lkl: add make_syscall_tbl.sh to create syscall definition

arch/lkl/Makefile

@@ -22,15 +22,22 @@ LDFLAGS_vmlinux += -r
 LKL_ENTRY_POINTS := lkl_start_kernel lkl_sys_halt lkl_syscall lkl_trigger_irq \
 	lkl_get_free_irq lkl_put_irq
 
+SYSCALL_TBL_HDR = arch/lkl/include/generated/asm/syscall_table.h
+UNISTD_HDR = arch/lkl/include/asm/unistd.h
+
 core-y += arch/lkl/kernel/
 
 all: lkl.o
 
 lkl.o: vmlinux
-	$(OBJCOPY) -R .eh_frame -R .syscall_defs $(foreach sym,$(LKL_ENTRY_POINTS),-G$(prefix)$(sym)) vmlinux lkl.o
+	$(OBJCOPY) -R .eh_frame $(foreach sym,$(LKL_ENTRY_POINTS),-G$(prefix)$(sym)) vmlinux lkl.o
+
+quiet_cmd_syscall_tbl = TBL     $<
+      cmd_syscall_tbl = $< systbl $(UNISTD_HDR) $@
 
-arch/lkl/include/generated/uapi/asm/syscall_defs.h: vmlinux
-	$(OBJCOPY) -j .syscall_defs -O binary --set-section-flags .syscall_defs=alloc $< $@
+$(SYSCALL_TBL_HDR): arch/lkl/scripts/make_syscall_tbl.sh $(UNISTD_HDR)
+	@mkdir -p arch/lkl/include/generated/asm/
+	$(call cmd,syscall_tbl)
 
 install: lkl.o __headers arch/lkl/include/generated/uapi/asm/syscall_defs.h
 	@echo "  INSTALL	$(INSTALL_PATH)/lib/lkl.o"
@@ -39,6 +46,8 @@ install: lkl.o __headers arch/lkl/include/generated/uapi/asm/syscall_defs.h
 		$(subst -j,-j$(NPROC),$(findstring -j,$(MAKEFLAGS))) \
 		$(INSTALL_PATH)/include
 
+archheaders: $(SYSCALL_TBL_HDR)
+
 archclean:
 	$(Q)$(MAKE) $(clean)=$(boot)
 

arch/lkl/include/asm/unistd.h

@@ -24,3 +24,83 @@
 	    __ASCII_MAP(x, __SC_ASCII, __VA_ARGS__) ")\\n\"\n"		\
 	    ".ascii \"#endif\\n\"\n"					\
 	    ".section .text\n");
+
+/*
+ * Unsupported system calls due to lack of support in LKL (e.g. related to
+ * virtual memory, signal, user processes). We also only support 64bit version
+ * of system calls where we have two version to keep the same API across 32 and
+ * 64 bit hosts.
+ */
+
+#define __IGNORE_restart_syscall
+#define __IGNORE_exit
+#define __IGNORE_fork
+#define __IGNORE_execve
+#define __IGNORE_ptrace
+#define __IGNORE_alarm
+#define __IGNORE_pause
+#define __IGNORE_kill
+#define __IGNORE_brk
+#define __IGNORE_uselib
+#define __IGNORE_swapon
+#define __IGNORE_mmap
+#define __IGNORE_munmap
+#define __IGNORE_swapoff
+#define __IGNORE_clone
+#define __IGNORE_mprotect
+#define __IGNORE_init_module
+#define __IGNORE_quotactl
+#define __IGNORE_msync
+#define __IGNORE_mlock
+#define __IGNORE_munlock
+#define __IGNORE_mlockall
+#define __IGNORE_munlockall
+#define __IGNORE_mremap
+#define __IGNORE_rt_sigreturn
+#define __IGNORE_rt_sigaction
+#define __IGNORE_rt_sigprocmask
+#define __IGNORE_rt_sigpending
+#define __IGNORE_rt_sigtimedwait
+#define __IGNORE_rt_sigqueueinfo
+#define __IGNORE_rt_sigsuspend
+#define __IGNORE_sigaltstack
+#define __IGNORE_vfork
+#define __IGNORE_mincore
+#define __IGNORE_madvise
+#define __IGNORE_getdents /* we use the 64 bit counter part instead */
+#define __IGNORE_tkill
+#define __IGNORE_exit_group
+#define __IGNORE_remap_file_pages
+#define __IGNORE_fstatfs /* we use the 64 bit counter part instead */
+#define __IGNORE_mbind
+#define __IGNORE_get_mempolicy
+#define __IGNORE_set_mempolicy
+#define __IGNORE_mq_open
+#define __IGNORE_mq_unlink
+#define __IGNORE_mq_timedsend
+#define __IGNORE_mq_timedreceive
+#define __IGNORE_mq_getsetattr
+#define __IGNORE_kexec_load
+#define __IGNORE_migrate_pages
+#define __IGNORE_unshare
+#define __IGNORE_set_robust_list
+#define __IGNORE_get_robust_list
+#define __IGNORE_sync_file_range
+#define __IGNORE_vmsplice
+#define __IGNORE_move_pages
+#define __IGNORE_mq_notify
+#define __IGNORE_delete_module
+#define __IGNORE_signalfd4
+#define __IGNORE_preadv /* we use the 64 bit counter part instead */
+#define __IGNORE_pwritev /* we use the 64 bit counter part instead */
+#define __IGNORE_rt_tgsigqueueinfo
+#define __IGNORE_perf_event_open
+#define __IGNORE_setns
+#define __IGNORE_process_vm_readv
+#define __IGNORE_process_vm_writev
+#define __IGNORE_kcmp
+#define __IGNORE_finit_module
+#define __IGNORE_seccomp
+#define __IGNORE_memfd_create
+#define __IGNORE_bpf
+#define __IGNORE_execveat

