* [PATCH] [RFC]RISC-V:[gprofng] Mimal support gprofng for riscv
@ 2024-06-30 9:37 XYenChi
2024-07-01 19:39 ` Vladimir Mezentsev
0 siblings, 1 reply; 7+ messages in thread
From: XYenChi @ 2024-06-30 9:37 UTC (permalink / raw)
To: binutils
Cc: vladimir.mezentsev, ruud.vanderpas, jiawei, shihua, shiyulong,
oriachiuan
Minimal support RISC-V. Test with qemu.
Result show:
[xyenchi@qemu-riscv build]$ ./gprofng/src/gp-display-text -fu1.er/p -head test.2.er/
Functions sorted by metric: Exclusive Total CPU Time
Excl. Total Incl. Total Name
CPU CPU
sec. % sec. %
0. 0. 0. 0. <Total>
ID Sel PID Experiment
== === ====== ==========
1 yes 145674 test.1.er/
Experiment: test.1.er/
No errors
No warnings
No archive command run
Target command (64-bit): 'echo 1'
Process pid 145674, ppid 399, pgrp 145674, sid 392
Current working directory: /home/xyenchi/binutils-gdb/build
Collector version: `2.41.50'; experiment version 12.4 (64-bit)
Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture `riscv64'
0 CPUs, clock speed 0 MHz.
Memory: 4096267 pages @ 4096 = 16001 MB.
Data collection parameters:
Clock-profiling, interval = 10007 microsecs.
Periodic sampling, 1 secs.
Follow descendant processes from: fork|exec|combo
Experiment started Thu Jun 27 16:23:20 2024
Experiment Ended: 0.040909200
Data Collection Duration: 0.040909200
If build with `allow_undefined_flag=true`, result will be:
[xyenchi@qemu-riscv build]$ ./gprofng/src/gp-display-text -fu1.er/p -head test.2.er/
Functions sorted by metric: Exclusive Total CPU Time
Excl. Total Incl. Total Name
CPU CPU
sec. % sec. %
0. 0. 0. 0. <Total>
ID Sel PID Experiment
== === ====== ==========
1 yes 145674 test.1.er/
Experiment: test.1.er/
No errors
No warnings
No archive command run
Target command (64-bit): 'echo 1'
Process pid 145674, ppid 399, pgrp 145674, sid 392
Current working directory: /home/xyenchi/binutils-gdb/build
Collector version: `2.41.50'; experiment version 12.4 (64-bit)
Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture `riscv64'
0 CPUs, clock speed 0 MHz.
Memory: 4096267 pages @ 4096 = 16001 MB.
Data collection parameters:
Clock-profiling, interval = 10007 microsecs.
Periodic sampling, 1 secs.
Follow descendant processes from: fork|exec|combo
Experiment started Thu Jun 27 16:23:20 2024
Experiment Ended: 0.040909200
Data Collection Duration: 0.040909200
ChangeLog:
Minimal support gprofng for RISC-V.
2024-06-28 Yixuan Chen <chenyixuan@iscasc.ac.cn>
* configure: Add RISC-V configure.
* configure.ac: Add RISC-V configure.
Unknown ChangeLog:
2024-06-28 Yixuan Chen <chenyixuan@iscasc.ac.cn>
* gprofng/common/core_pcbe.c (core_pcbe_init): Add RISC-V vendor.
(defined): Add RISC-V condition.
* gprofng/common/cpuid.c (defined): Add head file to use hwprobe.
(my_cpuid): Add riscv hwprobe.
* gprofng/common/gp-defs.h (TOK_A_RISCV): Add RISC-V.
(defined): Add RISC-V.
(ARCH_RISCV): Add RISC-V.
* gprofng/common/hwc_cpus.h: Add RISC-V vendor id.
* gprofng/common/hwcfuncs.h (HW_INTERVAL_TYPE): fix typo.
* gprofng/configure: Add RISC-V.
* gprofng/configure.ac: Add RISC-V.
* gprofng/libcollector/hwprofile.h (ARCH): Add riscv register calling.
(CONTEXT_PC): Add riscv register calling.
(CONTEXT_FP): Add riscv register calling.
(CONTEXT_SP): Add riscv register calling.
(SETFUNCTIONCONTEXT): Add riscv uc_mcontext calling.
* gprofng/libcollector/libcol_util.c (__collector_util_init): fix libc calling
* gprofng/libcollector/libcol_util.h (__collector_cas_32): Add RISC-V.
(ARCH):
* gprofng/libcollector/unwind.c (ARCH): Add RISC-V.
(GET_PC): Add riscv register calling.
(GET_SP): Add riscv register calling.
(GET_FP): Add riscv register calling.
(FILL_CONTEXT): Add riscv uc_mcontext calling.
* gprofng/src/DbeSession.cc (ARCH):
* gprofng/src/Disasm.cc (Disasm::disasm_open): Add RISC-V.
* gprofng/src/Experiment.cc (Experiment::ExperimentHandler::startElement): Add RISC-V.
* gprofng/src/checks.cc (ARCH): Add RISC-V.
* gprofng/src/collctrl.cc (_SC_CPUID_MAX): Add if not define _SC_CPUID_MAX condition. Set CPU frequency as 1000 to test but now not show at the test result, will fix with better way to get cpu frequency.
(Coll_Ctrl::Coll_Ctrl):
(defined):
* gprofng/src/dbe_types.h (enum Platform_t):Add RISC-V.
* gprofng/testsuite/gprofng.display/mttest/gethrtime.c: Add RISC-V.
---
configure | 2 +-
configure.ac | 2 +-
gprofng/common/core_pcbe.c | 5 ++-
gprofng/common/cpuid.c | 31 +++++++++++++++++--
gprofng/common/gp-defs.h | 5 ++-
gprofng/common/hwc_cpus.h | 7 +++++
gprofng/common/hwcfuncs.h | 2 +-
gprofng/configure | 4 +++
gprofng/configure.ac | 4 +++
gprofng/libcollector/hwprofile.h | 10 ++++++
gprofng/libcollector/libcol_util.c | 11 ++++++-
gprofng/libcollector/libcol_util.h | 10 +++---
gprofng/libcollector/unwind.c | 16 ++++++++--
gprofng/src/DbeSession.cc | 2 ++
gprofng/src/Disasm.cc | 2 ++
gprofng/src/Experiment.cc | 2 ++
gprofng/src/checks.cc | 4 +++
gprofng/src/collctrl.cc | 26 +++++++++-------
gprofng/src/dbe_types.h | 3 +-
.../gprofng.display/mttest/gethrtime.c | 2 ++
20 files changed, 123 insertions(+), 27 deletions(-)
diff --git a/configure b/configure
index dd743c58663..e3cbad5752b 100755
--- a/configure
+++ b/configure
@@ -3145,7 +3145,7 @@ fi
if test "$enable_gprofng" = "yes"; then
case "${target}" in
- x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux*)
+ x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux* | riscv64-*-linux*)
configdirs="$configdirs gprofng"
;;
esac
diff --git a/configure.ac b/configure.ac
index 01cfd017273..e63f27e6a29 100644
--- a/configure.ac
+++ b/configure.ac
@@ -412,7 +412,7 @@ enable_gprofng=$enableval,
enable_gprofng=yes)
if test "$enable_gprofng" = "yes"; then
case "${target}" in
- x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux*)
+ x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux* | riscv64-*-linux*)
configdirs="$configdirs gprofng"
;;
esac
diff --git a/gprofng/common/core_pcbe.c b/gprofng/common/core_pcbe.c
index b71d46f4e31..ef2a3f83d42 100644
--- a/gprofng/common/core_pcbe.c
+++ b/gprofng/common/core_pcbe.c
@@ -2754,6 +2754,9 @@ core_pcbe_init (void)
return 0;
case X86_VENDOR_Intel:
break;
+ case ANDES_VENDOR_ID:
+ case SIFIVE_VENDOR_ID:
+ case THEAD_VENDOR_ID:
default:
return -1;
}
@@ -2915,7 +2918,7 @@ core_pcbe_impl_name (void)
static const char *
core_pcbe_cpuref (void)
{
-#if defined(__aarch64__)
+#if defined(__aarch64__) || defined(__riscv)
return "";
#elif defined(__i386__) || defined(__x86_64)
switch (cpuid_getmodel ())
diff --git a/gprofng/common/cpuid.c b/gprofng/common/cpuid.c
index fd98b30c639..c29b6612afe 100644
--- a/gprofng/common/cpuid.c
+++ b/gprofng/common/cpuid.c
@@ -42,6 +42,11 @@ __get_cpuid (unsigned int op ATTRIBUTE_UNUSED, unsigned int *eax,
Tprintf (DBG_LT0, "cpuid.c:%d read_cpuid_id() MIDR_EL1=0x%016x\n", __LINE__, *eax);
return res;
}
+#elif defined(__riscv)
+#include <sched.h>
+#include <sys/syscall.h>
+#include <unistd.h>
+#include <asm/hwprobe.h>
#endif
/*
@@ -104,7 +109,7 @@ my_cpuid (unsigned int op, cpuid_regs_t *regs)
TprintfT (DBG_LT1, "my_cpuid: __get_cpuid(0x%x, 0x%x, 0x%x, 0x%x, 0x%x) returns %d\n",
op, regs->eax, regs->ebx, regs->ecx, regs->edx, ret);
return ret;
-}
+}
#endif
static cpuid_info_t *
@@ -180,8 +185,30 @@ get_cpuid_info ()
cpi->cpi_model += CPI_MODEL_XTD (regs.eax) << 4;
break;
}
+#elif defined(__riscv)
+ #ifndef __riscv_hwprobe
+ cpi->cpi_vendor = 0;
+ cpi->cpi_family = 0;
+ cpi->cpi_model = 0;
+ #else
+ struct riscv_hwprobe res;
+ res.key = RISCV_HWPROBE_KEY_MVENDORID;
+ cpu_set_t cpu_set;
+ int __riscv_hwprobe (struct riscv_hwprobe *pairs, \
+ long pair_count, long cpu_count, \
+ unsigned long *cpus, unsigned long flags) \
+ {
+ return syscall(__NR_riscv_hwprobe, pairs, pair_count, cpu_count, cpus, flags);
+ }
+ CPU_ZERO(&cpu_set);
+ CPU_SET(0, &cpu_set);
+ long ret = __riscv_hwprobe(&res, 1, 1, &cpu_set, 0);
+ cpi->cpi_vendor = res.value;
+ cpi->cpi_family = 0;
+ cpi->cpi_model = 0;
+ #endif
#endif
- return cpi;
+ return cpi;
}
static inline uint_t
diff --git a/gprofng/common/gp-defs.h b/gprofng/common/gp-defs.h
index 7cef5550696..891c4d32cf1 100644
--- a/gprofng/common/gp-defs.h
+++ b/gprofng/common/gp-defs.h
@@ -32,6 +32,7 @@
*/
#define ARCH(x) TOK_A_##x(ARCH)
#define TOK_A_Aarch64(x) x##_Aarch64
+#define TOK_A_RISCV(x) x##_RISCV
#define TOK_A_SPARC(x) x##_SPARC
#define TOK_A_Intel(x) x##_Intel
@@ -45,11 +46,13 @@
#define ARCH_Intel 1
#elif defined(__aarch64__)
#define ARCH_Aarch64 1
+#elif defined(riscv) || defined(__riscv)
+#define ARCH_RISCV 1
#else
#error "Undefined platform"
#endif
-#if defined(__sparcv9) || defined(__x86_64) || defined(__aarch64__)
+#if defined(__sparcv9) || defined(__x86_64) || defined(__aarch64__) || defined(__riscv)
#define WSIZE_64 1
#else
#define WSIZE_32 1
diff --git a/gprofng/common/hwc_cpus.h b/gprofng/common/hwc_cpus.h
index be820819dd3..b121e4bfb97 100644
--- a/gprofng/common/hwc_cpus.h
+++ b/gprofng/common/hwc_cpus.h
@@ -111,6 +111,13 @@ enum {
ARM_CPU_IMP_QCOM = 0x51
};
+// rscv Constants from arch/riscv/include/asm/vendorid_list.h
+enum {
+ ANDES_VENDOR_ID =0x31e,
+ SIFIVE_VENDOR_ID =0x489,
+ THEAD_VENDOR_ID =0x5b7
+};
+
#define AARCH64_VENDORSTR_ARM "ARM"
/* strings below must match those returned by cpc_getcpuver() */
diff --git a/gprofng/common/hwcfuncs.h b/gprofng/common/hwcfuncs.h
index a5f64f2a33f..1a6ee7e30ca 100644
--- a/gprofng/common/hwcfuncs.h
+++ b/gprofng/common/hwcfuncs.h
@@ -98,7 +98,7 @@ typedef struct { /* supplementary data fields */
#define HW_INTERVAL_MAX UINT64_MAX
#define HW_INTERVAL_PRESET(x) (HW_INTERVAL_MAX - ((uint64_t)(x) - 1))
-#define HW_INTERVAL_TYPE(x) ((uint64_t) (x)
+#define HW_INTERVAL_TYPE(x) ((uint64_t) (x))
/* parsing */
#define HWCFUNCS_MAX_ATTRS 20
diff --git a/gprofng/configure b/gprofng/configure
index 1c6a99291be..9ac7a651809 100755
--- a/gprofng/configure
+++ b/gprofng/configure
@@ -15788,6 +15788,10 @@ build_src=
build_src=true
build_collector=true
;;
+ riscv*-*-linux*)
+ build_src=true
+ build_collector=true
+ ;;
esac
# Check whether --enable-gprofng-tools was given.
if test "${enable_gprofng_tools+set}" = set; then :
diff --git a/gprofng/configure.ac b/gprofng/configure.ac
index 6b8fe262748..b4f2981a487 100644
--- a/gprofng/configure.ac
+++ b/gprofng/configure.ac
@@ -63,6 +63,10 @@ build_src=
build_src=true
build_collector=true
;;
+ riscv*-*-linux*)
+ build_src=true
+ build_collector=true
+ ;;
esac
AC_ARG_ENABLE(gprofng-tools,
AS_HELP_STRING([--disable-gprofng-tools], [do not build gprofng/src directory]),
diff --git a/gprofng/libcollector/hwprofile.h b/gprofng/libcollector/hwprofile.h
index 23422dd51fa..6517452eef7 100644
--- a/gprofng/libcollector/hwprofile.h
+++ b/gprofng/libcollector/hwprofile.h
@@ -84,6 +84,16 @@ typedef struct MHwcntr_packet
(ucp)->uc_mcontext.regs[CONTEXT_PC] = (greg_t)(funcp); \
(ucp)->uc_mcontext.regs[CONTEXT_SP] = 0; \
(ucp)->uc_mcontext.regs[CONTEXT_FP] = 0;
+
+#elif ARCH(RISCV)
+#define CONTEXT_PC REG_PC
+#define CONTEXT_FP 8
+#define CONTEXT_SP 2
+#define SETFUNCTIONCONTEXT(ucp,funcp) \
+ (ucp)->uc_mcontext.__gregs[CONTEXT_PC] = (greg_t)(funcp); \
+ (ucp)->uc_mcontext.__gregs[CONTEXT_FP] = 0; \
+ (ucp)->uc_mcontext.__gregs[CONTEXT_SP] = 0;
+
#endif /* ARCH() */
#endif
diff --git a/gprofng/libcollector/libcol_util.c b/gprofng/libcollector/libcol_util.c
index a8802d4a9b0..8ee1c56ff45 100644
--- a/gprofng/libcollector/libcol_util.c
+++ b/gprofng/libcollector/libcol_util.c
@@ -91,7 +91,7 @@ __collector_gettid ()
#endif
__asm__ __volatile__(syscall_instr
: "=a" (r) : "0" (__NR_gettid)
- : syscall_clobber);
+ : syscall_clobber);
#else
r = syscall (__NR_gettid);
#endif
@@ -1459,7 +1459,10 @@ __collector_util_init ()
else if ((ptr = dlvsym (libc, "fopen", "GLIBC_2.0")) != NULL)
__collector_util_funcs.fopen = ptr;
else
+ {
ptr = dlsym (libc, "fopen");
+ if(ptr) __collector_util_funcs.fopen = ptr;
+ }
if (__collector_util_funcs.fopen == NULL)
{
CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT fopen: %s\n", dlerror ());
@@ -1475,7 +1478,10 @@ __collector_util_init ()
else if ((ptr = dlvsym (libc, "popen", "GLIBC_2.0")) != NULL)
__collector_util_funcs.popen = ptr;
else
+ {
ptr = dlsym (libc, "popen");
+ if(ptr) __collector_util_funcs.popen = ptr;
+ }
if (__collector_util_funcs.popen == NULL)
{
CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT popen: %s\n", dlerror ());
@@ -1491,7 +1497,10 @@ __collector_util_init ()
else if ((ptr = dlvsym (libc, "fclose", "GLIBC_2.0")) != NULL)
__collector_util_funcs.fclose = ptr;
else
+ {
ptr = dlsym (libc, "fclose");
+ if(ptr) __collector_util_funcs.fclose = ptr;
+ }
if (__collector_util_funcs.fclose == NULL)
{
CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT fclose: %s\n", dlerror ());
diff --git a/gprofng/libcollector/libcol_util.h b/gprofng/libcollector/libcol_util.h
index c21b4a46c4b..6ac84db05ab 100644
--- a/gprofng/libcollector/libcol_util.h
+++ b/gprofng/libcollector/libcol_util.h
@@ -209,10 +209,10 @@ static __attribute__ ((always_inline)) inline uint32_t
__collector_cas_32 (volatile uint32_t *pdata, uint32_t old, uint32_t new)
{
uint32_t r;
- __asm__ __volatile__("lock; cmpxchgl %2, %1"
- : "=a" (r), "=m" (*pdata) : "r" (new),
- "a" (old), "m" (*pdata));
- return r;
+ __asm__ __volatile__("lock; cmpxchgl %2, %1"
+ : "=a" (r), "=m" (*pdata) : "r" (new),
+ "a" (old), "m" (*pdata));
+ return r;
}
/**
* This function enables a compare and swap operation to occur atomically.
@@ -270,7 +270,7 @@ __collector_cas_ptr (void *mem, void *cmp, void *new)
return r;
}
-#elif ARCH(Aarch64)
+#elif ARCH(Aarch64) || ARCH(RISCV)
static __attribute__ ((always_inline)) inline uint32_t
__collector_inc_32 (volatile uint32_t *ptr)
{
diff --git a/gprofng/libcollector/unwind.c b/gprofng/libcollector/unwind.c
index ff2f7aa9a7e..1244a453457 100644
--- a/gprofng/libcollector/unwind.c
+++ b/gprofng/libcollector/unwind.c
@@ -186,8 +186,14 @@ memory_error_func (int status ATTRIBUTE_UNUSED, bfd_vma addr ATTRIBUTE_UNUSED,
#define GET_PC(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[15])
#define GET_SP(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[13])
#define GET_FP(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[14])
+
+#elif ARCH(RISCV)
+#define GET_PC(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[REG_PC])
+#define GET_SP(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[2])
+#define GET_FP(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[8])
#endif /* ARCH() */
+
/*
* FILL_CONTEXT() for all platforms
* Could use getcontext() except:
@@ -230,13 +236,19 @@ memory_error_func (int status ATTRIBUTE_UNUSED, bfd_vma addr ATTRIBUTE_UNUSED,
context->uc_stack.ss_size = 0x100000; \
}
-#elif ARCH(Aarch64)
+#elif ARCH(Aarch64)
#define FILL_CONTEXT(context) \
{ CALL_UTIL (getcontext) (context); \
context->uc_mcontext.sp = (__u64) __builtin_frame_address(0); \
}
-#endif /* ARCH() */
+#elif ARCH(RISCV)
+#define FILL_CONTEXT(context) \
+ { CALL_UTIL(getcontext)(context); \
+ context->uc_mcontext.__gregs[2] = (uint64_t) __builtin_frame_address(0); \
+ }
+
+#endif/* ARCH() */
static int
getByteInstruction (unsigned char *p)
diff --git a/gprofng/src/DbeSession.cc b/gprofng/src/DbeSession.cc
index 86541d97f2e..3c218b5fc91 100644
--- a/gprofng/src/DbeSession.cc
+++ b/gprofng/src/DbeSession.cc
@@ -94,6 +94,8 @@ Platform_t DbeSession::platform =
Sparc;
#elif ARCH(Aarch64)
Aarch64;
+#elif ARCH(RISCV)
+ RISCV;
#else // ARCH(Intel)
Intel;
#endif
diff --git a/gprofng/src/Disasm.cc b/gprofng/src/Disasm.cc
index 9c7e59a590b..19e6f15ff6f 100644
--- a/gprofng/src/Disasm.cc
+++ b/gprofng/src/Disasm.cc
@@ -208,6 +208,7 @@ Disasm::disasm_open ()
case Amd64:
need_swap_endian = (DbeSession::platform == Sparc);
break;
+ case RISCV:
case Sparcv8plus:
case Sparcv9:
case Sparc:
@@ -246,6 +247,7 @@ Disasm::disasm_open ()
dis_info.arch = bfd_arch_i386;
dis_info.mach = bfd_mach_x86_64;
break;
+ case RISCV:
case Sparcv8plus:
case Sparcv9:
case Sparc:
diff --git a/gprofng/src/Experiment.cc b/gprofng/src/Experiment.cc
index 1378ad5ce07..ac78276d23c 100644
--- a/gprofng/src/Experiment.cc
+++ b/gprofng/src/Experiment.cc
@@ -542,6 +542,8 @@ Experiment::ExperimentHandler::startElement (char*, char*, char *qName, Attribut
exp->platform = Intel;
else if (strcmp (str, "aarch64") == 0)
exp->platform = Aarch64;
+ else if (strcmp (str, "riscv64") == 0)
+ exp->platform = RISCV;
else
exp->platform = Sparc;
exp->need_swap_endian = (DbeSession::platform == Sparc) ?
diff --git a/gprofng/src/checks.cc b/gprofng/src/checks.cc
index 094c3bbc60a..2f9fbd7f714 100644
--- a/gprofng/src/checks.cc
+++ b/gprofng/src/checks.cc
@@ -332,6 +332,10 @@ collect::check_executable_arch (Elf *elf)
case EM_AARCH64:
is_64 = true;
break;
+#elif ARCH(RISCV)
+ case EM_RISCV:
+ is_64 = true;
+ break;
#endif
default:
return EXEC_ELF_ARCH;
diff --git a/gprofng/src/collctrl.cc b/gprofng/src/collctrl.cc
index 5d68b689a64..1b04e52b5cc 100644
--- a/gprofng/src/collctrl.cc
+++ b/gprofng/src/collctrl.cc
@@ -51,9 +51,9 @@ extern const char *strsignal (int);
#endif
// _SC_CPUID_MAX is not available on 2.6/2.7
-#ifndef _SC_CPUID_MAX
-#define _SC_CPUID_MAX 517
-#endif
+
+// #define _SC_CPUID_MAX 517
+// #endif
const char *get_fstype (char *);
@@ -74,15 +74,16 @@ Coll_Ctrl::Coll_Ctrl (int _interactive, bool _defHWC, bool _kernelHWC)
default_stem = strdup ("test");
/* get CPU count and processor clock rate */
+ #ifndef _SC_CPUID_MAX
+ ncpus = sysconf (_SC_NPROCESSORS_CONF);
+ /* add 2048 to count, since on some systems CPUID does not start at zero */
+ ncpumax = ncpus + 2048;
+ #elif
ncpumax = sysconf (_SC_CPUID_MAX);
- if (ncpumax == -1)
- {
- ncpus = sysconf (_SC_NPROCESSORS_CONF);
- /* add 2048 to count, since on some systems CPUID does not start at zero */
- ncpumax = ncpus + 2048;
- }
- ncpus = 0;
- cpu_clk_freq = 0;
+ #endif
+
+ //ncpus = 0;
+ //cpu_clk_freq = 0;
// On Linux, read /proc/cpuinfo to get CPU count and clock rate
// Note that parsing is different on SPARC and x86
@@ -115,6 +116,9 @@ Coll_Ctrl::Coll_Ctrl (int _interactive, bool _defHWC, bool _kernelHWC)
#elif defined(__aarch64__)
asm volatile("mrs %0, cntfrq_el0" : "=r" (cpu_clk_freq));
+#elif defined(__riscv)
+ cpu_clk_freq = 1000;
+
#else
FILE *procf = fopen ("/proc/cpuinfo", "r");
if (procf != NULL)
diff --git a/gprofng/src/dbe_types.h b/gprofng/src/dbe_types.h
index 9fa842ce85d..c67d421566e 100644
--- a/gprofng/src/dbe_types.h
+++ b/gprofng/src/dbe_types.h
@@ -42,7 +42,8 @@ enum Platform_t
Sparcv8plus,
Java,
Amd64,
- Aarch64
+ Aarch64,
+ RISCV
};
enum WSize_t
diff --git a/gprofng/testsuite/gprofng.display/mttest/gethrtime.c b/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
index da39821027f..5e2cc3dc97b 100644
--- a/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
+++ b/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
@@ -32,6 +32,8 @@
#define SPARC 1
#elif defined(__aarch64__)
#define Aarch64 1
+#elif defined(__riscv)
+#define RISCV 1
#else
#define Intel 1
#endif
--
2.44.0
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH] [RFC]RISC-V:[gprofng] Mimal support gprofng for riscv
2024-06-30 9:37 [PATCH] [RFC]RISC-V:[gprofng] Mimal support gprofng for riscv XYenChi
@ 2024-07-01 19:39 ` Vladimir Mezentsev
2024-07-03 12:38 ` Yixuan Chen
0 siblings, 1 reply; 7+ messages in thread
From: Vladimir Mezentsev @ 2024-07-01 19:39 UTC (permalink / raw)
To: XYenChi, binutils; +Cc: ruud.vanderpas, jiawei, shihua, shiyulong
[-- Attachment #1: Type: text/plain, Size: 23002 bytes --]
Thank you for your work on gprofng.
