virtio-balloon: Drop huge page restriction

While the traditional balloon device would not be able to reclaim memory
when back by huge pages, it could still technically be used to to
restrict memory usage in the guest.

With the addition of hinting and reporting, they report ranges in bigger
sizes (4mb by default). Because of this, it is possible for the host
reclaim huge pages backing the guest.

Updates the performance tests for the balloon when back by huge pages,
added varients to the size reduction tests to ensure hinting and
reporting can reduce the RSS of the guest.

Move the inflation test to performance to ensure it runs sequentially in
CI otherwise the host can be exhausted of huge pages.

Signed-off-by: Jack Thomson <jackabt@amazon.com>

Author: JackThomson2

src/vmm/src/resources.rs

@@ -22,9 +22,7 @@ use crate::vmm_config::boot_source::{
 use crate::vmm_config::drive::*;
 use crate::vmm_config::entropy::*;
 use crate::vmm_config::instance_info::InstanceInfo;
-use crate::vmm_config::machine_config::{
-    HugePageConfig, MachineConfig, MachineConfigError, MachineConfigUpdate,
-};
+use crate::vmm_config::machine_config::{MachineConfig, MachineConfigError, MachineConfigUpdate};
 use crate::vmm_config::metrics::{MetricsConfig, MetricsConfigError, init_metrics};
 use crate::vmm_config::mmds::{MmdsConfig, MmdsConfigError};
 use crate::vmm_config::net::*;
@@ -255,9 +253,6 @@ impl VmResources {
             return Err(MachineConfigError::IncompatibleBalloonSize);
         }
 
-        if self.balloon.get().is_some() && updated.huge_pages != HugePageConfig::None {
-            return Err(MachineConfigError::BalloonAndHugePages);
-        }
         self.machine_config = updated;
 
         Ok(())
@@ -314,10 +309,6 @@ impl VmResources {
             return Err(BalloonConfigError::TooManyPagesRequested);
         }
 
-        if self.machine_config.huge_pages != HugePageConfig::None {
-            return Err(BalloonConfigError::HugePages);
-        }
-
         self.balloon.set(config)
     }
 

src/vmm/src/vmm_config/balloon.rs

@@ -20,8 +20,6 @@ pub enum BalloonConfigError {
     TooManyPagesRequested,
     /// Error creating the balloon device: {0}
     CreateFailure(crate::devices::virtio::balloon::BalloonError),
-    /// Firecracker's huge pages support is incompatible with memory ballooning.
-    HugePages,
 }
 
 /// This struct represents the strongly typed equivalent of the json body

src/vmm/src/vmm_config/machine_config.rs

@@ -29,8 +29,6 @@ pub enum MachineConfigError {
     SmtNotSupported,
     /// Could not determine host kernel version when checking hugetlbfs compatibility
     KernelVersion,
-    /// Firecracker's huge pages support is incompatible with memory ballooning.
-    BalloonAndHugePages,
 }
 
 /// Describes the possible (huge)page configurations for a microVM's memory.

tests/conftest.py

@@ -34,7 +34,7 @@
 from framework import defs, utils
 from framework.artifacts import disks, kernel_params
 from framework.defs import DEFAULT_BINARY_DIR
-from framework.microvm import MicroVMFactory, SnapshotType
+from framework.microvm import HugePagesConfig, MicroVMFactory, SnapshotType
 from framework.properties import global_props
 from framework.utils_cpu_templates import (
     custom_cpu_templates_params,
@@ -594,6 +594,15 @@ def pci_enabled(request):
     yield request.param
 
 
+@pytest.fixture(
+    params=[HugePagesConfig.NONE, HugePagesConfig.HUGETLBFS_2MB],
+    ids=["NO_HUGE_PAGES", "2M_HUGE_PAGES"],
+)
+def huge_pages(request):
+    """Fixture that allows configuring whether a microVM will have huge pages enabled or not"""
+    yield request.param
+
+
 def uvm_booted(
     microvm_factory,
     guest_kernel,

tests/framework/microvm.py

@@ -23,10 +23,11 @@
 from collections import namedtuple
 from dataclasses import dataclass
 from enum import Enum, auto
-from functools import lru_cache
+from functools import cached_property, lru_cache
 from pathlib import Path
 from typing import Optional
 
+import psutil
 from tenacity import Retrying, retry, stop_after_attempt, wait_fixed
 
 import host_tools.cargo_build as build_tools
@@ -472,7 +473,7 @@ def state(self):
         """Get the InstanceInfo property and return the state field."""
         return self.api.describe.get().json()["state"]
 
-    @property
+    @cached_property
     def firecracker_pid(self):
         """Return Firecracker's PID
 
@@ -491,6 +492,11 @@ def firecracker_pid(self):
             with attempt:
                 return int(self.jailer.pid_file.read_text(encoding="ascii"))
 
+    @cached_property
+    def ps(self):
+        """Returns a handle to the psutil.Process for this VM"""
+        return psutil.Process(self.firecracker_pid)
+
     @property
     def dimensions(self):
         """Gets a default set of cloudwatch dimensions describing the configuration of this microvm"""

tests/framework/utils.py

@@ -14,10 +14,12 @@
 import typing
 from collections import defaultdict, namedtuple
 from contextlib import contextmanager
+from pathlib import Path
 from typing import Dict
 
 import psutil
 import semver
+from packaging import version
 from tenacity import (
     Retrying,
     retry,
@@ -259,6 +261,48 @@ def get_free_mem_ssh(ssh_connection):
     raise Exception("Available memory not found in `/proc/meminfo")
 
 
+def get_stable_rss_mem_by_pid(process, percentage_delta=1):
+    """
+    Get the RSS memory that a guest uses, given the pid of the guest.
+
+    Wait till the fluctuations in RSS drop below percentage_delta.
+    Or print a warning if this does not happen.
+    """
+
+    # All values are reported as KiB
+
+    def get_rss_from_pmap():
+        """Returns current memory utilization in KiB, including used HugeTLBFS"""
+
+        proc_status = Path("/proc", str(process.pid), "status").read_text("utf-8")
+        for line in proc_status.splitlines():
+            if line.startswith("HugetlbPages:"):  # entry is in KiB
+                hugetlbfs_usage = int(line.split()[1])
+                break
+        else:
+            assert False, f"HugetlbPages not found in {str(proc_status)}"
+        return hugetlbfs_usage + process.memory_info().rss // 1024
+
+    first_rss = 0
+    second_rss = 0
+    for _ in range(5):
+        first_rss = get_rss_from_pmap()
+        time.sleep(1)
+        second_rss = get_rss_from_pmap()
+        abs_diff = abs(first_rss - second_rss)
+        abs_delta = abs_diff / first_rss * 100
+        print(
+            f"RSS readings: old: {first_rss} new: {second_rss} abs_diff: {abs_diff} abs_delta: {abs_delta}"
+        )
+        if abs_delta < percentage_delta:
+            return second_rss
+
+        time.sleep(1)
+
+    print("WARNING: RSS readings did not stabilize")
+    return second_rss
+
+
 def _format_output_message(proc, stdout, stderr):
     output_message = f"\n[{proc.pid}] Command:\n{proc.args}"
     # Append stdout/stderr to the output message
@@ -417,6 +461,11 @@ def get_kernel_version(level=2):
     return linux_version
 
 
+def supports_hugetlbfs_discard():
+    """Returns True if the kernel supports hugetlbfs discard"""
+    return version.parse(get_kernel_version()) >= version.parse("5.18.0")
+
+
 def generate_mmds_session_token(
     ssh_connection, ipv4_address, token_ttl, imds_compat=False
 ):