arch/armv7-r: userspace PMU access

This change enables performance monitoring unit (PMU) access from userspace
on ARMv7-R architecture by adding CONFIG_ARCH_HAVE_PERF_EVENTS_USER_ACCESS
support and building PMU code for userspace when needed.

Signed-off-by: hujun5 <hujun5@xiaomi.com>

Documentation/platforms/arm/qemu/boards/qemu-armv7r/index.rst

@@ -41,7 +41,7 @@ NSH (Flat Build)
 Configuring NuttX and compile::
 
      $ ./tools/configure.sh -l qemu-armv7r:nsh
-     $ make -j`nproc`
+     $ make -j$(nproc)
 
 Running with qemu::
 
@@ -58,7 +58,7 @@ with MPU support::
 
   $ cd nuttx
   $ ./tools/configure.sh qemu-armv7r:pnsh
-  $ make -j`nproc`
+  $ make -j$(nproc)
 
 Running with qemu (note: both nuttx and nuttx_user must be loaded)::
 
@@ -106,4 +106,93 @@ The nuttx ELF image can be debugged with QEMU.
 
      $ arm-none-eabi-gdb -tui --eval-command='target remote localhost:1234' nuttx
      (gdb) add-symbol-file nuttx_user
-     (gdb) c
+     (gdb) c
+
+=====================
+Userspace PMU Access
+=====================
+
+Overview
+========
+
+This document describes how to enable and use Performance Monitoring Unit (PMU)
+access from userspace applications on the ARM v7-R QEMU board. The PMU provides
+hardware performance counters that can be used to analyze and profile application
+performance.
+
+1. ARM v7-R QEMU board support
+2. NuttX kernel with PMU support enabled
+3. User-space access permissions configured
+
+Kernel Configuration
+--------------------
+
+To enable PMU support and userspace access, add the following configuration
+options to your defconfig::
+
+     +CONFIG_ARCH_PERF_EVENTS_USER_ACCESS=y
+     +CONFIG_ARCH_PERF_EVENTS=y
+
+
+Userspace API
+
+**perf_gettime**
+================
+
+Basic Usage
+-----------
+
+.. code-block:: c
+
+    clock_t start, end;
+    unsigned long frequency;
+    unsigned long cycles;
+
+    frequency = perf_getfreq();
+    if (frequency == 0) {
+        printf("ERROR: Performance frequency not available\n");
+        return;
+    }
+
+    printf("Operation Profiling Results:\n");
+    printf("CPU Frequency: %lu Hz\n\n", frequency);
+    printf("%-40s | %-12s | %-12s\n", "Operation", "Cycles", "Time (us)");
+    printf("%-40s-+-%-12s-+-%-12s\n", 
+           "----------------------------------------",
+           "------------", "------------");
+
+    start = perf_gettime();
+    result = 1;
+    for (int i = 1; i < 1000; i++) {
+        result *= i;
+    }
+
+    end = perf_gettime();
+    cycles = end - start;
+    printf("%-40s | %12lu | %12lu\n", "Multiplication (1K times)", cycles, (cycles * 1000000UL) / frequency);
+
+    start = perf_gettime();
+    result = 1000000;
+    for (int i = 1; i < 1000; i++) {
+        result /= (i + 1);
+    }
+    end = perf_gettime();
+    cycles = end - start;
+
+    printf("%-40s | %12lu | %12lu\n", "Division (1K times)", cycles, (cycles * 1000000UL) / frequency);
+
+Testing with QEMU
+=================
+
+PNSH (Protected) Configuration
+-------------------------------
+
+For protected build with userspace PMU access::
+
+     $ ./tools/configure.sh qemu-armv7r:pnsh
+     $ make -j$(nproc)
+
+2. Run QEMU (load both kernel and userspace)::
+
+     $ qemu-system-arm -M virt -semihosting -nographic -cpu cortex-r5f -device loader,file=nuttx_user -device loader,file=nuttx
+

arch/arm/Kconfig

@@ -1005,6 +1005,7 @@ config ARCH_ARMV7R
 	default n
 	select ARCH_HAVE_CPUINFO
 	select ARCH_HAVE_PERF_EVENTS
+	select ARCH_HAVE_PERF_EVENTS_USER_ACCESS
 
 config ARCH_CORTEXR4
 	bool

arch/arm/src/armv7-r/CMakeLists.txt

@@ -81,3 +81,8 @@ if(CONFIG_SMP)
 endif()
 
 target_sources(arch PRIVATE ${SRCS})
+
+if(NOT CONFIG_BUILD_FLAT AND CONFIG_ARCH_HAVE_PERF_EVENTS_USER_ACCESS)
+  target_sources(arch_interface PRIVATE arm_perf.c)
+  target_include_directories(arch_interface PRIVATE ${NUTTX_CHIP_ABS_DIR})
+endif()

arch/arm/src/armv7-r/Make.defs

@@ -67,3 +67,7 @@ ifeq ($(CONFIG_SMP),y)
   CMN_CSRCS += arm_cpustart.c arm_smpcall.c
   CMN_CSRCS += arm_cpuidlestack.c arm_scu.c
 endif
+
+ifeq ($(CONFIG_ARCH_HAVE_PERF_EVENTS_USER_ACCESS),y)
+  CMN_UCSRCS += arm_perf.c
+endif

arch/arm/src/armv7-r/arm_perf.c

@@ -33,6 +33,7 @@
 
 #ifdef CONFIG_ARCH_HAVE_PERF_EVENTS
 
+#if defined(CONFIG_BUILD_FLAT) || defined(__KERNEL__)
 /****************************************************************************
  * Private Data
  ****************************************************************************/
@@ -84,11 +85,6 @@ unsigned long up_perf_getfreq(void)
   return g_cpu_freq;
 }
 
-clock_t up_perf_gettime(void)
-{
-  return cp15_pmu_rdccr();
-}
-
 void up_perf_convert(clock_t elapsed, struct timespec *ts)
 {
   clock_t left;
@@ -97,4 +93,11 @@ void up_perf_convert(clock_t elapsed, struct timespec *ts)
   left        = elapsed - ts->tv_sec * g_cpu_freq;
   ts->tv_nsec = NSEC_PER_SEC * (uint64_t)left / g_cpu_freq;
 }
+#endif /* CONFIG_BUILD_FLAT || __KERNEL__ */
+
+clock_t up_perf_gettime(void)
+{
+  return cp15_pmu_rdccr();
+}
+
 #endif

sched/clock/clock_perf.c

@@ -31,7 +31,7 @@
 #include <nuttx/spinlock.h>
 #include <nuttx/wdog.h>
 
-#ifndef CONFIG_ARCH_HAVE_PERF_EVENTS_USER_ACCESS
+#ifndef CONFIG_ARCH_PERF_EVENTS_USER_ACCESS
 
 /****************************************************************************
  * Preprocessors
@@ -134,7 +134,7 @@ clock_t perf_gettime(void)
 }
 
 #  endif
-#endif /* !CONFIG_ARCH_HAVE_PERF_EVENTS_USER_ACCESS */
+#endif /* !CONFIG_ARCH_PERF_EVENTS_USER_ACCESS */
 
 #if defined(CONFIG_ALARM_ARCH) || defined (CONFIG_TIMER_ARCH) || \
     defined(CONFIG_ARCH_PERF_EVENTS)