selftests: Fixes for 32bit arch
Subsystems (3)
v2 cover selftests: Fixes for 32bit arch
Hi,
Here are the 2nd version of kselftest fixes some on 32bit arch (e.g. arm). In this version, I updated [1/5] to make va_max 1GB instead of 3GB, according to Alexey's comment.
When I built the ksefltest on arm, I hit some 32bit related warnings. Here are the patches to fix those issues.
- [1/5] va_max was set 2^32 even on 32bit arch. This can make
va_max == 0 and always fail. Make it 3GB on 32bit.
- [2/5] Some VM tests requires 64bit user space, which should not run on 32bit arch.
- [3/5] For counting the size of large file, we should use size_t instead of unsinged long.
- [4/5] Gcc warns printf format for size_t and int64_t on 32bit arch. Use %llu and cast it.
- [5/5] Gcc warns __u64 and pointer type castings. It should once translated to unsigned long.
Thank you,
---
Masami Hiramatsu (5):
selftests: proc: Make va_max 1GB on 32bit arch
selftests: vm: Build/Run 64bit tests only on 64bit arch
selftests: net: Use size_t and ssize_t for counting file size
selftests: net: Fix printf format warnings on arm
selftests: sync: Fix cast warnings on arm
tools/testing/selftests/net/so_txtime.c | 4 ++-- tools/testing/selftests/net/tcp_mmap.c | 8 ++++---- tools/testing/selftests/net/udpgso.c | 3 ++- tools/testing/selftests/net/udpgso_bench_tx.c | 3 ++- .../selftests/proc/proc-self-map-files-002.c | 11 ++++++++++- tools/testing/selftests/sync/sync.c | 6 +++--- tools/testing/selftests/vm/Makefile | 5 +++++ tools/testing/selftests/vm/run_vmtests | 10 ++++++++++ 8 files changed, 38 insertions(+), 12 deletions(-)
--
Masami Hiramatsu (Linaro) <[email protected]>
PATCH v2 1/5 selftests: proc: Make va_max 1GB on 32bit arch
Currently proc-self-map-files-002.c sets va_max (max test address of user virtual address) to 4GB, but it is too big for 32bit arch and 1UL << 32 is overflow on 32bit long.
Make va_max 1GB on 32bit arch like i386 and arm.
---
Changes in v2:
- Make the va_max 1GB according to Alexey's comment.
--- .../selftests/proc/proc-self-map-files-002.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)
tools/testing/selftests/proc/proc-self-map-files-002.c +10 -1
@@ -22,6 +22,7 @@#include <unistd.h> #include <sys/mman.h> #include <stdlib.h> +#include <asm/bitsperlong.h> static void pass (const char *fmt,unsigned long a,unsigned long b) {
@@ -44,10 +45,18 @@ static void fail(const char *fmt, unsigned long a, unsigned long b)exit (1 ); } +#if __BITS_PER_LONG ==32 +# define VA_MAX (1UL << 30) +#elif __BITS_PER_LONG ==64 +# define VA_MAX (1UL << 32) +#else +# define VA_MAX 0 +#endif +int main (void ) {const int PAGE_SIZE =sysconf (_SC_PAGESIZE); -const unsigned long va_max =1UL <<32 ; +const unsigned long va_max =VA_MAX ;unsigned long va;void *p;int fd;
PATCH v2 2/5 selftests: vm: Build/Run 64bit tests only on 64bit arch
In reply to Masami HiramatsuSome virtual address range tests requires 64bit address space, and we can not build and run those tests on the 32bit machine.
Filter the 64bit architectures in Makefile and run_vmtests, so that those tests are built/run only on 64bit archs.
