Skip to contentSkip to conversation

signal handling support for nolibc

PATCH Superseded Benjamin Berg 3 participants 13 messages Jul 10, 2025 – Jul 14, 2025

Subsystems (3)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

v2 cover signal handling support for nolibc

From: Benjamin Berg <[email protected]>

Hi,

This patchset adds signal handling to nolibc. Initially, I would like to use this for tests. But in the long run, the goal is to use nolibc for the UML kernel itself. In both cases, signal handling will be needed.

v2 contains some bugfixes and has a better test coverage. Also addressed are various review comments.

Benjamin

Benjamin Berg (4):
selftests/nolibc: fix EXPECT_NZ macro selftests/nolibc: validate order of constructor calls tools/nolibc: add more generic bitmask macros for FD_* tools/nolibc: add signal support

 tools/include/nolibc/arch-arm.h               |   7 +
 tools/include/nolibc/arch-arm64.h             |   3 +
 tools/include/nolibc/arch-loongarch.h         |   3 +
 tools/include/nolibc/arch-m68k.h              |  10 ++
 tools/include/nolibc/arch-mips.h              |   3 +
 tools/include/nolibc/arch-powerpc.h           |   8 +
 tools/include/nolibc/arch-riscv.h             |   3 +
 tools/include/nolibc/arch-s390.h              |   8 +-
 tools/include/nolibc/arch-sh.h                |   5 +
 tools/include/nolibc/arch-sparc.h             |  47 ++++++
 tools/include/nolibc/arch-x86.h               |  13 ++
 tools/include/nolibc/signal.h                 | 103 ++++++++++++
 tools/include/nolibc/sys.h                    |   2 +-
 tools/include/nolibc/time.h                   |   3 +-
 tools/include/nolibc/types.h                  |  81 ++++++----
 .../selftests/nolibc/nolibc-test-linkage.c    |  17 +-
 tools/testing/selftests/nolibc/nolibc-test.c  | 150 +++++++++++++++++-
 17 files changed, 422 insertions(+), 44 deletions(-)
Signature · 1 line
2.50.0
#1 lore raw

PATCH v2 1/4 selftests/nolibc: fix EXPECT_NZ macro

From: Benjamin Berg <[email protected]>

The expect non-zero macro was incorrect and never used. Fix its definition.

---
 tools/testing/selftests/nolibc/nolibc-test.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/testing/selftests/nolibc/nolibc-test.c +2 -2
index a297ee0d6d07..97efc98b6a3d 100644
@@ -196,8 +196,8 @@ int expect_zr(int expr, int llen)
 }
 
 
-#define EXPECT_NZ(cond, expr, val)			\
-	do { if (!(cond)) result(llen, SKIPPED); else ret += expect_nz(expr, llen; } while (0)
+#define EXPECT_NZ(cond, expr)				\
+	do { if (!(cond)) result(llen, SKIPPED); else ret += expect_nz(expr, llen); } while (0)
 
 static __attribute__((unused))
 int expect_nz(int expr, int llen)
Signature · 1 line
2.50.0
#2 lore raw

Re: [PATCH v2 1/4] selftests/nolibc: fix EXPECT_NZ macro

Quoted 4 lines · Benjamin Berg

From: Benjamin Berg <[email protected]>

The expect non-zero macro was incorrect and never used. Fix its definition.

Missing sign-off.
A Fixes would also be nice.

(...)

#3 lore raw

PATCH v2 2/4 selftests/nolibc: validate order of constructor calls

In reply to Benjamin Berg

From: Benjamin Berg <[email protected]>

Add new helpers to track multiple steps as bits in an integer. Store each step in a bit and use the lowest bit to store whether all steps occurred in the correct order and only once.

Use this for the constructor tests.

---

v2:
- Newly added patch

---
 .../selftests/nolibc/nolibc-test-linkage.c    | 17 ++++++++--
 tools/testing/selftests/nolibc/nolibc-test.c  | 31 ++++++++++++++++---
 2 files changed, 42 insertions(+), 6 deletions(-)
tools/testing/selftests/nolibc/nolibc-test-linkage.c +15 -2
index 0636d1b6e808..4435f389570b 100644
@@ -4,6 +4,19 @@
 
 #include <errno.h>
 
+/*
+ * Set BIT(step + 1), BIT(0) shows whether all steps ran once and in order
+ *
+ * Copied from nolibc-test.c.
+ */
+#define MARK_STEP_DONE(val, step) do {					\
+	if ((val) == 0 && (step) == 0)					\
+		(val) |= 0x1;						\
+	else if (!(val & (1 << (step))) || (val) & (1 << ((step) + 1)))	\
+		(val) &= ~0x1;						\
+	(val) |= 1 << ((step) + 1);					\
+	} while (0)
+
 void *linkage_test_errno_addr(void)
 {
 	return &errno;
@@ -14,11 +27,11 @@ int linkage_test_constructor_test_value = 0;
 __attribute__((constructor))
 static void constructor1(void)
 {
-	linkage_test_constructor_test_value |= 1 << 0;
+	MARK_STEP_DONE(linkage_test_constructor_test_value, 0);
 }
 
 __attribute__((constructor))
 static void constructor2(void)
 {
-	linkage_test_constructor_test_value |= 1 << 1;
+	MARK_STEP_DONE(linkage_test_constructor_test_value, 1);
 }
tools/testing/selftests/nolibc/nolibc-test.c +27 -4
index 97efc98b6a3d..d612150d2ea3 100644
@@ -72,6 +72,15 @@ static const int is_nolibc =
 #endif
 ;
 
+/* Set BIT(step + 1), BIT(0) shows whether all steps ran once and in order */
+#define MARK_STEP_DONE(val, step) do {					\
+	if ((val) == 0 && (step) == 0)					\
+		(val) |= 0x1;						\
+	else if ((val) & (1 << ((step) + 1)) || !(val & (1 << (step))))	\
+		(val) &= ~0x1;						\
+	(val) |= 1 << ((step) + 1);					\
+	} while (0)
+
 /* definition of a series of tests */
 struct test {
 	const char *name;              /* test name */
@@ -389,6 +398,20 @@ int expect_syserr2(int expr, int expret, int experr1, int experr2, int llen)
 }
 
 
+#define EXPECT_STEPS(cond, expr, num_steps)			\
+	do { if (!(cond)) result(llen, SKIPPED); else ret += expect_steps(expr, llen, num_steps); } while (0)
+
+static __attribute__((unused))
+int expect_steps(uint64_t expr, int llen, int num_steps)
+{
+	int ret = !(expr == ((uint64_t)1 << (num_steps + 1)) - 1);
+
+	llen += printf(" = %llx ", (long long)expr);
+	result(llen, ret ? FAIL : OK);
+	return ret;
+}
+
+
 #define EXPECT_PTRZR(cond, expr)				\
 	do { if (!(cond)) result(llen, SKIPPED); else ret += expect_ptrzr(expr, llen); } while (0)
 
