Skip to contentSkip to conversation

selftests: sud_test: return correct emulated syscall value on RISC-V

PATCH Superseded Clément Léger 2 participants 4 messages Sep 13, 2023 – Nov 9, 2023

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 selftests: sud_test: return correct emulated syscall value on RISC-V

Currently, the sud_test expects the emulated syscall to return the emulated syscall number. This assumption only works on architectures were the syscall calling convention use the same register for syscall number/syscall return value. This is not the case for RISC-V and thus the return value must be also emulated using the provided ucontext.

---
 tools/testing/selftests/syscall_user_dispatch/sud_test.c | 8 ++++++++
 1 file changed, 8 insertions(+)
tools/testing/selftests/syscall_user_dispatch/sud_test.c +8 -0
index b5d592d4099e..1b5553c19700 100644
@@ -158,6 +158,14 @@ static void handle_sigsys(int sig, siginfo_t *info, void *ucontext)
 
 	/* In preparation for sigreturn. */
 	SYSCALL_DISPATCH_OFF(glob_sel);
+
+	/*
+	 * Modify interrupted context returned value according to syscall
+	 * calling convention
+	 */
+#if defined(__riscv)
+	((ucontext_t*)ucontext)->uc_mcontext.__gregs[REG_A0] = MAGIC_SYSCALL_1;
+#endif
 }
 
 TEST(dispatch_and_return)
Signature · 1 line
2.40.1
Reviewed-by Acked-by on v1 1/1
#1 lore raw
Quoted 20 lines · On Wed, 13 Sep 2023 07:07:11 PDT (-0700), [email protected] wrote:

Currently, the sud_test expects the emulated syscall to return the emulated syscall number. This assumption only works on architectures were the syscall calling convention use the same register for syscall number/syscall return value. This is not the case for RISC-V and thus the return value must be also emulated using the provided ucontext.

---
 tools/testing/selftests/syscall_user_dispatch/sud_test.c | 8 ++++++++
 1 file changed, 8 insertions(+)
tools/testing/selftests/syscall_user_dispatch/sud_test.c +1 -0
index b5d592d4099e..1b5553c19700 100644
@@ -158,6 +158,14 @@ static void handle_sigsys(int sig, siginfo_t *info, void *ucontext)
 
 	/* In preparation for sigreturn. */
 	SYSCALL_DISPATCH_OFF(glob_sel);
+

tools/testing/selftests/syscall_user_dispatch/sud_test.c static void handle_sigsys(int sig, siginfo_t *info, void *ucontext) · line 171

Quoted 10 lines
tools/testing/selftests/syscall_user_dispatch/sud_test.c +7 -0
+	/*
+	 * Modify interrupted context returned value according to syscall
+	 * calling convention
+	 */
+#if defined(__riscv)
+	((ucontext_t*)ucontext)->uc_mcontext.__gregs[REG_A0] = MAGIC_SYSCALL_1;
+#endif
 }
 
 TEST(dispatch_and_return)

I'm not sure if I'm just tired, but it took me a while to figure out why this was necessary. I think this is a better explanation:

   diff --git a/tools/testing/selftests/syscall_user_dispatch/sud_test.c b/tools/testing/selftests/syscall_user_dispatch/sud_test.c
   index b5d592d4099e..a913fd90cfa3 100644
   --- a/tools/testing/selftests/syscall_user_dispatch/sud_test.c
   +++ b/tools/testing/selftests/syscall_user_dispatch/sud_test.c
   @@ -158,6 +158,16 @@ static void handle_sigsys(int sig, siginfo_t *info, void *ucontext)
    	/* In preparation for sigreturn. */
    	SYSCALL_DISPATCH_OFF(glob_sel);
   +	/*
   +	 * The tests for argument handling assume that `syscall(x) == x`.  This
   +	 * is a NOP on x86 because the syscall number is passed in %rax, which
   +	 * happens to also be the function ABI return register.  Other
   +	 * architectures may need to swizzle the arguments around.
   +	 */
   +#if defined(__riscv)
   +	(ucontext_t*)ucontext)->uc_mcontext.__gregs[REG_A0] =
   +		(ucontext_t*)ucontext)->uc_mcontext.__gregs[REG_A7];
   +#endif
    }
   
    TEST(dispatch_and_return)

but also

as I agree this is correct.

also: wouldn't arm64 also need to move x8 into x0 here, for essentially the same reason as we do?

#2 lore raw
Quoted 33 lines · Palmer Dabbelt
Quoted 32 lines · On Wed, 13 Sep 2023 07:07:11 PDT (-0700), [email protected] wrote:

Currently, the sud_test expects the emulated syscall to return the emulated syscall number. This assumption only works on architectures were the syscall calling convention use the same register for syscall number/syscall return value. This is not the case for RISC-V and thus the return value must be also emulated using the provided ucontext.

---
 tools/testing/selftests/syscall_user_dispatch/sud_test.c | 8 ++++++++

 1 file changed, 8 insertions(+)
tools/testing/selftests/syscall_user_dispatch/sud_test.c +8 -0
index b5d592d4099e..1b5553c19700 100644
@@ -158,6 +158,14 @@ static void handle_sigsys(int sig, siginfo_t
 *info, void *ucontext)
 
      /* In preparation for sigreturn. */
      SYSCALL_DISPATCH_OFF(glob_sel);
+
+    /*
+     * Modify interrupted context returned value according to syscall
+     * calling convention
+     */
+#if defined(__riscv)
+    ((ucontext_t*)ucontext)->uc_mcontext.__gregs[REG_A0] =
 MAGIC_SYSCALL_1;
+#endif
  }
Quoted 4 lines
Quoted 1 line

 TEST(dispatch_and_return)

I'm not sure if I'm just tired, but it took me a while to figure out why this was necessary.  I think this is a better explanation:

I think it's because this mechanism does not behave like other syscalls at all and the classic calling convention does not really apply...