--- tools/testing/selftests/vm/Makefile | 5 +++++ tools/testing/selftests/vm/run_vmtests | 10 ++++++++++ 2 files changed, 15 insertions(+)
tools/testing/selftests/vm/Makefile +5 -0
@@ -1,5 +1,7 @@ # SPDX-License-Identifier: GPL-2.0 # Makefile for vm selftests +uname_M := $(shell uname -m 2>/dev/null || echo not) +ARCH ?= $(shell echo $(uname_M) | sed -e 's/aarch64.*/arm64/') CFLAGS = -Wall -I ../../../../usr/include $(EXTRA_CFLAGS) LDLIBS = -lrt
@@ -16,8 +18,11 @@ TEST_GEN_FILES += on-fault-limit TEST_GEN_FILES += thuge-gen TEST_GEN_FILES += transhuge-stress TEST_GEN_FILES += userfaultfd + +ifneq (,$(filter $(ARCH),arm64 ia64 mips64 parisc64 ppc64 riscv64 s390x sh64 sparc64 x86_64)) TEST_GEN_FILES += va_128TBswitch TEST_GEN_FILES += virtual_address_range +endif TEST_PROGS := run_vmtests
tools/testing/selftests/vm/run_vmtests +10 -0
@@ -58,6 +58,14 @@ else exit 1 fi +#filter 64bit architectures +ARCH64STR="arm64 ia64 mips64 parisc64 ppc64 riscv64 s390x sh64 sparc64 x86_64" +if [ -z $ARCH ]; then + ARCH=`uname -m 2>/dev/null | sed -e 's/aarch64.*/arm64/'` +fi +VADDR64=0 +echo "$ARCH64STR" | grep $ARCH && VADDR64=1 + mkdir $mnt mount -t hugetlbfs none $mnt
@@ -189,6 +197,7 @@ else echo "[PASS]" fi +if [ $VADDR64 -ne 0 ]; then echo "-----------------------------" echo "running virtual_address_range" echo "-----------------------------"
@@ -210,6 +219,7 @@ if [ $? -ne 0 ]; then else echo "[PASS]" fi +fi # VADDR64 echo "------------------------------------" echo "running vmalloc stability smoke test"
PATCH v2 3/5 selftests: net: Use size_t and ssize_t for counting file size
In reply to Masami HiramatsuUse size_t and ssize_t correctly for counting send file size instead of unsigned long and long, because long is 32bit on 32bit arch, which is not enough for counting long file size (>4GB).
--- tools/testing/selftests/net/tcp_mmap.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
tools/testing/selftests/net/tcp_mmap.c +4 -4
@@ -71,7 +71,7 @@#define MSG_ZEROCOPY 0x4000000#endif -#define FILE_SZ (1UL << 35) +#define FILE_SZ (1ULL << 35)static int cfg_family =AF_INET6 ;static socklen_t cfg_alen =sizeof (struct sockaddr_in6 );static int cfg_port =8787 ;
@@ -155,7 +155,7 @@ void *child_thread(void *arg)socklen_t zc_len =sizeof (zc);int res; - zc.address = (__u64 )addr; + zc.address = (__u64)((unsigned long )addr); zc.length = chunk_size; zc.recv_skip_hint =0 ; res = getsockopt(fd, IPPROTO_TCP, TCP_ZEROCOPY_RECEIVE,
@@ -302,7 +302,7 @@ int main(int argc, char *argv[]) {struct sockaddr_storage listenaddr, addr;unsigned int max_pacing_rate =0 ; -unsigned long total =0 ; +size_t total =0 ;char *host =NULL ;int fd, c, on =1 ;char *buffer;
@@ -417,7 +417,7 @@ int main(int argc, char *argv[]) zflg =0 ; }while (total <FILE_SZ ) { -long wr =FILE_SZ - total; +ssize_t wr =FILE_SZ - total;if (wr > chunk_size) wr = chunk_size;
PATCH v2 4/5 selftests: net: Fix printf format warnings on arm
In reply to Masami HiramatsuFix printf format warnings on arm (and other 32bit arch).
- udpgso.c and udpgso_bench_tx use %lu for size_t but it should be unsigned long long on 32bit arch.
- so_txtime.c uses %ld for int64_t, but it should be unsigned long long on 32bit arch.
