Merge pull request #4772 from natali-rs1985/T7887

T7887: system_option: Add validation for memory hugepage-count

5 files changed, 91 insertions, 23 deletions

diff --git a/smoketest/config-tests/vpp b/smoketest/config-tests/vpp
index e42568965..12ea24884 100644
--- a/smoketest/config-tests/vpp
+++ b/smoketest/config-tests/vpp
@@ -15,7 +15,7 @@ set system console device ttyS0 speed '115200'
 set system host-name 'r16'
 set system login user vyos authentication encrypted-password '$6$rounds=656000$FZBlGpWsmSrV2Rvq$YPNAPtk4k6u99FAMxR6cw4DUPCgOomwCgRZRSO5rAoJK8RlMSCkVAFVF3ozL/3mZMfxcuCnwvd5HX6f9V5KzO.'
 set system name-server '203.0.113.1'
-set system option kernel memory hugepage-size 2M hugepage-count '2048'
+set system option kernel memory hugepage-size 2M hugepage-count '1800'
 set system option time-format '24-hour'
 set system syslog local facility all level 'info'
 set system syslog local facility local7 level 'debug'
diff --git a/smoketest/configs/vpp b/smoketest/configs/vpp
index 472c2919b..7e82d04ee 100644
--- a/smoketest/configs/vpp
+++ b/smoketest/configs/vpp
@@ -45,7 +45,7 @@ system {
         kernel {
             memory {
                 hugepage-size 2M {
-                    hugepage-count "2048"
+                    hugepage-count "1800"
                 }
             }
         }
diff --git a/smoketest/scripts/cli/test_system_option.py b/smoketest/scripts/cli/test_system_option.py
index 4fb318fd8..48935f555 100755
--- a/smoketest/scripts/cli/test_system_option.py
+++ b/smoketest/scripts/cli/test_system_option.py
@@ -105,11 +105,6 @@ class TestSystemOption(VyOSUnitTestSHIM.TestCase):
         isolate_cpus = '1,2,3'
         nohz_full = '2'
         rcu_no_cbs = '1,2,4-5'
-        default_hp_size = '2M'
-        hp_size_1g = '1G'
-        hp_size_2m = '2M'
-        hp_count_1g = '2'
-        hp_count_2m = '512'
 
         self.cli_set(['system', 'option', 'kernel', 'cpu', 'disable-nmi-watchdog'])
         self.cli_set(['system', 'option', 'kernel', 'cpu', 'isolate-cpus', isolate_cpus])
@@ -121,9 +116,6 @@ class TestSystemOption(VyOSUnitTestSHIM.TestCase):
         self.cli_set(['system', 'option', 'kernel', 'disable-power-saving'])
         self.cli_set(['system', 'option', 'kernel', 'disable-softlockup'])
         self.cli_set(['system', 'option', 'kernel', 'memory', 'disable-numa-balancing'])
-        self.cli_set(['system', 'option', 'kernel', 'memory', 'default-hugepage-size', default_hp_size])
-        self.cli_set(['system', 'option', 'kernel', 'memory', 'hugepage-size', hp_size_1g, 'hugepage-count', hp_count_1g])
-        self.cli_set(['system', 'option', 'kernel', 'memory', 'hugepage-size', hp_size_2m, 'hugepage-count', hp_count_2m])
         self.cli_set(['system', 'option', 'kernel', 'quiet'])
 
         self.cli_set(['system', 'option', 'kernel', 'amd-pstate-driver', amd_pstate_mode])
@@ -147,9 +139,6 @@ class TestSystemOption(VyOSUnitTestSHIM.TestCase):
         self.assertIn(f' isolcpus={isolate_cpus}', tmp)
         self.assertIn(f' nohz_full={nohz_full}', tmp)
         self.assertIn(f' rcu_nocbs={rcu_no_cbs}', tmp)
-        self.assertIn(f' default_hugepagesz={default_hp_size}', tmp)
-        self.assertIn(f' hugepagesz={hp_size_1g} hugepages={hp_count_1g}', tmp)
-        self.assertIn(f' hugepagesz={hp_size_2m} hugepages={hp_count_2m}', tmp)
         self.assertIn(' numa_balancing=disable', tmp)
 
         if cpu_vendor == 'AuthenticAMD':
diff --git a/smoketest/scripts/cli/test_vpp.py b/smoketest/scripts/cli/test_vpp.py
index 8cc2e86e7..4738db485 100755
--- a/smoketest/scripts/cli/test_vpp.py
+++ b/smoketest/scripts/cli/test_vpp.py
@@ -31,6 +31,7 @@ from vyos.utils.process import process_named_running
 from vyos.utils.file import read_file
 from vyos.utils.process import rc_cmd
 from vyos.utils.system import sysctl_read
+from vyos.system import image
 
 PROCESS_NAME = 'vpp_main'
 VPP_CONF = '/run/vpp/vpp.conf'
@@ -1477,6 +1478,39 @@ class TestVPP(VyOSUnitTestSHIM.TestCase):
         self.cli_delete(['interfaces', 'ethernet', interface, 'vif'])
         self.cli_commit()
 