@@ -690,14 +713,14 @@ int expect_strtox(int llen, void *func, const char *input, int base, intmax_t ex
 __attribute__((constructor))
 static void constructor1(void)
 {
-	constructor_test_value |= 1 << 0;
+	MARK_STEP_DONE(constructor_test_value, 0);
 }
 
 __attribute__((constructor))
 static void constructor2(int argc, char **argv, char **envp)
 {
 	if (argc && argv && envp)
-		constructor_test_value |= 1 << 1;
+		MARK_STEP_DONE(constructor_test_value, 1);
 }
 
 int run_startup(int min, int max)
@@ -736,9 +759,9 @@ int run_startup(int min, int max)
 		CASE_TEST(environ_HOME);     EXPECT_PTRNZ(1, getenv("HOME")); break;
 		CASE_TEST(auxv_addr);        EXPECT_PTRGT(test_auxv != (void *)-1, test_auxv, brk); break;
 		CASE_TEST(auxv_AT_UID);      EXPECT_EQ(1, getauxval(AT_UID), getuid()); break;
-		CASE_TEST(constructor);      EXPECT_EQ(is_nolibc, constructor_test_value, 0x3); break;
+		CASE_TEST(constructor);      EXPECT_STEPS(is_nolibc, constructor_test_value, 2); break;
 		CASE_TEST(linkage_errno);    EXPECT_PTREQ(1, linkage_test_errno_addr(), &errno); break;
-		CASE_TEST(linkage_constr);   EXPECT_EQ(1, linkage_test_constructor_test_value, 0x3); break;
+		CASE_TEST(linkage_constr);   EXPECT_STEPS(1, linkage_test_constructor_test_value, 2); break;
 		case __LINE__:
 			return ret; /* must be last */
 		/* note: do not set any defaults so as to permit holes above */
Signature · 1 line
2.50.0
#4 lore raw

Re: [PATCH v2 2/4] selftests/nolibc: validate order of constructor calls

Quoted 7 lines · Benjamin Berg

From: Benjamin Berg <[email protected]>

Add new helpers to track multiple steps as bits in an integer. Store each step in a bit and use the lowest bit to store whether all steps occurred in the correct order and only once.

Use this for the constructor tests.

We actually stopped validating the constructor execution order in commit a782d45c867c ("selftests/nolibc: stop testing constructor order") Sorry for my misinformation.

I think we can keep the scheme for the signal tests. but I would prefer a function instead of a macro.

(...)

#5 lore raw

PATCH v2 3/4 tools/nolibc: add more generic bitmask macros for FD_*

In reply to Benjamin Berg

From: Benjamin Berg <[email protected]>

The FD_* macros are assuming a specific type for the bitmask. Add new macros that introspect the type of the passed variable in order to know the size of the bitmask. This way the same macros can be used for other purposes.

---

v2:
- Rename the macros for consistency and to mark them internal
- Fix shift to use the correct element type

---
 tools/include/nolibc/types.h | 72 ++++++++++++++++++++----------------
 1 file changed, 40 insertions(+), 32 deletions(-)
tools/include/nolibc/types.h +40 -32
index 16c6e9ec9451..f7f2ddf41e89 100644
@@ -115,48 +115,56 @@
 #define EXIT_SUCCESS 0
 #define EXIT_FAILURE 1
 
-#define FD_SETIDXMASK (8 * sizeof(unsigned long))
-#define FD_SETBITMASK (8 * sizeof(unsigned long)-1)
-
-/* for select() */
-typedef struct {
-	unsigned long fds[(FD_SETSIZE + FD_SETBITMASK) / FD_SETIDXMASK];
-} fd_set;
-
-#define FD_CLR(fd, set) do {						\
-		fd_set *__set = (set);					\
-		int __fd = (fd);					\
-		if (__fd >= 0)						\
-			__set->fds[__fd / FD_SETIDXMASK] &=		\
-				~(1U << (__fd & FD_SETBITMASK));	\
+#define __NOLIBC_BITMASK_CLEAR(num, set) do {				     \
+		__typeof__(set) *__set = &(set);			     \
+		int __num = (num);					     \
+		__typeof__(**__set) __bit = 1;				     \
+		if (__num >= 0 && __num < 8 * (ssize_t)sizeof(*__set))	     \
+			(*__set)[__num / (8 * sizeof(**__set))] &=	     \
+				~(__bit << (__num % (8 * sizeof(**__set)))); \
 	} while (0)
 
-#define FD_SET(fd, set) do {						\
-		fd_set *__set = (set);					\
-		int __fd = (fd);					\
-		if (__fd >= 0)						\
-			__set->fds[__fd / FD_SETIDXMASK] |=		\
-				1 << (__fd & FD_SETBITMASK);		\
+#define __NOLIBC_BITMASK_SET(num, set) do {				  \
+		__typeof__(set) *__set = &(set);			  \
+		int __num = (num);					  \
+		__typeof__(**__set) __bit = 1;				  \
+		if (__num >= 0 && __num < 8 * (ssize_t)sizeof(*__set))	  \
+			(*__set)[__num / (8 * sizeof(**__set))] |=	  \
+				__bit << (__num % (8 * sizeof(**__set))); \
 	} while (0)
 
-#define FD_ISSET(fd, set) ({						\
-			fd_set *__set = (set);				\
-			int __fd = (fd);				\
-		int __r = 0;						\
-		if (__fd >= 0)						\
-			__r = !!(__set->fds[__fd / FD_SETIDXMASK] &	\
-1U << (__fd & FD_SETBITMASK));						\
-		__r;							\
+#define __NOLIBC_BITMASK_TEST(num, set) ({				  \
+		__typeof__(set) *__set = &(set);			  \
+		int __num = (num);					  \
+		__typeof__(**__set) __r = 0;				  \
+		__typeof__(**__set) __bit = 1;				  \
+		if (__num >= 0 && __num < 8 * (ssize_t)sizeof(*__set))	  \
+			__r = (*__set)[__num / (8 * sizeof(**__set))] &	  \
+			      (__bit << (__num % (8 * sizeof(**__set)))); \
+		!!__r;							  \
 	})
 
-#define FD_ZERO(set) do {						\
-		fd_set *__set = (set);					\
+#define __NOLIBC_BITMASK_ZERO(set) do {					\
+		__typeof__(set) *__set = &(set);			\
 		int __idx;						\
-		int __size = (FD_SETSIZE+FD_SETBITMASK) / FD_SETIDXMASK;\
+		int __size = sizeof(*__set) / sizeof(**__set);		\
 		for (__idx = 0; __idx < __size; __idx++)		\
-			__set->fds[__idx] = 0;				\
+			(*__set)[__idx] = 0;				\
 	} while (0)
 