arch/lkl/kernel/Makefile

@@ -1,3 +1,13 @@
-extra-y := vmlinux.lds
+SYSCALL_DEFS_HDR := ../include/generated/uapi/asm/syscall_defs.h
+UNISTD_HDR = arch/lkl/include/asm/unistd.h
 
+extra-y := vmlinux.lds $(SYSCALL_DEFS_HDR)
 obj-y = setup.o threads.o irq.o time.o syscalls.o misc.o mem.o console.o
+
+
+quiet_cmd_syscall_def = CALL    $<
+      cmd_syscall_def = $< sysdef $(UNISTD_HDR) arch/lkl/include/generated/uapi/asm/syscall_defs.h
+
+arch/lkl/kernel/$(SYSCALL_DEFS_HDR): arch/lkl/scripts/make_syscall_tbl.sh $(UNISTD_HDR)
+	@mkdir -p arch/lkl/include/generated/uapi/asm/
+	$(call cmd,syscall_def)

arch/lkl/kernel/syscalls.c

@@ -19,6 +19,8 @@ typedef long (*syscall_handler_t)(long arg1, ...);
 syscall_handler_t syscall_table[__NR_syscalls] = {
 	[0 ... __NR_syscalls - 1] =  (syscall_handler_t)sys_ni_syscall,
 #include <asm/unistd.h>
+	/* override __IGNORE-ed syscalls */
+#include <asm/syscall_table.h>
 };
 