See my comments below.
On 6/30/24 02:37, XYenChi wrote:
> Minimal support RISC-V. Test with qemu.
>
> Result show:
> [xyenchi@qemu-riscv build]$ ./gprofng/src/gp-display-text -fu1.er/p -head test.2.er/
> Functions sorted by metric: Exclusive Total CPU Time
>
> Excl. Total Incl. Total Name
> CPU CPU
> sec. % sec. %
> 0. 0. 0. 0. <Total>
>
> ID Sel PID Experiment
> == === ====== ==========
> 1 yes 145674 test.1.er/
> Experiment: test.1.er/
> No errors
> No warnings
> No archive command run
>
> Target command (64-bit): 'echo 1'
> Process pid 145674, ppid 399, pgrp 145674, sid 392
> Current working directory: /home/xyenchi/binutils-gdb/build
> Collector version: `2.41.50'; experiment version 12.4 (64-bit)
It looks like you are using an old version of the binutils-gdb source.
The current version is 2.42.50.
> Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture `riscv64'
> 0 CPUs, clock speed 0 MHz.
> Memory: 4096267 pages @ 4096 = 16001 MB.
> Data collection parameters:
> Clock-profiling, interval = 10007 microsecs.
> Periodic sampling, 1 secs.
> Follow descendant processes from: fork|exec|combo
>
> Experiment started Thu Jun 27 16:23:20 2024
>
> Experiment Ended: 0.040909200
> Data Collection Duration: 0.040909200
>
> If build with `allow_undefined_flag=true`, result will be:
>
> [xyenchi@qemu-riscv build]$ ./gprofng/src/gp-display-text -fu1.er/p -head test.2.er/
> Functions sorted by metric: Exclusive Total CPU Time
>
> Excl. Total Incl. Total Name
> CPU CPU
> sec. % sec. %
> 0. 0. 0. 0. <Total>
>
> ID Sel PID Experiment
> == === ====== ==========
> 1 yes 145674 test.1.er/
> Experiment: test.1.er/
> No errors
> No warnings
> No archive command run
>
> Target command (64-bit): 'echo 1'
> Process pid 145674, ppid 399, pgrp 145674, sid 392
> Current working directory: /home/xyenchi/binutils-gdb/build
> Collector version: `2.41.50'; experiment version 12.4 (64-bit)
> Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture `riscv64'
> 0 CPUs, clock speed 0 MHz.
> Memory: 4096267 pages @ 4096 = 16001 MB.
> Data collection parameters:
> Clock-profiling, interval = 10007 microsecs.
> Periodic sampling, 1 secs.
> Follow descendant processes from: fork|exec|combo
>
> Experiment started Thu Jun 27 16:23:20 2024
>
> Experiment Ended: 0.040909200
> Data Collection Duration: 0.040909200
>
> ChangeLog:
> Minimal support gprofng for RISC-V.
>
> 2024-06-28 Yixuan Chen<chenyixuan@iscasc.ac.cn>
>
> * configure: Add RISC-V configure.
> * configure.ac: Add RISC-V configure.
>
> Unknown ChangeLog:
>
> 2024-06-28 Yixuan Chen<chenyixuan@iscasc.ac.cn>
>
> * gprofng/common/core_pcbe.c (core_pcbe_init): Add RISC-V vendor.
> (defined): Add RISC-V condition.
> * gprofng/common/cpuid.c (defined): Add head file to use hwprobe.
> (my_cpuid): Add riscv hwprobe.
> * gprofng/common/gp-defs.h (TOK_A_RISCV): Add RISC-V.
> (defined): Add RISC-V.
> (ARCH_RISCV): Add RISC-V.
> * gprofng/common/hwc_cpus.h: Add RISC-V vendor id.
> * gprofng/common/hwcfuncs.h (HW_INTERVAL_TYPE): fix typo.
> * gprofng/configure: Add RISC-V.
> * gprofng/configure.ac: Add RISC-V.
> * gprofng/libcollector/hwprofile.h (ARCH): Add riscv register calling.
> (CONTEXT_PC): Add riscv register calling.
> (CONTEXT_FP): Add riscv register calling.
> (CONTEXT_SP): Add riscv register calling.
> (SETFUNCTIONCONTEXT): Add riscv uc_mcontext calling.
> * gprofng/libcollector/libcol_util.c (__collector_util_init): fix libc calling
> * gprofng/libcollector/libcol_util.h (__collector_cas_32): Add RISC-V.
> (ARCH):
> * gprofng/libcollector/unwind.c (ARCH): Add RISC-V.
> (GET_PC): Add riscv register calling.
> (GET_SP): Add riscv register calling.
> (GET_FP): Add riscv register calling.
> (FILL_CONTEXT): Add riscv uc_mcontext calling.
> * gprofng/src/DbeSession.cc (ARCH):
> * gprofng/src/Disasm.cc (Disasm::disasm_open): Add RISC-V.
> * gprofng/src/Experiment.cc (Experiment::ExperimentHandler::startElement): Add RISC-V.
> * gprofng/src/checks.cc (ARCH): Add RISC-V.
> * gprofng/src/collctrl.cc (_SC_CPUID_MAX): Add if not define _SC_CPUID_MAX condition. Set CPU frequency as 1000 to test but now not show at the test result, will fix with better way to get cpu frequency.
> (Coll_Ctrl::Coll_Ctrl):
> (defined):
> * gprofng/src/dbe_types.h (enum Platform_t):Add RISC-V.
> * gprofng/testsuite/gprofng.display/mttest/gethrtime.c: Add RISC-V.
>
> ---
> configure | 2 +-
> configure.ac | 2 +-
> gprofng/common/core_pcbe.c | 5 ++-
> gprofng/common/cpuid.c | 31 +++++++++++++++++--
> gprofng/common/gp-defs.h | 5 ++-
> gprofng/common/hwc_cpus.h | 7 +++++
> gprofng/common/hwcfuncs.h | 2 +-
> gprofng/configure | 4 +++
> gprofng/configure.ac | 4 +++
> gprofng/libcollector/hwprofile.h | 10 ++++++
> gprofng/libcollector/libcol_util.c | 11 ++++++-
> gprofng/libcollector/libcol_util.h | 10 +++---
> gprofng/libcollector/unwind.c | 16 ++++++++--
> gprofng/src/DbeSession.cc | 2 ++
> gprofng/src/Disasm.cc | 2 ++
> gprofng/src/Experiment.cc | 2 ++
> gprofng/src/checks.cc | 4 +++
> gprofng/src/collctrl.cc | 26 +++++++++-------
> gprofng/src/dbe_types.h | 3 +-
> .../gprofng.display/mttest/gethrtime.c | 2 ++
> 20 files changed, 123 insertions(+), 27 deletions(-)
>
> diff --git a/configure b/configure
> index dd743c58663..e3cbad5752b 100755
> --- a/configure
> +++ b/configure
> @@ -3145,7 +3145,7 @@ fi
>
> if test "$enable_gprofng" = "yes"; then
> case "${target}" in
> - x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux*)
> + x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux* | riscv64-*-linux*)
> configdirs="$configdirs gprofng"
> ;;
> esac
> diff --git a/configure.ac b/configure.ac
> index 01cfd017273..e63f27e6a29 100644
> --- a/configure.ac
> +++ b/configure.ac
> @@ -412,7 +412,7 @@ enable_gprofng=$enableval,
> enable_gprofng=yes)
> if test "$enable_gprofng" = "yes"; then
> case "${target}" in
> - x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux*)
> + x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux* | riscv64-*-linux*)
> configdirs="$configdirs gprofng"
> ;;
> esac
> diff --git a/gprofng/common/core_pcbe.c b/gprofng/common/core_pcbe.c
> index b71d46f4e31..ef2a3f83d42 100644
> --- a/gprofng/common/core_pcbe.c
> +++ b/gprofng/common/core_pcbe.c
> @@ -2754,6 +2754,9 @@ core_pcbe_init (void)
> return 0;
> case X86_VENDOR_Intel:
> break;
> + case ANDES_VENDOR_ID:
The formatting is wrong.
See a coding standard:
https://sourceware.org/gdb/wiki/Internals%20GDB-C-Coding-Standards#Whitespaces
/*Code indentation*//
// Lines should be indented with a mix of tabs and spaces. 8 spaces
should be replaced with Tab.
Vim users could add the following to their ~/.vimrc://
//// set autoindent tabstop=8 shiftwidth=2//
/
> + case SIFIVE_VENDOR_ID:
> + case THEAD_VENDOR_ID:
> default:
> return -1;
> }
> @@ -2915,7 +2918,7 @@ core_pcbe_impl_name (void)
> static const char *
> core_pcbe_cpuref (void)
> {
> -#if defined(__aarch64__)
> +#if defined(__aarch64__) || defined(__riscv)
> return "";
> #elif defined(__i386__) || defined(__x86_64)
> switch (cpuid_getmodel ())
> diff --git a/gprofng/common/cpuid.c b/gprofng/common/cpuid.c
> index fd98b30c639..c29b6612afe 100644
> --- a/gprofng/common/cpuid.c
> +++ b/gprofng/common/cpuid.c
> @@ -42,6 +42,11 @@ __get_cpuid (unsigned int op ATTRIBUTE_UNUSED, unsigned int *eax,
> Tprintf (DBG_LT0, "cpuid.c:%d read_cpuid_id() MIDR_EL1=0x%016x\n", __LINE__, *eax);
> return res;
> }
> +#elif defined(__riscv)
> +#include <sched.h>
> +#include <sys/syscall.h>
> +#include <unistd.h>
> +#include <asm/hwprobe.h>
> #endif
>
> /*
> @@ -104,7 +109,7 @@ my_cpuid (unsigned int op, cpuid_regs_t *regs)
> TprintfT (DBG_LT1, "my_cpuid: __get_cpuid(0x%x, 0x%x, 0x%x, 0x%x, 0x%x) returns %d\n",
> op, regs->eax, regs->ebx, regs->ecx, regs->edx, ret);
> return ret;
> -}
> +}
> #endif
>
> static cpuid_info_t *
> @@ -180,8 +185,30 @@ get_cpuid_info ()
> cpi->cpi_model += CPI_MODEL_XTD (regs.eax) << 4;
> break;
> }
> +#elif defined(__riscv)
> + #ifndef __riscv_hwprobe
> + cpi->cpi_vendor = 0;
> + cpi->cpi_family = 0;
> + cpi->cpi_model = 0;
> + #else
> + struct riscv_hwprobe res;
> + res.key = RISCV_HWPROBE_KEY_MVENDORID;
> + cpu_set_t cpu_set;
> + int __riscv_hwprobe (struct riscv_hwprobe *pairs, \
> + long pair_count, long cpu_count, \
> + unsigned long *cpus, unsigned long flags) \
> + {
> + return syscall(__NR_riscv_hwprobe, pairs, pair_count, cpu_count, cpus, flags);
> + }
> + CPU_ZERO(&cpu_set);
> + CPU_SET(0, &cpu_set);
> + long ret = __riscv_hwprobe(&res, 1, 1, &cpu_set, 0);
> + cpi->cpi_vendor = res.value;
> + cpi->cpi_family = 0;
> + cpi->cpi_model = 0;
> + #endif
> #endif
> - return cpi;
> + return cpi;
> }
Could you also add a code into read_cpuinfo() (gprofng/src/collctrl.cc
line ~85)
to read cpu_model, cpu_family, etc from /proc/cpu_info.
>
> static inline uint_t
> diff --git a/gprofng/common/gp-defs.h b/gprofng/common/gp-defs.h
> index 7cef5550696..891c4d32cf1 100644
> --- a/gprofng/common/gp-defs.h
> +++ b/gprofng/common/gp-defs.h
> @@ -32,6 +32,7 @@
> */
> #define ARCH(x) TOK_A_##x(ARCH)
> #define TOK_A_Aarch64(x) x##_Aarch64
> +#define TOK_A_RISCV(x) x##_RISCV
> #define TOK_A_SPARC(x) x##_SPARC
> #define TOK_A_Intel(x) x##_Intel
>
> @@ -45,11 +46,13 @@
> #define ARCH_Intel 1
> #elif defined(__aarch64__)
> #define ARCH_Aarch64 1
> +#elif defined(riscv) || defined(__riscv)
> +#define ARCH_RISCV 1
> #else
> #error "Undefined platform"
> #endif
>
> -#if defined(__sparcv9) || defined(__x86_64) || defined(__aarch64__)
> +#if defined(__sparcv9) || defined(__x86_64) || defined(__aarch64__) || defined(__riscv)
> #define WSIZE_64 1
> #else
> #define WSIZE_32 1
> diff --git a/gprofng/common/hwc_cpus.h b/gprofng/common/hwc_cpus.h
> index be820819dd3..b121e4bfb97 100644
> --- a/gprofng/common/hwc_cpus.h
> +++ b/gprofng/common/hwc_cpus.h
> @@ -111,6 +111,13 @@ enum {
> ARM_CPU_IMP_QCOM = 0x51
> };
>
> +// rscv Constants from arch/riscv/include/asm/vendorid_list.h
> +enum {
> + ANDES_VENDOR_ID =0x31e,
> + SIFIVE_VENDOR_ID =0x489,
> + THEAD_VENDOR_ID =0x5b7
> +};
> +
> #define AARCH64_VENDORSTR_ARM "ARM"
>
> /* strings below must match those returned by cpc_getcpuver() */
> diff --git a/gprofng/common/hwcfuncs.h b/gprofng/common/hwcfuncs.h
> index a5f64f2a33f..1a6ee7e30ca 100644
> --- a/gprofng/common/hwcfuncs.h
> +++ b/gprofng/common/hwcfuncs.h
> @@ -98,7 +98,7 @@ typedef struct { /* supplementary data fields */
>
> #define HW_INTERVAL_MAX UINT64_MAX
> #define HW_INTERVAL_PRESET(x) (HW_INTERVAL_MAX - ((uint64_t)(x) - 1))
> -#define HW_INTERVAL_TYPE(x) ((uint64_t) (x)
> +#define HW_INTERVAL_TYPE(x) ((uint64_t) (x))
HW_INTERVAL_TYPE is not used. Just remove this line.
>
> /* parsing */
> #define HWCFUNCS_MAX_ATTRS 20
> diff --git a/gprofng/configure b/gprofng/configure
> index 1c6a99291be..9ac7a651809 100755
> --- a/gprofng/configure
> +++ b/gprofng/configure
> @@ -15788,6 +15788,10 @@ build_src=
> build_src=true
> build_collector=true
> ;;
> + riscv*-*-linux*)
> + build_src=true
> + build_collector=true
> + ;;
> esac
> # Check whether --enable-gprofng-tools was given.
> if test "${enable_gprofng_tools+set}" = set; then :
> diff --git a/gprofng/configure.ac b/gprofng/configure.ac
> index 6b8fe262748..b4f2981a487 100644
> --- a/gprofng/configure.ac
> +++ b/gprofng/configure.ac
> @@ -63,6 +63,10 @@ build_src=
> build_src=true
> build_collector=true
> ;;
> + riscv*-*-linux*)
> + build_src=true
> + build_collector=true
> + ;;
> esac
> AC_ARG_ENABLE(gprofng-tools,
> AS_HELP_STRING([--disable-gprofng-tools], [do not build gprofng/src directory]),
> diff --git a/gprofng/libcollector/hwprofile.h b/gprofng/libcollector/hwprofile.h
> index 23422dd51fa..6517452eef7 100644
> --- a/gprofng/libcollector/hwprofile.h
> +++ b/gprofng/libcollector/hwprofile.h
> @@ -84,6 +84,16 @@ typedef struct MHwcntr_packet
> (ucp)->uc_mcontext.regs[CONTEXT_PC] = (greg_t)(funcp); \
> (ucp)->uc_mcontext.regs[CONTEXT_SP] = 0; \
> (ucp)->uc_mcontext.regs[CONTEXT_FP] = 0;
> +
> +#elif ARCH(RISCV)
> +#define CONTEXT_PC REG_PC
> +#define CONTEXT_FP 8
> +#define CONTEXT_SP 2
> +#define SETFUNCTIONCONTEXT(ucp,funcp) \
> + (ucp)->uc_mcontext.__gregs[CONTEXT_PC] = (greg_t)(funcp); \
> + (ucp)->uc_mcontext.__gregs[CONTEXT_FP] = 0; \
> + (ucp)->uc_mcontext.__gregs[CONTEXT_SP] = 0;
> +
> #endif /* ARCH() */
>
> #endif
> diff --git a/gprofng/libcollector/libcol_util.c b/gprofng/libcollector/libcol_util.c
> index a8802d4a9b0..8ee1c56ff45 100644
> --- a/gprofng/libcollector/libcol_util.c
> +++ b/gprofng/libcollector/libcol_util.c
> @@ -91,7 +91,7 @@ __collector_gettid ()
> #endif
> __asm__ __volatile__(syscall_instr
> : "=a" (r) : "0" (__NR_gettid)
> - : syscall_clobber);
> + : syscall_clobber);
> #else
> r = syscall (__NR_gettid);
> #endif
> @@ -1459,7 +1459,10 @@ __collector_util_init ()
> else if ((ptr = dlvsym (libc, "fopen", "GLIBC_2.0")) != NULL)
> __collector_util_funcs.fopen = ptr;
> else
> + {
> ptr = dlsym (libc, "fopen");
> + if(ptr) __collector_util_funcs.fopen = ptr;
> + }
> if (__collector_util_funcs.fopen == NULL)
> {
> CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT fopen: %s\n", dlerror ());
> @@ -1475,7 +1478,10 @@ __collector_util_init ()
> else if ((ptr = dlvsym (libc, "popen", "GLIBC_2.0")) != NULL)
> __collector_util_funcs.popen = ptr;
> else
> + {
> ptr = dlsym (libc, "popen");
> + if(ptr) __collector_util_funcs.popen = ptr;
> + }
> if (__collector_util_funcs.popen == NULL)
> {
> CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT popen: %s\n", dlerror ());
> @@ -1491,7 +1497,10 @@ __collector_util_init ()
> else if ((ptr = dlvsym (libc, "fclose", "GLIBC_2.0")) != NULL)
> __collector_util_funcs.fclose = ptr;
> else
> + {
> ptr = dlsym (libc, "fclose");
> + if(ptr) __collector_util_funcs.fclose = ptr;
> + }
> if (__collector_util_funcs.fclose == NULL)
> {
> CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT fclose: %s\n", dlerror ());
> diff --git a/gprofng/libcollector/libcol_util.h b/gprofng/libcollector/libcol_util.h
> index c21b4a46c4b..6ac84db05ab 100644
> --- a/gprofng/libcollector/libcol_util.h
> +++ b/gprofng/libcollector/libcol_util.h
> @@ -209,10 +209,10 @@ static __attribute__ ((always_inline)) inline uint32_t
> __collector_cas_32 (volatile uint32_t *pdata, uint32_t old, uint32_t new)
> {
> uint32_t r;
> - __asm__ __volatile__("lock; cmpxchgl %2, %1"
> - : "=a" (r), "=m" (*pdata) : "r" (new),
> - "a" (old), "m" (*pdata));
> - return r;
> + __asm__ __volatile__("lock; cmpxchgl %2, %1"
> + : "=a" (r), "=m" (*pdata) : "r" (new),
> + "a" (old), "m" (*pdata));
> + return r;
> }
> /**
> * This function enables a compare and swap operation to occur atomically.
> @@ -270,7 +270,7 @@ __collector_cas_ptr (void *mem, void *cmp, void *new)
> return r;
> }
>
> -#elif ARCH(Aarch64)
> +#elif ARCH(Aarch64) || ARCH(RISCV)
> static __attribute__ ((always_inline)) inline uint32_t
> __collector_inc_32 (volatile uint32_t *ptr)
> {
> diff --git a/gprofng/libcollector/unwind.c b/gprofng/libcollector/unwind.c
> index ff2f7aa9a7e..1244a453457 100644
> --- a/gprofng/libcollector/unwind.c
> +++ b/gprofng/libcollector/unwind.c
> @@ -186,8 +186,14 @@ memory_error_func (int status ATTRIBUTE_UNUSED, bfd_vma addr ATTRIBUTE_UNUSED,
> #define GET_PC(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[15])
> #define GET_SP(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[13])
> #define GET_FP(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[14])
> +
> +#elif ARCH(RISCV)
> +#define GET_PC(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[REG_PC])
> +#define GET_SP(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[2])
> +#define GET_FP(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[8])
> #endif /* ARCH() */
>
> +
> /*
> * FILL_CONTEXT() for all platforms
> * Could use getcontext() except:
> @@ -230,13 +236,19 @@ memory_error_func (int status ATTRIBUTE_UNUSED, bfd_vma addr ATTRIBUTE_UNUSED,
> context->uc_stack.ss_size = 0x100000; \
> }
>
> -#elif ARCH(Aarch64)
> +#elif ARCH(Aarch64)
> #define FILL_CONTEXT(context) \
> { CALL_UTIL (getcontext) (context); \
> context->uc_mcontext.sp = (__u64) __builtin_frame_address(0); \
> }
>
> -#endif /* ARCH() */
> +#elif ARCH(RISCV)
> +#define FILL_CONTEXT(context) \
> + { CALL_UTIL(getcontext)(context); \
> + context->uc_mcontext.__gregs[2] = (uint64_t) __builtin_frame_address(0); \
> + }
> +
> +#endif/* ARCH() */
>
> static int
> getByteInstruction (unsigned char *p)
> diff --git a/gprofng/src/DbeSession.cc b/gprofng/src/DbeSession.cc
> index 86541d97f2e..3c218b5fc91 100644
> --- a/gprofng/src/DbeSession.cc
> +++ b/gprofng/src/DbeSession.cc
> @@ -94,6 +94,8 @@ Platform_t DbeSession::platform =
> Sparc;
> #elif ARCH(Aarch64)
> Aarch64;
> +#elif ARCH(RISCV)
> + RISCV;
> #else // ARCH(Intel)
> Intel;
> #endif
> diff --git a/gprofng/src/Disasm.cc b/gprofng/src/Disasm.cc
> index 9c7e59a590b..19e6f15ff6f 100644
> --- a/gprofng/src/Disasm.cc
> +++ b/gprofng/src/Disasm.cc
> @@ -208,6 +208,7 @@ Disasm::disasm_open ()
> case Amd64:
> need_swap_endian = (DbeSession::platform == Sparc);
> break;
> + case RISCV:
> case Sparcv8plus:
> case Sparcv9:
> case Sparc:
> @@ -246,6 +247,7 @@ Disasm::disasm_open ()
> dis_info.arch = bfd_arch_i386;
> dis_info.mach = bfd_mach_x86_64;
> break;
> + case RISCV:
> case Sparcv8plus:
> case Sparcv9:
> case Sparc:
> diff --git a/gprofng/src/Experiment.cc b/gprofng/src/Experiment.cc
> index 1378ad5ce07..ac78276d23c 100644
> --- a/gprofng/src/Experiment.cc
> +++ b/gprofng/src/Experiment.cc
> @@ -542,6 +542,8 @@ Experiment::ExperimentHandler::startElement (char*, char*, char *qName, Attribut
> exp->platform = Intel;
> else if (strcmp (str, "aarch64") == 0)
> exp->platform = Aarch64;
> + else if (strcmp (str, "riscv64") == 0)
> + exp->platform = RISCV;
The formatting is wrong.
It looks like in your environment the tab size is 4 spaces, but in the
GNU standard it is 8.
> else
> exp->platform = Sparc;
> exp->need_swap_endian = (DbeSession::platform == Sparc) ?
> diff --git a/gprofng/src/checks.cc b/gprofng/src/checks.cc
> index 094c3bbc60a..2f9fbd7f714 100644
> --- a/gprofng/src/checks.cc
> +++ b/gprofng/src/checks.cc
> @@ -332,6 +332,10 @@ collect::check_executable_arch (Elf *elf)
> case EM_AARCH64:
> is_64 = true;
> break;
> +#elif ARCH(RISCV)
> + case EM_RISCV:
> + is_64 = true;
> + break;
Again formatting.
> #endif
> default:
> return EXEC_ELF_ARCH;
> diff --git a/gprofng/src/collctrl.cc b/gprofng/src/collctrl.cc
> index 5d68b689a64..1b04e52b5cc 100644
> --- a/gprofng/src/collctrl.cc
> +++ b/gprofng/src/collctrl.cc
> @@ -51,9 +51,9 @@ extern const char *strsignal (int);
> #endif
>
> // _SC_CPUID_MAX is not available on 2.6/2.7
> -#ifndef _SC_CPUID_MAX
> -#define _SC_CPUID_MAX 517
> -#endif
> +
> +// #define _SC_CPUID_MAX 517
> +// #endif
These lines are missing in latest sources.
The latest sources are here: https://sourceware.org/git/binutils-gdb.git.
The 2.43 release will be on Sunday 14th July.
It will be good if your fixes are included in this release.
