Allow default behavior for ARM64 machines and skip high-memory VMs (#3498)

* Add sufficient sized data-disk in kdump test

* Move print_additional_info just before panic

* Use execute_async and remove kill_on_timeout parameter

* Fix linter errors by moving the check into a new internal function

* Remove whitespace

* Use black format

* Allow default behavior for ARM64 machines and skip high-memory VMs

* Fix flake8 error

* Fix linter error

lisa/tools/free.py

@@ -83,6 +83,9 @@ class Free(Tool):
     def get_free_memory_gb(self) -> int:
         return self._get_field_bytes_kib("Mem", "free") >> 20
 
+    def get_total_memory_gb(self) -> int:
+        return self._get_field_bytes_kib("Mem", "total") >> 20
+
     def get_total_memory(self) -> str:
         """
         Returns total memory in power of 1000 with unit

lisa/tools/kdump.py

@@ -18,7 +18,7 @@ from lisa.tools.lscpu import Lscpu
 from lisa.tools.make import Make
 from lisa.tools.sysctl import Sysctl
 from lisa.tools.tar import Tar
-from lisa.util import LisaException, UnsupportedDistroException
+from lisa.util import LisaException, SkippedException, UnsupportedDistroException
 
 from .kernel_config import KernelConfig
 
@@ -608,26 +608,6 @@ class KdumpCBLMariner(KdumpBase):
         """
         This method enables the kdump service.
         """
-        kdump_conf = "/etc/kdump.conf"
-        sed = self.node.tools[Sed]
-        # Remove force_no_rebuild=1 if present
-        sed.substitute(
-            match_lines="^force_no_rebuild",
-            regexp="force_no_rebuild",
-            replacement="#force_no_rebuild",
-            file=kdump_conf,
-            sudo=True,
-        )
-        # Set mariner_2_initrd_use_suffix. Otherwise it will replace
-        # the original initrd file which will cause a reboot-loop
-        sed.substitute(
-            match_lines="mariner_2_initrd_use_suffix",
-            regexp="#mariner_2_initrd_use_suffix",
-            replacement="mariner_2_initrd_use_suffix",
-            file=kdump_conf,
-            sudo=True,
-        )
-
         # Check for sufficient core numbers
         self.ensure_nr_cpus()
 
@@ -668,7 +648,42 @@ class KdumpCBLMariner(KdumpBase):
         """
         If the system memory size is bigger than 1T, the default size of /var/crash
         may not be enough to store the dump file, need to change the dump path
+
+        path option is not supported by default initrd. Regenerated initrd will not
+        boot properly in ARM64. So, skip the test if it is ARM64 and Mariner-2.0
         """
+        if (
+            self.node.os.information.version.major == 2
+            and isinstance(self.node.os, Posix)
+            and self.node.os.get_kernel_information().hardware_platform == "aarch64"
+        ):
+            raise SkippedException(
+                "path option is not supported in Mariner-2.0 for ARM64. "
+                "kdump will not work well on high memory Mariner-2.0 systems"
+            )
+
+        # Update forc_rebuild before changing the dump path. Otherwise the default
+        # initrd will not honor the path
+        kdump_conf = "/etc/kdump.conf"
+        sed = self.node.tools[Sed]
+        # Remove force_no_rebuild=1 if present
+        sed.substitute(
+            match_lines="^force_no_rebuild",
+            regexp="force_no_rebuild",
+            replacement="#force_no_rebuild",
+            file=kdump_conf,
+            sudo=True,
+        )
+        # Set mariner_2_initrd_use_suffix. Otherwise it will replace
+        # the original initrd file which will cause a reboot-loop
+        sed.substitute(
+            match_lines="mariner_2_initrd_use_suffix",
+            regexp="#mariner_2_initrd_use_suffix",
+            replacement="mariner_2_initrd_use_suffix",
+            file=kdump_conf,
+            sudo=True,
+        )
+
         self.node.execute(
             f"mkdir -p {dump_path}",
             expected_exit_code=0,

microsoft/testsuites/kdump/kdumpcrash.py

@@ -25,7 +25,7 @@ from lisa import (
 from lisa.features import Disk, SerialConsole
 from lisa.features.security_profile import CvmDisabled
 from lisa.operating_system import BSD, Redhat, Windows
-from lisa.tools import Dmesg, Echo, KdumpBase, KernelConfig, Lscpu, Stat
+from lisa.tools import Df, Dmesg, Echo, KdumpBase, KernelConfig, Lscpu, Stat
 from lisa.tools.free import Free
 from lisa.util.perf_timer import create_timer
 from lisa.util.shell import try_connect
@@ -270,19 +270,12 @@ class KdumpCrash(TestSuite):
 
     def _is_system_with_more_memory(self, node: Node) -> bool:
         free = node.tools[Free]
-        total_memory = free.get_total_memory()
-        # Return true when system memory is 10 GiB higher than the OS disk size
-        if "T" in total_memory or (
-            "G" in total_memory
-            and (
-                node.capability.disk
-                and isinstance(node.capability.disk.os_disk_size, int)
-                and (
-                    float(total_memory.strip("G"))
-                    > (node.capability.disk.os_disk_size - 10)
-                )
-            )
-        ):
+        total_memory_in_gb = free.get_total_memory_gb()
+
+        df = node.tools[Df]
+        available_space_in_os_disk = df.get_filesystem_available_space("/", True)
+
+        if total_memory_in_gb > available_space_in_os_disk:
             return True
         return False
 
@@ -300,8 +293,8 @@ class KdumpCrash(TestSuite):
             self.crash_kernel = "auto"
 
         if self._is_system_with_more_memory(node):
-            # System memory is more os disk size, need to change the dump path
-            # and increase the timeout duration
+            # As system memory is more than free os disk size, need to
+            # change the dump path and increase the timeout duration
             kdump.config_resource_disk_dump_path(
                 self._get_resource_disk_dump_path(node)
             )