--- tools/testing/selftests/net/so_txtime.c | 4 ++-- tools/testing/selftests/net/udpgso.c | 3 ++- tools/testing/selftests/net/udpgso_bench_tx.c | 3 ++- 3 files changed, 6 insertions(+), 4 deletions(-)
tools/testing/selftests/net/so_txtime.c +2 -2
@@ -105,8 +105,8 @@ static void do_recv_one(int fdr, struct timed_send *ts) tstop = (gettime_ns () - glob_tstart) /1000 ; texpect = ts->delay_us >=0 ? ts->delay_us :0 ; -fprintf (stderr,"payload:%c delay:%ld expected:%ld (us)\n" , - rbuf[0 ], tstop, texpect); +fprintf (stderr,"payload:%c delay:%lld expected:%lld (us)\n" , + rbuf[0 ], (long long )tstop, (long long )texpect);if (rbuf[0 ] != ts->data)error (1 ,0 ,"payload mismatch. expected %c" , ts->data);
tools/testing/selftests/net/udpgso.c +2 -1
@@ -448,7 +448,8 @@ static bool __send_one(int fd, struct msghdr *msg, int flags)if (ret ==-1 )error (1 , errno,"sendmsg" );if (ret != msg->msg_iov->iov_len) -error (1 ,0 ,"sendto: %d != %lu" , ret, msg->msg_iov->iov_len); +error (1 ,0 ,"sendto: %d != %llu" , ret, + (unsigned long long )msg->msg_iov->iov_len);if (msg->msg_flags)error (1 ,0 ,"sendmsg: return flags 0x%x\n" , msg->msg_flags);
tools/testing/selftests/net/udpgso_bench_tx.c +2 -1
@@ -405,7 +405,8 @@ static int send_udp_segment(int fd, char *data)if (ret ==-1 )error (1 , errno,"sendmsg" );if (ret != iov.iov_len) -error (1 ,0 ,"sendmsg: %u != %lu\n" , ret, iov.iov_len); +error (1 ,0 ,"sendmsg: %u != %llu\n" , ret, + (unsigned long long )iov.iov_len);return 1 ; }
PATCH v2 5/5 selftests: sync: Fix cast warnings on arm
In reply to Masami HiramatsuFix warnings on __u64 and pointer translation on arm and other 32bit architectures. Since the pointer is 32bits on those archs, we should not directly cast those types.
--- tools/testing/selftests/sync/sync.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
tools/testing/selftests/sync/sync.c +3 -3
@@ -109,7 +109,7 @@ static struct sync_file_info *sync_file_info(int fd)return NULL ; } - info->sync_fence_info = (uint64_t )fence_info; + info->sync_fence_info = (uint64_t )(unsigned long )fence_info; err =ioctl (fd,SYNC_IOC_FILE_INFO , info);if (err <0 ) {
@@ -124,7 +124,7 @@ static struct sync_file_info *sync_file_info(int fd)static void sync_file_info_free (struct sync_file_info *info) { -free ((void *)info->sync_fence_info); +free ((void *)(unsigned long )info->sync_fence_info);free (info); }
@@ -152,7 +152,7 @@ int sync_fence_count_with_status(int fd, int status)if (!info)return -1 ; - fence_info = (struct sync_fence_info *)info->sync_fence_info; + fence_info = (struct sync_fence_info *)(unsigned long )info->sync_fence_info;for (i =0 ; i < info->num_fences ; i++) {if (fence_info[i].status == status) count++;
Re: [BUGFIX PATCH v2 1/5] selftests: proc: Make va_max 1GB on 32bit arch
In reply to Masami HiramatsuQuoted 9 lines · Masami Hiramatsu
Currently
proc-self-map-files-002.csetsva_max(max test address of user virtual address) to 4GB, but it is too big for 32bit arch and 1UL << 32 is overflow on 32bit long.Make
va_max1GB on 32bit arch like i386 and arm.tools/testing/selftests/proc/proc-self-map-files-002.c +3
-0+#if __BITS_PER_LONG ==32 +# define VA_MAX (1UL << 30) +#elif __BITS_PER_LONG ==64
tools/testing/selftests/proc/proc-self-map-files-002.c static void fail(const char *fmt, unsigned long a, unsigned long b) · line 59
Quoted 10 lines
tools/testing/selftests/proc/proc-self-map-files-002.c +6
-1+# define VA_MAX (1UL << 32) +#else +# define VA_MAX 0 +#endif +int main (void ) {const int PAGE_SIZE =sysconf (_SC_PAGESIZE); -const unsigned long va_max =1UL <<32 ; +const unsigned long va_max =VA_MAX ;
No, just make it like 1MB unconditionally.