-Vladimir
>
> const char *get_fstype (char *);
>
> @@ -74,15 +74,16 @@ Coll_Ctrl::Coll_Ctrl (int _interactive, bool _defHWC, bool _kernelHWC)
> default_stem = strdup ("test");
>
> /* get CPU count and processor clock rate */
> + #ifndef _SC_CPUID_MAX
> + ncpus = sysconf (_SC_NPROCESSORS_CONF);
> + /* add 2048 to count, since on some systems CPUID does not start at zero */
> + ncpumax = ncpus + 2048;
> + #elif
> ncpumax = sysconf (_SC_CPUID_MAX);
> - if (ncpumax == -1)
> - {
> - ncpus = sysconf (_SC_NPROCESSORS_CONF);
> - /* add 2048 to count, since on some systems CPUID does not start at zero */
> - ncpumax = ncpus + 2048;
> - }
> - ncpus = 0;
> - cpu_clk_freq = 0;
> + #endif
> +
> + //ncpus = 0;
> + //cpu_clk_freq = 0;
>
> // On Linux, read /proc/cpuinfo to get CPU count and clock rate
> // Note that parsing is different on SPARC and x86
> @@ -115,6 +116,9 @@ Coll_Ctrl::Coll_Ctrl (int _interactive, bool _defHWC, bool _kernelHWC)
> #elif defined(__aarch64__)
> asm volatile("mrs %0, cntfrq_el0" : "=r" (cpu_clk_freq));
>
> +#elif defined(__riscv)
> + cpu_clk_freq = 1000;
> +
> #else
> FILE *procf = fopen ("/proc/cpuinfo", "r");
> if (procf != NULL)
> diff --git a/gprofng/src/dbe_types.h b/gprofng/src/dbe_types.h
> index 9fa842ce85d..c67d421566e 100644
> --- a/gprofng/src/dbe_types.h
> +++ b/gprofng/src/dbe_types.h
> @@ -42,7 +42,8 @@ enum Platform_t
> Sparcv8plus,
> Java,
> Amd64,
> - Aarch64
> + Aarch64,
> + RISCV
> };
>
> enum WSize_t
> diff --git a/gprofng/testsuite/gprofng.display/mttest/gethrtime.c b/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
> index da39821027f..5e2cc3dc97b 100644
> --- a/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
> +++ b/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
> @@ -32,6 +32,8 @@
> #define SPARC 1
> #elif defined(__aarch64__)
> #define Aarch64 1
> +#elif defined(__riscv)
> +#define RISCV 1
> #else
> #define Intel 1
> #endif
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH] [RFC]RISC-V:[gprofng] Mimal support gprofng for riscv
2024-07-01 19:39 ` Vladimir Mezentsev
@ 2024-07-03 12:38 ` Yixuan Chen
2024-07-03 20:10 ` Vladimir Mezentsev
0 siblings, 1 reply; 7+ messages in thread
From: Yixuan Chen @ 2024-07-03 12:38 UTC (permalink / raw)
To: Vladimir Mezentsev; +Cc: binutils, ruud.vanderpas, jiawei, shihua, shiyulong
[-- Attachment #1: Type: text/plain, Size: 26361 bytes --]
Hi Vladimir,
Thank you very much for your guide.
Now I use the binutils-gdb master branch to port. But the `Collector
version : `2.41.50'` doesn't change with binutils-gdb version.
The third line of the log could see the binutils version.
Here comes the log:
```
Making all in gp-display-html
make[4]: Entering directory
'/home/xyenchi/binutils-gdb/build/gprofng/gp-display-html'
sed -e 's/BINUTILS_VERSION/2.42.50/' < ../../../gprofng/gp-display-html/
gp-display-html.in > gp-display-html
chmod +x gp-display-html
make[4]: Leaving directory
'/home/xyenchi/binutils-gdb/build/gprofng/gp-display-html'
Making all in doc
make[4]: Entering directory '/home/xyenchi/binutils-gdb/build/gprofng/doc'
make[4]: Nothing to be done for 'all'.
make[4]: Leaving directory '/home/xyenchi/binutils-gdb/build/gprofng/doc'
make[4]: Entering directory '/home/xyenchi/binutils-gdb/build/gprofng'
make[4]: Leaving directory '/home/xyenchi/binutils-gdb/build/gprofng'
make[3]: Leaving directory '/home/xyenchi/binutils-gdb/build/gprofng'
make[2]: Leaving directory '/home/xyenchi/binutils-gdb/build/gprofng'
make[1]: Leaving directory '/home/xyenchi/binutils-gdb/build'
[xyenchi@qemu-riscv build]$ rm -rf test.1.er/
[xyenchi@qemu-riscv build]$ ./gprofng/src/gp-collect-app echo 1
Creating experiment directory test.1.er (Process ID: 127275) ...
1
./gprofng/src/gp-display-text -func -exp
-head test.1.er/v build]$ ./gprofng/src/gp-display-text -func -exp -head
test.1.er/
Functions sorted by metric: Exclusive Total CPU Time
Excl. Total Incl. Total Name
CPU CPU
sec. % sec. %
0. 0. 0. 0. <Total>
ID Sel PID Experiment
== === ====== ==========
1 yes 127275 test.1.er/
Experiment: test.1.er/
No errors
No warnings
No archive command run
Target command (64-bit): 'echo 1'
Process pid 127275, ppid 399, pgrp 127275, sid 391
Current working directory: /home/xyenchi/binutils-gdb/build
Collector version: `2.41.50'; experiment version 12.4 (64-bit)
Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture
`riscv64'
0 CPUs, clock speed 0 MHz.
Memory: 4096267 pages @ 4096 = 16001 MB.
Data collection parameters:
Clock-profiling, interval = 10007 microsecs.
Periodic sampling, 1 secs.
Follow descendant processes from: fork|exec|combo
Experiment started Wed Jul 3 20:09:17 2024
Experiment Ended: 0.022871300
Data Collection Duration: 0.022871300
```
I only find this code to change the collector version:
https://github.com/bminor/binutils-gdb/blob/afa87be0fc751c158d666a48a9077d26d6ec8666/gprofng/src/Experiment.cc#L1224
but I don't known how to fix, could you give me some suggestion?
Best regards,
Yixuan Chen
Vladimir Mezentsev <vladimir.mezentsev@oracle.com> 于2024年7月2日周二 03:39写道:
> Thank you for your work on gprofng.
> See my comments below.
>
>
> On 6/30/24 02:37, XYenChi wrote:
>
> Minimal support RISC-V. Test with qemu.
>
> Result show:
> [xyenchi@qemu-riscv build]$ ./gprofng/src/gp-display-text -fu1.er/p -head test.2.er/
> Functions sorted by metric: Exclusive Total CPU Time
>
> Excl. Total Incl. Total Name
> CPU CPU
> sec. % sec. %
> 0. 0. 0. 0. <Total>
>
> ID Sel PID Experiment
> == === ====== ==========
> 1 yes 145674 test.1.er/
> Experiment: test.1.er/
> No errors
> No warnings
> No archive command run
>
> Target command (64-bit): 'echo 1'
> Process pid 145674, ppid 399, pgrp 145674, sid 392
> Current working directory: /home/xyenchi/binutils-gdb/build
> Collector version: `2.41.50'; experiment version 12.4 (64-bit)
>
>
> It looks like you are using an old version of the binutils-gdb source.
> The current version is 2.42.50.
>
>
> Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture `riscv64'
> 0 CPUs, clock speed 0 MHz.
> Memory: 4096267 pages @ 4096 = 16001 MB.
> Data collection parameters:
> Clock-profiling, interval = 10007 microsecs.
> Periodic sampling, 1 secs.
> Follow descendant processes from: fork|exec|combo
>
> Experiment started Thu Jun 27 16:23:20 2024
>
> Experiment Ended: 0.040909200
> Data Collection Duration: 0.040909200
>
> If build with `allow_undefined_flag=true`, result will be:
>
> [xyenchi@qemu-riscv build]$ ./gprofng/src/gp-display-text -fu1.er/p -head test.2.er/
> Functions sorted by metric: Exclusive Total CPU Time
>
> Excl. Total Incl. Total Name
> CPU CPU
> sec. % sec. %
> 0. 0. 0. 0. <Total>
>
> ID Sel PID Experiment
> == === ====== ==========
> 1 yes 145674 test.1.er/
> Experiment: test.1.er/
> No errors
> No warnings
> No archive command run
>
> Target command (64-bit): 'echo 1'
> Process pid 145674, ppid 399, pgrp 145674, sid 392
> Current working directory: /home/xyenchi/binutils-gdb/build
> Collector version: `2.41.50'; experiment version 12.4 (64-bit)
> Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture `riscv64'
> 0 CPUs, clock speed 0 MHz.
> Memory: 4096267 pages @ 4096 = 16001 MB.
> Data collection parameters:
> Clock-profiling, interval = 10007 microsecs.
> Periodic sampling, 1 secs.
> Follow descendant processes from: fork|exec|combo
>
> Experiment started Thu Jun 27 16:23:20 2024
>
> Experiment Ended: 0.040909200
> Data Collection Duration: 0.040909200
>
> ChangeLog:
> Minimal support gprofng for RISC-V.
>
> 2024-06-28 Yixuan Chen <chenyixuan@iscasc.ac.cn> <chenyixuan@iscasc.ac.cn>
>
> * configure: Add RISC-V configure.
> * configure.ac: Add RISC-V configure.
>
> Unknown ChangeLog:
>
> 2024-06-28 Yixuan Chen <chenyixuan@iscasc.ac.cn> <chenyixuan@iscasc.ac.cn>
>
> * gprofng/common/core_pcbe.c (core_pcbe_init): Add RISC-V vendor.
> (defined): Add RISC-V condition.
> * gprofng/common/cpuid.c (defined): Add head file to use hwprobe.
> (my_cpuid): Add riscv hwprobe.
> * gprofng/common/gp-defs.h (TOK_A_RISCV): Add RISC-V.
> (defined): Add RISC-V.
> (ARCH_RISCV): Add RISC-V.
> * gprofng/common/hwc_cpus.h: Add RISC-V vendor id.
> * gprofng/common/hwcfuncs.h (HW_INTERVAL_TYPE): fix typo.
> * gprofng/configure: Add RISC-V.
> * gprofng/configure.ac: Add RISC-V.
> * gprofng/libcollector/hwprofile.h (ARCH): Add riscv register calling.
> (CONTEXT_PC): Add riscv register calling.
> (CONTEXT_FP): Add riscv register calling.
> (CONTEXT_SP): Add riscv register calling.
> (SETFUNCTIONCONTEXT): Add riscv uc_mcontext calling.
> * gprofng/libcollector/libcol_util.c (__collector_util_init): fix libc calling
> * gprofng/libcollector/libcol_util.h (__collector_cas_32): Add RISC-V.
> (ARCH):
> * gprofng/libcollector/unwind.c (ARCH): Add RISC-V.
> (GET_PC): Add riscv register calling.
> (GET_SP): Add riscv register calling.
> (GET_FP): Add riscv register calling.
> (FILL_CONTEXT): Add riscv uc_mcontext calling.
> * gprofng/src/DbeSession.cc (ARCH):
> * gprofng/src/Disasm.cc (Disasm::disasm_open): Add RISC-V.
> * gprofng/src/Experiment.cc (Experiment::ExperimentHandler::startElement): Add RISC-V.
> * gprofng/src/checks.cc (ARCH): Add RISC-V.
> * gprofng/src/collctrl.cc (_SC_CPUID_MAX): Add if not define _SC_CPUID_MAX condition. Set CPU frequency as 1000 to test but now not show at the test result, will fix with better way to get cpu frequency.
> (Coll_Ctrl::Coll_Ctrl):
> (defined):
> * gprofng/src/dbe_types.h (enum Platform_t):Add RISC-V.
> * gprofng/testsuite/gprofng.display/mttest/gethrtime.c: Add RISC-V.
>
> ---
> configure | 2 +-
> configure.ac | 2 +-
> gprofng/common/core_pcbe.c | 5 ++-
> gprofng/common/cpuid.c | 31 +++++++++++++++++--
> gprofng/common/gp-defs.h | 5 ++-
> gprofng/common/hwc_cpus.h | 7 +++++
> gprofng/common/hwcfuncs.h | 2 +-
> gprofng/configure | 4 +++
> gprofng/configure.ac | 4 +++
> gprofng/libcollector/hwprofile.h | 10 ++++++
> gprofng/libcollector/libcol_util.c | 11 ++++++-
> gprofng/libcollector/libcol_util.h | 10 +++---
> gprofng/libcollector/unwind.c | 16 ++++++++--
> gprofng/src/DbeSession.cc | 2 ++
> gprofng/src/Disasm.cc | 2 ++
> gprofng/src/Experiment.cc | 2 ++
> gprofng/src/checks.cc | 4 +++
> gprofng/src/collctrl.cc | 26 +++++++++-------
> gprofng/src/dbe_types.h | 3 +-
> .../gprofng.display/mttest/gethrtime.c | 2 ++
> 20 files changed, 123 insertions(+), 27 deletions(-)
>
> diff --git a/configure b/configure
> index dd743c58663..e3cbad5752b 100755
> --- a/configure
> +++ b/configure
> @@ -3145,7 +3145,7 @@ fi
>
> if test "$enable_gprofng" = "yes"; then
> case "${target}" in
> - x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux*)
> + x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux* | riscv64-*-linux*)
> configdirs="$configdirs gprofng"
> ;;
> esac
> diff --git a/configure.ac b/configure.ac
> index 01cfd017273..e63f27e6a29 100644
> --- a/configure.ac
> +++ b/configure.ac
> @@ -412,7 +412,7 @@ enable_gprofng=$enableval,
> enable_gprofng=yes)
> if test "$enable_gprofng" = "yes"; then
> case "${target}" in
> - x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux*)
> + x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux* | riscv64-*-linux*)
> configdirs="$configdirs gprofng"
> ;;
> esac
> diff --git a/gprofng/common/core_pcbe.c b/gprofng/common/core_pcbe.c
> index b71d46f4e31..ef2a3f83d42 100644
> --- a/gprofng/common/core_pcbe.c
> +++ b/gprofng/common/core_pcbe.c
> @@ -2754,6 +2754,9 @@ core_pcbe_init (void)
> return 0;
> case X86_VENDOR_Intel:
> break;
> + case ANDES_VENDOR_ID:
>
>
> The formatting is wrong.
> See a coding standard:
> https://sourceware.org/gdb/wiki/Internals%20GDB-C-Coding-Standards#Whitespaces
> * Code indentation*
>
> * Lines should be indented with a mix of tabs and spaces. 8 spaces should
> be replaced with Tab. Vim users could add the following to their
> ~/.vimrc:*
> * set autoindent tabstop=8 shiftwidth=2*
>
>
> + case SIFIVE_VENDOR_ID:
> + case THEAD_VENDOR_ID:
> default:
> return -1;
> }
> @@ -2915,7 +2918,7 @@ core_pcbe_impl_name (void)
> static const char *
> core_pcbe_cpuref (void)
> {
> -#if defined(__aarch64__)
> +#if defined(__aarch64__) || defined(__riscv)
> return "";
> #elif defined(__i386__) || defined(__x86_64)
> switch (cpuid_getmodel ())
> diff --git a/gprofng/common/cpuid.c b/gprofng/common/cpuid.c
> index fd98b30c639..c29b6612afe 100644
> --- a/gprofng/common/cpuid.c
> +++ b/gprofng/common/cpuid.c
> @@ -42,6 +42,11 @@ __get_cpuid (unsigned int op ATTRIBUTE_UNUSED, unsigned int *eax,
> Tprintf (DBG_LT0, "cpuid.c:%d read_cpuid_id() MIDR_EL1=0x%016x\n", __LINE__, *eax);
> return res;
> }
> +#elif defined(__riscv)
> +#include <sched.h>
> +#include <sys/syscall.h>
> +#include <unistd.h>
> +#include <asm/hwprobe.h>
> #endif
>
> /*
> @@ -104,7 +109,7 @@ my_cpuid (unsigned int op, cpuid_regs_t *regs)
> TprintfT (DBG_LT1, "my_cpuid: __get_cpuid(0x%x, 0x%x, 0x%x, 0x%x, 0x%x) returns %d\n",
> op, regs->eax, regs->ebx, regs->ecx, regs->edx, ret);
> return ret;
> -}
> +}
> #endif
>
> static cpuid_info_t *
> @@ -180,8 +185,30 @@ get_cpuid_info ()
> cpi->cpi_model += CPI_MODEL_XTD (regs.eax) << 4;
> break;
> }
> +#elif defined(__riscv)
> + #ifndef __riscv_hwprobe
> + cpi->cpi_vendor = 0;
> + cpi->cpi_family = 0;
> + cpi->cpi_model = 0;
> + #else
> + struct riscv_hwprobe res;
> + res.key = RISCV_HWPROBE_KEY_MVENDORID;
> + cpu_set_t cpu_set;
> + int __riscv_hwprobe (struct riscv_hwprobe *pairs, \
> + long pair_count, long cpu_count, \
> + unsigned long *cpus, unsigned long flags) \
> + {
> + return syscall(__NR_riscv_hwprobe, pairs, pair_count, cpu_count, cpus, flags);
> + }
> + CPU_ZERO(&cpu_set);
> + CPU_SET(0, &cpu_set);
> + long ret = __riscv_hwprobe(&res, 1, 1, &cpu_set, 0);
> + cpi->cpi_vendor = res.value;
> + cpi->cpi_family = 0;
> + cpi->cpi_model = 0;
> + #endif
> #endif
> - return cpi;
> + return cpi;
> }
>
>
> Could you also add a code into read_cpuinfo() (gprofng/src/collctrl.cc
> line ~85)
> to read cpu_model, cpu_family, etc from /proc/cpu_info.
>
>
>
> static inline uint_t
> diff --git a/gprofng/common/gp-defs.h b/gprofng/common/gp-defs.h
> index 7cef5550696..891c4d32cf1 100644
> --- a/gprofng/common/gp-defs.h
> +++ b/gprofng/common/gp-defs.h
> @@ -32,6 +32,7 @@
> */
> #define ARCH(x) TOK_A_##x(ARCH)
> #define TOK_A_Aarch64(x) x##_Aarch64
> +#define TOK_A_RISCV(x) x##_RISCV
> #define TOK_A_SPARC(x) x##_SPARC
> #define TOK_A_Intel(x) x##_Intel
>
> @@ -45,11 +46,13 @@
> #define ARCH_Intel 1
> #elif defined(__aarch64__)
> #define ARCH_Aarch64 1
> +#elif defined(riscv) || defined(__riscv)
> +#define ARCH_RISCV 1
> #else
> #error "Undefined platform"
> #endif
>
> -#if defined(__sparcv9) || defined(__x86_64) || defined(__aarch64__)
> +#if defined(__sparcv9) || defined(__x86_64) || defined(__aarch64__) || defined(__riscv)
> #define WSIZE_64 1
> #else
> #define WSIZE_32 1
> diff --git a/gprofng/common/hwc_cpus.h b/gprofng/common/hwc_cpus.h
> index be820819dd3..b121e4bfb97 100644
> --- a/gprofng/common/hwc_cpus.h
> +++ b/gprofng/common/hwc_cpus.h
> @@ -111,6 +111,13 @@ enum {
> ARM_CPU_IMP_QCOM = 0x51
> };
>
> +// rscv Constants from arch/riscv/include/asm/vendorid_list.h
> +enum {
> + ANDES_VENDOR_ID =0x31e,
> + SIFIVE_VENDOR_ID =0x489,
> + THEAD_VENDOR_ID =0x5b7
> +};
> +
> #define AARCH64_VENDORSTR_ARM "ARM"
>
> /* strings below must match those returned by cpc_getcpuver() */
> diff --git a/gprofng/common/hwcfuncs.h b/gprofng/common/hwcfuncs.h
> index a5f64f2a33f..1a6ee7e30ca 100644
> --- a/gprofng/common/hwcfuncs.h
> +++ b/gprofng/common/hwcfuncs.h
> @@ -98,7 +98,7 @@ typedef struct { /* supplementary data fields */
>
> #define HW_INTERVAL_MAX UINT64_MAX
> #define HW_INTERVAL_PRESET(x) (HW_INTERVAL_MAX - ((uint64_t)(x) - 1))
> -#define HW_INTERVAL_TYPE(x) ((uint64_t) (x)
> +#define HW_INTERVAL_TYPE(x) ((uint64_t) (x))
>
>
> HW_INTERVAL_TYPE is not used. Just remove this line.
>
>
>
>
> /* parsing */
> #define HWCFUNCS_MAX_ATTRS 20
> diff --git a/gprofng/configure b/gprofng/configure
> index 1c6a99291be..9ac7a651809 100755
> --- a/gprofng/configure
> +++ b/gprofng/configure
> @@ -15788,6 +15788,10 @@ build_src=
> build_src=true
> build_collector=true
> ;;
> + riscv*-*-linux*)
> + build_src=true
> + build_collector=true
> + ;;
> esac
> # Check whether --enable-gprofng-tools was given.
> if test "${enable_gprofng_tools+set}" = set; then :
> diff --git a/gprofng/configure.ac b/gprofng/configure.ac
> index 6b8fe262748..b4f2981a487 100644
> --- a/gprofng/configure.ac
> +++ b/gprofng/configure.ac
> @@ -63,6 +63,10 @@ build_src=
> build_src=true
> build_collector=true
> ;;
> + riscv*-*-linux*)
> + build_src=true
> + build_collector=true
> + ;;
> esac
> AC_ARG_ENABLE(gprofng-tools,
> AS_HELP_STRING([--disable-gprofng-tools], [do not build gprofng/src directory]),
> diff --git a/gprofng/libcollector/hwprofile.h b/gprofng/libcollector/hwprofile.h
> index 23422dd51fa..6517452eef7 100644
> --- a/gprofng/libcollector/hwprofile.h
> +++ b/gprofng/libcollector/hwprofile.h
> @@ -84,6 +84,16 @@ typedef struct MHwcntr_packet
> (ucp)->uc_mcontext.regs[CONTEXT_PC] = (greg_t)(funcp); \
> (ucp)->uc_mcontext.regs[CONTEXT_SP] = 0; \
> (ucp)->uc_mcontext.regs[CONTEXT_FP] = 0;
> +
> +#elif ARCH(RISCV)
> +#define CONTEXT_PC REG_PC
> +#define CONTEXT_FP 8
> +#define CONTEXT_SP 2
> +#define SETFUNCTIONCONTEXT(ucp,funcp) \
> + (ucp)->uc_mcontext.__gregs[CONTEXT_PC] = (greg_t)(funcp); \
> + (ucp)->uc_mcontext.__gregs[CONTEXT_FP] = 0; \
> + (ucp)->uc_mcontext.__gregs[CONTEXT_SP] = 0;
> +
> #endif /* ARCH() */
>
> #endif
> diff --git a/gprofng/libcollector/libcol_util.c b/gprofng/libcollector/libcol_util.c
> index a8802d4a9b0..8ee1c56ff45 100644
> --- a/gprofng/libcollector/libcol_util.c
> +++ b/gprofng/libcollector/libcol_util.c
> @@ -91,7 +91,7 @@ __collector_gettid ()
> #endif
> __asm__ __volatile__(syscall_instr
> : "=a" (r) : "0" (__NR_gettid)
> - : syscall_clobber);
> + : syscall_clobber);
> #else
> r = syscall (__NR_gettid);
> #endif
> @@ -1459,7 +1459,10 @@ __collector_util_init ()
> else if ((ptr = dlvsym (libc, "fopen", "GLIBC_2.0")) != NULL)
> __collector_util_funcs.fopen = ptr;
> else
> + {
> ptr = dlsym (libc, "fopen");
> + if(ptr) __collector_util_funcs.fopen = ptr;
> + }
> if (__collector_util_funcs.fopen == NULL)
> {
> CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT fopen: %s\n", dlerror ());
> @@ -1475,7 +1478,10 @@ __collector_util_init ()
> else if ((ptr = dlvsym (libc, "popen", "GLIBC_2.0")) != NULL)
> __collector_util_funcs.popen = ptr;
> else
> + {
> ptr = dlsym (libc, "popen");
> + if(ptr) __collector_util_funcs.popen = ptr;
> + }
> if (__collector_util_funcs.popen == NULL)
> {
> CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT popen: %s\n", dlerror ());
> @@ -1491,7 +1497,10 @@ __collector_util_init ()
> else if ((ptr = dlvsym (libc, "fclose", "GLIBC_2.0")) != NULL)
> __collector_util_funcs.fclose = ptr;
> else
> + {
> ptr = dlsym (libc, "fclose");
> + if(ptr) __collector_util_funcs.fclose = ptr;
> + }
> if (__collector_util_funcs.fclose == NULL)
> {
> CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT fclose: %s\n", dlerror ());
> diff --git a/gprofng/libcollector/libcol_util.h b/gprofng/libcollector/libcol_util.h
> index c21b4a46c4b..6ac84db05ab 100644
> --- a/gprofng/libcollector/libcol_util.h
> +++ b/gprofng/libcollector/libcol_util.h
> @@ -209,10 +209,10 @@ static __attribute__ ((always_inline)) inline uint32_t
> __collector_cas_32 (volatile uint32_t *pdata, uint32_t old, uint32_t new)
> {
> uint32_t r;
> - __asm__ __volatile__("lock; cmpxchgl %2, %1"
> - : "=a" (r), "=m" (*pdata) : "r" (new),
> - "a" (old), "m" (*pdata));
> - return r;
> + __asm__ __volatile__("lock; cmpxchgl %2, %1"
> + : "=a" (r), "=m" (*pdata) : "r" (new),
> + "a" (old), "m" (*pdata));
> + return r;
> }
> /**
> * This function enables a compare and swap operation to occur atomically.