Quoted 10 lines

   diff --git a/tools/testing/selftests/syscall_user_dispatch/sud_test.c b/tools/testing/selftests/syscall_user_dispatch/sud_test.c index b5d592d4099e..a913fd90cfa3 100644
   --- a/tools/testing/selftests/syscall_user_dispatch/sud_test.c +++ b/tools/testing/selftests/syscall_user_dispatch/sud_test.c @@ -158,6 +158,16 @@ static void handle_sigsys(int sig, siginfo_t *info, void *ucontext)
           /* In preparation for sigreturn. */ SYSCALL_DISPATCH_OFF(glob_sel);

Quoted 8 lines

   +    /*
   +     * The tests for argument handling assume that syscall(x) == x.  This
   +     * is a NOP on x86 because the syscall number is passed in %rax, which
   +     * happens to also be the function ABI return register.  Other +     * architectures may need to swizzle the arguments around. +     */

Indeed, that is more clear. Should I send a v2 ?

Quoted 9 lines

   +#if defined(__riscv)
   +    (ucontext_t*)ucontext)->uc_mcontext.__gregs[REG_A0] = +        (ucontext_t*)ucontext)->uc_mcontext.__gregs[REG_A7]; +#endif
    }
       TEST(dispatch_and_return)

but also

Quoted 7 lines

as I agree this is correct.

also: wouldn't arm64 also need to move x8 into x0 here, for essentially the same reason as we do?

Yes, as well as for a bunch of other architectures. I suspect this has only been tested on x86. AFAIK, this feature is mainly for wine usage which then makes sense for x86 and games.

Thanks,

Clément

_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv

#3 lore raw
Quoted 36 lines · On Thu, 09 Nov 2023 00:22:46 PST (-0800), [email protected] wrote:
Quoted 33 lines · Palmer Dabbelt
Quoted 32 lines · On Wed, 13 Sep 2023 07:07:11 PDT (-0700), [email protected] wrote:

Currently, the sud_test expects the emulated syscall to return the emulated syscall number. This assumption only works on architectures were the syscall calling convention use the same register for syscall number/syscall return value. This is not the case for RISC-V and thus the return value must be also emulated using the provided ucontext.

---
 tools/testing/selftests/syscall_user_dispatch/sud_test.c | 8 ++++++++

 1 file changed, 8 insertions(+)
tools/testing/selftests/syscall_user_dispatch/sud_test.c +8 -0
index b5d592d4099e..1b5553c19700 100644
@@ -158,6 +158,14 @@ static void handle_sigsys(int sig, siginfo_t
 *info, void *ucontext)
 
      /* In preparation for sigreturn. */
      SYSCALL_DISPATCH_OFF(glob_sel);
+
+    /*
+     * Modify interrupted context returned value according to syscall
+     * calling convention
+     */
+#if defined(__riscv)
+    ((ucontext_t*)ucontext)->uc_mcontext.__gregs[REG_A0] =
 MAGIC_SYSCALL_1;
+#endif
  }
Quoted 7 lines
Quoted 4 lines
Quoted 1 line

 TEST(dispatch_and_return)

I'm not sure if I'm just tired, but it took me a while to figure out why this was necessary.  I think this is a better explanation:

I think it's because this mechanism does not behave like other syscalls at all and the classic calling convention does not really apply...

Yep. I also got tripped up because I mis-read the docs and though SIGSYS was only for some error case (where it's actually for all the intercepted syscalls).

Quoted 18 lines
Quoted 9 lines

   diff --git a/tools/testing/selftests/syscall_user_dispatch/sud_test.c b/tools/testing/selftests/syscall_user_dispatch/sud_test.c index b5d592d4099e..a913fd90cfa3 100644
   --- a/tools/testing/selftests/syscall_user_dispatch/sud_test.c +++ b/tools/testing/selftests/syscall_user_dispatch/sud_test.c @@ -158,6 +158,16 @@ static void handle_sigsys(int sig, siginfo_t *info, void *ucontext)
           /* In preparation for sigreturn. */ SYSCALL_DISPATCH_OFF(glob_sel);

Quoted 8 lines

   +    /*
   +     * The tests for argument handling assume that syscall(x) == x.  This
   +     * is a NOP on x86 because the syscall number is passed in %rax, which
   +     * happens to also be the function ABI return register.  Other +     * architectures may need to swizzle the arguments around. +     */

Quoted 1 line

Indeed, that is more clear. Should I send a v2 ?

I would, but +Thomas as it looks like he's the one taking patches for this.

Quoted 13 lines
Quoted 13 lines

   +#if defined(__riscv)
   +    (ucontext_t*)ucontext)->uc_mcontext.__gregs[REG_A0] = +        (ucontext_t*)ucontext)->uc_mcontext.__gregs[REG_A7]; +#endif
    }
       TEST(dispatch_and_return)

but also

Quoted 8 lines
Quoted 4 lines

as I agree this is correct.

also: wouldn't arm64 also need to move x8 into x0 here, for essentially the same reason as we do?

Yes, as well as for a bunch of other architectures. I suspect this has only been tested on x86. AFAIK, this feature is mainly for wine usage which then makes sense for x86 and games.

Ya, makes sense -- I'd just looked at Arm to double-check my understanding here, as we usually don't find bugs in generic code before Arm does.

Quoted 4 lines

Thanks,

Clément