock_getres()
+extern "C" int __kernel_clock_getres(clockid_t clock, struct timespec* res) {
+  int ret = 0;
+
+  switch (clock) {
+    case CLOCK_REALTIME:
+    case CLOCK_MONOTONIC: {
+      res->tv_sec = 0;
+      res->tv_nsec = 1;
+      break;
+    }
+
+    default:
+      ret = sys_clock_getres(clock, res);
+      break;
+  }
+
+  return ret;
+}
+
+// __kernel_rt_sigreturn() implements gettimeofday()
+extern "C" int __kernel_rt_sigreturn(unsigned long unused) {
+  // No optimizations yet, just make the real system call.
+  return sys_rt_sigreturn();
+}
+
+#else
+#error "unsupported architecture"
+#endif
 }  // namespace vdso
diff --git a/vdso/vdso.lds b/vdso/vdso_amd64.lds
index 166779931..166779931 100644
--- a/vdso/vdso.lds
+++ b/vdso/vdso_amd64.lds
diff --git a/vdso/vdso_arm64.lds b/vdso/vdso_arm64.lds
new file mode 100644
index 000000000..19f8efa01
--- /dev/null
+++ b/vdso/vdso_arm64.lds
@@ -0,0 +1,99 @@
+/*
+ * Linker script for the VDSO.
+ *
+ * The VDSO is essentially a normal ELF shared library that is mapped into the
+ * address space of the process that is going to use it. The address of the
+ * VDSO is passed to the runtime linker in the AT_SYSINFO_EHDR entry of the aux
+ * vector.
+ *
+ * There are, however, three ways in which the VDSO differs from a normal
+ * shared library:
+ *
+ * - The runtime linker does not attempt to process any relocations for the
+ *   VDSO so it is the responsibility of whoever loads the VDSO into the
+ *   address space to do this if necessary. Because of this restriction we are
+ *   careful to ensure that the VDSO does not need to have any relocations
+ *   applied to it.
+ *
+ * - Although the VDSO is position independent and would normally be linked at
+ *   virtual address 0, the Linux kernel VDSO is actually linked at a non zero
+ *   virtual address and the code in the system runtime linker that handles the
+ *   VDSO expects this to be the case so we have to explicitly link this VDSO
+ *   at a non zero address. The actual address is arbitrary, but we use the
+ *   same one as the Linux kernel VDSO.
+ *
+ * - The VDSO will be directly mmapped by the sentry, rather than going through
+ *   a normal ELF loading process. The VDSO must be carefully constructed such
+ *   that the layout in the ELF file is identical to the layout in memory.
+ */
+
+VDSO_PRELINK = 0xffffffffff700000;
+
+OUTPUT_FORMAT("elf64-littleaarch64", "elf64-bigaarch64", "elf64-littleaarch64")
+OUTPUT_ARCH(aarch64)
+
+SECTIONS {
+  /* The parameter page is mapped just before the VDSO. */
+  _params = VDSO_PRELINK - 0x1000;
+
+  . = VDSO_PRELINK + SIZEOF_HEADERS;
+
+  .hash          : { *(.hash) }             :text
+  .gnu.hash      : { *(.gnu.hash) }
+  .dynsym        : { *(.dynsym) }
+  .dynstr        : { *(.dynstr) }
+  .gnu.version   : { *(.gnu.version) }
+  .gnu.version_d : { *(.gnu.version_d) }
+  .gnu.version_r : { *(.gnu.version_r) }
+
+  .note          : { *(.note.*) }           :text  :note
+
+  .eh_frame_hdr  : { *(.eh_frame_hdr) }     :text  :eh_frame_hdr
+  .eh_frame      : { KEEP (*(.eh_frame)) }  :text
+
+  .dynamic       : { *(.dynamic) }          :text  :dynamic
+
+  .rodata        : { *(.rodata*) }          :text
+
+  .altinstructions       : { *(.altinstructions) }
+  .altinstr_replacement  : { *(.altinstr_replacement) }
+
+  /*
+   * TODO: Remove this alignment? Then the VDSO would fit
+   * in a single page.
+   */
+  . = ALIGN(0x1000);
+  .text          : { *(.text*) }            :text    =0xd503201f
+
+  /*
+   * N.B. There is no data/bss section. This VDSO neither needs nor uses a data
+   * section. We omit it entirely because some gcc/clang and gold/bfd version
+   * combinations struggle to handle an empty data PHDR segment (internal
+   * linker assertion failures result).
+   *
+   * If the VDSO does incorrectly include a data section, the linker will
+   * include it in the text segment. check_vdso.py looks for this degenerate
+   * case.
+   */
+}
+
+PHDRS {
+  text          PT_LOAD     FLAGS(5) FILEHDR PHDRS;  /* PF_R | PF_X */
+  dynamic       PT_DYNAMIC  FLAGS(4);                /* PF_R */
+  note          PT_NOTE     FLAGS(4);                /* PF_R */
+  eh_frame_hdr  PT_GNU_EH_FRAME;
+}
+
+/*
+ * Define the symbols that are to be exported.
+ */
+VERSION {
+  LINUX_2.6.39 {
+  global:
+   __kernel_clock_getres;
+   __kernel_clock_gettime;
+   __kernel_gettimeofday;
+   __kernel_rt_sigreturn;
+  local: *;
+  };
+}
diff --git a/vdso/vdso_time.cc b/vdso/vdso_time.cc
index a59771bff..9fc262f60 100644
--- a/vdso/vdso_time.cc
+++ b/vdso/vdso_time.cc
@@ -55,12 +55,26 @@ struct params {
 //
 // So instead, we use inline assembly with a construct that seems to have wide
 // compatibility across many toolchains.
+#if __x86_64__
+
+inline struct params* get_params() {
+  struct params* p = nullptr;
+  asm("leaq _params(%%rip), %0" : "=r"(p) : :);
+  return p;
+}
+
+#elif __aarch64__
+
 inline struct params* get_params() {
   struct params* p = nullptr;
-  asm volatile("leaq _params(%%rip), %0" : "=r"(p) : :);
+  asm("adr %0, _params" : "=r"(p) : :);
   return p;
 }
 
+#else
+#error "unsupported architecture"
+#endif
+
 namespace vdso {
 
 const uint64_t kNsecsPerSec = 1000000000UL;