> @@ -270,7 +270,7 @@ __collector_cas_ptr (void *mem, void *cmp, void *new)
> return r;
> }
>
> -#elif ARCH(Aarch64)
> +#elif ARCH(Aarch64) || ARCH(RISCV)
> static __attribute__ ((always_inline)) inline uint32_t
> __collector_inc_32 (volatile uint32_t *ptr)
> {
> diff --git a/gprofng/libcollector/unwind.c b/gprofng/libcollector/unwind.c
> index ff2f7aa9a7e..1244a453457 100644
> --- a/gprofng/libcollector/unwind.c
> +++ b/gprofng/libcollector/unwind.c
> @@ -186,8 +186,14 @@ memory_error_func (int status ATTRIBUTE_UNUSED, bfd_vma addr ATTRIBUTE_UNUSED,
> #define GET_PC(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[15])
> #define GET_SP(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[13])
> #define GET_FP(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[14])
> +
> +#elif ARCH(RISCV)
> +#define GET_PC(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[REG_PC])
> +#define GET_SP(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[2])
> +#define GET_FP(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[8])
> #endif /* ARCH() */
>
> +
> /*
> * FILL_CONTEXT() for all platforms
> * Could use getcontext() except:
> @@ -230,13 +236,19 @@ memory_error_func (int status ATTRIBUTE_UNUSED, bfd_vma addr ATTRIBUTE_UNUSED,
> context->uc_stack.ss_size = 0x100000; \
> }
>
> -#elif ARCH(Aarch64)
> +#elif ARCH(Aarch64)
> #define FILL_CONTEXT(context) \
> { CALL_UTIL (getcontext) (context); \
> context->uc_mcontext.sp = (__u64) __builtin_frame_address(0); \
> }
>
> -#endif /* ARCH() */
> +#elif ARCH(RISCV)
> +#define FILL_CONTEXT(context) \
> + { CALL_UTIL(getcontext)(context); \
> + context->uc_mcontext.__gregs[2] = (uint64_t) __builtin_frame_address(0); \
> + }
> +
> +#endif/* ARCH() */
>
> static int
> getByteInstruction (unsigned char *p)
> diff --git a/gprofng/src/DbeSession.cc b/gprofng/src/DbeSession.cc
> index 86541d97f2e..3c218b5fc91 100644
> --- a/gprofng/src/DbeSession.cc
> +++ b/gprofng/src/DbeSession.cc
> @@ -94,6 +94,8 @@ Platform_t DbeSession::platform =
> Sparc;
> #elif ARCH(Aarch64)
> Aarch64;
> +#elif ARCH(RISCV)
> + RISCV;
> #else // ARCH(Intel)
> Intel;
> #endif
> diff --git a/gprofng/src/Disasm.cc b/gprofng/src/Disasm.cc
> index 9c7e59a590b..19e6f15ff6f 100644
> --- a/gprofng/src/Disasm.cc
> +++ b/gprofng/src/Disasm.cc
> @@ -208,6 +208,7 @@ Disasm::disasm_open ()
> case Amd64:
> need_swap_endian = (DbeSession::platform == Sparc);
> break;
> + case RISCV:
> case Sparcv8plus:
> case Sparcv9:
> case Sparc:
> @@ -246,6 +247,7 @@ Disasm::disasm_open ()
> dis_info.arch = bfd_arch_i386;
> dis_info.mach = bfd_mach_x86_64;
> break;
> + case RISCV:
> case Sparcv8plus:
> case Sparcv9:
> case Sparc:
> diff --git a/gprofng/src/Experiment.cc b/gprofng/src/Experiment.cc
> index 1378ad5ce07..ac78276d23c 100644
> --- a/gprofng/src/Experiment.cc
> +++ b/gprofng/src/Experiment.cc
> @@ -542,6 +542,8 @@ Experiment::ExperimentHandler::startElement (char*, char*, char *qName, Attribut
> exp->platform = Intel;
> else if (strcmp (str, "aarch64") == 0)
> exp->platform = Aarch64;
> + else if (strcmp (str, "riscv64") == 0)
> + exp->platform = RISCV;
>
>
> The formatting is wrong.
> It looks like in your environment the tab size is 4 spaces, but in the GNU
> standard it is 8.
>
> else
> exp->platform = Sparc;
> exp->need_swap_endian = (DbeSession::platform == Sparc) ?
> diff --git a/gprofng/src/checks.cc b/gprofng/src/checks.cc
> index 094c3bbc60a..2f9fbd7f714 100644
> --- a/gprofng/src/checks.cc
> +++ b/gprofng/src/checks.cc
> @@ -332,6 +332,10 @@ collect::check_executable_arch (Elf *elf)
> case EM_AARCH64:
> is_64 = true;
> break;
> +#elif ARCH(RISCV)
> + case EM_RISCV:
> + is_64 = true;
> + break;
>
>
> Again formatting.
>
>
> #endif
> default:
> return EXEC_ELF_ARCH;
> diff --git a/gprofng/src/collctrl.cc b/gprofng/src/collctrl.cc
> index 5d68b689a64..1b04e52b5cc 100644
> --- a/gprofng/src/collctrl.cc
> +++ b/gprofng/src/collctrl.cc
> @@ -51,9 +51,9 @@ extern const char *strsignal (int);
> #endif
>
> // _SC_CPUID_MAX is not available on 2.6/2.7
> -#ifndef _SC_CPUID_MAX
> -#define _SC_CPUID_MAX 517
> -#endif
> +
> +// #define _SC_CPUID_MAX 517
> +// #endif
>
>
> These lines are missing in latest sources.
> The latest sources are here: https://sourceware.org/git/binutils-gdb.git.
>
>
> The 2.43 release will be on Sunday 14th July.
> It will be good if your fixes are included in this release.
>
> -Vladimir
>
>
>
> const char *get_fstype (char *);
>
> @@ -74,15 +74,16 @@ Coll_Ctrl::Coll_Ctrl (int _interactive, bool _defHWC, bool _kernelHWC)
> default_stem = strdup ("test");
>
> /* get CPU count and processor clock rate */
> + #ifndef _SC_CPUID_MAX
> + ncpus = sysconf (_SC_NPROCESSORS_CONF);
> + /* add 2048 to count, since on some systems CPUID does not start at zero */
> + ncpumax = ncpus + 2048;
> + #elif
> ncpumax = sysconf (_SC_CPUID_MAX);
> - if (ncpumax == -1)
> - {
> - ncpus = sysconf (_SC_NPROCESSORS_CONF);
> - /* add 2048 to count, since on some systems CPUID does not start at zero */
> - ncpumax = ncpus + 2048;
> - }
> - ncpus = 0;
> - cpu_clk_freq = 0;
> + #endif
> +
> + //ncpus = 0;
> + //cpu_clk_freq = 0;
>
> // On Linux, read /proc/cpuinfo to get CPU count and clock rate
> // Note that parsing is different on SPARC and x86
> @@ -115,6 +116,9 @@ Coll_Ctrl::Coll_Ctrl (int _interactive, bool _defHWC, bool _kernelHWC)
> #elif defined(__aarch64__)
> asm volatile("mrs %0, cntfrq_el0" : "=r" (cpu_clk_freq));
>
> +#elif defined(__riscv)
> + cpu_clk_freq = 1000;
> +
> #else
> FILE *procf = fopen ("/proc/cpuinfo", "r");
> if (procf != NULL)
> diff --git a/gprofng/src/dbe_types.h b/gprofng/src/dbe_types.h
> index 9fa842ce85d..c67d421566e 100644
> --- a/gprofng/src/dbe_types.h
> +++ b/gprofng/src/dbe_types.h
> @@ -42,7 +42,8 @@ enum Platform_t
> Sparcv8plus,
> Java,
> Amd64,
> - Aarch64
> + Aarch64,
> + RISCV
> };
>
> enum WSize_t
> diff --git a/gprofng/testsuite/gprofng.display/mttest/gethrtime.c b/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
> index da39821027f..5e2cc3dc97b 100644
> --- a/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
> +++ b/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
> @@ -32,6 +32,8 @@
> #define SPARC 1
> #elif defined(__aarch64__)
> #define Aarch64 1
> +#elif defined(__riscv)
> +#define RISCV 1
> #else
> #define Intel 1
> #endif
>
>
>
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH] [RFC]RISC-V:[gprofng] Mimal support gprofng for riscv
2024-07-03 12:38 ` Yixuan Chen
@ 2024-07-03 20:10 ` Vladimir Mezentsev
2024-07-04 9:21 ` Yixuan Chen
0 siblings, 1 reply; 7+ messages in thread
From: Vladimir Mezentsev @ 2024-07-03 20:10 UTC (permalink / raw)
To: Yixuan Chen; +Cc: binutils, ruud.vanderpas, jiawei, shihua, shiyulong
[-- Attachment #1: Type: text/plain, Size: 30693 bytes --]
Hi Yixuan Chen,
On 7/3/24 05:38, Yixuan Chen wrote:
> Hi Vladimir,
>
> Thank you very much for your guide.
>
> Now I use the binutils-gdb master branch to port. But the `Collector
> version : `2.41.50'` doesn't change with binutils-gdb version.
> The third line of the log could see the binutils version.
>
> Here comes the log:
> ```
> Making all in gp-display-html
> make[4]: Entering directory
> '/home/xyenchi/binutils-gdb/build/gprofng/gp-display-html'
> sed -e 's/BINUTILS_VERSION/2.42.50/' <
> ../../../gprofng/gp-display-html/gp-display-html.in
> <http://gp-display-html.in> > gp-display-html
> chmod +x gp-display-html
> make[4]: Leaving directory
> '/home/xyenchi/binutils-gdb/build/gprofng/gp-display-html'
> Making all in doc
> make[4]: Entering directory '/home/xyenchi/binutils-gdb/build/gprofng/doc'
> make[4]: Nothing to be done for 'all'.
> make[4]: Leaving directory '/home/xyenchi/binutils-gdb/build/gprofng/doc'
> make[4]: Entering directory '/home/xyenchi/binutils-gdb/build/gprofng'
> make[4]: Leaving directory '/home/xyenchi/binutils-gdb/build/gprofng'
> make[3]: Leaving directory '/home/xyenchi/binutils-gdb/build/gprofng'
> make[2]: Leaving directory '/home/xyenchi/binutils-gdb/build/gprofng'
> make[1]: Leaving directory '/home/xyenchi/binutils-gdb/build'
> [xyenchi@qemu-riscv build]$ rm -rf test.1.er/ <http://test.1.er/>
> [xyenchi@qemu-riscv build]$ ./gprofng/src/gp-collect-app echo 1
This is wrong.
You are using the built version of gprofng. But you must use the
installed version.
For example:
cd /home/xyenchi/binutils-gdb/build/
rm -rf *
../configure --prefix=`pwd`/INSTALL [THE_OTHER_CONFIGURE_OPTIONS]
make
make install
/home/xyenchi/binutils-gdb/build/INSTALL/bin/gprofng collect app -O
test.1.er echo 1
*We recommend using `gprofng collect app` instead of calling
gp-collect-app directly.*
/home/xyenchi/binutils-gdb/build/INSTALL/bin/gprofng display text -func
-exp -head test.1.er/ <http://test.1.er/>
The problem is:
gp-collect-app sets
LD_PRELOAD=<PATH_TO_GPROFNG_LIB_DIR>/libgp-collector.so before starting
the user application.
It looks like you did the installation for the old sources and did
not reinstall gprofng.
In this case, you used the old libgp-collector.so (2.41.50).
Thank you for fixing gprofng.
-Vladimir
> Creating experiment directory test.1.er <http://test.1.er> (Process
> ID: 127275) ...
> 1 ./gprofng/src/gp-display-text -func -exp -head test.1.er/v
> <http://test.1.er/v> build]$ ./gprofng/src/gp-display-text -func -exp
> -head test.1.er/ <http://test.1.er/>
> Functions sorted by metric: Exclusive Total CPU Time
>
> Excl. Total Incl. Total Name
> CPU CPU
> sec. % sec. %
> 0. 0. 0. 0. <Total>
>
> ID Sel PID Experiment
> == === ====== ==========
> 1 yes 127275 test.1.er/ <http://test.1.er/>
> Experiment: test.1.er/ <http://test.1.er/>
> No errors
> No warnings
> No archive command run
>
> Target command (64-bit): 'echo 1'
> Process pid 127275, ppid 399, pgrp 127275, sid 391
> Current working directory: /home/xyenchi/binutils-gdb/build
> Collector version: `2.41.50'; experiment version 12.4 (64-bit)
> Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture
> `riscv64'
> 0 CPUs, clock speed 0 MHz.
> Memory: 4096267 pages @ 4096 = 16001 MB.
> Data collection parameters:
> Clock-profiling, interval = 10007 microsecs.
> Periodic sampling, 1 secs.
> Follow descendant processes from: fork|exec|combo
>
> Experiment started Wed Jul 3 20:09:17 2024
>
> Experiment Ended: 0.022871300
> Data Collection Duration: 0.022871300
> ```
>
> I only find this code to change the collector
> version:https://github.com/bminor/binutils-gdb/blob/afa87be0fc751c158d666a48a9077d26d6ec8666/gprofng/src/Experiment.cc#L1224
> but I don't known how to fix, could you give me some suggestion?
>
> Best regards,
> Yixuan Chen
>
> Vladimir Mezentsev <vladimir.mezentsev@oracle.com> 于2024年7月2日周二
> 03:39写道:
>
> Thank you for your work on gprofng.
> See my comments below.
>
>
> On 6/30/24 02:37, XYenChi wrote:
>> Minimal support RISC-V. Test with qemu.
>>
>> Result show:
>> [xyenchi@qemu-riscv build]$ ./gprofng/src/gp-display-text -fu1.er/p <http://fu1.er/p> -headtest.2.er/ <http://test.2.er/>
>> Functions sorted by metric: Exclusive Total CPU Time
>>
>> Excl. Total Incl. Total Name
>> CPU CPU
>> sec. % sec. %
>> 0. 0. 0. 0. <Total>
>>
>> ID Sel PID Experiment
>> == === ====== ==========
>> 1 yes 145674test.1.er/ <http://test.1.er/>
>> Experiment:test.1.er/ <http://test.1.er/>
>> No errors
>> No warnings
>> No archive command run
>>
>> Target command (64-bit): 'echo 1'
>> Process pid 145674, ppid 399, pgrp 145674, sid 392
>> Current working directory: /home/xyenchi/binutils-gdb/build
>> Collector version: `2.41.50'; experiment version 12.4 (64-bit)
>
> It looks like you are using an old version of the binutils-gdb source.
> The current version is 2.42.50.
>
>
>> Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture `riscv64'
>> 0 CPUs, clock speed 0 MHz.
>> Memory: 4096267 pages @ 4096 = 16001 MB.
>> Data collection parameters:
>> Clock-profiling, interval = 10007 microsecs.
>> Periodic sampling, 1 secs.
>> Follow descendant processes from: fork|exec|combo
>>
>> Experiment started Thu Jun 27 16:23:20 2024
>>
>> Experiment Ended: 0.040909200
>> Data Collection Duration: 0.040909200
>>
>> If build with `allow_undefined_flag=true`, result will be:
>>
>> [xyenchi@qemu-riscv build]$ ./gprofng/src/gp-display-text -fu1.er/p <http://fu1.er/p> -headtest.2.er/ <http://test.2.er/>
>> Functions sorted by metric: Exclusive Total CPU Time
>>
>> Excl. Total Incl. Total Name
>> CPU CPU
>> sec. % sec. %
>> 0. 0. 0. 0. <Total>
>>
>> ID Sel PID Experiment
>> == === ====== ==========
>> 1 yes 145674test.1.er/ <http://test.1.er/>
>> Experiment:test.1.er/ <http://test.1.er/>
>> No errors
>> No warnings
>> No archive command run
>>
>> Target command (64-bit): 'echo 1'
>> Process pid 145674, ppid 399, pgrp 145674, sid 392
>> Current working directory: /home/xyenchi/binutils-gdb/build
>> Collector version: `2.41.50'; experiment version 12.4 (64-bit)
>> Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture `riscv64'
>> 0 CPUs, clock speed 0 MHz.
>> Memory: 4096267 pages @ 4096 = 16001 MB.
>> Data collection parameters:
>> Clock-profiling, interval = 10007 microsecs.
>> Periodic sampling, 1 secs.
>> Follow descendant processes from: fork|exec|combo
>>
>> Experiment started Thu Jun 27 16:23:20 2024
>>
>> Experiment Ended: 0.040909200
>> Data Collection Duration: 0.040909200
>>
>> ChangeLog:
>> Minimal support gprofng for RISC-V.
>>
>> 2024-06-28 Yixuan Chen<chenyixuan@iscasc.ac.cn> <mailto:chenyixuan@iscasc.ac.cn>
>>
>> * configure: Add RISC-V configure.
>> *configure.ac <http://configure.ac>: Add RISC-V configure.
>>
>> Unknown ChangeLog:
>>
>> 2024-06-28 Yixuan Chen<chenyixuan@iscasc.ac.cn> <mailto:chenyixuan@iscasc.ac.cn>
>>
>> * gprofng/common/core_pcbe.c (core_pcbe_init): Add RISC-V vendor.
>> (defined): Add RISC-V condition.
>> * gprofng/common/cpuid.c (defined): Add head file to use hwprobe.
>> (my_cpuid): Add riscv hwprobe.
>> * gprofng/common/gp-defs.h (TOK_A_RISCV): Add RISC-V.
>> (defined): Add RISC-V.
>> (ARCH_RISCV): Add RISC-V.
>> * gprofng/common/hwc_cpus.h: Add RISC-V vendor id.
>> * gprofng/common/hwcfuncs.h (HW_INTERVAL_TYPE): fix typo.
>> * gprofng/configure: Add RISC-V.
>> * gprofng/configure.ac <http://configure.ac>: Add RISC-V.
>> * gprofng/libcollector/hwprofile.h (ARCH): Add riscv register calling.
>> (CONTEXT_PC): Add riscv register calling.
>> (CONTEXT_FP): Add riscv register calling.
>> (CONTEXT_SP): Add riscv register calling.
>> (SETFUNCTIONCONTEXT): Add riscv uc_mcontext calling.
>> * gprofng/libcollector/libcol_util.c (__collector_util_init): fix libc calling
>> * gprofng/libcollector/libcol_util.h (__collector_cas_32): Add RISC-V.
>> (ARCH):
>> * gprofng/libcollector/unwind.c (ARCH): Add RISC-V.
>> (GET_PC): Add riscv register calling.
>> (GET_SP): Add riscv register calling.
>> (GET_FP): Add riscv register calling.
>> (FILL_CONTEXT): Add riscv uc_mcontext calling.
>> * gprofng/src/DbeSession.cc (ARCH):
>> * gprofng/src/Disasm.cc (Disasm::disasm_open): Add RISC-V.
>> * gprofng/src/Experiment.cc (Experiment::ExperimentHandler::startElement): Add RISC-V.
>> * gprofng/src/checks.cc (ARCH): Add RISC-V.
>> * gprofng/src/collctrl.cc (_SC_CPUID_MAX): Add if not define _SC_CPUID_MAX condition. Set CPU frequency as 1000 to test but now not show at the test result, will fix with better way to get cpu frequency.
>> (Coll_Ctrl::Coll_Ctrl):
>> (defined):
>> * gprofng/src/dbe_types.h (enum Platform_t):Add RISC-V.
>> * gprofng/testsuite/gprofng.display/mttest/gethrtime.c: Add RISC-V.
>>
>> ---
>> configure | 2 +-
>> configure.ac <http://configure.ac> | 2 +-
>> gprofng/common/core_pcbe.c | 5 ++-
>> gprofng/common/cpuid.c | 31 +++++++++++++++++--
>> gprofng/common/gp-defs.h | 5 ++-
>> gprofng/common/hwc_cpus.h | 7 +++++
>> gprofng/common/hwcfuncs.h | 2 +-
>> gprofng/configure | 4 +++
>> gprofng/configure.ac <http://configure.ac> | 4 +++
>> gprofng/libcollector/hwprofile.h | 10 ++++++
>> gprofng/libcollector/libcol_util.c | 11 ++++++-
>> gprofng/libcollector/libcol_util.h | 10 +++---
>> gprofng/libcollector/unwind.c | 16 ++++++++--
>> gprofng/src/DbeSession.cc | 2 ++
>> gprofng/src/Disasm.cc | 2 ++
>> gprofng/src/Experiment.cc | 2 ++
>> gprofng/src/checks.cc | 4 +++
>> gprofng/src/collctrl.cc | 26 +++++++++-------
>> gprofng/src/dbe_types.h | 3 +-
>> .../gprofng.display/mttest/gethrtime.c | 2 ++
>> 20 files changed, 123 insertions(+), 27 deletions(-)
>>
>> diff --git a/configure b/configure
>> index dd743c58663..e3cbad5752b 100755
>> --- a/configure
>> +++ b/configure
>> @@ -3145,7 +3145,7 @@ fi
>>
>> if test "$enable_gprofng" = "yes"; then
>> case "${target}" in
>> - x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux*)
>> + x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux* | riscv64-*-linux*)
>> configdirs="$configdirs gprofng"
>> ;;
>> esac
>> diff --git a/configure.ac <http://configure.ac> b/configure.ac <http://configure.ac>
>> index 01cfd017273..e63f27e6a29 100644
>> --- a/configure.ac <http://configure.ac>
>> +++ b/configure.ac <http://configure.ac>
>> @@ -412,7 +412,7 @@ enable_gprofng=$enableval,
>> enable_gprofng=yes)
>> if test "$enable_gprofng" = "yes"; then
>> case "${target}" in
>> - x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux*)
>> + x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux* | riscv64-*-linux*)
>> configdirs="$configdirs gprofng"
>> ;;
>> esac
>> diff --git a/gprofng/common/core_pcbe.c b/gprofng/common/core_pcbe.c
>> index b71d46f4e31..ef2a3f83d42 100644
>> --- a/gprofng/common/core_pcbe.c
>> +++ b/gprofng/common/core_pcbe.c
>> @@ -2754,6 +2754,9 @@ core_pcbe_init (void)
>> return 0;
>> case X86_VENDOR_Intel:
>> break;
>> + case ANDES_VENDOR_ID:
>
> The formatting is wrong.
> See a coding standard:
> https://sourceware.org/gdb/wiki/Internals%20GDB-C-Coding-Standards#Whitespaces
> /*Code indentation*//
> // Lines should be indented with a mix of tabs and spaces. 8
> spaces should be replaced with Tab.
> Vim users could add the following to their ~/.vimrc://
> // set autoindent tabstop=8 shiftwidth=2//
> /
>
>> + case SIFIVE_VENDOR_ID:
>> + case THEAD_VENDOR_ID:
>> default:
>> return -1;
>> }
>> @@ -2915,7 +2918,7 @@ core_pcbe_impl_name (void)
>> static const char *
>> core_pcbe_cpuref (void)
>> {
>> -#if defined(__aarch64__)
>> +#if defined(__aarch64__) || defined(__riscv)
>> return "";
>> #elif defined(__i386__) || defined(__x86_64)
>> switch (cpuid_getmodel ())
>> diff --git a/gprofng/common/cpuid.c b/gprofng/common/cpuid.c
>> index fd98b30c639..c29b6612afe 100644
>> --- a/gprofng/common/cpuid.c
>> +++ b/gprofng/common/cpuid.c
>> @@ -42,6 +42,11 @@ __get_cpuid (unsigned int op ATTRIBUTE_UNUSED, unsigned int *eax,
>> Tprintf (DBG_LT0, "cpuid.c:%d read_cpuid_id() MIDR_EL1=0x%016x\n", __LINE__, *eax);
>> return res;
>> }
>> +#elif defined(__riscv)
>> +#include <sched.h>
>> +#include <sys/syscall.h>
>> +#include <unistd.h>
>> +#include <asm/hwprobe.h>
>> #endif
>>
>> /*
>> @@ -104,7 +109,7 @@ my_cpuid (unsigned int op, cpuid_regs_t *regs)
>> TprintfT (DBG_LT1, "my_cpuid: __get_cpuid(0x%x, 0x%x, 0x%x, 0x%x, 0x%x) returns %d\n",
>> op, regs->eax, regs->ebx, regs->ecx, regs->edx, ret);
>> return ret;
>> -}
>> +}
>> #endif
>>
>> static cpuid_info_t *
>> @@ -180,8 +185,30 @@ get_cpuid_info ()
>> cpi->cpi_model += CPI_MODEL_XTD (regs.eax) << 4;
>> break;
>> }
>> +#elif defined(__riscv)
>> + #ifndef __riscv_hwprobe
>> + cpi->cpi_vendor = 0;
>> + cpi->cpi_family = 0;
>> + cpi->cpi_model = 0;
>> + #else
>> + struct riscv_hwprobe res;
>> + res.key = RISCV_HWPROBE_KEY_MVENDORID;
>> + cpu_set_t cpu_set;
>> + int __riscv_hwprobe (struct riscv_hwprobe *pairs, \
>> + long pair_count, long cpu_count, \
>> + unsigned long *cpus, unsigned long flags) \
>> + {
>> + return syscall(__NR_riscv_hwprobe, pairs, pair_count, cpu_count, cpus, flags);
>> + }
>> + CPU_ZERO(&cpu_set);
>> + CPU_SET(0, &cpu_set);
>> + long ret = __riscv_hwprobe(&res, 1, 1, &cpu_set, 0);
>> + cpi->cpi_vendor = res.value;
>> + cpi->cpi_family = 0;
>> + cpi->cpi_model = 0;
>> + #endif
>> #endif
>> - return cpi;
>> + return cpi;
>> }
>
> Could you also add a code into read_cpuinfo()
> (gprofng/src/collctrl.cc line ~85)
> to read cpu_model, cpu_family, etc from /proc/cpu_info.