+#define FD_SETIDXMASK (8 * sizeof(unsigned long))
+#define FD_SETBITMASK (8 * sizeof(unsigned long)-1)
+
+/* for select() */
+typedef struct {
+	unsigned long fds[(FD_SETSIZE + FD_SETBITMASK) / FD_SETIDXMASK];
+} fd_set;
+
+#define FD_CLR(fd, set) __NOLIBC_BITMASK_CLEAR(fd, (set)->fds)
+#define FD_SET(fd, set) __NOLIBC_BITMASK_SET(fd, (set)->fds)
+#define FD_ISSET(fd, set) __NOLIBC_BITMASK_TEST(fd, (set)->fds)
+#define FD_ZERO(set) __NOLIBC_BITMASK_ZERO((set)->fds)
+
 /* for getdents64() */
 struct linux_dirent64 {
 	uint64_t       d_ino;
Signature · 1 line
2.50.0
#6 lore raw

PATCH v2 4/4 tools/nolibc: add signal support

In reply to Benjamin Berg

From: Benjamin Berg <[email protected]>

Add support for sigaction() and implement the normal sa_mask helpers.

On many architectures, linux/signal.h pulls in compatibility definitions for the old sigaction syscall instead of rt_sigaction. However, the kernel can be compiled without support for this compatibility syscall and it also results in sa_mask to be too small for realtime signals.

To work around this, the includes are handled separately for each architecture. This way either linux/signal.h or the asm-generic headers can be used to get the correct definition for the rt_sigaction syscall including sigset_t.

---

v2:
- Use newly added macros to check signal emission order
- Add tests for sigset handling
- Restore the default handler after signal test
- make signal_check variable static

v1:
- Update architecture support (adding sh)
- Move sparc sys_rt_sigaction logic into its header
- Add sig_atomic_t
- Use new BITSET_* macros
- Move test into syscall suite
- Various other small changes

---
 tools/include/nolibc/arch-arm.h              |   7 ++
 tools/include/nolibc/arch-arm64.h            |   3 +
 tools/include/nolibc/arch-loongarch.h        |   3 +
 tools/include/nolibc/arch-m68k.h             |  10 ++
 tools/include/nolibc/arch-mips.h             |   3 +
 tools/include/nolibc/arch-powerpc.h          |   8 ++
 tools/include/nolibc/arch-riscv.h            |   3 +
 tools/include/nolibc/arch-s390.h             |   8 +-
 tools/include/nolibc/arch-sh.h               |   5 +
 tools/include/nolibc/arch-sparc.h            |  47 ++++++++
 tools/include/nolibc/arch-x86.h              |  13 +++
 tools/include/nolibc/signal.h                | 103 +++++++++++++++++
 tools/include/nolibc/sys.h                   |   2 +-
 tools/include/nolibc/time.h                  |   3 +-
 tools/include/nolibc/types.h                 |   9 ++
 tools/testing/selftests/nolibc/nolibc-test.c | 115 +++++++++++++++++++
 16 files changed, 338 insertions(+), 4 deletions(-)
tools/include/nolibc/arch-arm.h +7 -0
index 1f66e7e5a444..1faf6c2dbeb8 100644
@@ -10,6 +10,13 @@
 #include "compiler.h"
 #include "crt.h"
 
+/* Needed to get the correct struct sigaction definition */
+#define SA_RESTORER	0x04000000
+
+/* Avoid linux/signal.h, it has an incorrect _NSIG and sigset_t */
+#include <asm-generic/signal.h>
+#include <asm-generic/siginfo.h>
+
 /* Syscalls for ARM in ARM or Thumb modes :
  *   - registers are 32-bit
  *   - stack is 8-byte aligned
tools/include/nolibc/arch-arm64.h +3 -0
index 02a3f74c8ec8..ad14fc0ae5cb 100644
@@ -10,6 +10,9 @@
 #include "compiler.h"
 #include "crt.h"
 
+/* Architecture has a usable linux/signal.h */
+#include <linux/signal.h>
+
 /* Syscalls for ARM64 :
  *   - registers are 64-bit
  *   - stack is 16-byte aligned
tools/include/nolibc/arch-loongarch.h +3 -0
index 5511705303ea..68d60d04ef59 100644
@@ -10,6 +10,9 @@
 #include "compiler.h"
 #include "crt.h"
 
+/* Architecture has a usable linux/signal.h */
+#include <linux/signal.h>
+
 /* Syscalls for LoongArch :
  *   - stack is 16-byte aligned
  *   - syscall number is passed in a7
tools/include/nolibc/arch-m68k.h +10 -0
index 6dac1845f298..981b4cc55a69 100644
@@ -13,6 +13,16 @@
 #include "compiler.h"
 #include "crt.h"
 
+/*
+ * Needed to get the correct struct sigaction definition. m68k does not use
+ * sa_restorer, but it is included in the structure.
+ */
+#define SA_RESTORER	0x04000000
+
+/* Avoid linux/signal.h, it has an incorrect _NSIG and sigset_t */
+#include <asm-generic/signal.h>
+#include <asm-generic/siginfo.h>
+
 #define _NOLIBC_SYSCALL_CLOBBERLIST "memory"
 
 #define my_syscall0(num)                                                      \
tools/include/nolibc/arch-mips.h +3 -0
index 0cbac63b249a..fb2f503f151f 100644
@@ -14,6 +14,9 @@
 #error Unsupported MIPS ABI
 #endif
 
+/* Architecture has a usable linux/signal.h */
+#include <linux/signal.h>
+
 /* Syscalls for MIPS ABI O32 :
  *   - WARNING! there's always a delayed slot!
  *   - WARNING again, the syntax is different, registers take a '$' and numbers
tools/include/nolibc/arch-powerpc.h +8 -0
index 204564bbcd32..c846a7ddcf3c 100644
@@ -10,6 +10,14 @@
 #include "compiler.h"
 #include "crt.h"
 
+/* Needed to get the correct struct sigaction definition */
+#define SA_RESTORER	0x04000000
+#define _NOLIBC_ARCH_NEEDS_SA_RESTORER
+
+/* Avoid linux/signal.h, it has an incorrect _NSIG and sigset_t */
+#include <asm-generic/signal.h>
+#include <asm-generic/siginfo.h>
+
 /* Syscalls for PowerPC :
  *   - stack is 16-byte aligned
  *   - syscall number is passed in r0
tools/include/nolibc/arch-riscv.h +3 -0
index 885383a86c38..709e6a262d9a 100644
@@ -10,6 +10,9 @@
 #include "compiler.h"
 #include "crt.h"
 
