[PATCH v3 15/44] gdb, infrun, ze: allow resuming unavailable threads

[Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>]

From: Markus Metzger <markus.t.metzger@intel.com>

On GPUs, threads may not always respond to interrupt requests when we try
to stabilize threads on an all-stop target.  In order to still be able to
support all-stop model on those targets, we allow this case and model such
threads as unavailable.

They do exist but we cannot interact with them.  In particular, they do
not have registers, including PC.

In some cases, all threads on a target may be unavailable.  This can
happen when we currently do not have any work scheduled on that device.
One particular instance of this is during startup when we newly attach to
the device.

We still need to allow 'continuing' those threads to make GDB listen to
events from that target.

Update gdb.threads/killed-outside.exp by adding another case to the
expected messages when killing the inferior.  It already contained cases
with and without thread exited messages.  Add another one for an exited
thread but no pc-not-available message.
---
 gdb/infrun.c                                 | 11 ++++++++++-
 gdb/testsuite/gdb.threads/killed-outside.exp |  4 ++++
 2 files changed, 14 insertions(+), 1 deletion(-)

diff --git a/gdb/infrun.c b/gdb/infrun.c
index dc55e17b98a13dac14171a795574344ec2a3c545..0ab642abfe246d87179c296401b5cf5a2c3003dd 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -2730,7 +2730,16 @@ resume_1 (enum gdb_signal sig)
       step = false;
     }
 
-  CORE_ADDR pc = regcache_read_pc (regcache);
+  /* Allow resuming threads that do not have registers available.
+
+     This may happen on GPUs for threads that we tried to stop but that
+     did not respond.  We model them as unavailable threads.
+
+     PC will be zero in that case and we do not expect any breakpoints at
+     this address so we're not adding any extra checks.  */
+  CORE_ADDR pc = step
+    ? regcache_read_pc (regcache)
+    : regcache_read_pc_protected (regcache);
 
   infrun_debug_printf ("step=%d, signal=%s, trap_expected=%d, "
 		       "current thread [%s] at %s",
diff --git a/gdb/testsuite/gdb.threads/killed-outside.exp b/gdb/testsuite/gdb.threads/killed-outside.exp
index 9edb8d251b5737dbc5dfd55ce0245beb7be3a458..b0ca478bc3d42540057322bc30c22cdc07792665 100644
--- a/gdb/testsuite/gdb.threads/killed-outside.exp
+++ b/gdb/testsuite/gdb.threads/killed-outside.exp
@@ -58,4 +58,8 @@ gdb_test_multiple "continue" "prompt after first continue" {
     -re -wrap ".*$killed_msg.*$no_longer_exists_msg" {
 	pass $gdb_test_name
     }
+    -re "Continuing\.\r\n\r\n$killed_msg\r\n$no_longer_exists_msg\r\n$gdb_prompt $" {
+	pass $gdb_test_name
+	gdb_test "continue" $not_being_run_msg "second continue"
+    }
 }

-- 
2.34.1

Intel Deutschland GmbH
Registered Address: Am Campeon 10, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Sean Fennelly, Jeffrey Schneiderman, Tiffany Doon Silva
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928