>
>
>>
>> static inline uint_t
>> diff --git a/gprofng/common/gp-defs.h b/gprofng/common/gp-defs.h
>> index 7cef5550696..891c4d32cf1 100644
>> --- a/gprofng/common/gp-defs.h
>> +++ b/gprofng/common/gp-defs.h
>> @@ -32,6 +32,7 @@
>> */
>> #define ARCH(x) TOK_A_##x(ARCH)
>> #define TOK_A_Aarch64(x) x##_Aarch64
>> +#define TOK_A_RISCV(x) x##_RISCV
>> #define TOK_A_SPARC(x) x##_SPARC
>> #define TOK_A_Intel(x) x##_Intel
>>
>> @@ -45,11 +46,13 @@
>> #define ARCH_Intel 1
>> #elif defined(__aarch64__)
>> #define ARCH_Aarch64 1
>> +#elif defined(riscv) || defined(__riscv)
>> +#define ARCH_RISCV 1
>> #else
>> #error "Undefined platform"
>> #endif
>>
>> -#if defined(__sparcv9) || defined(__x86_64) || defined(__aarch64__)
>> +#if defined(__sparcv9) || defined(__x86_64) || defined(__aarch64__) || defined(__riscv)
>> #define WSIZE_64 1
>> #else
>> #define WSIZE_32 1
>> diff --git a/gprofng/common/hwc_cpus.h b/gprofng/common/hwc_cpus.h
>> index be820819dd3..b121e4bfb97 100644
>> --- a/gprofng/common/hwc_cpus.h
>> +++ b/gprofng/common/hwc_cpus.h
>> @@ -111,6 +111,13 @@ enum {
>> ARM_CPU_IMP_QCOM = 0x51
>> };
>>
>> +// rscv Constants from arch/riscv/include/asm/vendorid_list.h
>> +enum {
>> + ANDES_VENDOR_ID =0x31e,
>> + SIFIVE_VENDOR_ID =0x489,
>> + THEAD_VENDOR_ID =0x5b7
>> +};
>> +
>> #define AARCH64_VENDORSTR_ARM "ARM"
>>
>> /* strings below must match those returned by cpc_getcpuver() */
>> diff --git a/gprofng/common/hwcfuncs.h b/gprofng/common/hwcfuncs.h
>> index a5f64f2a33f..1a6ee7e30ca 100644
>> --- a/gprofng/common/hwcfuncs.h
>> +++ b/gprofng/common/hwcfuncs.h
>> @@ -98,7 +98,7 @@ typedef struct { /* supplementary data fields */
>>
>> #define HW_INTERVAL_MAX UINT64_MAX
>> #define HW_INTERVAL_PRESET(x) (HW_INTERVAL_MAX - ((uint64_t)(x) - 1))
>> -#define HW_INTERVAL_TYPE(x) ((uint64_t) (x)
>> +#define HW_INTERVAL_TYPE(x) ((uint64_t) (x))
>
> HW_INTERVAL_TYPE is not used. Just remove this line.
>
>
>
>>
>> /* parsing */
>> #define HWCFUNCS_MAX_ATTRS 20
>> diff --git a/gprofng/configure b/gprofng/configure
>> index 1c6a99291be..9ac7a651809 100755
>> --- a/gprofng/configure
>> +++ b/gprofng/configure
>> @@ -15788,6 +15788,10 @@ build_src=
>> build_src=true
>> build_collector=true
>> ;;
>> + riscv*-*-linux*)
>> + build_src=true
>> + build_collector=true
>> + ;;
>> esac
>> # Check whether --enable-gprofng-tools was given.
>> if test "${enable_gprofng_tools+set}" = set; then :
>> diff --git a/gprofng/configure.ac <http://configure.ac> b/gprofng/configure.ac <http://configure.ac>
>> index 6b8fe262748..b4f2981a487 100644
>> --- a/gprofng/configure.ac <http://configure.ac>
>> +++ b/gprofng/configure.ac <http://configure.ac>
>> @@ -63,6 +63,10 @@ build_src=
>> build_src=true
>> build_collector=true
>> ;;
>> + riscv*-*-linux*)
>> + build_src=true
>> + build_collector=true
>> + ;;
>> esac
>> AC_ARG_ENABLE(gprofng-tools,
>> AS_HELP_STRING([--disable-gprofng-tools], [do not build gprofng/src directory]),
>> diff --git a/gprofng/libcollector/hwprofile.h b/gprofng/libcollector/hwprofile.h
>> index 23422dd51fa..6517452eef7 100644
>> --- a/gprofng/libcollector/hwprofile.h
>> +++ b/gprofng/libcollector/hwprofile.h
>> @@ -84,6 +84,16 @@ typedef struct MHwcntr_packet
>> (ucp)->uc_mcontext.regs[CONTEXT_PC] = (greg_t)(funcp); \
>> (ucp)->uc_mcontext.regs[CONTEXT_SP] = 0; \
>> (ucp)->uc_mcontext.regs[CONTEXT_FP] = 0;
>> +
>> +#elif ARCH(RISCV)
>> +#define CONTEXT_PC REG_PC
>> +#define CONTEXT_FP 8
>> +#define CONTEXT_SP 2
>> +#define SETFUNCTIONCONTEXT(ucp,funcp) \
>> + (ucp)->uc_mcontext.__gregs[CONTEXT_PC] = (greg_t)(funcp); \
>> + (ucp)->uc_mcontext.__gregs[CONTEXT_FP] = 0; \
>> + (ucp)->uc_mcontext.__gregs[CONTEXT_SP] = 0;
>> +
>> #endif /* ARCH() */
>>
>> #endif
>> diff --git a/gprofng/libcollector/libcol_util.c b/gprofng/libcollector/libcol_util.c
>> index a8802d4a9b0..8ee1c56ff45 100644
>> --- a/gprofng/libcollector/libcol_util.c
>> +++ b/gprofng/libcollector/libcol_util.c
>> @@ -91,7 +91,7 @@ __collector_gettid ()
>> #endif
>> __asm__ __volatile__(syscall_instr
>> : "=a" (r) : "0" (__NR_gettid)
>> - : syscall_clobber);
>> + : syscall_clobber);
>> #else
>> r = syscall (__NR_gettid);
>> #endif
>> @@ -1459,7 +1459,10 @@ __collector_util_init ()
>> else if ((ptr = dlvsym (libc, "fopen", "GLIBC_2.0")) != NULL)
>> __collector_util_funcs.fopen = ptr;
>> else
>> + {
>> ptr = dlsym (libc, "fopen");
>> + if(ptr) __collector_util_funcs.fopen = ptr;
>> + }
>> if (__collector_util_funcs.fopen == NULL)
>> {
>> CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT fopen: %s\n", dlerror ());
>> @@ -1475,7 +1478,10 @@ __collector_util_init ()
>> else if ((ptr = dlvsym (libc, "popen", "GLIBC_2.0")) != NULL)
>> __collector_util_funcs.popen = ptr;
>> else
>> + {
>> ptr = dlsym (libc, "popen");
>> + if(ptr) __collector_util_funcs.popen = ptr;
>> + }
>> if (__collector_util_funcs.popen == NULL)
>> {
>> CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT popen: %s\n", dlerror ());
>> @@ -1491,7 +1497,10 @@ __collector_util_init ()
>> else if ((ptr = dlvsym (libc, "fclose", "GLIBC_2.0")) != NULL)
>> __collector_util_funcs.fclose = ptr;
>> else
>> + {
>> ptr = dlsym (libc, "fclose");
>> + if(ptr) __collector_util_funcs.fclose = ptr;
>> + }
>> if (__collector_util_funcs.fclose == NULL)
>> {
>> CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT fclose: %s\n", dlerror ());
>> diff --git a/gprofng/libcollector/libcol_util.h b/gprofng/libcollector/libcol_util.h
>> index c21b4a46c4b..6ac84db05ab 100644
>> --- a/gprofng/libcollector/libcol_util.h
>> +++ b/gprofng/libcollector/libcol_util.h
>> @@ -209,10 +209,10 @@ static __attribute__ ((always_inline)) inline uint32_t
>> __collector_cas_32 (volatile uint32_t *pdata, uint32_t old, uint32_t new)
>> {
>> uint32_t r;
>> - __asm__ __volatile__("lock; cmpxchgl %2, %1"
>> - : "=a" (r), "=m" (*pdata) : "r" (new),
>> - "a" (old), "m" (*pdata));
>> - return r;
>> + __asm__ __volatile__("lock; cmpxchgl %2, %1"
>> + : "=a" (r), "=m" (*pdata) : "r" (new),
>> + "a" (old), "m" (*pdata));
>> + return r;
>> }
>> /**
>> * This function enables a compare and swap operation to occur atomically.
>> @@ -270,7 +270,7 @@ __collector_cas_ptr (void *mem, void *cmp, void *new)
>> return r;
>> }
>>
>> -#elif ARCH(Aarch64)
>> +#elif ARCH(Aarch64) || ARCH(RISCV)
>> static __attribute__ ((always_inline)) inline uint32_t
>> __collector_inc_32 (volatile uint32_t *ptr)
>> {
>> diff --git a/gprofng/libcollector/unwind.c b/gprofng/libcollector/unwind.c
>> index ff2f7aa9a7e..1244a453457 100644
>> --- a/gprofng/libcollector/unwind.c
>> +++ b/gprofng/libcollector/unwind.c
>> @@ -186,8 +186,14 @@ memory_error_func (int status ATTRIBUTE_UNUSED, bfd_vma addr ATTRIBUTE_UNUSED,
>> #define GET_PC(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[15])
>> #define GET_SP(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[13])
>> #define GET_FP(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[14])
>> +
>> +#elif ARCH(RISCV)
>> +#define GET_PC(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[REG_PC])
>> +#define GET_SP(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[2])
>> +#define GET_FP(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[8])
>> #endif /* ARCH() */
>>
>> +
>> /*
>> * FILL_CONTEXT() for all platforms
>> * Could use getcontext() except:
>> @@ -230,13 +236,19 @@ memory_error_func (int status ATTRIBUTE_UNUSED, bfd_vma addr ATTRIBUTE_UNUSED,
>> context->uc_stack.ss_size = 0x100000; \
>> }
>>
>> -#elif ARCH(Aarch64)
>> +#elif ARCH(Aarch64)
>> #define FILL_CONTEXT(context) \
>> { CALL_UTIL (getcontext) (context); \
>> context->uc_mcontext.sp = (__u64) __builtin_frame_address(0); \
>> }
>>
>> -#endif /* ARCH() */
>> +#elif ARCH(RISCV)
>> +#define FILL_CONTEXT(context) \
>> + { CALL_UTIL(getcontext)(context); \
>> + context->uc_mcontext.__gregs[2] = (uint64_t) __builtin_frame_address(0); \
>> + }
>> +
>> +#endif/* ARCH() */
>>
>> static int
>> getByteInstruction (unsigned char *p)
>> diff --git a/gprofng/src/DbeSession.cc b/gprofng/src/DbeSession.cc
>> index 86541d97f2e..3c218b5fc91 100644
>> --- a/gprofng/src/DbeSession.cc
>> +++ b/gprofng/src/DbeSession.cc
>> @@ -94,6 +94,8 @@ Platform_t DbeSession::platform =
>> Sparc;
>> #elif ARCH(Aarch64)
>> Aarch64;
>> +#elif ARCH(RISCV)
>> + RISCV;
>> #else // ARCH(Intel)
>> Intel;
>> #endif
>> diff --git a/gprofng/src/Disasm.cc b/gprofng/src/Disasm.cc
>> index 9c7e59a590b..19e6f15ff6f 100644
>> --- a/gprofng/src/Disasm.cc
>> +++ b/gprofng/src/Disasm.cc
>> @@ -208,6 +208,7 @@ Disasm::disasm_open ()
>> case Amd64:
>> need_swap_endian = (DbeSession::platform == Sparc);
>> break;
>> + case RISCV:
>> case Sparcv8plus:
>> case Sparcv9:
>> case Sparc:
>> @@ -246,6 +247,7 @@ Disasm::disasm_open ()
>> dis_info.arch = bfd_arch_i386;
>> dis_info.mach = bfd_mach_x86_64;
>> break;
>> + case RISCV:
>> case Sparcv8plus:
>> case Sparcv9:
>> case Sparc:
>> diff --git a/gprofng/src/Experiment.cc b/gprofng/src/Experiment.cc
>> index 1378ad5ce07..ac78276d23c 100644
>> --- a/gprofng/src/Experiment.cc
>> +++ b/gprofng/src/Experiment.cc
>> @@ -542,6 +542,8 @@ Experiment::ExperimentHandler::startElement (char*, char*, char *qName, Attribut
>> exp->platform = Intel;
>> else if (strcmp (str, "aarch64") == 0)
>> exp->platform = Aarch64;
>> + else if (strcmp (str, "riscv64") == 0)
>> + exp->platform = RISCV;
>
> The formatting is wrong.
> It looks like in your environment the tab size is 4 spaces, but in
> the GNU standard it is 8.
>
>> else
>> exp->platform = Sparc;
>> exp->need_swap_endian = (DbeSession::platform == Sparc) ?
>> diff --git a/gprofng/src/checks.cc b/gprofng/src/checks.cc
>> index 094c3bbc60a..2f9fbd7f714 100644
>> --- a/gprofng/src/checks.cc
>> +++ b/gprofng/src/checks.cc
>> @@ -332,6 +332,10 @@ collect::check_executable_arch (Elf *elf)
>> case EM_AARCH64:
>> is_64 = true;
>> break;
>> +#elif ARCH(RISCV)
>> + case EM_RISCV:
>> + is_64 = true;
>> + break;
>
> Again formatting.
>
>
>> #endif
>> default:
>> return EXEC_ELF_ARCH;
>> diff --git a/gprofng/src/collctrl.cc b/gprofng/src/collctrl.cc
>> index 5d68b689a64..1b04e52b5cc 100644
>> --- a/gprofng/src/collctrl.cc
>> +++ b/gprofng/src/collctrl.cc
>> @@ -51,9 +51,9 @@ extern const char *strsignal (int);
>> #endif
>>
>> // _SC_CPUID_MAX is not available on 2.6/2.7
>> -#ifndef _SC_CPUID_MAX
>> -#define _SC_CPUID_MAX 517
>> -#endif
>> +
>> +// #define _SC_CPUID_MAX 517
>> +// #endif
>
> These lines are missing in latest sources.
> The latest sources are here:
> https://sourceware.org/git/binutils-gdb.git.
>
>
> The 2.43 release will be on Sunday 14th July.
> It will be good if your fixes are included in this release.
>
> -Vladimir
>
>
>>
>> const char *get_fstype (char *);
>>
>> @@ -74,15 +74,16 @@ Coll_Ctrl::Coll_Ctrl (int _interactive, bool _defHWC, bool _kernelHWC)
>> default_stem = strdup ("test");
>>
>> /* get CPU count and processor clock rate */
>> + #ifndef _SC_CPUID_MAX
>> + ncpus = sysconf (_SC_NPROCESSORS_CONF);
>> + /* add 2048 to count, since on some systems CPUID does not start at zero */
>> + ncpumax = ncpus + 2048;
>> + #elif
>> ncpumax = sysconf (_SC_CPUID_MAX);
>> - if (ncpumax == -1)
>> - {
>> - ncpus = sysconf (_SC_NPROCESSORS_CONF);
>> - /* add 2048 to count, since on some systems CPUID does not start at zero */
>> - ncpumax = ncpus + 2048;
>> - }
>> - ncpus = 0;
>> - cpu_clk_freq = 0;
>> + #endif
>> +
>> + //ncpus = 0;
>> + //cpu_clk_freq = 0;
>>
>> // On Linux, read /proc/cpuinfo to get CPU count and clock rate
>> // Note that parsing is different on SPARC and x86
>> @@ -115,6 +116,9 @@ Coll_Ctrl::Coll_Ctrl (int _interactive, bool _defHWC, bool _kernelHWC)
>> #elif defined(__aarch64__)
>> asm volatile("mrs %0, cntfrq_el0" : "=r" (cpu_clk_freq));
>>
>> +#elif defined(__riscv)
>> + cpu_clk_freq = 1000;
>> +
>> #else
>> FILE *procf = fopen ("/proc/cpuinfo", "r");
>> if (procf != NULL)
>> diff --git a/gprofng/src/dbe_types.h b/gprofng/src/dbe_types.h
>> index 9fa842ce85d..c67d421566e 100644
>> --- a/gprofng/src/dbe_types.h
>> +++ b/gprofng/src/dbe_types.h
>> @@ -42,7 +42,8 @@ enum Platform_t
>> Sparcv8plus,
>> Java,
>> Amd64,
>> - Aarch64
>> + Aarch64,
>> + RISCV
>> };
>>
>> enum WSize_t
>> diff --git a/gprofng/testsuite/gprofng.display/mttest/gethrtime.c b/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
>> index da39821027f..5e2cc3dc97b 100644
>> --- a/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
>> +++ b/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
>> @@ -32,6 +32,8 @@
>> #define SPARC 1
>> #elif defined(__aarch64__)
>> #define Aarch64 1
>> +#elif defined(__riscv)
>> +#define RISCV 1
>> #else
>> #define Intel 1
>> #endif
>
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH] [RFC]RISC-V:[gprofng] Mimal support gprofng for riscv
2024-07-03 20:10 ` Vladimir Mezentsev
@ 2024-07-04 9:21 ` Yixuan Chen
0 siblings, 0 replies; 7+ messages in thread
From: Yixuan Chen @ 2024-07-04 9:21 UTC (permalink / raw)
To: Vladimir Mezentsev; +Cc: binutils, ruud.vanderpas, jiawei, shihua, shiyulong
[-- Attachment #1: Type: text/plain, Size: 28438 bytes --]
It works well with your method. I sent a patch v2, fixing the format issue
and removing the useless definition.
Thank you so much!
Best regards,
Yixuan Chen
Vladimir Mezentsev <vladimir.mezentsev@oracle.com> 于2024年7月4日周四 04:10写道:
> Hi Yixuan Chen,
>
> On 7/3/24 05:38, Yixuan Chen wrote:
>
> Hi Vladimir,
>
> Thank you very much for your guide.
>
> Now I use the binutils-gdb master branch to port. But the `Collector
> version : `2.41.50'` doesn't change with binutils-gdb version.
> The third line of the log could see the binutils version.
>
> Here comes the log:
> ```
> Making all in gp-display-html
> make[4]: Entering directory
> '/home/xyenchi/binutils-gdb/build/gprofng/gp-display-html'
> sed -e 's/BINUTILS_VERSION/2.42.50/' < ../../../gprofng/gp-display-html/
> gp-display-html.in > gp-display-html
> chmod +x gp-display-html
> make[4]: Leaving directory
> '/home/xyenchi/binutils-gdb/build/gprofng/gp-display-html'
> Making all in doc
> make[4]: Entering directory '/home/xyenchi/binutils-gdb/build/gprofng/doc'
> make[4]: Nothing to be done for 'all'.
> make[4]: Leaving directory '/home/xyenchi/binutils-gdb/build/gprofng/doc'
> make[4]: Entering directory '/home/xyenchi/binutils-gdb/build/gprofng'
> make[4]: Leaving directory '/home/xyenchi/binutils-gdb/build/gprofng'
> make[3]: Leaving directory '/home/xyenchi/binutils-gdb/build/gprofng'
> make[2]: Leaving directory '/home/xyenchi/binutils-gdb/build/gprofng'
> make[1]: Leaving directory '/home/xyenchi/binutils-gdb/build'
> [xyenchi@qemu-riscv build]$ rm -rf test.1.er/
> [xyenchi@qemu-riscv build]$ ./gprofng/src/gp-collect-app echo 1
>
>
> This is wrong.
> You are using the built version of gprofng. But you must use the installed
> version.
> For example:
> cd /home/xyenchi/binutils-gdb/build/
> rm -rf *
> ../configure --prefix=`pwd`/INSTALL [THE_OTHER_CONFIGURE_OPTIONS]
> make
> make install
> /home/xyenchi/binutils-gdb/build/INSTALL/bin/gprofng collect app -O
> test.1.er echo 1
> *We recommend using `gprofng collect app` instead of calling
> gp-collect-app directly.*
> /home/xyenchi/binutils-gdb/build/INSTALL/bin/gprofng display text -func
> -exp -head test.1.er/
>
>
> The problem is:
> gp-collect-app sets
> LD_PRELOAD=<PATH_TO_GPROFNG_LIB_DIR>/libgp-collector.so before starting the
> user application.
> It looks like you did the installation for the old sources and did not
> reinstall gprofng.
> In this case, you used the old libgp-collector.so (2.41.50).
>
>
> Thank you for fixing gprofng.
> -Vladimir
>
>
> Creating experiment directory test.1.er (Process ID: 127275) ...
> 1
> ./gprofng/src/gp-display-text -func -exp
> -head test.1.er/v build]$ ./gprofng/src/gp-display-text -func -exp -head
> test.1.er/
> Functions sorted by metric: Exclusive Total CPU Time
>
> Excl. Total Incl. Total Name
> CPU CPU
> sec. % sec. %
> 0. 0. 0. 0. <Total>
>
> ID Sel PID Experiment
> == === ====== ==========
> 1 yes 127275 test.1.er/
> Experiment: test.1.er/
> No errors
> No warnings
> No archive command run
>
> Target command (64-bit): 'echo 1'
> Process pid 127275, ppid 399, pgrp 127275, sid 391
> Current working directory: /home/xyenchi/binutils-gdb/build
> Collector version: `2.41.50'; experiment version 12.4 (64-bit)
> Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture
> `riscv64'
> 0 CPUs, clock speed 0 MHz.
> Memory: 4096267 pages @ 4096 = 16001 MB.
> Data collection parameters:
> Clock-profiling, interval = 10007 microsecs.
> Periodic sampling, 1 secs.
> Follow descendant processes from: fork|exec|combo
>
> Experiment started Wed Jul 3 20:09:17 2024
>
> Experiment Ended: 0.022871300
> Data Collection Duration: 0.022871300
> ```
>
> I only find this code to change the collector version:
> https://github.com/bminor/binutils-gdb/blob/afa87be0fc751c158d666a48a9077d26d6ec8666/gprofng/src/Experiment.cc#L1224
> but I don't known how to fix, could you give me some suggestion?
>
> Best regards,
> Yixuan Chen
>
> Vladimir Mezentsev <vladimir.mezentsev@oracle.com> 于2024年7月2日周二 03:39写道:
>
>> Thank you for your work on gprofng.
>> See my comments below.
>>
>>
>> On 6/30/24 02:37, XYenChi wrote:
>>
>> Minimal support RISC-V. Test with qemu.
>>
>> Result show:
>> [xyenchi@qemu-riscv build]$ ./gprofng/src/gp-display-text -fu1.er/p -head test.2.er/
>> Functions sorted by metric: Exclusive Total CPU Time
>>
>> Excl. Total Incl. Total Name
>> CPU CPU
>> sec. % sec. %
>> 0. 0. 0. 0. <Total>
>>
>> ID Sel PID Experiment
>> == === ====== ==========
>> 1 yes 145674 test.1.er/
>> Experiment: test.1.er/
>> No errors
>> No warnings
>> No archive command run
>>
>> Target command (64-bit): 'echo 1'
>> Process pid 145674, ppid 399, pgrp 145674, sid 392
>> Current working directory: /home/xyenchi/binutils-gdb/build
>> Collector version: `2.41.50'; experiment version 12.4 (64-bit)
>>
>>
>> It looks like you are using an old version of the binutils-gdb source.
>> The current version is 2.42.50.
>>
>>
>> Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture `riscv64'
>> 0 CPUs, clock speed 0 MHz.
>> Memory: 4096267 pages @ 4096 = 16001 MB.
>> Data collection parameters:
>> Clock-profiling, interval = 10007 microsecs.
>> Periodic sampling, 1 secs.
>> Follow descendant processes from: fork|exec|combo
>>
>> Experiment started Thu Jun 27 16:23:20 2024
>>
>> Experiment Ended: 0.040909200
>> Data Collection Duration: 0.040909200
>>
>> If build with `allow_undefined_flag=true`, result will be:
>>
>> [xyenchi@qemu-riscv build]$ ./gprofng/src/gp-display-text -fu1.er/p -head test.2.er/
>> Functions sorted by metric: Exclusive Total CPU Time
>>
>> Excl. Total Incl. Total Name
>> CPU CPU
>> sec. % sec. %
>> 0. 0. 0. 0. <Total>
>>
>> ID Sel PID Experiment
>> == === ====== ==========
>> 1 yes 145674 test.1.er/
>> Experiment: test.1.er/
>> No errors
>> No warnings
>> No archive command run
>>
>> Target command (64-bit): 'echo 1'
>> Process pid 145674, ppid 399, pgrp 145674, sid 392
>> Current working directory: /home/xyenchi/binutils-gdb/build
>> Collector version: `2.41.50'; experiment version 12.4 (64-bit)
>> Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture `riscv64'
>> 0 CPUs, clock speed 0 MHz.
>> Memory: 4096267 pages @ 4096 = 16001 MB.
>> Data collection parameters:
>> Clock-profiling, interval = 10007 microsecs.
>> Periodic sampling, 1 secs.
>> Follow descendant processes from: fork|exec|combo
>>
>> Experiment started Thu Jun 27 16:23:20 2024
>>
>> Experiment Ended: 0.040909200
>> Data Collection Duration: 0.040909200
>>
>> ChangeLog:
>> Minimal support gprofng for RISC-V.
>>
>> 2024-06-28 Yixuan Chen <chenyixuan@iscasc.ac.cn> <chenyixuan@iscasc.ac.cn>
>>
>> * configure: Add RISC-V configure.
>> * configure.ac: Add RISC-V configure.
>>
>> Unknown ChangeLog:
>>
>> 2024-06-28 Yixuan Chen <chenyixuan@iscasc.ac.cn> <chenyixuan@iscasc.ac.cn>
>>
>> * gprofng/common/core_pcbe.c (core_pcbe_init): Add RISC-V vendor.
>> (defined): Add RISC-V condition.
>> * gprofng/common/cpuid.c (defined): Add head file to use hwprobe.
>> (my_cpuid): Add riscv hwprobe.