+/* Architecture has a usable linux/signal.h */
+#include <linux/signal.h>
+
 /* Syscalls for RISCV :
  *   - stack is 16-byte aligned
  *   - syscall number is passed in a7
tools/include/nolibc/arch-s390.h +7 -1
index df4c3cc713ac..0dccb6d1ad64 100644
@@ -5,13 +5,19 @@
 
 #ifndef _NOLIBC_ARCH_S390_H
 #define _NOLIBC_ARCH_S390_H
-#include <linux/signal.h>
 #include <linux/unistd.h>
 
 #include "compiler.h"
 #include "crt.h"
 #include "std.h"
 
+/* Needed to get the correct struct sigaction definition */
+#define SA_RESTORER	0x04000000
+
+/* Avoid linux/signal.h, it has an incorrect _NSIG and sigset_t */
+#include <asm-generic/signal.h>
+#include <asm-generic/siginfo.h>
+
 /* Syscalls for s390:
  *   - registers are 64-bit
  *   - syscall number is passed in r1
tools/include/nolibc/arch-sh.h +5 -0
index a96b8914607e..3378afc78e26 100644
@@ -7,9 +7,14 @@
 #ifndef _NOLIBC_ARCH_SH_H
 #define _NOLIBC_ARCH_SH_H
 
+#include <linux/unistd.h>
+
 #include "compiler.h"
 #include "crt.h"
 
+/* Architecture has a usable linux/signal.h */
+#include <linux/signal.h>
+
 /*
  * Syscalls for SuperH:
  *   - registers are 32bit wide
tools/include/nolibc/arch-sparc.h +47 -0

57 lines not on this page · raw message

tools/include/nolibc/arch-x86.h +13 -0
index d3efc0c3b8ad..1fe75203d834 100644
@@ -10,8 +10,21 @@
 #include "compiler.h"
 #include "crt.h"
 
+/* Needed to get the correct struct sigaction definition */
+#define SA_RESTORER	0x04000000
+
+/* Restorer must be set on x86 for both 32 and 64 bit */
+#define _NOLIBC_ARCH_NEEDS_SA_RESTORER
+
+/* Avoid linux/signal.h, it has an incorrect _NSIG and sigset_t */
+#include <asm-generic/signal.h>
+#include <asm-generic/siginfo.h>
+
 #if !defined(__x86_64__)
 
+/* On i386 we need to set SA_SIGINFO to use rt_sigreturn */
+#define _NOLIBC_ARCH_FORCE_SIG_FLAGS SA_SIGINFO
+
 /* Syscalls for i386 :
  *   - mostly similar to x86_64
  *   - registers are 32-bit
tools/include/nolibc/signal.h +103 -0

113 lines not on this page · raw message

tools/include/nolibc/sys.h +1 -1
index 295e71d34aba..73b935576561 100644
@@ -14,7 +14,6 @@
 
 /* system includes */
 #include <linux/unistd.h>
-#include <linux/signal.h>  /* for SIGCHLD */
 #include <linux/termios.h>
 #include <linux/mman.h>
 #include <linux/fs.h>
@@ -24,6 +23,7 @@
 #include <linux/fcntl.h> /* for O_* and AT_* */
 #include <linux/sched.h> /* for clone_args */
 #include <linux/stat.h>  /* for statx() */
+/* signal definitions are included by arch.h */
 
 #include "errno.h"
 #include "stdarg.h"
tools/include/nolibc/time.h +1 -2
index d02bc44d2643..103574f76515 100644
@@ -14,9 +14,8 @@
 #include "arch.h"
 #include "types.h"
 #include "sys.h"
-
-#include <linux/signal.h>
 #include <linux/time.h>
+/* signal definitions are included by arch.h */
 
 static __inline__
 void __nolibc_timespec_user_to_kernel(const struct timespec *ts, struct __kernel_timespec *kts)
tools/include/nolibc/types.h +9 -0
index f7f2ddf41e89..7e205386b72c 100644
@@ -152,6 +152,15 @@
 			(*__set)[__idx] = 0;				\
 	} while (0)
 
+#define __NOLIBC_BITMASK_FILL(set) do {					\
+		__typeof__(set) *__set = &(set);			\
+		int __idx;						\
+		int __size = sizeof(*__set) / sizeof(**__set);		\
+		__typeof__(**__set) __zero = 0;				\
+		for (__idx = 0; __idx < __size; __idx++)		\
+			(*__set)[__idx] = ~__zero;			\
+	} while (0)
+
 #define FD_SETIDXMASK (8 * sizeof(unsigned long))
 #define FD_SETBITMASK (8 * sizeof(unsigned long)-1)
 
tools/testing/selftests/nolibc/nolibc-test.c +115 -0

127 lines not on this page · raw message

Signature · 1 line
2.50.0
#7 lore raw

Re: [PATCH v2 4/4] tools/nolibc: add signal support

Quoted 9 lines · Benjamin Berg

From: Benjamin Berg <[email protected]>

Add support for sigaction() and implement the normal sa_mask helpers.

On many architectures, linux/signal.h pulls in compatibility definitions for the old sigaction syscall instead of rt_sigaction. However, the kernel can be compiled without support for this compatibility syscall and it also results in sa_mask to be too small for realtime signals.

Quoted 4 lines

To work around this, the includes are handled separately for each architecture. This way either linux/signal.h or the asm-generic headers can be used to get the correct definition for the rt_sigaction syscall including sigset_t.

I checked this against my WIP alpha support and there this scheme breaks. linux/signal.h provides the old compat types but the asm-generic variant provides an incorrect SIGCHLD.

Any ideas?

Quoted 36 lines

---

v2:
- Use newly added macros to check signal emission order
- Add tests for sigset handling
- Restore the default handler after signal test
- make signal_check variable static

v1:
- Update architecture support (adding sh)
- Move sparc sys_rt_sigaction logic into its header
- Add sig_atomic_t
- Use new BITSET_* macros
- Move test into syscall suite
- Various other small changes

---
 tools/include/nolibc/arch-arm.h              |   7 ++
 tools/include/nolibc/arch-arm64.h            |   3 +
 tools/include/nolibc/arch-loongarch.h        |   3 +
 tools/include/nolibc/arch-m68k.h             |  10 ++
 tools/include/nolibc/arch-mips.h             |   3 +
 tools/include/nolibc/arch-powerpc.h          |   8 ++
 tools/include/nolibc/arch-riscv.h            |   3 +
 tools/include/nolibc/arch-s390.h             |   8 +-
 tools/include/nolibc/arch-sh.h               |   5 +
 tools/include/nolibc/arch-sparc.h            |  47 ++++++++
 tools/include/nolibc/arch-x86.h              |  13 +++
 tools/include/nolibc/signal.h                | 103 +++++++++++++++++
 tools/include/nolibc/sys.h                   |   2 +-
 tools/include/nolibc/time.h                  |   3 +-
 tools/include/nolibc/types.h                 |   9 ++
 tools/testing/selftests/nolibc/nolibc-test.c | 115 +++++++++++++++++++
 16 files changed, 338 insertions(+), 4 deletions(-)

(...)

tools/include/nolibc/signal.h line 17

Quoted 9 lines
tools/include/nolibc/signal.h +1 -0
index ac13e53ac31d..16b8b17496bc 100644
@@ -14,6 +14,14 @@
 #include "arch.h"
 #include "types.h"
 #include "sys.h"
+#include "string.h"

Unnecessary now as memset() is not used anymore.