 static struct syscall_queue {

arch/lkl/scripts/make_syscall_tbl.sh

@@ -0,0 +1,123 @@
+#!/bin/bash
+
+#set -x
+
+# requires: bash-4.0 (hashtable), cpp, grep, sed, cat
+
+##
+## Usage: $0
+## ARGV[0]: (OPT: "systbl" oe "sysdef")
+## ARGV[1]: (IN: asm/unistd.h)
+## ARGV[2]: (OUT: syscall_table.h(systbl) or syscall_defs.h (sysdef))
+##
+
+mode=$1
+in=$2
+out=$3
+
+if [ "$mode" != "systbl" ] && [ "$mode" != "sysdef" ] ; then
+	echo "Usage: $0 [systbl|sysdef] (unistd.h) (output file)"
+	exit
+fi
+
+echo "" > $out
+
+# obtain the blacklist of syscall
+declare -A ignore_syscall
+
+##
+## step 1: create ignored syscall table entries based on
+##
+while read def nr ; do
+	name=$(echo $nr | sed "s/__IGNORE_/__NR_/")
+	ignore_syscall[$name]="ignored"
+
+if [ $mode = "systbl" ] ; then
+	cat << EOF  >> $out
+#define __SYSCALL_STE_${name}(nr,sym) [nr] = (syscall_handler_t)sys_ni_syscall,
+__SYSCALL_STE_${name}(${name},dummy)
+EOF
+fi
+done < <(${CROSS_COMPILE}cpp ${LINUXINCLUDE} -dM -fdirectives-only $in | grep '__IGNORE_')
+
+##
+## stop if systbl is specified
+##
+if [ $mode = "systbl" ] ; then
+	exit
+fi
+
+
+##
+## step 2-1: grep SYSCALL_DEFINE in whole tree with specific
+## directories ($SYSCALL_DIRS)
+##
+SYSCALL_DIRS="block drivers/pci drivers/char fs init ipc kernel mm net"
+SYSCALL_FILES=$(find ${SYSCALL_DIRS}  -name \*.c | \
+		       xargs grep -E "^SYSCALL_DEFINE[0-6]" -l)
+
+##
+## step 2-2: dump lkl syscall definitions from preprocessor's ouptut
+##
+set -f
+for file in ${SYSCALL_FILES} ;
+do
+	while read syscall; do
+		func=$(echo $syscall | sed "s/SYSCALL_DEFINE.*(\([^,]*\).*/\1/")
+		#echo "syscall= "$syscall
+		#echo "func= "$func
+
+		# 0) pull the system call name
+		funcname=$(echo $func | sed "s/(.*//" | sed "s/long sys_//")
+		#echo "funcname ="$funcname
+
+		# 1) skip if the syscall is unsupported
+		if [ "${ignore_syscall[__NR_$funcname]}" = "ignored" ] ; then
+		#echo $funcname " ignore= " "${ignore_syscall[__NR_$funcname]}"
+			continue
+		fi
+
+		# 2) count the number of argument
+		func_l=/$func/
+		IFS=','
+		set -- $func_l
+		argc=0
+		if [ $(echo $func | sed "s/.*(//") != "void)" ] ; then
+			argc=`expr $(($#-1)) + 1`
+		fi
+		#echo "argc = $argc"
+		IFS=' '
+
+		# 3) pull arg name template
+		args=$(echo $func | sed "s/.*(/(/" | sed "s/__lkl__kernel_//g"\
+			      | sed "s/lkl_//g" | sed "s/[()]//g")
+		OIFS="$IFS"; IFS=','
+		arg=($args); IFS="$OIFS"
+		#echo "arg ="$arg > /dev/stdout
+
+		param=""
+		for i in {1..6} ; do
+			if [ $i -gt ${argc} ] ; then
+				break
+			fi
+			if [ $i -ne 1 ] ; then
+				param="${param}, "
+			fi
+			nparam=$(echo ${arg[$((i-1))]} | \
+					sed "s/\(.*\) \(.*\)$/\1, \2/")
+			param="${param}${nparam}"
+		done
+		#echo ${param} > /dev/stdout
+		IFS=' '
+
+		# 4) give a template based on funcname, args
+		cat << EOF >> $out
+
+#ifdef __NR_$funcname
+SYSCALL_DEFINE$argc(_$funcname, $param);
+#endif /* __NR_$funcname */
+EOF
+	done < <(${CROSS_COMPILE}cpp -P ${LINUXINCLUDE} $file 2>/dev/null \
+			| grep "long sys_" | sed "s/^; / /" \
+			| sed "s/\(.*\) __attribute.*/\1/" | grep -vE ";|,$")
+done

include/linux/syscalls.h

@@ -181,7 +181,6 @@ extern struct trace_event_functions exit_syscall_print_funcs;
 
 #define SYSCALL_DEFINE0(sname)					\
 	SYSCALL_METADATA(_##sname, 0);				\
-	__SYSCALL_DEFINE_ARCH(0, _##sname);			\
 	asmlinkage long sys_##sname(void)
 
 #define SYSCALL_DEFINE1(name, ...) SYSCALL_DEFINEx(1, _##name, __VA_ARGS__)
@@ -193,7 +192,6 @@ extern struct trace_event_functions exit_syscall_print_funcs;
 
 #define SYSCALL_DEFINEx(x, sname, ...)				\
 	SYSCALL_METADATA(sname, x, __VA_ARGS__)			\
-	__SYSCALL_DEFINE_ARCH(x, sname, __VA_ARGS__)		\
 	__SYSCALL_DEFINEx(x, sname, __VA_ARGS__)
 
 #define __PROTECT(...) asmlinkage_protect(__VA_ARGS__)

tools/lkl/tests/boot.c

@@ -420,6 +420,19 @@ static int test_umount(char *str, int len)
 	return 0;
 }
 
+static int test_syscall(char *str, int len)
+{
+	long ret, params[6];
+
+	ret = lkl_syscall(__lkl__NR_fork, params);
+
+	snprintf(str, len, "%ld", ret);
+	if (ret == -LKL_ENOSYS)
+		return 1;
+
+	return 0;
+}
+
 static struct cl_option *find_short_opt(char name)
 {
 	struct cl_option *opt;
@@ -516,6 +529,7 @@ int main(int argc, char **argv)
 	TEST(opendir);
 	TEST(getdents64);
 	TEST(umount);
+	TEST(syscall);
 
 	lkl_sys_halt();