>> * gprofng/common/gp-defs.h (TOK_A_RISCV): Add RISC-V.
>> (defined): Add RISC-V.
>> (ARCH_RISCV): Add RISC-V.
>> * gprofng/common/hwc_cpus.h: Add RISC-V vendor id.
>> * gprofng/common/hwcfuncs.h (HW_INTERVAL_TYPE): fix typo.
>> * gprofng/configure: Add RISC-V.
>> * gprofng/configure.ac: Add RISC-V.
>> * gprofng/libcollector/hwprofile.h (ARCH): Add riscv register calling.
>> (CONTEXT_PC): Add riscv register calling.
>> (CONTEXT_FP): Add riscv register calling.
>> (CONTEXT_SP): Add riscv register calling.
>> (SETFUNCTIONCONTEXT): Add riscv uc_mcontext calling.
>> * gprofng/libcollector/libcol_util.c (__collector_util_init): fix libc calling
>> * gprofng/libcollector/libcol_util.h (__collector_cas_32): Add RISC-V.
>> (ARCH):
>> * gprofng/libcollector/unwind.c (ARCH): Add RISC-V.
>> (GET_PC): Add riscv register calling.
>> (GET_SP): Add riscv register calling.
>> (GET_FP): Add riscv register calling.
>> (FILL_CONTEXT): Add riscv uc_mcontext calling.
>> * gprofng/src/DbeSession.cc (ARCH):
>> * gprofng/src/Disasm.cc (Disasm::disasm_open): Add RISC-V.
>> * gprofng/src/Experiment.cc (Experiment::ExperimentHandler::startElement): Add RISC-V.
>> * gprofng/src/checks.cc (ARCH): Add RISC-V.
>> * gprofng/src/collctrl.cc (_SC_CPUID_MAX): Add if not define _SC_CPUID_MAX condition. Set CPU frequency as 1000 to test but now not show at the test result, will fix with better way to get cpu frequency.
>> (Coll_Ctrl::Coll_Ctrl):
>> (defined):
>> * gprofng/src/dbe_types.h (enum Platform_t):Add RISC-V.
>> * gprofng/testsuite/gprofng.display/mttest/gethrtime.c: Add RISC-V.
>>
>> ---
>> configure | 2 +-
>> configure.ac | 2 +-
>> gprofng/common/core_pcbe.c | 5 ++-
>> gprofng/common/cpuid.c | 31 +++++++++++++++++--
>> gprofng/common/gp-defs.h | 5 ++-
>> gprofng/common/hwc_cpus.h | 7 +++++
>> gprofng/common/hwcfuncs.h | 2 +-
>> gprofng/configure | 4 +++
>> gprofng/configure.ac | 4 +++
>> gprofng/libcollector/hwprofile.h | 10 ++++++
>> gprofng/libcollector/libcol_util.c | 11 ++++++-
>> gprofng/libcollector/libcol_util.h | 10 +++---
>> gprofng/libcollector/unwind.c | 16 ++++++++--
>> gprofng/src/DbeSession.cc | 2 ++
>> gprofng/src/Disasm.cc | 2 ++
>> gprofng/src/Experiment.cc | 2 ++
>> gprofng/src/checks.cc | 4 +++
>> gprofng/src/collctrl.cc | 26 +++++++++-------
>> gprofng/src/dbe_types.h | 3 +-
>> .../gprofng.display/mttest/gethrtime.c | 2 ++
>> 20 files changed, 123 insertions(+), 27 deletions(-)
>>
>> diff --git a/configure b/configure
>> index dd743c58663..e3cbad5752b 100755
>> --- a/configure
>> +++ b/configure
>> @@ -3145,7 +3145,7 @@ fi
>>
>> if test "$enable_gprofng" = "yes"; then
>> case "${target}" in
>> - x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux*)
>> + x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux* | riscv64-*-linux*)
>> configdirs="$configdirs gprofng"
>> ;;
>> esac
>> diff --git a/configure.ac b/configure.ac
>> index 01cfd017273..e63f27e6a29 100644
>> --- a/configure.ac
>> +++ b/configure.ac
>> @@ -412,7 +412,7 @@ enable_gprofng=$enableval,
>> enable_gprofng=yes)
>> if test "$enable_gprofng" = "yes"; then
>> case "${target}" in
>> - x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux*)
>> + x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux* | riscv64-*-linux*)
>> configdirs="$configdirs gprofng"
>> ;;
>> esac
>> diff --git a/gprofng/common/core_pcbe.c b/gprofng/common/core_pcbe.c
>> index b71d46f4e31..ef2a3f83d42 100644
>> --- a/gprofng/common/core_pcbe.c
>> +++ b/gprofng/common/core_pcbe.c
>> @@ -2754,6 +2754,9 @@ core_pcbe_init (void)
>> return 0;
>> case X86_VENDOR_Intel:
>> break;
>> + case ANDES_VENDOR_ID:
>>
>>
>> The formatting is wrong.
>> See a coding standard:
>> https://sourceware.org/gdb/wiki/Internals%20GDB-C-Coding-Standards#Whitespaces
>> * Code indentation*
>>
>> * Lines should be indented with a mix of tabs and spaces. 8 spaces
>> should be replaced with Tab. Vim users could add the following to their
>> ~/.vimrc:*
>> * set autoindent tabstop=8 shiftwidth=2*
>>
>>
>> + case SIFIVE_VENDOR_ID:
>> + case THEAD_VENDOR_ID:
>> default:
>> return -1;
>> }
>> @@ -2915,7 +2918,7 @@ core_pcbe_impl_name (void)
>> static const char *
>> core_pcbe_cpuref (void)
>> {
>> -#if defined(__aarch64__)
>> +#if defined(__aarch64__) || defined(__riscv)
>> return "";
>> #elif defined(__i386__) || defined(__x86_64)
>> switch (cpuid_getmodel ())
>> diff --git a/gprofng/common/cpuid.c b/gprofng/common/cpuid.c
>> index fd98b30c639..c29b6612afe 100644
>> --- a/gprofng/common/cpuid.c
>> +++ b/gprofng/common/cpuid.c
>> @@ -42,6 +42,11 @@ __get_cpuid (unsigned int op ATTRIBUTE_UNUSED, unsigned int *eax,
>> Tprintf (DBG_LT0, "cpuid.c:%d read_cpuid_id() MIDR_EL1=0x%016x\n", __LINE__, *eax);
>> return res;
>> }
>> +#elif defined(__riscv)
>> +#include <sched.h>
>> +#include <sys/syscall.h>
>> +#include <unistd.h>
>> +#include <asm/hwprobe.h>
>> #endif
>>
>> /*
>> @@ -104,7 +109,7 @@ my_cpuid (unsigned int op, cpuid_regs_t *regs)
>> TprintfT (DBG_LT1, "my_cpuid: __get_cpuid(0x%x, 0x%x, 0x%x, 0x%x, 0x%x) returns %d\n",
>> op, regs->eax, regs->ebx, regs->ecx, regs->edx, ret);
>> return ret;
>> -}
>> +}
>> #endif
>>
>> static cpuid_info_t *
>> @@ -180,8 +185,30 @@ get_cpuid_info ()
>> cpi->cpi_model += CPI_MODEL_XTD (regs.eax) << 4;
>> break;
>> }
>> +#elif defined(__riscv)
>> + #ifndef __riscv_hwprobe
>> + cpi->cpi_vendor = 0;
>> + cpi->cpi_family = 0;
>> + cpi->cpi_model = 0;
>> + #else
>> + struct riscv_hwprobe res;
>> + res.key = RISCV_HWPROBE_KEY_MVENDORID;
>> + cpu_set_t cpu_set;
>> + int __riscv_hwprobe (struct riscv_hwprobe *pairs, \
>> + long pair_count, long cpu_count, \
>> + unsigned long *cpus, unsigned long flags) \
>> + {
>> + return syscall(__NR_riscv_hwprobe, pairs, pair_count, cpu_count, cpus, flags);
>> + }
>> + CPU_ZERO(&cpu_set);
>> + CPU_SET(0, &cpu_set);
>> + long ret = __riscv_hwprobe(&res, 1, 1, &cpu_set, 0);
>> + cpi->cpi_vendor = res.value;
>> + cpi->cpi_family = 0;
>> + cpi->cpi_model = 0;
>> + #endif
>> #endif
>> - return cpi;
>> + return cpi;
>> }
>>
>>
>> Could you also add a code into read_cpuinfo() (gprofng/src/collctrl.cc
>> line ~85)
>> to read cpu_model, cpu_family, etc from /proc/cpu_info.
>>
>>
>>
>> static inline uint_t
>> diff --git a/gprofng/common/gp-defs.h b/gprofng/common/gp-defs.h
>> index 7cef5550696..891c4d32cf1 100644
>> --- a/gprofng/common/gp-defs.h
>> +++ b/gprofng/common/gp-defs.h
>> @@ -32,6 +32,7 @@
>> */
>> #define ARCH(x) TOK_A_##x(ARCH)
>> #define TOK_A_Aarch64(x) x##_Aarch64
>> +#define TOK_A_RISCV(x) x##_RISCV
>> #define TOK_A_SPARC(x) x##_SPARC
>> #define TOK_A_Intel(x) x##_Intel
>>
>> @@ -45,11 +46,13 @@
>> #define ARCH_Intel 1
>> #elif defined(__aarch64__)
>> #define ARCH_Aarch64 1
>> +#elif defined(riscv) || defined(__riscv)
>> +#define ARCH_RISCV 1
>> #else
>> #error "Undefined platform"
>> #endif
>>
>> -#if defined(__sparcv9) || defined(__x86_64) || defined(__aarch64__)
>> +#if defined(__sparcv9) || defined(__x86_64) || defined(__aarch64__) || defined(__riscv)
>> #define WSIZE_64 1
>> #else
>> #define WSIZE_32 1
>> diff --git a/gprofng/common/hwc_cpus.h b/gprofng/common/hwc_cpus.h
>> index be820819dd3..b121e4bfb97 100644
>> --- a/gprofng/common/hwc_cpus.h
>> +++ b/gprofng/common/hwc_cpus.h
>> @@ -111,6 +111,13 @@ enum {
>> ARM_CPU_IMP_QCOM = 0x51
>> };
>>
>> +// rscv Constants from arch/riscv/include/asm/vendorid_list.h
>> +enum {
>> + ANDES_VENDOR_ID =0x31e,
>> + SIFIVE_VENDOR_ID =0x489,
>> + THEAD_VENDOR_ID =0x5b7
>> +};
>> +
>> #define AARCH64_VENDORSTR_ARM "ARM"
>>
>> /* strings below must match those returned by cpc_getcpuver() */
>> diff --git a/gprofng/common/hwcfuncs.h b/gprofng/common/hwcfuncs.h
>> index a5f64f2a33f..1a6ee7e30ca 100644
>> --- a/gprofng/common/hwcfuncs.h
>> +++ b/gprofng/common/hwcfuncs.h
>> @@ -98,7 +98,7 @@ typedef struct { /* supplementary data fields */
>>
>> #define HW_INTERVAL_MAX UINT64_MAX
>> #define HW_INTERVAL_PRESET(x) (HW_INTERVAL_MAX - ((uint64_t)(x) - 1))
>> -#define HW_INTERVAL_TYPE(x) ((uint64_t) (x)
>> +#define HW_INTERVAL_TYPE(x) ((uint64_t) (x))
>>
>>
>> HW_INTERVAL_TYPE is not used. Just remove this line.
>>
>>
>>
>>
>> /* parsing */
>> #define HWCFUNCS_MAX_ATTRS 20
>> diff --git a/gprofng/configure b/gprofng/configure
>> index 1c6a99291be..9ac7a651809 100755
>> --- a/gprofng/configure
>> +++ b/gprofng/configure
>> @@ -15788,6 +15788,10 @@ build_src=
>> build_src=true
>> build_collector=true
>> ;;
>> + riscv*-*-linux*)
>> + build_src=true
>> + build_collector=true
>> + ;;
>> esac
>> # Check whether --enable-gprofng-tools was given.
>> if test "${enable_gprofng_tools+set}" = set; then :
>> diff --git a/gprofng/configure.ac b/gprofng/configure.ac
>> index 6b8fe262748..b4f2981a487 100644
>> --- a/gprofng/configure.ac
>> +++ b/gprofng/configure.ac
>> @@ -63,6 +63,10 @@ build_src=
>> build_src=true
>> build_collector=true
>> ;;
>> + riscv*-*-linux*)
>> + build_src=true
>> + build_collector=true
>> + ;;
>> esac
>> AC_ARG_ENABLE(gprofng-tools,
>> AS_HELP_STRING([--disable-gprofng-tools], [do not build gprofng/src directory]),
>> diff --git a/gprofng/libcollector/hwprofile.h b/gprofng/libcollector/hwprofile.h
>> index 23422dd51fa..6517452eef7 100644
>> --- a/gprofng/libcollector/hwprofile.h
>> +++ b/gprofng/libcollector/hwprofile.h
>> @@ -84,6 +84,16 @@ typedef struct MHwcntr_packet
>> (ucp)->uc_mcontext.regs[CONTEXT_PC] = (greg_t)(funcp); \
>> (ucp)->uc_mcontext.regs[CONTEXT_SP] = 0; \
>> (ucp)->uc_mcontext.regs[CONTEXT_FP] = 0;
>> +
>> +#elif ARCH(RISCV)
>> +#define CONTEXT_PC REG_PC
>> +#define CONTEXT_FP 8
>> +#define CONTEXT_SP 2
>> +#define SETFUNCTIONCONTEXT(ucp,funcp) \
>> + (ucp)->uc_mcontext.__gregs[CONTEXT_PC] = (greg_t)(funcp); \
>> + (ucp)->uc_mcontext.__gregs[CONTEXT_FP] = 0; \
>> + (ucp)->uc_mcontext.__gregs[CONTEXT_SP] = 0;
>> +
>> #endif /* ARCH() */
>>
>> #endif
>> diff --git a/gprofng/libcollector/libcol_util.c b/gprofng/libcollector/libcol_util.c
>> index a8802d4a9b0..8ee1c56ff45 100644
>> --- a/gprofng/libcollector/libcol_util.c
>> +++ b/gprofng/libcollector/libcol_util.c
>> @@ -91,7 +91,7 @@ __collector_gettid ()
>> #endif
>> __asm__ __volatile__(syscall_instr
>> : "=a" (r) : "0" (__NR_gettid)
>> - : syscall_clobber);
>> + : syscall_clobber);
>> #else
>> r = syscall (__NR_gettid);
>> #endif
>> @@ -1459,7 +1459,10 @@ __collector_util_init ()
>> else if ((ptr = dlvsym (libc, "fopen", "GLIBC_2.0")) != NULL)
>> __collector_util_funcs.fopen = ptr;
>> else
>> + {
>> ptr = dlsym (libc, "fopen");
>> + if(ptr) __collector_util_funcs.fopen = ptr;
>> + }
>> if (__collector_util_funcs.fopen == NULL)
>> {
>> CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT fopen: %s\n", dlerror ());
>> @@ -1475,7 +1478,10 @@ __collector_util_init ()
>> else if ((ptr = dlvsym (libc, "popen", "GLIBC_2.0")) != NULL)
>> __collector_util_funcs.popen = ptr;
>> else
>> + {
>> ptr = dlsym (libc, "popen");
>> + if(ptr) __collector_util_funcs.popen = ptr;
>> + }
>> if (__collector_util_funcs.popen == NULL)
>> {
>> CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT popen: %s\n", dlerror ());
>> @@ -1491,7 +1497,10 @@ __collector_util_init ()
>> else if ((ptr = dlvsym (libc, "fclose", "GLIBC_2.0")) != NULL)
>> __collector_util_funcs.fclose = ptr;
>> else
>> + {
>> ptr = dlsym (libc, "fclose");
>> + if(ptr) __collector_util_funcs.fclose = ptr;
>> + }
>> if (__collector_util_funcs.fclose == NULL)
>> {
>> CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT fclose: %s\n", dlerror ());
>> diff --git a/gprofng/libcollector/libcol_util.h b/gprofng/libcollector/libcol_util.h
>> index c21b4a46c4b..6ac84db05ab 100644
>> --- a/gprofng/libcollector/libcol_util.h
>> +++ b/gprofng/libcollector/libcol_util.h
>> @@ -209,10 +209,10 @@ static __attribute__ ((always_inline)) inline uint32_t
>> __collector_cas_32 (volatile uint32_t *pdata, uint32_t old, uint32_t new)
>> {
>> uint32_t r;
>> - __asm__ __volatile__("lock; cmpxchgl %2, %1"
>> - : "=a" (r), "=m" (*pdata) : "r" (new),
>> - "a" (old), "m" (*pdata));
>> - return r;
>> + __asm__ __volatile__("lock; cmpxchgl %2, %1"
>> + : "=a" (r), "=m" (*pdata) : "r" (new),
>> + "a" (old), "m" (*pdata));
>> + return r;
>> }
>> /**
>> * This function enables a compare and swap operation to occur atomically.
>> @@ -270,7 +270,7 @@ __collector_cas_ptr (void *mem, void *cmp, void *new)
>> return r;
>> }
>>
>> -#elif ARCH(Aarch64)
>> +#elif ARCH(Aarch64) || ARCH(RISCV)
>> static __attribute__ ((always_inline)) inline uint32_t
>> __collector_inc_32 (volatile uint32_t *ptr)
>> {
>> diff --git a/gprofng/libcollector/unwind.c b/gprofng/libcollector/unwind.c
>> index ff2f7aa9a7e..1244a453457 100644
>> --- a/gprofng/libcollector/unwind.c
>> +++ b/gprofng/libcollector/unwind.c
>> @@ -186,8 +186,14 @@ memory_error_func (int status ATTRIBUTE_UNUSED, bfd_vma addr ATTRIBUTE_UNUSED,
>> #define GET_PC(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[15])
>> #define GET_SP(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[13])
>> #define GET_FP(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[14])
>> +
>> +#elif ARCH(RISCV)
>> +#define GET_PC(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[REG_PC])
>> +#define GET_SP(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[2])
>> +#define GET_FP(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[8])
>> #endif /* ARCH() */
>>
>> +
>> /*
>> * FILL_CONTEXT() for all platforms
>> * Could use getcontext() except:
>> @@ -230,13 +236,19 @@ memory_error_func (int status ATTRIBUTE_UNUSED, bfd_vma addr ATTRIBUTE_UNUSED,
>> context->uc_stack.ss_size = 0x100000; \
>> }
>>
>> -#elif ARCH(Aarch64)
>> +#elif ARCH(Aarch64)
>> #define FILL_CONTEXT(context) \
>> { CALL_UTIL (getcontext) (context); \
>> context->uc_mcontext.sp = (__u64) __builtin_frame_address(0); \
>> }
>>
>> -#endif /* ARCH() */
>> +#elif ARCH(RISCV)
>> +#define FILL_CONTEXT(context) \
>> + { CALL_UTIL(getcontext)(context); \
>> + context->uc_mcontext.__gregs[2] = (uint64_t) __builtin_frame_address(0); \
>> + }
>> +
>> +#endif/* ARCH() */
>>
>> static int
>> getByteInstruction (unsigned char *p)
>> diff --git a/gprofng/src/DbeSession.cc b/gprofng/src/DbeSession.cc
>> index 86541d97f2e..3c218b5fc91 100644
>> --- a/gprofng/src/DbeSession.cc
>> +++ b/gprofng/src/DbeSession.cc
>> @@ -94,6 +94,8 @@ Platform_t DbeSession::platform =
>> Sparc;
>> #elif ARCH(Aarch64)
>> Aarch64;
>> +#elif ARCH(RISCV)
>> + RISCV;
>> #else // ARCH(Intel)
>> Intel;
>> #endif
>> diff --git a/gprofng/src/Disasm.cc b/gprofng/src/Disasm.cc
>> index 9c7e59a590b..19e6f15ff6f 100644
>> --- a/gprofng/src/Disasm.cc
>> +++ b/gprofng/src/Disasm.cc
>> @@ -208,6 +208,7 @@ Disasm::disasm_open ()
>> case Amd64:
>> need_swap_endian = (DbeSession::platform == Sparc);
>> break;
>> + case RISCV:
>> case Sparcv8plus:
>> case Sparcv9:
>> case Sparc:
>> @@ -246,6 +247,7 @@ Disasm::disasm_open ()
>> dis_info.arch = bfd_arch_i386;
>> dis_info.mach = bfd_mach_x86_64;
>> break;
>> + case RISCV:
>> case Sparcv8plus:
>> case Sparcv9:
>> case Sparc:
>> diff --git a/gprofng/src/Experiment.cc b/gprofng/src/Experiment.cc
>> index 1378ad5ce07..ac78276d23c 100644
>> --- a/gprofng/src/Experiment.cc
>> +++ b/gprofng/src/Experiment.cc
>> @@ -542,6 +542,8 @@ Experiment::ExperimentHandler::startElement (char*, char*, char *qName, Attribut
>> exp->platform = Intel;
>> else if (strcmp (str, "aarch64") == 0)
>> exp->platform = Aarch64;
>> + else if (strcmp (str, "riscv64") == 0)
>> + exp->platform = RISCV;
>>
>>
>> The formatting is wrong.
>> It looks like in your environment the tab size is 4 spaces, but in the
>> GNU standard it is 8.
>>
>> else
>> exp->platform = Sparc;
>> exp->need_swap_endian = (DbeSession::platform == Sparc) ?
>> diff --git a/gprofng/src/checks.cc b/gprofng/src/checks.cc
>> index 094c3bbc60a..2f9fbd7f714 100644
>> --- a/gprofng/src/checks.cc
>> +++ b/gprofng/src/checks.cc
>> @@ -332,6 +332,10 @@ collect::check_executable_arch (Elf *elf)
>> case EM_AARCH64:
>> is_64 = true;
>> break;
>> +#elif ARCH(RISCV)
>> + case EM_RISCV:
>> + is_64 = true;
>> + break;
>>
>>
>> Again formatting.
>>
>>
>> #endif
>> default:
>> return EXEC_ELF_ARCH;
>> diff --git a/gprofng/src/collctrl.cc b/gprofng/src/collctrl.cc
>> index 5d68b689a64..1b04e52b5cc 100644
>> --- a/gprofng/src/collctrl.cc
>> +++ b/gprofng/src/collctrl.cc
>> @@ -51,9 +51,9 @@ extern const char *strsignal (int);
>> #endif
>>
>> // _SC_CPUID_MAX is not available on 2.6/2.7
>> -#ifndef _SC_CPUID_MAX
>> -#define _SC_CPUID_MAX 517
>> -#endif
>> +
>> +// #define _SC_CPUID_MAX 517
>> +// #endif
>>
>>
>> These lines are missing in latest sources.
>> The latest sources are here: https://sourceware.org/git/binutils-gdb.git.
>>
>>
>> The 2.43 release will be on Sunday 14th July.
>> It will be good if your fixes are included in this release.
>>
>> -Vladimir
>>
>>
>>
>> const char *get_fstype (char *);
>>
>> @@ -74,15 +74,16 @@ Coll_Ctrl::Coll_Ctrl (int _interactive, bool _defHWC, bool _kernelHWC)
>> default_stem = strdup ("test");
>>
>> /* get CPU count and processor clock rate */
>> + #ifndef _SC_CPUID_MAX
>> + ncpus = sysconf (_SC_NPROCESSORS_CONF);
>> + /* add 2048 to count, since on some systems CPUID does not start at zero */
>> + ncpumax = ncpus + 2048;
>> + #elif
>> ncpumax = sysconf (_SC_CPUID_MAX);
>> - if (ncpumax == -1)
>> - {
>> - ncpus = sysconf (_SC_NPROCESSORS_CONF);
>> - /* add 2048 to count, since on some systems CPUID does not start at zero */
>> - ncpumax = ncpus + 2048;
>> - }
>> - ncpus = 0;
>> - cpu_clk_freq = 0;
>> + #endif
>> +
>> + //ncpus = 0;
>> + //cpu_clk_freq = 0;
>>
>> // On Linux, read /proc/cpuinfo to get CPU count and clock rate
>> // Note that parsing is different on SPARC and x86
>> @@ -115,6 +116,9 @@ Coll_Ctrl::Coll_Ctrl (int _interactive, bool _defHWC, bool _kernelHWC)
>> #elif defined(__aarch64__)
>> asm volatile("mrs %0, cntfrq_el0" : "=r" (cpu_clk_freq));
>>
>> +#elif defined(__riscv)
>> + cpu_clk_freq = 1000;
>> +
>> #else
>> FILE *procf = fopen ("/proc/cpuinfo", "r");
>> if (procf != NULL)
>> diff --git a/gprofng/src/dbe_types.h b/gprofng/src/dbe_types.h
>> index 9fa842ce85d..c67d421566e 100644
>> --- a/gprofng/src/dbe_types.h
>> +++ b/gprofng/src/dbe_types.h
>> @@ -42,7 +42,8 @@ enum Platform_t
>> Sparcv8plus,
>> Java,
>> Amd64,
>> - Aarch64
>> + Aarch64,
>> + RISCV
>> };
>>
>> enum WSize_t
>> diff --git a/gprofng/testsuite/gprofng.display/mttest/gethrtime.c b/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
>> index da39821027f..5e2cc3dc97b 100644
>> --- a/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
>> +++ b/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
>> @@ -32,6 +32,8 @@
>> #define SPARC 1
>> #elif defined(__aarch64__)
>> #define Aarch64 1
>> +#elif defined(__riscv)
>> +#define RISCV 1
>> #else
>> #define Intel 1
>> #endif
>>
>>
>>
>
^ permalink raw reply [flat|nested] 7+ messages in thread
* [PATCH] [RFC]RISC-V:[gprofng] Mimal support gprofng for riscv
@ 2024-07-01 3:11 chenyixuan
0 siblings, 0 replies; 7+ messages in thread
From: chenyixuan @ 2024-07-01 3:11 UTC (permalink / raw)
To: binutils
Cc: vladimir.mezentsev, ruud.vanderpas, shiyulong, oriachiuan,
shihua, jiawei
From: XYenChi <oriachiuan@gmail.com>
Minimal support RISC-V. Test with qemu.