Quoted 7 lines
tools/include/nolibc/signal.h +7 -0
+/* other signal definitions are included by arch.h */
+
+/* The kernel headers do not provide a sig_atomic_t definition */
+#ifndef __sig_atomic_t_defined
+#define __sig_atomic_t_defined 1
+typedef int sig_atomic_t;
+#endif

(...)

tools/testing/selftests/nolibc/nolibc-test.c int test_namespace(void) · line 1296

Quoted 9 lines
tools/testing/selftests/nolibc/nolibc-test.c +1 -0
index d612150d2ea3..fcd44b27cd5e 100644
@@ -1293,6 +1293,120 @@ int test_namespace(void)
 	return ret;
 }
 
+sig_atomic_t signal_check;

Still not static :-)

Quoted 59 lines
tools/testing/selftests/nolibc/nolibc-test.c +59 -0
+
+static void sighandler(int signum)
+{
+	if (signum == SIGUSR1) {
+		kill(getpid(), SIGUSR2);
+		/* The second step has not run because SIGUSR2 is masked */
+		MARK_STEP_DONE(signal_check, 0);
+	} else {
+		MARK_STEP_DONE(signal_check, 1);
+	}
+}
+
+int test_signals(int test_idx)
+{
+	struct sigaction sa = {
+		.sa_flags = 0,
+		.sa_handler = sighandler,
+	};
+	struct sigaction sa_old = {
+		/* Anything other than SIG_DFL */
+		.sa_handler = sighandler,
+	};
+	int llen; /* line length */
+	int ret = 0;
+	int res;
+
+	signal_check = 0;
+
+#ifdef NOLIBC
+	/* Do some checks on sa_mask handling */
+	sigfillset(&sa.sa_mask);
+	llen = printf("    sa_mask.sig[0] (full): ");
+	EXPECT_EQ(1, sa.sa_mask.sig[0],
+		     ~(__typeof__(sa.sa_mask.sig[0]))0);
+	llen = printf("    sa_mask.sig[%d] (full): ", (int)_NSIG_WORDS - 1);
+	EXPECT_EQ(1, sa.sa_mask.sig[_NSIG_WORDS - 1],
+		     ~(__typeof__(sa.sa_mask.sig[0]))0);
+
+	sigemptyset(&sa.sa_mask);
+	llen = printf("    sa_mask.sig[0] (empty): ");
+	EXPECT_EQ(1, sa.sa_mask.sig[0], 0);
+	llen = printf("    sa_mask.sig[%d] (empty): ", (int)_NSIG_WORDS - 1);
+	EXPECT_EQ(1, sa.sa_mask.sig[_NSIG_WORDS - 1], 0);
+
+	/* SIGUSR2 is always in the first word */
+	sigaddset(&sa.sa_mask, SIGUSR2);
+	llen = printf("    sa_mask.sig[0] (SIGUSR2 set): ");
+	EXPECT_EQ(1, sa.sa_mask.sig[0], 1 << (SIGUSR2 - 1));
+
+	llen = printf("    sa_mask.sig[0] (test SIGUSR2): ");
+	EXPECT_NZ(1, sigismember(&sa.sa_mask, SIGUSR2));
+
+	sigdelset(&sa.sa_mask, SIGUSR2);
+	llen = printf("    sa_mask.sig[0] (SIGUSR2 unset): ");
+	EXPECT_ZR(1, sigismember(&sa.sa_mask, SIGUSR2));
+
+	/* _NSIG is the highest valid number and may not be in the first word */
+	sigaddset(&sa.sa_mask, _NSIG);
+	llen = printf("    sa_mask.sig[%d] (_NSIG set): ", (int)_NSIG_WORDS - 1);

tools/testing/selftests/nolibc/nolibc-test.c int test_namespace(void) · line 1365

Quoted 10 lines
tools/testing/selftests/nolibc/nolibc-test.c +10 -0
+	EXPECT_EQ(1, sa.sa_mask.sig[_NSIG_WORDS - 1],
+		     1UL << (_NSIG - (_NSIG_WORDS - 1) * _NSIG_BPW - 1));
+
+	llen = printf("    sa_mask.sig[%d] (test _NSIG): ", (int)_NSIG_WORDS - 1);
+	EXPECT_NZ(1, sigismember(&sa.sa_mask, _NSIG));
+
+	sigdelset(&sa.sa_mask, _NSIG);
+	llen = printf("    sa_mask.sig[%d] (_NSIG unset): ", (int)_NSIG_WORDS - 1);
+	EXPECT_ZR(1, sigismember(&sa.sa_mask, _NSIG));
+#endif

This is more testing than expected, but that's good.
Could you move it to a dedicated test function, gated on "is_nolibc" and the #ifdef.

Quoted 25 lines
tools/testing/selftests/nolibc/nolibc-test.c +25 -0
+
+	/* sa_mask is empty at this point, set SIGUSR2 to verify masking */
+	sigaddset(&sa.sa_mask, SIGUSR2);
+
+	res = sigaction(SIGUSR1, &sa, &sa_old);
+	llen = printf("    register SIGUSR1: %d", res);
+	EXPECT_SYSZR(1, res);
+	if (res)
+		goto out;
+
+	llen = printf("    sa_old.sa_handler: SIG_DFL (%p)", SIG_DFL);
+	EXPECT_PTREQ(1, SIG_DFL, sa_old.sa_handler);
+	if (res)
+		goto out;
+
+	res = sigaction(SIGUSR2, &sa, NULL);
+	llen = printf("    register SIGUSR2: %d", res);
+	EXPECT_SYSZR(1, res);
+	if (res)
+		goto out;
+
+	/* Trigger the first signal. */
+	kill(getpid(), SIGUSR1);
+
+	/* Check the two signal handlers ran in the expected order */

tools/testing/selftests/nolibc/nolibc-test.c int test_namespace(void) · line 1400

Quoted 10 lines
tools/testing/selftests/nolibc/nolibc-test.c +10 -0
+	llen = printf("    signal emission: ");
+	EXPECT_STEPS(1, signal_check, 2);
+
+out:
+	sa.sa_handler = SIG_DFL;
+	res = sigaction(SIGUSR1, &sa, NULL);
+	llen = printf("    restore SIGUSR1: %d", res);
+	EXPECT_SYSZR(1, res);
+
+	res = sigaction(SIGUSR1, &sa, NULL);

SIGUSR2?

