[Valgrind-developers] [valgrind] s390x: Remove a fixs390. Add an explanation.

[Florian K. <fk...@so...>]

https://sourceware.org/git/gitweb.cgi?p=valgrind.git;h=c164b5714172e7678ab258545c2bc0127f21a318

commit c164b5714172e7678ab258545c2bc0127f21a318
Author: Florian Krohm <fl...@ei...>
Date:   Wed May 28 21:07:44 2025 +0000

    s390x: Remove a fixs390. Add an explanation.

Diff:
---
 VEX/priv/host_s390_defs.h | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/VEX/priv/host_s390_defs.h b/VEX/priv/host_s390_defs.h
index dcdb6713e9..48fbac7644 100644
--- a/VEX/priv/host_s390_defs.h
+++ b/VEX/priv/host_s390_defs.h
@@ -706,8 +706,18 @@ typedef struct {
          s390_amode   *guest_IA;
       } xassisted;
       struct {
-         /* fixs390: I don't think these are really needed
-            as the gsp and the offset are fixed  no ? */
+         /* Note: these fields are needed. Here's why:
+            These fields are amodes for accessing the host_EvC_COUNTER and
+            host_EvC_FAILADDR fields in the guest state.
+            When guest and host architecture are both s390x then we know that
+            the displacement in evcheck::counter is
+            offsetof(VexGuestS390XState, host_EvC_COUNTER) and likewise for
+            the displacement in evcheck::fail_addr. There would be no point
+            to build these amodes in the first place because we could just
+            hardwire the displacements in s390_insn_evcheck_emit.
+            However in a multi-arch setting the amodes point to the
+            host_EvC_COUNTER/FAILADDR fields in a *different* guest state and
+            those offsets are not known. So we do need to build the amodes. */
          s390_amode   *counter;    /* dispatch counter */
          s390_amode   *fail_addr;
       } evcheck;