This is not intended to cover all address space, just large enough part (larger than reasonable vm.mmap_min_addr)
Re: [BUGFIX PATCH v2 1/5] selftests: proc: Make va_max 1GB on 32bit arch
Quoted 7 lines · On Mon, 21 Oct 2019 20:30:53 +0300 Alexey Dobriyan <[email protected]> wrote:
Quoted 5 lines · Masami Hiramatsu
Currently
proc-self-map-files-002.csetsva_max(max test address of user virtual address) to 4GB, but it is too big for 32bit arch and 1UL << 32 is overflow on 32bit long.Make
va_max1GB on 32bit arch like i386 and arm.
Quoted 15 lines
tools/testing/selftests/proc/proc-self-map-files-002.c static void fail(const char *fmt, unsigned long a, unsigned long b) · line 59
Quoted 13 lines
tools/testing/selftests/proc/proc-self-map-files-002.c +9
-1+#if __BITS_PER_LONG ==32 +# define VA_MAX (1UL << 30) +#elif __BITS_PER_LONG ==64 +# define VA_MAX (1UL << 32) +#else +# define VA_MAX 0 +#endif +int main (void ) {const int PAGE_SIZE =sysconf (_SC_PAGESIZE); -const unsigned long va_max =1UL <<32 ; +const unsigned long va_max =VA_MAX ;No, just make it like 1MB unconditionally.
Ah, I sse. BTW, would you mean 1GB?
Quoted 2 lines
This is not intended to cover all address space, just large enough part (larger than reasonable vm.mmap_min_addr)
Then, should we better to check the /proc/sys/vm/mmap_min_addr?
Thank you,
Signature · 1 line
Masami Hiramatsu <[email protected]>
Re: [BUGFIX PATCH v2 1/5] selftests: proc: Make va_max 1GB on 32bit arch
Quoted 26 lines · On Wed, 23 Oct 2019 10:56:18 +0900 Masami Hiramatsu <[email protected]> wrote:
Quoted 7 lines · On Mon, 21 Oct 2019 20:30:53 +0300 Alexey Dobriyan <[email protected]> wrote:
Quoted 5 lines · Masami Hiramatsu
Currently
proc-self-map-files-002.csetsva_max(max test address of user virtual address) to 4GB, but it is too big for 32bit arch and 1UL << 32 is overflow on 32bit long.Make
va_max1GB on 32bit arch like i386 and arm.Quoted 15 lines
tools/testing/selftests/proc/proc-self-map-files-002.c static void fail(const char *fmt, unsigned long a, unsigned long b) · line 59
Quoted 13 lines
tools/testing/selftests/proc/proc-self-map-files-002.c +9
-1+#if __BITS_PER_LONG ==32 +# define VA_MAX (1UL << 30) +#elif __BITS_PER_LONG ==64 +# define VA_MAX (1UL << 32) +#else +# define VA_MAX 0 +#endif +int main (void ) {const int PAGE_SIZE =sysconf (_SC_PAGESIZE); -const unsigned long va_max =1UL <<32 ; +const unsigned long va_max =VA_MAX ;No, just make it like 1MB unconditionally.
Quoted 1 line
Ah, I sse. BTW, would you mean 1GB?
I understand that 1MB will be good enough, since vm.mmap_min_addr is 64KB by default (except for arm/arm64 which is 32KB).
OK, I'll update and resend.
Thank you,
Quoted 10 lines
Quoted 3 lines
This is not intended to cover all address space, just large enough part (larger than reasonable vm.mmap_min_addr)
Then, should we better to check the
/proc/sys/vm/mmap_min_addr?Thank you,
Signature · 1 line
Masami Hiramatsu <[email protected]>
Signature · 1 line
Masami Hiramatsu <[email protected]>