Quoted 23 lines
tools/testing/selftests/nolibc/nolibc-test.c +10 -0
+	llen = printf("    restore SIGUSR2: %d", res);
+	EXPECT_SYSZR(1, res);
+
+	llen = printf("%d %s", test_idx, "sigaction");
+	EXPECT_EQ(1, res, 0);
+
+	return ret;
+}
+
 /* Run syscall tests between IDs <min> and <max>.
  * Return 0 on success, non-zero on failure.
  */
@@ -1421,6 +1535,7 @@ int run_syscall(int min, int max)
 		CASE_TEST(syscall_noargs);    EXPECT_SYSEQ(1, syscall(__NR_getpid), getpid()); break;
 		CASE_TEST(syscall_args);      EXPECT_SYSER(1, syscall(__NR_statx, 0, NULL, 0, 0, NULL), -1, EFAULT); break;
 		CASE_TEST(namespace);         EXPECT_SYSZR(euid0 && proc, test_namespace()); break;
+		case __LINE__:                ret += test_signals(test); break;
 		case __LINE__:
 			return ret; /* must be last */
 		/* note: do not set any defaults so as to permit holes above */
Signature · 1 line
2.50.0
#8 lore raw

Re: [PATCH v2 4/4] tools/nolibc: add signal support

Quoted 20 lines · Thomas Weißschuh
Quoted 12 lines · Benjamin Berg

From: Benjamin Berg <[email protected]>

Add support for sigaction() and implement the normal sa_mask helpers.

On many architectures, linux/signal.h pulls in compatibility definitions
for the old sigaction syscall instead of rt_sigaction. However, the kernel can be compiled without support for this compatibility syscall
and it also results in sa_mask to be too small for realtime signals.

Quoted 6 lines

To work around this, the includes are handled separately for each architecture. This way either linux/signal.h or the asm-generic headers
can be used to get the correct definition for the rt_sigaction syscall
including sigset_t.

Quoted 5 lines

I checked this against my WIP alpha support and there this scheme breaks. linux/signal.h provides the old compat types but the asm-generic variant provides an incorrect SIGCHLD.

Any ideas?

I had a quick look, and I don't have a good idea really. For sparc there was at least a #define that permitted to get the correct version. But here, there is nothing.

Probably the best is to just copy in a modified version of signal.h with the correct definitions present. i.e. replace the "ifndef __KERNEL__" sections with something reasonable.

Quoted 65 lines
Quoted 39 lines

---

v2:
- Use newly added macros to check signal emission order
- Add tests for sigset handling
- Restore the default handler after signal test
- make signal_check variable static

v1:
- Update architecture support (adding sh)
- Move sparc sys_rt_sigaction logic into its header
- Add sig_atomic_t
- Use new BITSET_* macros
- Move test into syscall suite
- Various other small changes
---
 tools/include/nolibc/arch-arm.h              |   7 ++ tools/include/nolibc/arch-arm64.h            |   3 + tools/include/nolibc/arch-loongarch.h        |   3 + tools/include/nolibc/arch-m68k.h             |  10 ++ tools/include/nolibc/arch-mips.h             |   3 + tools/include/nolibc/arch-powerpc.h          |   8 ++ tools/include/nolibc/arch-riscv.h            |   3 + tools/include/nolibc/arch-s390.h             |   8 +- tools/include/nolibc/arch-sh.h               |   5 + tools/include/nolibc/arch-sparc.h            |  47 ++++++++ tools/include/nolibc/arch-x86.h              |  13 +++ tools/include/nolibc/signal.h                | 103
+++++++++++++++++
 tools/include/nolibc/sys.h                   |   2 +- tools/include/nolibc/time.h                  |   3 +- tools/include/nolibc/types.h                 |   9 ++ tools/testing/selftests/nolibc/nolibc-test.c | 115
+++++++++++++++++++

 16 files changed, 338 insertions(+), 4 deletions(-)

(...)

tools/include/nolibc/signal.h line 17

Quoted 10 lines
tools/include/nolibc/signal.h +1 -0
index ac13e53ac31d..16b8b17496bc 100644
@@ -14,6 +14,14 @@
  #include "arch.h"
  #include "types.h"
  #include "sys.h"
+#include "string.h"

Unnecessary now as memset() is not used anymore.

Quoted 7 lines
tools/include/nolibc/signal.h +7 -0
+/* other signal definitions are included by arch.h */
+
+/* The kernel headers do not provide a sig_atomic_t definition */
+#ifndef __sig_atomic_t_defined
+#define __sig_atomic_t_defined 1
+typedef int sig_atomic_t;
+#endif

(...)

Quoted 12 lines
Quoted 10 lines
tools/testing/selftests/nolibc/nolibc-test.c +0 -0
index d612150d2ea3..fcd44b27cd5e 100644
@@ -1293,6 +1293,120 @@ int test_namespace(void)
  	return ret;
  }


+sig_atomic_t signal_check;

Still not static :-)

Eeks.

Quoted 74 lines
Quoted 74 lines
tools/testing/selftests/nolibc/nolibc-test.c +6 -0
+
+static void sighandler(int signum)
+{
+	if (signum == SIGUSR1) {
+		kill(getpid(), SIGUSR2);
+		/* The second step has not run because SIGUSR2 is

masked */

tools/testing/selftests/nolibc/nolibc-test.c +29 -0
+		MARK_STEP_DONE(signal_check, 0);
+	} else {
+		MARK_STEP_DONE(signal_check, 1);
+	}
+}
+
+int test_signals(int test_idx)
+{
+	struct sigaction sa = {
+		.sa_flags = 0,
+		.sa_handler = sighandler,
+	};
+	struct sigaction sa_old = {
+		/* Anything other than SIG_DFL */
+		.sa_handler = sighandler,
+	};
+	int llen; /* line length */
+	int ret = 0;
+	int res;
+
+	signal_check = 0;
+
+#ifdef NOLIBC
+	/* Do some checks on sa_mask handling */
+	sigfillset(&sa.sa_mask);
+	llen = printf("    sa_mask.sig[0] (full): ");
+	EXPECT_EQ(1, sa.sa_mask.sig[0],
+		     ~(__typeof__(sa.sa_mask.sig[0]))0);
+	llen = printf("    sa_mask.sig[%d] (full): ",

(int)_NSIG_WORDS - 1);

tools/testing/selftests/nolibc/nolibc-test.c +7 -0
+	EXPECT_EQ(1, sa.sa_mask.sig[_NSIG_WORDS - 1],
+		     ~(__typeof__(sa.sa_mask.sig[0]))0);
+
+	sigemptyset(&sa.sa_mask);
+	llen = printf("    sa_mask.sig[0] (empty): ");
+	EXPECT_EQ(1, sa.sa_mask.sig[0], 0);
+	llen = printf("    sa_mask.sig[%d] (empty): ",