Result show:
[xyenchi@qemu-riscv build]$ ./gprofng/src/gp-display-text -fu1.er/p -head test.2.er/
Functions sorted by metric: Exclusive Total CPU Time
Excl. Total Incl. Total Name
CPU CPU
sec. % sec. %
0. 0. 0. 0. <Total>
ID Sel PID Experiment
== === ====== ==========
1 yes 145674 test.1.er/
Experiment: test.1.er/
No errors
No warnings
No archive command run
Target command (64-bit): 'echo 1'
Process pid 145674, ppid 399, pgrp 145674, sid 392
Current working directory: /home/xyenchi/binutils-gdb/build
Collector version: `2.41.50'; experiment version 12.4 (64-bit)
Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture `riscv64'
0 CPUs, clock speed 0 MHz.
Memory: 4096267 pages @ 4096 = 16001 MB.
Data collection parameters:
Clock-profiling, interval = 10007 microsecs.
Periodic sampling, 1 secs.
Follow descendant processes from: fork|exec|combo
Experiment started Thu Jun 27 16:23:20 2024
Experiment Ended: 0.040909200
Data Collection Duration: 0.040909200
If build with `allow_undefined_flag=true`, result will be:
[xyenchi@qemu-riscv build]$ ./gprofng/src/gp-display-text -fu1.er/p -head test.2.er/
Functions sorted by metric: Exclusive Total CPU Time
Excl. Total Incl. Total Name
CPU CPU
sec. % sec. %
0. 0. 0. 0. <Total>
ID Sel PID Experiment
== === ====== ==========
1 yes 145674 test.1.er/
Experiment: test.1.er/
No errors
No warnings
No archive command run
Target command (64-bit): 'echo 1'
Process pid 145674, ppid 399, pgrp 145674, sid 392
Current working directory: /home/xyenchi/binutils-gdb/build
Collector version: `2.41.50'; experiment version 12.4 (64-bit)
Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture `riscv64'
0 CPUs, clock speed 0 MHz.
Memory: 4096267 pages @ 4096 = 16001 MB.
Data collection parameters:
Clock-profiling, interval = 10007 microsecs.
Periodic sampling, 1 secs.
Follow descendant processes from: fork|exec|combo
Experiment started Thu Jun 27 16:23:20 2024
Experiment Ended: 0.040909200
Data Collection Duration: 0.040909200
ChangeLog:
Minimal support gprofng for RISC-V.
2024-06-28 Yixuan Chen <chenyixuan@iscasc.ac.cn>
* configure: Add RISC-V configure.
* configure.ac: Add RISC-V configure.
Unknown ChangeLog:
2024-06-28 Yixuan Chen <chenyixuan@iscasc.ac.cn>
* gprofng/common/core_pcbe.c (core_pcbe_init): Add RISC-V vendor.
(defined): Add RISC-V condition.
* gprofng/common/cpuid.c (defined): Add head file to use hwprobe.
(my_cpuid): Add riscv hwprobe.
* gprofng/common/gp-defs.h (TOK_A_RISCV): Add RISC-V.
(defined): Add RISC-V.
(ARCH_RISCV): Add RISC-V.
* gprofng/common/hwc_cpus.h: Add RISC-V vendor id.
* gprofng/common/hwcfuncs.h (HW_INTERVAL_TYPE): fix typo.
* gprofng/configure: Add RISC-V.
* gprofng/configure.ac: Add RISC-V.
* gprofng/libcollector/hwprofile.h (ARCH): Add riscv register calling.
(CONTEXT_PC): Add riscv register calling.
(CONTEXT_FP): Add riscv register calling.
(CONTEXT_SP): Add riscv register calling.
(SETFUNCTIONCONTEXT): Add riscv uc_mcontext calling.
* gprofng/libcollector/libcol_util.c (__collector_util_init): fix libc calling
* gprofng/libcollector/libcol_util.h (__collector_cas_32): Add RISC-V.
(ARCH):
* gprofng/libcollector/unwind.c (ARCH): Add RISC-V.
(GET_PC): Add riscv register calling.
(GET_SP): Add riscv register calling.
(GET_FP): Add riscv register calling.
(FILL_CONTEXT): Add riscv uc_mcontext calling.
* gprofng/src/DbeSession.cc (ARCH):
* gprofng/src/Disasm.cc (Disasm::disasm_open): Add RISC-V.
* gprofng/src/Experiment.cc (Experiment::ExperimentHandler::startElement): Add RISC-V.
* gprofng/src/checks.cc (ARCH): Add RISC-V.
* gprofng/src/collctrl.cc (_SC_CPUID_MAX): Add if not define _SC_CPUID_MAX condition. Set CPU frequency as 1000 to test but now not show at the test result, will fix with better way to get cpu frequency.
(Coll_Ctrl::Coll_Ctrl):
(defined):
* gprofng/src/dbe_types.h (enum Platform_t):Add RISC-V.
* gprofng/testsuite/gprofng.display/mttest/gethrtime.c: Add RISC-V.
---
configure | 2 +-
configure.ac | 2 +-
gprofng/common/core_pcbe.c | 5 ++-
gprofng/common/cpuid.c | 31 +++++++++++++++++--
gprofng/common/gp-defs.h | 5 ++-
gprofng/common/hwc_cpus.h | 7 +++++
gprofng/common/hwcfuncs.h | 2 +-
gprofng/configure | 4 +++
gprofng/configure.ac | 4 +++
gprofng/libcollector/hwprofile.h | 10 ++++++
gprofng/libcollector/libcol_util.c | 11 ++++++-
gprofng/libcollector/libcol_util.h | 10 +++---
gprofng/libcollector/unwind.c | 16 ++++++++--
gprofng/src/DbeSession.cc | 2 ++
gprofng/src/Disasm.cc | 2 ++
gprofng/src/Experiment.cc | 2 ++
gprofng/src/checks.cc | 4 +++
gprofng/src/collctrl.cc | 26 +++++++++-------
gprofng/src/dbe_types.h | 3 +-
.../gprofng.display/mttest/gethrtime.c | 2 ++
20 files changed, 123 insertions(+), 27 deletions(-)
diff --git a/configure b/configure
index dd743c58663..e3cbad5752b 100755
--- a/configure
+++ b/configure
@@ -3145,7 +3145,7 @@ fi
if test "$enable_gprofng" = "yes"; then
case "${target}" in
- x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux*)
+ x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux* | riscv64-*-linux*)
configdirs="$configdirs gprofng"
;;
esac
diff --git a/configure.ac b/configure.ac
index 01cfd017273..e63f27e6a29 100644
--- a/configure.ac
+++ b/configure.ac
@@ -412,7 +412,7 @@ enable_gprofng=$enableval,
enable_gprofng=yes)
if test "$enable_gprofng" = "yes"; then
case "${target}" in
- x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux*)
+ x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux* | riscv64-*-linux*)
configdirs="$configdirs gprofng"
;;
esac
diff --git a/gprofng/common/core_pcbe.c b/gprofng/common/core_pcbe.c
index b71d46f4e31..ef2a3f83d42 100644
--- a/gprofng/common/core_pcbe.c
+++ b/gprofng/common/core_pcbe.c
@@ -2754,6 +2754,9 @@ core_pcbe_init (void)
return 0;
case X86_VENDOR_Intel:
break;
+ case ANDES_VENDOR_ID:
+ case SIFIVE_VENDOR_ID:
+ case THEAD_VENDOR_ID:
default:
return -1;
}
@@ -2915,7 +2918,7 @@ core_pcbe_impl_name (void)
static const char *
core_pcbe_cpuref (void)
{
-#if defined(__aarch64__)
+#if defined(__aarch64__) || defined(__riscv)
return "";
#elif defined(__i386__) || defined(__x86_64)
switch (cpuid_getmodel ())
diff --git a/gprofng/common/cpuid.c b/gprofng/common/cpuid.c
index fd98b30c639..c29b6612afe 100644
--- a/gprofng/common/cpuid.c
+++ b/gprofng/common/cpuid.c
@@ -42,6 +42,11 @@ __get_cpuid (unsigned int op ATTRIBUTE_UNUSED, unsigned int *eax,
Tprintf (DBG_LT0, "cpuid.c:%d read_cpuid_id() MIDR_EL1=0x%016x\n", __LINE__, *eax);
return res;
}
+#elif defined(__riscv)
+#include <sched.h>
+#include <sys/syscall.h>
+#include <unistd.h>
+#include <asm/hwprobe.h>
#endif
/*
@@ -104,7 +109,7 @@ my_cpuid (unsigned int op, cpuid_regs_t *regs)
TprintfT (DBG_LT1, "my_cpuid: __get_cpuid(0x%x, 0x%x, 0x%x, 0x%x, 0x%x) returns %d\n",
op, regs->eax, regs->ebx, regs->ecx, regs->edx, ret);
return ret;
-}
+}
#endif
static cpuid_info_t *
@@ -180,8 +185,30 @@ get_cpuid_info ()
cpi->cpi_model += CPI_MODEL_XTD (regs.eax) << 4;
break;
}
+#elif defined(__riscv)
+ #ifndef __riscv_hwprobe
+ cpi->cpi_vendor = 0;
+ cpi->cpi_family = 0;
+ cpi->cpi_model = 0;
+ #else
+ struct riscv_hwprobe res;
+ res.key = RISCV_HWPROBE_KEY_MVENDORID;
+ cpu_set_t cpu_set;
+ int __riscv_hwprobe (struct riscv_hwprobe *pairs, \
+ long pair_count, long cpu_count, \
+ unsigned long *cpus, unsigned long flags) \
+ {
+ return syscall(__NR_riscv_hwprobe, pairs, pair_count, cpu_count, cpus, flags);
+ }
+ CPU_ZERO(&cpu_set);
+ CPU_SET(0, &cpu_set);
+ long ret = __riscv_hwprobe(&res, 1, 1, &cpu_set, 0);
+ cpi->cpi_vendor = res.value;
+ cpi->cpi_family = 0;
+ cpi->cpi_model = 0;
+ #endif
#endif
- return cpi;
+ return cpi;
}
static inline uint_t
diff --git a/gprofng/common/gp-defs.h b/gprofng/common/gp-defs.h
index 7cef5550696..891c4d32cf1 100644
--- a/gprofng/common/gp-defs.h
+++ b/gprofng/common/gp-defs.h
@@ -32,6 +32,7 @@
*/
#define ARCH(x) TOK_A_##x(ARCH)
#define TOK_A_Aarch64(x) x##_Aarch64
+#define TOK_A_RISCV(x) x##_RISCV
#define TOK_A_SPARC(x) x##_SPARC
#define TOK_A_Intel(x) x##_Intel
@@ -45,11 +46,13 @@
#define ARCH_Intel 1
#elif defined(__aarch64__)
#define ARCH_Aarch64 1
+#elif defined(riscv) || defined(__riscv)
+#define ARCH_RISCV 1
#else
#error "Undefined platform"
#endif
-#if defined(__sparcv9) || defined(__x86_64) || defined(__aarch64__)
+#if defined(__sparcv9) || defined(__x86_64) || defined(__aarch64__) || defined(__riscv)
#define WSIZE_64 1
#else
#define WSIZE_32 1
diff --git a/gprofng/common/hwc_cpus.h b/gprofng/common/hwc_cpus.h
index be820819dd3..b121e4bfb97 100644
--- a/gprofng/common/hwc_cpus.h
+++ b/gprofng/common/hwc_cpus.h
@@ -111,6 +111,13 @@ enum {
ARM_CPU_IMP_QCOM = 0x51
};
+// rscv Constants from arch/riscv/include/asm/vendorid_list.h
+enum {
+ ANDES_VENDOR_ID =0x31e,
+ SIFIVE_VENDOR_ID =0x489,
+ THEAD_VENDOR_ID =0x5b7
+};
+
#define AARCH64_VENDORSTR_ARM "ARM"
/* strings below must match those returned by cpc_getcpuver() */
diff --git a/gprofng/common/hwcfuncs.h b/gprofng/common/hwcfuncs.h
index a5f64f2a33f..1a6ee7e30ca 100644
--- a/gprofng/common/hwcfuncs.h
+++ b/gprofng/common/hwcfuncs.h
@@ -98,7 +98,7 @@ typedef struct { /* supplementary data fields */
#define HW_INTERVAL_MAX UINT64_MAX
#define HW_INTERVAL_PRESET(x) (HW_INTERVAL_MAX - ((uint64_t)(x) - 1))
-#define HW_INTERVAL_TYPE(x) ((uint64_t) (x)
+#define HW_INTERVAL_TYPE(x) ((uint64_t) (x))
/* parsing */
#define HWCFUNCS_MAX_ATTRS 20
diff --git a/gprofng/configure b/gprofng/configure
index 1c6a99291be..9ac7a651809 100755
--- a/gprofng/configure
+++ b/gprofng/configure
@@ -15788,6 +15788,10 @@ build_src=
build_src=true
build_collector=true
;;
+ riscv*-*-linux*)
+ build_src=true
+ build_collector=true
+ ;;
esac
# Check whether --enable-gprofng-tools was given.
if test "${enable_gprofng_tools+set}" = set; then :
diff --git a/gprofng/configure.ac b/gprofng/configure.ac
index 6b8fe262748..b4f2981a487 100644
--- a/gprofng/configure.ac
+++ b/gprofng/configure.ac
@@ -63,6 +63,10 @@ build_src=
build_src=true
build_collector=true
;;
+ riscv*-*-linux*)
+ build_src=true
+ build_collector=true
+ ;;
esac
AC_ARG_ENABLE(gprofng-tools,
AS_HELP_STRING([--disable-gprofng-tools], [do not build gprofng/src directory]),
diff --git a/gprofng/libcollector/hwprofile.h b/gprofng/libcollector/hwprofile.h
index 23422dd51fa..6517452eef7 100644
--- a/gprofng/libcollector/hwprofile.h
+++ b/gprofng/libcollector/hwprofile.h
@@ -84,6 +84,16 @@ typedef struct MHwcntr_packet
(ucp)->uc_mcontext.regs[CONTEXT_PC] = (greg_t)(funcp); \
(ucp)->uc_mcontext.regs[CONTEXT_SP] = 0; \
(ucp)->uc_mcontext.regs[CONTEXT_FP] = 0;
+
+#elif ARCH(RISCV)
+#define CONTEXT_PC REG_PC
+#define CONTEXT_FP 8
+#define CONTEXT_SP 2
+#define SETFUNCTIONCONTEXT(ucp,funcp) \
+ (ucp)->uc_mcontext.__gregs[CONTEXT_PC] = (greg_t)(funcp); \
+ (ucp)->uc_mcontext.__gregs[CONTEXT_FP] = 0; \
+ (ucp)->uc_mcontext.__gregs[CONTEXT_SP] = 0;
+
#endif /* ARCH() */
#endif
diff --git a/gprofng/libcollector/libcol_util.c b/gprofng/libcollector/libcol_util.c
index a8802d4a9b0..8ee1c56ff45 100644
--- a/gprofng/libcollector/libcol_util.c
+++ b/gprofng/libcollector/libcol_util.c
@@ -91,7 +91,7 @@ __collector_gettid ()
#endif
__asm__ __volatile__(syscall_instr
: "=a" (r) : "0" (__NR_gettid)
- : syscall_clobber);
+ : syscall_clobber);
#else
r = syscall (__NR_gettid);
#endif
@@ -1459,7 +1459,10 @@ __collector_util_init ()
else if ((ptr = dlvsym (libc, "fopen", "GLIBC_2.0")) != NULL)
__collector_util_funcs.fopen = ptr;
else
+ {
ptr = dlsym (libc, "fopen");
+ if(ptr) __collector_util_funcs.fopen = ptr;
+ }
if (__collector_util_funcs.fopen == NULL)
{
CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT fopen: %s\n", dlerror ());
@@ -1475,7 +1478,10 @@ __collector_util_init ()
else if ((ptr = dlvsym (libc, "popen", "GLIBC_2.0")) != NULL)
__collector_util_funcs.popen = ptr;
else
+ {
ptr = dlsym (libc, "popen");
+ if(ptr) __collector_util_funcs.popen = ptr;
+ }
if (__collector_util_funcs.popen == NULL)
{
CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT popen: %s\n", dlerror ());
@@ -1491,7 +1497,10 @@ __collector_util_init ()
else if ((ptr = dlvsym (libc, "fclose", "GLIBC_2.0")) != NULL)
__collector_util_funcs.fclose = ptr;
else
+ {
ptr = dlsym (libc, "fclose");
+ if(ptr) __collector_util_funcs.fclose = ptr;
+ }
if (__collector_util_funcs.fclose == NULL)
{
CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT fclose: %s\n", dlerror ());
diff --git a/gprofng/libcollector/libcol_util.h b/gprofng/libcollector/libcol_util.h
index c21b4a46c4b..6ac84db05ab 100644
--- a/gprofng/libcollector/libcol_util.h
+++ b/gprofng/libcollector/libcol_util.h
@@ -209,10 +209,10 @@ static __attribute__ ((always_inline)) inline uint32_t
__collector_cas_32 (volatile uint32_t *pdata, uint32_t old, uint32_t new)
{
uint32_t r;
- __asm__ __volatile__("lock; cmpxchgl %2, %1"
- : "=a" (r), "=m" (*pdata) : "r" (new),
- "a" (old), "m" (*pdata));
- return r;
+ __asm__ __volatile__("lock; cmpxchgl %2, %1"
+ : "=a" (r), "=m" (*pdata) : "r" (new),
+ "a" (old), "m" (*pdata));
+ return r;
}
/**
* This function enables a compare and swap operation to occur atomically.
@@ -270,7 +270,7 @@ __collector_cas_ptr (void *mem, void *cmp, void *new)
return r;
}
-#elif ARCH(Aarch64)
+#elif ARCH(Aarch64) || ARCH(RISCV)
static __attribute__ ((always_inline)) inline uint32_t
__collector_inc_32 (volatile uint32_t *ptr)
{
diff --git a/gprofng/libcollector/unwind.c b/gprofng/libcollector/unwind.c
index ff2f7aa9a7e..1244a453457 100644
--- a/gprofng/libcollector/unwind.c
+++ b/gprofng/libcollector/unwind.c
@@ -186,8 +186,14 @@ memory_error_func (int status ATTRIBUTE_UNUSED, bfd_vma addr ATTRIBUTE_UNUSED,
#define GET_PC(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[15])
#define GET_SP(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[13])
#define GET_FP(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[14])
+
+#elif ARCH(RISCV)
+#define GET_PC(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[REG_PC])
+#define GET_SP(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[2])
+#define GET_FP(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[8])
#endif /* ARCH() */
+
/*
* FILL_CONTEXT() for all platforms
* Could use getcontext() except:
@@ -230,13 +236,19 @@ memory_error_func (int status ATTRIBUTE_UNUSED, bfd_vma addr ATTRIBUTE_UNUSED,
context->uc_stack.ss_size = 0x100000; \
}
-#elif ARCH(Aarch64)
+#elif ARCH(Aarch64)
#define FILL_CONTEXT(context) \
{ CALL_UTIL (getcontext) (context); \
context->uc_mcontext.sp = (__u64) __builtin_frame_address(0); \
}
-#endif /* ARCH() */
+#elif ARCH(RISCV)
+#define FILL_CONTEXT(context) \
+ { CALL_UTIL(getcontext)(context); \
+ context->uc_mcontext.__gregs[2] = (uint64_t) __builtin_frame_address(0); \
+ }
+
+#endif/* ARCH() */
static int
getByteInstruction (unsigned char *p)
diff --git a/gprofng/src/DbeSession.cc b/gprofng/src/DbeSession.cc
index 86541d97f2e..3c218b5fc91 100644
--- a/gprofng/src/DbeSession.cc
+++ b/gprofng/src/DbeSession.cc
@@ -94,6 +94,8 @@ Platform_t DbeSession::platform =
Sparc;
#elif ARCH(Aarch64)
Aarch64;
+#elif ARCH(RISCV)
+ RISCV;
#else // ARCH(Intel)
Intel;
#endif
diff --git a/gprofng/src/Disasm.cc b/gprofng/src/Disasm.cc
index 9c7e59a590b..19e6f15ff6f 100644
--- a/gprofng/src/Disasm.cc
+++ b/gprofng/src/Disasm.cc
@@ -208,6 +208,7 @@ Disasm::disasm_open ()
case Amd64:
need_swap_endian = (DbeSession::platform == Sparc);
break;
+ case RISCV:
case Sparcv8plus:
case Sparcv9:
case Sparc:
@@ -246,6 +247,7 @@ Disasm::disasm_open ()
dis_info.arch = bfd_arch_i386;
dis_info.mach = bfd_mach_x86_64;
break;
+ case RISCV:
case Sparcv8plus:
case Sparcv9:
case Sparc:
diff --git a/gprofng/src/Experiment.cc b/gprofng/src/Experiment.cc
index 1378ad5ce07..ac78276d23c 100644
--- a/gprofng/src/Experiment.cc
+++ b/gprofng/src/Experiment.cc
@@ -542,6 +542,8 @@ Experiment::ExperimentHandler::startElement (char*, char*, char *qName, Attribut
exp->platform = Intel;
else if (strcmp (str, "aarch64") == 0)
exp->platform = Aarch64;
+ else if (strcmp (str, "riscv64") == 0)
+ exp->platform = RISCV;
else
exp->platform = Sparc;
exp->need_swap_endian = (DbeSession::platform == Sparc) ?
diff --git a/gprofng/src/checks.cc b/gprofng/src/checks.cc
index 094c3bbc60a..2f9fbd7f714 100644
--- a/gprofng/src/checks.cc
+++ b/gprofng/src/checks.cc
@@ -332,6 +332,10 @@ collect::check_executable_arch (Elf *elf)
case EM_AARCH64:
is_64 = true;
break;
+#elif ARCH(RISCV)
+ case EM_RISCV:
+ is_64 = true;
+ break;
#endif
default:
return EXEC_ELF_ARCH;
diff --git a/gprofng/src/collctrl.cc b/gprofng/src/collctrl.cc
index 5d68b689a64..1b04e52b5cc 100644
--- a/gprofng/src/collctrl.cc
+++ b/gprofng/src/collctrl.cc
@@ -51,9 +51,9 @@ extern const char *strsignal (int);
#endif
// _SC_CPUID_MAX is not available on 2.6/2.7
-#ifndef _SC_CPUID_MAX
-#define _SC_CPUID_MAX 517
-#endif
+
+// #define _SC_CPUID_MAX 517
+// #endif
const char *get_fstype (char *);
@@ -74,15 +74,16 @@ Coll_Ctrl::Coll_Ctrl (int _interactive, bool _defHWC, bool _kernelHWC)
default_stem = strdup ("test");
/* get CPU count and processor clock rate */
+ #ifndef _SC_CPUID_MAX
+ ncpus = sysconf (_SC_NPROCESSORS_CONF);
+ /* add 2048 to count, since on some systems CPUID does not start at zero */
+ ncpumax = ncpus + 2048;
+ #elif
ncpumax = sysconf (_SC_CPUID_MAX);
- if (ncpumax == -1)
- {
- ncpus = sysconf (_SC_NPROCESSORS_CONF);
- /* add 2048 to count, since on some systems CPUID does not start at zero */
- ncpumax = ncpus + 2048;
- }
- ncpus = 0;
- cpu_clk_freq = 0;
+ #endif
+
+ //ncpus = 0;
+ //cpu_clk_freq = 0;
// On Linux, read /proc/cpuinfo to get CPU count and clock rate
// Note that parsing is different on SPARC and x86
@@ -115,6 +116,9 @@ Coll_Ctrl::Coll_Ctrl (int _interactive, bool _defHWC, bool _kernelHWC)
#elif defined(__aarch64__)
asm volatile("mrs %0, cntfrq_el0" : "=r" (cpu_clk_freq));
+#elif defined(__riscv)
+ cpu_clk_freq = 1000;
+
#else
FILE *procf = fopen ("/proc/cpuinfo", "r");
if (procf != NULL)
diff --git a/gprofng/src/dbe_types.h b/gprofng/src/dbe_types.h
index 9fa842ce85d..c67d421566e 100644
--- a/gprofng/src/dbe_types.h
+++ b/gprofng/src/dbe_types.h
@@ -42,7 +42,8 @@ enum Platform_t
Sparcv8plus,
Java,
Amd64,
- Aarch64
+ Aarch64,
+ RISCV
};
enum WSize_t
diff --git a/gprofng/testsuite/gprofng.display/mttest/gethrtime.c b/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
index da39821027f..5e2cc3dc97b 100644
--- a/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
+++ b/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
@@ -32,6 +32,8 @@
#define SPARC 1
#elif defined(__aarch64__)
#define Aarch64 1
+#elif defined(__riscv)
+#define RISCV 1
#else
#define Intel 1
#endif
--
2.44.0
^ permalink raw reply [flat|nested] 7+ messages in thread
* [PATCH] [RFC]RISC-V:[gprofng] Mimal support gprofng for riscv
@ 2024-06-28 10:28 XYenChi
0 siblings, 0 replies; 7+ messages in thread
From: XYenChi @ 2024-06-28 10:28 UTC (permalink / raw)
To: binutils
Cc: vladimir.mezentsev, ruud.vanderpas, jiawei, shihua, shiyulong,
oriachiuan
From: XYenChi <oriachiuan@gmail.com>
Minimal support RISC-V. Test with qemu.
