workqueue: Fix errors seen for newer (v6.4 and later) kernels.

LSE-373.

Signed-off-by: Imran Khan <[email protected]>

drgn_tools/workqueue.py

@@ -12,6 +12,7 @@
 from typing import Union
 
 from drgn import cast
+from drgn import IntegerLike
 from drgn import NULL
 from drgn import Object
 from drgn import Program
@@ -22,6 +23,7 @@
 from drgn.helpers.linux.list import list_empty
 from drgn.helpers.linux.list import list_for_each_entry
 from drgn.helpers.linux.percpu import per_cpu
+from drgn.helpers.linux.percpu import per_cpu_ptr
 from drgn.helpers.linux.pid import find_task
 
 
@@ -46,12 +48,41 @@
     "show_one_worker_pool",
     "is_task_a_worker",
     "find_worker_executing_work",
+    "workqueue_get_pwq",
 )
 
 
 _PF_WQ_WORKER = 0x00000020
 
 
+def _work_offq_pool_none(prog: Program) -> IntegerLike:
+    # Linux kernel commit afa4bb778e48 ("workqueue: clean up WORK_* constant
+    # types, clarify masking") (in v6.4) changed WORK_OFFQ_POOL_NONE from
+    # constants of type enum to constants of type unsigned long.
+    try:
+        val = prog["WORK_OFFQ_POOL_NONE"].value_()
+    except KeyError:
+        val = (
+            Object(prog, "unsigned long", 1).value_()
+            << prog["WORK_OFFQ_POOL_BITS"]
+        ) - 1
+    return val
+
+
+def _work_struct_wq_data_mask(prog: Program) -> IntegerLike:
+    # Linux kernel commit afa4bb778e48 ("workqueue: clean up WORK_* constant
+    # types, clarify masking") (in v6.4) changed WORK_STRUCT_WQ_DATA_MASK from
+    # constants of type enum to constants of type unsigned long.
+    try:
+        val = prog["WORK_STRUCT_WQ_DATA_MASK"].value_()
+    except KeyError:
+        val = ~(
+            (Object(prog, "unsigned long", 1) << prog["WORK_STRUCT_FLAG_BITS"])
+            - 1
+        )
+    return val
+
+
 def _print_work(work: Object) -> None:
     prog = work.prog_
     try:
@@ -63,6 +94,28 @@ def _print_work(work: Object) -> None:
     )
 
 
+def workqueue_get_pwq(workqueue: Object, cpu: int) -> Object:
+    """
+    Find pool_workqueue of a bound workqueue for a given CPU.
+
+    :param workqueue: ``struct workqueue_struct *``
+    :return: ``struct pool_workqueue *``.
+    """
+    # At first Linux kernel commit ee1ceef72754 ("workqueue: Rename wq->cpu_pwqs to
+    # wq->cpu_pwq") (in v6.6) renamed cpu_pwqs to cpu_pwq and then Linux kernel commit
+    # 687a9aa56f81("workqueue: Make per-cpu pool_workqueues allocated and released
+    # like unbound ones") (in v6.6) changed cpu_pwq to double pointer.
+    # As both of the changes were made in v6.6, there are no kernel versions
+    # with wq->cpu_pwq as a pointer. Still I have mentioned both the changes so that
+    # we can track both name change and type change of this member.
+    try:
+        pwq = per_cpu_ptr(workqueue.cpu_pwqs, cpu)
+    except AttributeError:
+        pwq = per_cpu_ptr(workqueue.cpu_pwq, cpu)[0]
+
+    return pwq
+
+
 def for_each_workqueue(prog: Program) -> Iterator[Object]:
     """
     Iterate over all workqueues in the system.
@@ -243,7 +296,7 @@ def get_work_pwq(work: Object) -> Object:
     if data & prog["WORK_STRUCT_PWQ"].value_():
         return cast(
             "struct pool_workqueue *",
-            data & _wq_data_mask(prog),
+            data & _work_struct_wq_data_mask(prog),
         )
     else:
         return NULL(work.prog_, "struct pool_workqueue *")
@@ -262,12 +315,12 @@ def get_work_pool(work: Object) -> Object:
     data = cast("unsigned long", work.data.counter.read_())
 
     if data & prog["WORK_STRUCT_PWQ"].value_():
-        pwq = data & _wq_data_mask(prog)
+        pwq = data & _work_struct_wq_data_mask(prog)
         pool = Object(prog, "struct pool_workqueue", address=pwq).pool
     else:
         pool_id = data >> prog["WORK_OFFQ_POOL_SHIFT"].value_()
 
-        if pool_id == prog["WORK_OFFQ_POOL_NONE"].value_():
+        if pool_id == _work_offq_pool_none(prog):
             return NULL(work.prog_, "struct worker_pool *")
 
         pool = idr_find(prog["worker_pool_idr"].address_of_(), pool_id)

tests/test_workqueue.py

@@ -78,11 +78,12 @@ def test_for_each_cpu_worker_pool(prog: drgn.Program) -> None:
 def test_for_each_pwq(prog: drgn.Program) -> None:
     workq = prog["system_wq"]
     pwqs = [pwq.value_() for pwq in wq.for_each_pwq(workq)]
-    cpu_pwqs = [
-        per_cpu_ptr(workq.cpu_pwqs, cpu).value_()
+    cpu_pwqs_attr = "cpu_pwqs" if hasattr(workq, "cpu_pwqs") else "cpu_pwq"
+    cpu_pwqs_list = [
+        per_cpu_ptr(getattr(workq, cpu_pwqs_attr), cpu).value_()
         for cpu in for_each_online_cpu(prog)
     ]
-    assert pwqs.sort() == cpu_pwqs.sort()
+    assert pwqs.sort() == cpu_pwqs_list.sort()
 
 
 def test_for_each_pending_work_on_cpu(prog: drgn.Program) -> None:
@@ -97,7 +98,7 @@ def test_for_each_pending_work_in_pool(prog: drgn.Program) -> None:
 
 
 def test_for_each_pending_work_of_pwq(prog: drgn.Program) -> None:
-    cpu_pwqs_0 = per_cpu_ptr(prog["system_wq"].cpu_pwqs, 0)
+    cpu_pwqs_0 = wq.workqueue_get_pwq(prog["system_wq"], 0)
     for work in wq.for_each_pending_work_of_pwq(cpu_pwqs_0):
         pass