(int)_NSIG_WORDS - 1);

tools/testing/selftests/nolibc/nolibc-test.c +15 -0
+	EXPECT_EQ(1, sa.sa_mask.sig[_NSIG_WORDS - 1], 0);
+
+	/* SIGUSR2 is always in the first word */
+	sigaddset(&sa.sa_mask, SIGUSR2);
+	llen = printf("    sa_mask.sig[0] (SIGUSR2 set): ");
+	EXPECT_EQ(1, sa.sa_mask.sig[0], 1 << (SIGUSR2 - 1));
+
+	llen = printf("    sa_mask.sig[0] (test SIGUSR2): ");
+	EXPECT_NZ(1, sigismember(&sa.sa_mask, SIGUSR2));
+
+	sigdelset(&sa.sa_mask, SIGUSR2);
+	llen = printf("    sa_mask.sig[0] (SIGUSR2 unset): ");
+	EXPECT_ZR(1, sigismember(&sa.sa_mask, SIGUSR2));
+
+	/* _NSIG is the highest valid number and may not be in the

first word */

tools/testing/selftests/nolibc/nolibc-test.c +2 -0
+	sigaddset(&sa.sa_mask, _NSIG);
+	llen = printf("    sa_mask.sig[%d] (_NSIG set): ",

(int)_NSIG_WORDS - 1);

tools/testing/selftests/nolibc/nolibc-test.c +4 -1
+	EXPECT_EQ(1, sa.sa_mask.sig[_NSIG_WORDS - 1],
+		     1UL << (_NSIG - (_NSIG_WORDS - 1) * _NSIG_BPW
- 1));
+
+	llen = printf("    sa_mask.sig[%d] (test _NSIG): ",

(int)_NSIG_WORDS - 1);

tools/testing/selftests/nolibc/nolibc-test.c +3 -0
+	EXPECT_NZ(1, sigismember(&sa.sa_mask, _NSIG));
+
+	sigdelset(&sa.sa_mask, _NSIG);
Quoted 8 lines

tools/testing/selftests/nolibc/nolibc-test.c int test_namespace(void) · line 1365

Quoted 4 lines
tools/testing/selftests/nolibc/nolibc-test.c +1 -0
+	llen = printf("    sa_mask.sig[%d] (_NSIG unset): ",

(int)_NSIG_WORDS - 1);

tools/testing/selftests/nolibc/nolibc-test.c +2 -0
+	EXPECT_ZR(1, sigismember(&sa.sa_mask, _NSIG));
+#endif

This is more testing than expected, but that's good.
Could you move it to a dedicated test function, gated on "is_nolibc" and the #ifdef.

It did find the bit shift type bug in the macros :-)

Benjamin

Quoted 71 lines
Quoted 29 lines
tools/testing/selftests/nolibc/nolibc-test.c +2 -0
+
+	/* sa_mask is empty at this point, set SIGUSR2 to verify

masking */

tools/testing/selftests/nolibc/nolibc-test.c +9 -0
+	sigaddset(&sa.sa_mask, SIGUSR2);
+
+	res = sigaction(SIGUSR1, &sa, &sa_old);
+	llen = printf("    register SIGUSR1: %d", res);
+	EXPECT_SYSZR(1, res);
+	if (res)
+		goto out;
+
+	llen = printf("    sa_old.sa_handler: SIG_DFL (%p)",

SIG_DFL);

tools/testing/selftests/nolibc/nolibc-test.c +14 -0
+	EXPECT_PTREQ(1, SIG_DFL, sa_old.sa_handler);
+	if (res)
+		goto out;
+
+	res = sigaction(SIGUSR2, &sa, NULL);
+	llen = printf("    register SIGUSR2: %d", res);
+	EXPECT_SYSZR(1, res);
+	if (res)
+		goto out;
+
+	/* Trigger the first signal. */
+	kill(getpid(), SIGUSR1);
+
+	/* Check the two signal handlers ran in the expected order

*/

tools/testing/selftests/nolibc/nolibc-test.c int test_namespace(void) · line 1400

Quoted 10 lines
tools/testing/selftests/nolibc/nolibc-test.c +10 -0
+	llen = printf("    signal emission: ");
+	EXPECT_STEPS(1, signal_check, 2);
+
+out:
+	sa.sa_handler = SIG_DFL;
+	res = sigaction(SIGUSR1, &sa, NULL);
+	llen = printf("    restore SIGUSR1: %d", res);
+	EXPECT_SYSZR(1, res);
+
+	res = sigaction(SIGUSR1, &sa, NULL);

SIGUSR2?

Quoted 28 lines
tools/testing/selftests/nolibc/nolibc-test.c +9 -0
+	llen = printf("    restore SIGUSR2: %d", res);
+	EXPECT_SYSZR(1, res);
+
+	llen = printf("%d %s", test_idx, "sigaction");
+	EXPECT_EQ(1, res, 0);
+
+	return ret;
+}
+

 /* Run syscall tests between IDs <min> and <max>.
  * Return 0 on success, non-zero on failure. */

tools/testing/selftests/nolibc/nolibc-test.c +0 -0
@@ -1421,6 +1535,7 @@ int run_syscall(int min, int max)
  		CASE_TEST(syscall_noargs);    EXPECT_SYSEQ(1,
 syscall(__NR_getpid), getpid()); break;
  		CASE_TEST(syscall_args);      EXPECT_SYSER(1,
 syscall(__NR_statx, 0, NULL, 0, 0, NULL), -1, EFAULT); break;

  CASE_TEST(namespace);         EXPECT_SYSZR(euid0 && proc, test_namespace()); break;

tools/testing/selftests/nolibc/nolibc-test.c +1 -0
+		case __LINE__:                ret +=

test_signals(test); break;
  case __LINE__:
  return ret; /* must be last */ /* note: do not set any defaults so as to permit holes above */

Signature · 1 line
2.50.0
#9 lore raw

Re: [PATCH v2 4/4] tools/nolibc: add signal support

Quoted 27 lines · Benjamin Berg
Quoted 20 lines · Thomas Weißschuh
Quoted 12 lines · Benjamin Berg

From: Benjamin Berg <[email protected]>

Add support for sigaction() and implement the normal sa_mask helpers.

On many architectures, linux/signal.h pulls in compatibility definitions
for the old sigaction syscall instead of rt_sigaction. However, the kernel can be compiled without support for this compatibility syscall
and it also results in sa_mask to be too small for realtime signals.

Quoted 6 lines

To work around this, the includes are handled separately for each architecture. This way either linux/signal.h or the asm-generic headers
can be used to get the correct definition for the rt_sigaction syscall
including sigset_t.