Result show:
[xyenchi@qemu-riscv build]$ ./gprofng/src/gp-display-text -fu1.er/p -head test.2.er/
Functions sorted by metric: Exclusive Total CPU Time
Excl. Total Incl. Total Name
CPU CPU
sec. % sec. %
0. 0. 0. 0. <Total>
ID Sel PID Experiment
== === ====== ==========
1 yes 145674 test.1.er/
Experiment: test.1.er/
No errors
No warnings
No archive command run
Target command (64-bit): 'echo 1'
Process pid 145674, ppid 399, pgrp 145674, sid 392
Current working directory: /home/xyenchi/binutils-gdb/build
Collector version: `2.41.50'; experiment version 12.4 (64-bit)
Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture `riscv64'
0 CPUs, clock speed 0 MHz.
Memory: 4096267 pages @ 4096 = 16001 MB.
Data collection parameters:
Clock-profiling, interval = 10007 microsecs.
Periodic sampling, 1 secs.
Follow descendant processes from: fork|exec|combo
Experiment started Thu Jun 27 16:23:20 2024
Experiment Ended: 0.040909200
Data Collection Duration: 0.040909200
If build with `allow_undefined_flag=true`, result will be:
[xyenchi@qemu-riscv build]$ ./gprofng/src/gp-display-text -fu1.er/p -head test.2.er/
Functions sorted by metric: Exclusive Total CPU Time
Excl. Total Incl. Total Name
CPU CPU
sec. % sec. %
0. 0. 0. 0. <Total>
ID Sel PID Experiment
== === ====== ==========
1 yes 145674 test.1.er/
Experiment: test.1.er/
No errors
No warnings
No archive command run
Target command (64-bit): 'echo 1'
Process pid 145674, ppid 399, pgrp 145674, sid 392
Current working directory: /home/xyenchi/binutils-gdb/build
Collector version: `2.41.50'; experiment version 12.4 (64-bit)
Host `qemu-riscv', OS `Linux 6.7.0-rc8', page size 4096, architecture `riscv64'
0 CPUs, clock speed 0 MHz.
Memory: 4096267 pages @ 4096 = 16001 MB.
Data collection parameters:
Clock-profiling, interval = 10007 microsecs.
Periodic sampling, 1 secs.
Follow descendant processes from: fork|exec|combo
Experiment started Thu Jun 27 16:23:20 2024
Experiment Ended: 0.040909200
Data Collection Duration: 0.040909200
ChangeLog:
Minimal support gprofng for RISC-V.
2024-06-28 Yixuan Chen <chenyixuan@iscasc.ac.cn>
* configure: Add RISC-V configure.
* configure.ac: Add RISC-V configure.
Unknown ChangeLog:
2024-06-28 Yixuan Chen <chenyixuan@iscasc.ac.cn>
* gprofng/common/core_pcbe.c (core_pcbe_init): Add RISC-V vendor.
(defined): Add RISC-V condition.
* gprofng/common/cpuid.c (defined): Add head file to use hwprobe.
(my_cpuid): Add riscv hwprobe.
* gprofng/common/gp-defs.h (TOK_A_RISCV): Add RISC-V.
(defined): Add RISC-V.
(ARCH_RISCV): Add RISC-V.
* gprofng/common/hwc_cpus.h: Add RISC-V vendor id.
* gprofng/common/hwcfuncs.h (HW_INTERVAL_TYPE): fix typo.
* gprofng/configure: Add RISC-V.
* gprofng/configure.ac: Add RISC-V.
* gprofng/libcollector/hwprofile.h (ARCH): Add riscv register calling.
(CONTEXT_PC): Add riscv register calling.
(CONTEXT_FP): Add riscv register calling.
(CONTEXT_SP): Add riscv register calling.
(SETFUNCTIONCONTEXT): Add riscv uc_mcontext calling.
* gprofng/libcollector/libcol_util.c (__collector_util_init): fix libc calling
* gprofng/libcollector/libcol_util.h (__collector_cas_32): Add RISC-V.
(ARCH):
* gprofng/libcollector/unwind.c (ARCH): Add RISC-V.
(GET_PC): Add riscv register calling.
(GET_SP): Add riscv register calling.
(GET_FP): Add riscv register calling.
(FILL_CONTEXT): Add riscv uc_mcontext calling.
* gprofng/src/DbeSession.cc (ARCH):
* gprofng/src/Disasm.cc (Disasm::disasm_open): Add RISC-V.
* gprofng/src/Experiment.cc (Experiment::ExperimentHandler::startElement): Add RISC-V.
* gprofng/src/checks.cc (ARCH): Add RISC-V.
* gprofng/src/collctrl.cc (_SC_CPUID_MAX): Add if not define _SC_CPUID_MAX condition. Set CPU frequency as 1000 to test but now not show at the test result, will fix with better way to get cpu frequency.
(Coll_Ctrl::Coll_Ctrl):
(defined):
* gprofng/src/dbe_types.h (enum Platform_t):Add RISC-V.
* gprofng/testsuite/gprofng.display/mttest/gethrtime.c: Add RISC-V.
---
configure | 2 +-
configure.ac | 2 +-
gprofng/common/core_pcbe.c | 5 ++-
gprofng/common/cpuid.c | 31 +++++++++++++++++--
gprofng/common/gp-defs.h | 5 ++-
gprofng/common/hwc_cpus.h | 7 +++++
gprofng/common/hwcfuncs.h | 2 +-
gprofng/configure | 4 +++
gprofng/configure.ac | 4 +++
gprofng/libcollector/hwprofile.h | 10 ++++++
gprofng/libcollector/libcol_util.c | 11 ++++++-
gprofng/libcollector/libcol_util.h | 10 +++---
gprofng/libcollector/unwind.c | 16 ++++++++--
gprofng/src/DbeSession.cc | 2 ++
gprofng/src/Disasm.cc | 2 ++
gprofng/src/Experiment.cc | 2 ++
gprofng/src/checks.cc | 4 +++
gprofng/src/collctrl.cc | 26 +++++++++-------
gprofng/src/dbe_types.h | 3 +-
.../gprofng.display/mttest/gethrtime.c | 2 ++
20 files changed, 123 insertions(+), 27 deletions(-)
diff --git a/configure b/configure
index dd743c58663..e3cbad5752b 100755
--- a/configure
+++ b/configure
@@ -3145,7 +3145,7 @@ fi
if test "$enable_gprofng" = "yes"; then
case "${target}" in
- x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux*)
+ x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux* | riscv64-*-linux*)
configdirs="$configdirs gprofng"
;;
esac
diff --git a/configure.ac b/configure.ac
index 01cfd017273..e63f27e6a29 100644
--- a/configure.ac
+++ b/configure.ac
@@ -412,7 +412,7 @@ enable_gprofng=$enableval,
enable_gprofng=yes)
if test "$enable_gprofng" = "yes"; then
case "${target}" in
- x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux*)
+ x86_64-*-linux* | i?86-*-linux* | aarch64-*-linux* | riscv64-*-linux*)
configdirs="$configdirs gprofng"
;;
esac
diff --git a/gprofng/common/core_pcbe.c b/gprofng/common/core_pcbe.c
index b71d46f4e31..ef2a3f83d42 100644
--- a/gprofng/common/core_pcbe.c
+++ b/gprofng/common/core_pcbe.c
@@ -2754,6 +2754,9 @@ core_pcbe_init (void)
return 0;
case X86_VENDOR_Intel:
break;
+ case ANDES_VENDOR_ID:
+ case SIFIVE_VENDOR_ID:
+ case THEAD_VENDOR_ID:
default:
return -1;
}
@@ -2915,7 +2918,7 @@ core_pcbe_impl_name (void)
static const char *
core_pcbe_cpuref (void)
{
-#if defined(__aarch64__)
+#if defined(__aarch64__) || defined(__riscv)
return "";
#elif defined(__i386__) || defined(__x86_64)
switch (cpuid_getmodel ())
diff --git a/gprofng/common/cpuid.c b/gprofng/common/cpuid.c
index fd98b30c639..c29b6612afe 100644
--- a/gprofng/common/cpuid.c
+++ b/gprofng/common/cpuid.c
@@ -42,6 +42,11 @@ __get_cpuid (unsigned int op ATTRIBUTE_UNUSED, unsigned int *eax,
Tprintf (DBG_LT0, "cpuid.c:%d read_cpuid_id() MIDR_EL1=0x%016x\n", __LINE__, *eax);
return res;
}
+#elif defined(__riscv)
+#include <sched.h>
+#include <sys/syscall.h>
+#include <unistd.h>
+#include <asm/hwprobe.h>
#endif
/*
@@ -104,7 +109,7 @@ my_cpuid (unsigned int op, cpuid_regs_t *regs)
TprintfT (DBG_LT1, "my_cpuid: __get_cpuid(0x%x, 0x%x, 0x%x, 0x%x, 0x%x) returns %d\n",
op, regs->eax, regs->ebx, regs->ecx, regs->edx, ret);
return ret;
-}
+}
#endif
static cpuid_info_t *
@@ -180,8 +185,30 @@ get_cpuid_info ()
cpi->cpi_model += CPI_MODEL_XTD (regs.eax) << 4;
break;
}
+#elif defined(__riscv)
+ #ifndef __riscv_hwprobe
+ cpi->cpi_vendor = 0;
+ cpi->cpi_family = 0;
+ cpi->cpi_model = 0;
+ #else
+ struct riscv_hwprobe res;
+ res.key = RISCV_HWPROBE_KEY_MVENDORID;
+ cpu_set_t cpu_set;
+ int __riscv_hwprobe (struct riscv_hwprobe *pairs, \
+ long pair_count, long cpu_count, \
+ unsigned long *cpus, unsigned long flags) \
+ {
+ return syscall(__NR_riscv_hwprobe, pairs, pair_count, cpu_count, cpus, flags);
+ }
+ CPU_ZERO(&cpu_set);
+ CPU_SET(0, &cpu_set);
+ long ret = __riscv_hwprobe(&res, 1, 1, &cpu_set, 0);
+ cpi->cpi_vendor = res.value;
+ cpi->cpi_family = 0;
+ cpi->cpi_model = 0;
+ #endif
#endif
- return cpi;
+ return cpi;
}
static inline uint_t
diff --git a/gprofng/common/gp-defs.h b/gprofng/common/gp-defs.h
index 7cef5550696..891c4d32cf1 100644
--- a/gprofng/common/gp-defs.h
+++ b/gprofng/common/gp-defs.h
@@ -32,6 +32,7 @@
*/
#define ARCH(x) TOK_A_##x(ARCH)
#define TOK_A_Aarch64(x) x##_Aarch64
+#define TOK_A_RISCV(x) x##_RISCV
#define TOK_A_SPARC(x) x##_SPARC
#define TOK_A_Intel(x) x##_Intel
@@ -45,11 +46,13 @@
#define ARCH_Intel 1
#elif defined(__aarch64__)
#define ARCH_Aarch64 1
+#elif defined(riscv) || defined(__riscv)
+#define ARCH_RISCV 1
#else
#error "Undefined platform"
#endif
-#if defined(__sparcv9) || defined(__x86_64) || defined(__aarch64__)
+#if defined(__sparcv9) || defined(__x86_64) || defined(__aarch64__) || defined(__riscv)
#define WSIZE_64 1
#else
#define WSIZE_32 1
diff --git a/gprofng/common/hwc_cpus.h b/gprofng/common/hwc_cpus.h
index be820819dd3..b121e4bfb97 100644
--- a/gprofng/common/hwc_cpus.h
+++ b/gprofng/common/hwc_cpus.h
@@ -111,6 +111,13 @@ enum {
ARM_CPU_IMP_QCOM = 0x51
};
+// rscv Constants from arch/riscv/include/asm/vendorid_list.h
+enum {
+ ANDES_VENDOR_ID =0x31e,
+ SIFIVE_VENDOR_ID =0x489,
+ THEAD_VENDOR_ID =0x5b7
+};
+
#define AARCH64_VENDORSTR_ARM "ARM"
/* strings below must match those returned by cpc_getcpuver() */
diff --git a/gprofng/common/hwcfuncs.h b/gprofng/common/hwcfuncs.h
index a5f64f2a33f..1a6ee7e30ca 100644
--- a/gprofng/common/hwcfuncs.h
+++ b/gprofng/common/hwcfuncs.h
@@ -98,7 +98,7 @@ typedef struct { /* supplementary data fields */
#define HW_INTERVAL_MAX UINT64_MAX
#define HW_INTERVAL_PRESET(x) (HW_INTERVAL_MAX - ((uint64_t)(x) - 1))
-#define HW_INTERVAL_TYPE(x) ((uint64_t) (x)
+#define HW_INTERVAL_TYPE(x) ((uint64_t) (x))
/* parsing */
#define HWCFUNCS_MAX_ATTRS 20
diff --git a/gprofng/configure b/gprofng/configure
index 1c6a99291be..9ac7a651809 100755
--- a/gprofng/configure
+++ b/gprofng/configure
@@ -15788,6 +15788,10 @@ build_src=
build_src=true
build_collector=true
;;
+ riscv*-*-linux*)
+ build_src=true
+ build_collector=true
+ ;;
esac
# Check whether --enable-gprofng-tools was given.
if test "${enable_gprofng_tools+set}" = set; then :
diff --git a/gprofng/configure.ac b/gprofng/configure.ac
index 6b8fe262748..b4f2981a487 100644
--- a/gprofng/configure.ac
+++ b/gprofng/configure.ac
@@ -63,6 +63,10 @@ build_src=
build_src=true
build_collector=true
;;
+ riscv*-*-linux*)
+ build_src=true
+ build_collector=true
+ ;;
esac
AC_ARG_ENABLE(gprofng-tools,
AS_HELP_STRING([--disable-gprofng-tools], [do not build gprofng/src directory]),
diff --git a/gprofng/libcollector/hwprofile.h b/gprofng/libcollector/hwprofile.h
index 23422dd51fa..6517452eef7 100644
--- a/gprofng/libcollector/hwprofile.h
+++ b/gprofng/libcollector/hwprofile.h
@@ -84,6 +84,16 @@ typedef struct MHwcntr_packet
(ucp)->uc_mcontext.regs[CONTEXT_PC] = (greg_t)(funcp); \
(ucp)->uc_mcontext.regs[CONTEXT_SP] = 0; \
(ucp)->uc_mcontext.regs[CONTEXT_FP] = 0;
+
+#elif ARCH(RISCV)
+#define CONTEXT_PC REG_PC
+#define CONTEXT_FP 8
+#define CONTEXT_SP 2
+#define SETFUNCTIONCONTEXT(ucp,funcp) \
+ (ucp)->uc_mcontext.__gregs[CONTEXT_PC] = (greg_t)(funcp); \
+ (ucp)->uc_mcontext.__gregs[CONTEXT_FP] = 0; \
+ (ucp)->uc_mcontext.__gregs[CONTEXT_SP] = 0;
+
#endif /* ARCH() */
#endif
diff --git a/gprofng/libcollector/libcol_util.c b/gprofng/libcollector/libcol_util.c
index a8802d4a9b0..8ee1c56ff45 100644
--- a/gprofng/libcollector/libcol_util.c
+++ b/gprofng/libcollector/libcol_util.c
@@ -91,7 +91,7 @@ __collector_gettid ()
#endif
__asm__ __volatile__(syscall_instr
: "=a" (r) : "0" (__NR_gettid)
- : syscall_clobber);
+ : syscall_clobber);
#else
r = syscall (__NR_gettid);
#endif
@@ -1459,7 +1459,10 @@ __collector_util_init ()
else if ((ptr = dlvsym (libc, "fopen", "GLIBC_2.0")) != NULL)
__collector_util_funcs.fopen = ptr;
else
+ {
ptr = dlsym (libc, "fopen");
+ if(ptr) __collector_util_funcs.fopen = ptr;
+ }
if (__collector_util_funcs.fopen == NULL)
{
CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT fopen: %s\n", dlerror ());
@@ -1475,7 +1478,10 @@ __collector_util_init ()
else if ((ptr = dlvsym (libc, "popen", "GLIBC_2.0")) != NULL)
__collector_util_funcs.popen = ptr;
else
+ {
ptr = dlsym (libc, "popen");
+ if(ptr) __collector_util_funcs.popen = ptr;
+ }
if (__collector_util_funcs.popen == NULL)
{
CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT popen: %s\n", dlerror ());
@@ -1491,7 +1497,10 @@ __collector_util_init ()
else if ((ptr = dlvsym (libc, "fclose", "GLIBC_2.0")) != NULL)
__collector_util_funcs.fclose = ptr;
else
+ {
ptr = dlsym (libc, "fclose");
+ if(ptr) __collector_util_funcs.fclose = ptr;
+ }
if (__collector_util_funcs.fclose == NULL)
{
CALL_UTIL (fprintf)(stderr, "COL_ERROR_UTIL_INIT fclose: %s\n", dlerror ());
diff --git a/gprofng/libcollector/libcol_util.h b/gprofng/libcollector/libcol_util.h
index c21b4a46c4b..6ac84db05ab 100644
--- a/gprofng/libcollector/libcol_util.h
+++ b/gprofng/libcollector/libcol_util.h
@@ -209,10 +209,10 @@ static __attribute__ ((always_inline)) inline uint32_t
__collector_cas_32 (volatile uint32_t *pdata, uint32_t old, uint32_t new)
{
uint32_t r;
- __asm__ __volatile__("lock; cmpxchgl %2, %1"
- : "=a" (r), "=m" (*pdata) : "r" (new),
- "a" (old), "m" (*pdata));
- return r;
+ __asm__ __volatile__("lock; cmpxchgl %2, %1"
+ : "=a" (r), "=m" (*pdata) : "r" (new),
+ "a" (old), "m" (*pdata));
+ return r;
}
/**
* This function enables a compare and swap operation to occur atomically.
@@ -270,7 +270,7 @@ __collector_cas_ptr (void *mem, void *cmp, void *new)
return r;
}
-#elif ARCH(Aarch64)
+#elif ARCH(Aarch64) || ARCH(RISCV)
static __attribute__ ((always_inline)) inline uint32_t
__collector_inc_32 (volatile uint32_t *ptr)
{
diff --git a/gprofng/libcollector/unwind.c b/gprofng/libcollector/unwind.c
index ff2f7aa9a7e..1244a453457 100644
--- a/gprofng/libcollector/unwind.c
+++ b/gprofng/libcollector/unwind.c
@@ -186,8 +186,14 @@ memory_error_func (int status ATTRIBUTE_UNUSED, bfd_vma addr ATTRIBUTE_UNUSED,
#define GET_PC(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[15])
#define GET_SP(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[13])
#define GET_FP(ctx) (((ucontext_t*)ctx)->uc_mcontext.regs[14])
+
+#elif ARCH(RISCV)
+#define GET_PC(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[REG_PC])
+#define GET_SP(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[2])
+#define GET_FP(ctx) (((ucontext_t*)ctx)->uc_mcontext.__gregs[8])
#endif /* ARCH() */
+
/*
* FILL_CONTEXT() for all platforms
* Could use getcontext() except:
@@ -230,13 +236,19 @@ memory_error_func (int status ATTRIBUTE_UNUSED, bfd_vma addr ATTRIBUTE_UNUSED,
context->uc_stack.ss_size = 0x100000; \
}
-#elif ARCH(Aarch64)
+#elif ARCH(Aarch64)
#define FILL_CONTEXT(context) \
{ CALL_UTIL (getcontext) (context); \
context->uc_mcontext.sp = (__u64) __builtin_frame_address(0); \
}
-#endif /* ARCH() */
+#elif ARCH(RISCV)
+#define FILL_CONTEXT(context) \
+ { CALL_UTIL(getcontext)(context); \
+ context->uc_mcontext.__gregs[2] = (uint64_t) __builtin_frame_address(0); \
+ }
+
+#endif/* ARCH() */
static int
getByteInstruction (unsigned char *p)
diff --git a/gprofng/src/DbeSession.cc b/gprofng/src/DbeSession.cc
index 86541d97f2e..3c218b5fc91 100644
--- a/gprofng/src/DbeSession.cc
+++ b/gprofng/src/DbeSession.cc
@@ -94,6 +94,8 @@ Platform_t DbeSession::platform =
Sparc;
#elif ARCH(Aarch64)
Aarch64;
+#elif ARCH(RISCV)
+ RISCV;
#else // ARCH(Intel)
Intel;
#endif
diff --git a/gprofng/src/Disasm.cc b/gprofng/src/Disasm.cc
index 9c7e59a590b..19e6f15ff6f 100644
--- a/gprofng/src/Disasm.cc
+++ b/gprofng/src/Disasm.cc
@@ -208,6 +208,7 @@ Disasm::disasm_open ()
case Amd64:
need_swap_endian = (DbeSession::platform == Sparc);
break;
+ case RISCV:
case Sparcv8plus:
case Sparcv9:
case Sparc:
@@ -246,6 +247,7 @@ Disasm::disasm_open ()
dis_info.arch = bfd_arch_i386;
dis_info.mach = bfd_mach_x86_64;
break;
+ case RISCV:
case Sparcv8plus:
case Sparcv9:
case Sparc:
diff --git a/gprofng/src/Experiment.cc b/gprofng/src/Experiment.cc
index 1378ad5ce07..ac78276d23c 100644
--- a/gprofng/src/Experiment.cc
+++ b/gprofng/src/Experiment.cc
@@ -542,6 +542,8 @@ Experiment::ExperimentHandler::startElement (char*, char*, char *qName, Attribut
exp->platform = Intel;
else if (strcmp (str, "aarch64") == 0)
exp->platform = Aarch64;
+ else if (strcmp (str, "riscv64") == 0)
+ exp->platform = RISCV;
else
exp->platform = Sparc;
exp->need_swap_endian = (DbeSession::platform == Sparc) ?
diff --git a/gprofng/src/checks.cc b/gprofng/src/checks.cc
index 094c3bbc60a..2f9fbd7f714 100644
--- a/gprofng/src/checks.cc
+++ b/gprofng/src/checks.cc
@@ -332,6 +332,10 @@ collect::check_executable_arch (Elf *elf)
case EM_AARCH64:
is_64 = true;
break;
+#elif ARCH(RISCV)
+ case EM_RISCV:
+ is_64 = true;
+ break;
#endif
default:
return EXEC_ELF_ARCH;
diff --git a/gprofng/src/collctrl.cc b/gprofng/src/collctrl.cc
index 5d68b689a64..1b04e52b5cc 100644
--- a/gprofng/src/collctrl.cc
+++ b/gprofng/src/collctrl.cc
@@ -51,9 +51,9 @@ extern const char *strsignal (int);
#endif
// _SC_CPUID_MAX is not available on 2.6/2.7
-#ifndef _SC_CPUID_MAX
-#define _SC_CPUID_MAX 517
-#endif
+
+// #define _SC_CPUID_MAX 517
+// #endif
const char *get_fstype (char *);
@@ -74,15 +74,16 @@ Coll_Ctrl::Coll_Ctrl (int _interactive, bool _defHWC, bool _kernelHWC)
default_stem = strdup ("test");
/* get CPU count and processor clock rate */
+ #ifndef _SC_CPUID_MAX
+ ncpus = sysconf (_SC_NPROCESSORS_CONF);
+ /* add 2048 to count, since on some systems CPUID does not start at zero */
+ ncpumax = ncpus + 2048;
+ #elif
ncpumax = sysconf (_SC_CPUID_MAX);
- if (ncpumax == -1)
- {
- ncpus = sysconf (_SC_NPROCESSORS_CONF);
- /* add 2048 to count, since on some systems CPUID does not start at zero */
- ncpumax = ncpus + 2048;
- }
- ncpus = 0;
- cpu_clk_freq = 0;
+ #endif
+
+ //ncpus = 0;
+ //cpu_clk_freq = 0;
// On Linux, read /proc/cpuinfo to get CPU count and clock rate
// Note that parsing is different on SPARC and x86
@@ -115,6 +116,9 @@ Coll_Ctrl::Coll_Ctrl (int _interactive, bool _defHWC, bool _kernelHWC)
#elif defined(__aarch64__)
asm volatile("mrs %0, cntfrq_el0" : "=r" (cpu_clk_freq));
+#elif defined(__riscv)
+ cpu_clk_freq = 1000;
+
#else
FILE *procf = fopen ("/proc/cpuinfo", "r");
if (procf != NULL)
diff --git a/gprofng/src/dbe_types.h b/gprofng/src/dbe_types.h
index 9fa842ce85d..c67d421566e 100644
--- a/gprofng/src/dbe_types.h
+++ b/gprofng/src/dbe_types.h
@@ -42,7 +42,8 @@ enum Platform_t
Sparcv8plus,
Java,
Amd64,
- Aarch64
+ Aarch64,
+ RISCV
};
enum WSize_t
diff --git a/gprofng/testsuite/gprofng.display/mttest/gethrtime.c b/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
index da39821027f..5e2cc3dc97b 100644
--- a/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
+++ b/gprofng/testsuite/gprofng.display/mttest/gethrtime.c
@@ -32,6 +32,8 @@
#define SPARC 1
#elif defined(__aarch64__)
#define Aarch64 1
+#elif defined(__riscv)
+#define RISCV 1
#else
#define Intel 1
#endif
--
2.44.0
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2024-07-04 9:22 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-06-30 9:37 [PATCH] [RFC]RISC-V:[gprofng] Mimal support gprofng for riscv XYenChi
2024-07-01 19:39 ` Vladimir Mezentsev
2024-07-03 12:38 ` Yixuan Chen
2024-07-03 20:10 ` Vladimir Mezentsev
2024-07-04 9:21 ` Yixuan Chen
-- strict thread matches above, loose matches on Subject: below --
2024-07-01 3:11 chenyixuan
2024-06-28 10:28 XYenChi
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).