[PATCH] accel/tcg/user-exec: Fix read-modify-write of code on s390 hosts

[PATCH] accel/tcg/user-exec: Fix read-modify-write of code on s390 hosts

[Ilya Leoshkevich]

x86_64 dotnet/runtime uses cmpxchg for code patching. When running it
under s390x qemu-linux user, cpu_signal_handler() does not recognize
this as a write and does not restore PAGE_WRITE cleared by
tb_page_add(), incorrectly forwarding the signal to the guest code.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
---
 accel/tcg/user-exec.c | 37 ++++++++++++++++++++++++++++++++-----
 1 file changed, 32 insertions(+), 5 deletions(-)

diff --git a/accel/tcg/user-exec.c b/accel/tcg/user-exec.c
index 90d1a2d327..587a0f1ef9 100644
--- a/accel/tcg/user-exec.c
+++ b/accel/tcg/user-exec.c
@@ -681,17 +681,24 @@ int cpu_signal_handler(int host_signum, void *pinfo,
     pc = uc->uc_mcontext.psw.addr;
 
     /* ??? On linux, the non-rt signal handler has 4 (!) arguments instead
-       of the normal 2 arguments.  The 3rd argument contains the "int_code"
-       from the hardware which does in fact contain the is_write value.
+       of the normal 2 arguments.  The 4th argument contains the "Translation-
+       Exception Identification for DAT Exceptions" from the hardware (aka
+       "int_parm_long"), which does in fact contain the is_write value.
        The rt signal handler, as far as I can tell, does not give this value
-       at all.  Not that we could get to it from here even if it were.  */
-    /* ??? This is not even close to complete, since it ignores all
-       of the read-modify-write instructions.  */
+       at all.  Not that we could get to it from here even if it were.
+       So fall back to parsing instructions.  Treat read-modify-write ones as
+       writes, which is not fully correct, but for tracking self-modifying code
+       this is better than treating them as reads.  Checking si_addr page flags
+       might be a viable improvement, albeit a racy one.  */
+    /* ??? This is not even close to complete.  */
     pinsn = (uint16_t *)pc;
     switch (pinsn[0] >> 8) {
     case 0x50: /* ST */
     case 0x42: /* STC */
     case 0x40: /* STH */
+    case 0xba: /* CS */
+    case 0xbb: /* CDS */
+    case 0xc8: /* CSST */
         is_write = 1;
         break;
     case 0xc4: /* RIL format insns */
@@ -715,7 +722,27 @@ int cpu_signal_handler(int host_signum, void *pinfo,
             is_write = 1;
         }
         break;
+    case 0xeb: /* RSY format insns */
+        switch (pinsn[2] & 0xff) {
+        case 0x14: /* CSY */
+        case 0x30: /* CSG */
+        case 0x31: /* CDSY */
+        case 0x3e: /* CDSG */
+        case 0xe4: /* LANG */
+        case 0xe6: /* LAOG */
+        case 0xe7: /* LAXG */
+        case 0xe8: /* LAAG */
+        case 0xea: /* LAALG */
+        case 0xf4: /* LAN */
+        case 0xf6: /* LAO */
+        case 0xf7: /* LAX */
+        case 0xfa: /* LAAL */
+        case 0xf8: /* LAA */
+            is_write = 1;
+        }
+        break;
     }
+
     return handle_cpu_signal(pc, info, is_write, &uc->uc_sigmask);
 }
 
-- 
2.31.1

Re: [PATCH] accel/tcg/user-exec: Fix read-modify-write of code on s390 hosts

[Richard Henderson]

On 8/3/21 9:54 AM, Ilya Leoshkevich wrote:
>       /* ??? On linux, the non-rt signal handler has 4 (!) arguments instead
> -       of the normal 2 arguments.  The 3rd argument contains the "int_code"
> -       from the hardware which does in fact contain the is_write value.
> +       of the normal 2 arguments.  The 4th argument contains the "Translation-
> +       Exception Identification for DAT Exceptions" from the hardware (aka
> +       "int_parm_long"), which does in fact contain the is_write value.
>          The rt signal handler, as far as I can tell, does not give this value
> -       at all.  Not that we could get to it from here even if it were.  */
> -    /* ??? This is not even close to complete, since it ignores all
> -       of the read-modify-write instructions.  */
> +       at all.  Not that we could get to it from here even if it were.
> +       So fall back to parsing instructions.  Treat read-modify-write ones as
> +       writes, which is not fully correct, but for tracking self-modifying code
> +       this is better than treating them as reads.  Checking si_addr page flags
> +       might be a viable improvement, albeit a racy one.  */
> +    /* ??? This is not even close to complete.  */

You should have gotten a checkpatch warning here.
Just convert the comment to

   /*
    * this style
    */

>       pinsn = (uint16_t *)pc;
>       switch (pinsn[0] >> 8) {
>       case 0x50: /* ST */
>       case 0x42: /* STC */
>       case 0x40: /* STH */
> +    case 0xba: /* CS */
> +    case 0xbb: /* CDS */
> +    case 0xc8: /* CSST */

CSST is format SSF; you're not checking enough bits to distinguish from LOAD PAIR DISJOINT.

Otherwise, looks good.

r~