Quoted 5 lines

I checked this against my WIP alpha support and there this scheme breaks. linux/signal.h provides the old compat types but the asm-generic variant provides an incorrect SIGCHLD.

Any ideas?

Quoted 7 lines

I had a quick look, and I don't have a good idea really. For sparc there was at least a #define that permitted to get the correct version. But here, there is nothing.

Probably the best is to just copy in a modified version of signal.h with the correct definitions present. i.e. replace the "ifndef __KERNEL__" sections with something reasonable.

But then we can just have our own copy for every architecture, no? To me that looks like the best solution right now. These UAPI headers and APIs are a mess...

Thomas

maintainer
#10 lore raw

Re: [PATCH v2 4/4] tools/nolibc: add signal support

Quoted 32 lines · Thomas Weißschuh
Quoted 31 lines · Benjamin Berg
Quoted 20 lines · Thomas Weißschuh
Quoted 12 lines · Benjamin Berg

From: Benjamin Berg <[email protected]>

Add support for sigaction() and implement the normal sa_mask helpers.

On many architectures, linux/signal.h pulls in compatibility definitions
for the old sigaction syscall instead of rt_sigaction. However, the kernel can be compiled without support for this compatibility syscall
and it also results in sa_mask to be too small for realtime signals.

Quoted 6 lines

To work around this, the includes are handled separately for each architecture. This way either linux/signal.h or the asm-generic headers
can be used to get the correct definition for the rt_sigaction syscall
including sigset_t.

Quoted 5 lines

I checked this against my WIP alpha support and there this scheme breaks. linux/signal.h provides the old compat types but the asm-generic variant provides an incorrect SIGCHLD.

Any ideas?

I had a quick look, and I don't have a good idea really. For sparc there was at least a #define that permitted to get the correct version. But here, there is nothing.

Quoted 6 lines
Quoted 3 lines

Probably the best is to just copy in a modified version of signal.h with the correct definitions present. i.e. replace the "ifndef __KERNEL__" sections with something reasonable.

But then we can just have our own copy for every architecture, no? To me that looks like the best solution right now.

I generally agree. Originally in nolibc, lots of stuff was copied and we only used very little of UAPI. As new architectures were supported, we took this opportunity for cleaning this up and relying more on UAPI, but that doesn't mean we need to get rid of our local definitions all the time, so if that helps, yeah clearly we can have our local copies and definitions.

Cheers,
Willy

#11 lore raw

Re: [PATCH v2 4/4] tools/nolibc: add signal support

Quoted 40 lines · Willy Tarreau
Quoted 32 lines · Thomas Weißschuh
Quoted 31 lines · Benjamin Berg
Quoted 20 lines · Thomas Weißschuh
Quoted 12 lines · Benjamin Berg

From: Benjamin Berg <[email protected]>

Add support for sigaction() and implement the normal sa_mask helpers.

On many architectures, linux/signal.h pulls in compatibility definitions
for the old sigaction syscall instead of rt_sigaction. However, the kernel can be compiled without support for this compatibility syscall
and it also results in sa_mask to be too small for realtime signals.

Quoted 6 lines

To work around this, the includes are handled separately for each architecture. This way either linux/signal.h or the asm-generic headers
can be used to get the correct definition for the rt_sigaction syscall
including sigset_t.

Quoted 5 lines

I checked this against my WIP alpha support and there this scheme breaks. linux/signal.h provides the old compat types but the asm-generic variant provides an incorrect SIGCHLD.

Any ideas?

I had a quick look, and I don't have a good idea really. For sparc there was at least a #define that permitted to get the correct version. But here, there is nothing.

Quoted 6 lines
Quoted 3 lines

Probably the best is to just copy in a modified version of signal.h with the correct definitions present. i.e. replace the "ifndef __KERNEL__" sections with something reasonable.

But then we can just have our own copy for every architecture, no? To me that looks like the best solution right now.

Quoted 6 lines

I generally agree. Originally in nolibc, lots of stuff was copied and we only used very little of UAPI. As new architectures were supported, we took this opportunity for cleaning this up and relying more on UAPI, but that doesn't mean we need to get rid of our local definitions all the time, so if that helps, yeah clearly we can have our local copies and definitions.

Not sure our own copy for every architecture is the best solution. Many of them work fine with the asm-generic headers, and using those where they apply seems pretty reasonable to me.

So, maybe that would be a good compromise? i.e. use asm-generic where possible and copy in a file otherwise. That seems slightly less convoluted and still avoids having to actually fix the architecture headers.

Benjamin

#12 lore raw

Re: [PATCH v2 4/4] tools/nolibc: add signal support

Quoted 48 lines · Benjamin Berg
Quoted 40 lines · Willy Tarreau
Quoted 32 lines · Thomas Weißschuh
Quoted 31 lines · Benjamin Berg
Quoted 20 lines · Thomas Weißschuh
Quoted 12 lines · Benjamin Berg

From: Benjamin Berg <[email protected]>

Add support for sigaction() and implement the normal sa_mask helpers.

On many architectures, linux/signal.h pulls in compatibility definitions
for the old sigaction syscall instead of rt_sigaction. However, the kernel can be compiled without support for this compatibility syscall
and it also results in sa_mask to be too small for realtime signals.

Quoted 6 lines

To work around this, the includes are handled separately for each architecture. This way either linux/signal.h or the asm-generic headers
can be used to get the correct definition for the rt_sigaction syscall
including sigset_t.

Quoted 5 lines

I checked this against my WIP alpha support and there this scheme breaks. linux/signal.h provides the old compat types but the asm-generic variant provides an incorrect SIGCHLD.

Any ideas?

I had a quick look, and I don't have a good idea really. For sparc there was at least a #define that permitted to get the correct version. But here, there is nothing.

Quoted 6 lines
Quoted 3 lines

Probably the best is to just copy in a modified version of signal.h with the correct definitions present. i.e. replace the "ifndef __KERNEL__" sections with something reasonable.

But then we can just have our own copy for every architecture, no? To me that looks like the best solution right now.

Quoted 6 lines

I generally agree. Originally in nolibc, lots of stuff was copied and we only used very little of UAPI. As new architectures were supported, we took this opportunity for cleaning this up and relying more on UAPI, but that doesn't mean we need to get rid of our local definitions all the time, so if that helps, yeah clearly we can have our local copies and definitions.

Quoted 3 lines

Not sure our own copy for every architecture is the best solution. Many of them work fine with the asm-generic headers, and using those where they apply seems pretty reasonable to me.

I meant *one shared* copy for all architectures.
It would essentially be the asm-generic definitions with some ifdeffery.

Quoted 4 lines

So, maybe that would be a good compromise? i.e. use asm-generic where possible and copy in a file otherwise. That seems slightly less convoluted and still avoids having to actually fix the architecture headers.

Does my proposal from above satisfy this?

Thomas