+    def test_20_kernel_options_hugepages(self):
+        default_hp_size = '2M'
+        hp_size_1g = '1G'
+        hp_size_2m = '2M'
+        hp_count_1g = '2'
+        hp_count_2m = '512'
+        memory_path = ['system', 'option', 'kernel', 'memory']
+
+        self.cli_set(memory_path + ['default-hugepage-size', default_hp_size])
+        self.cli_set(
+            memory_path + ['hugepage-size', hp_size_2m, 'hugepage-count', hp_count_2m]
+        )
+        self.cli_set(
+            memory_path + ['hugepage-size', hp_size_1g, 'hugepage-count', '2000']
+        )
+        # very big number of 1G hugepages, not enough memory for configuring them
+        # expect raise ConfigError
+        with self.assertRaises(ConfigSessionError):
+            self.cli_commit()
+
+        self.cli_set(
+            memory_path + ['hugepage-size', hp_size_1g, 'hugepage-count', hp_count_1g]
+        )
+        self.cli_commit()
+
+        # Read GRUB config file for current running image
+        tmp = read_file(
+            f'{image.grub.GRUB_DIR_VYOS_VERS}/{image.get_running_image()}.cfg'
+        )
+        self.assertIn(f' default_hugepagesz={default_hp_size}', tmp)
+        self.assertIn(f' hugepagesz={hp_size_1g} hugepages={hp_count_1g}', tmp)
+        self.assertIn(f' hugepagesz={hp_size_2m} hugepages={hp_count_2m}', tmp)
+
 
 if __name__ == '__main__':
     unittest.main(verbosity=2, failfast=VyOSUnitTestSHIM.TestCase.debug_on())
diff --git a/src/conf_mode/system_option.py b/src/conf_mode/system_option.py
index cb3d6ce9b..a0b3dd72c 100755
--- a/src/conf_mode/system_option.py
+++ b/src/conf_mode/system_option.py
@@ -15,6 +15,7 @@
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 
 import os
+import psutil
 
 from sys import exit
 from time import sleep
@@ -39,6 +40,8 @@ from vyos.configdep import call_dependents
 from vyos import ConfigError
 from vyos import airbag
 
+from vyos.vpp.config_resource_checks import memory as mem_check
+
 airbag.enable()
 
 curlrc_config = r'/etc/curlrc'
@@ -56,6 +59,21 @@ tuned_profiles = {
 }
 
 
+def _get_total_hugepages_and_memory(config):
+    unit_map = {'M': 1 << 20, 'G': 1 << 30}
+
+    total_pages = 0
+    total_bytes = 0
+
+    hp_sizes = config.get('hugepage_size', {})
+    for size_str, hp_config in hp_sizes.items():
+        pages = int(hp_config.get('hugepage_count', 0))
+        total_pages += pages
+        total_bytes += pages * int(size_str[:-1]) * unit_map[size_str[-1]]
+
+    return total_pages, total_bytes
+
+
 def get_config(config=None):
     if config:
         conf = config
@@ -116,6 +134,40 @@ def verify(options):
                 f'AMD pstate driver cannot be used with "{cpu_vendor}" CPU!'
             )
 
+        _, hp_memory_bytes = _get_total_hugepages_and_memory(
+            options['kernel'].get('memory', {})
+        )
+        if hp_memory_bytes:
+            memory = psutil.virtual_memory()
+            memory_total_bytes = memory.total
+
+            # Exclude hugepage usage from system "used" memory
+            hp_memory_used = sum(
+                p['memory'] for p in mem_check.get_hugepages_info().values()
+            )
+            memory_used_bytes = memory.used - hp_memory_used
+
+            # TODO: need to calculate how much memory is consumed for other services, tmpfs etc.
+            # for now we should leave at least 4 GB for system usage and other processes
+            min_system_reserved_gd = 4
+            memory_margin_gb = 1
+            reserved_bytes = max(
+                min_system_reserved_gd * 1024**3,
+                memory_used_bytes + memory_margin_gb * 1024**3,
+            )
+
+            available_for_hp_bytes = memory_total_bytes - reserved_bytes
+            if available_for_hp_bytes < hp_memory_bytes:
+                # For the error message, convert to GB and round to 1 decimal
+                hp_memory_gb = round(hp_memory_bytes / 1024**3, 1)
+                available_for_hp_gb = max(0, round(available_for_hp_bytes / 1024**3, 1))
+                reserved_gb = round(reserved_bytes / 1024**3, 1)
+                raise ConfigError(
+                    f'Configured hugepages require {hp_memory_gb} GB of memory, but only '
+                    f'{available_for_hp_gb:.1f} GB is available '
+                    f'({reserved_gb} GB is reserved for system usage and services)'
+                )
+
     return None
 
 
@@ -269,16 +321,9 @@ def apply(options):
             write_file(kernel_dynamic_debug, f'module {module} -p')
 
     if 'resource_limits' in options:
-        unit_map = {'M': 1 << 20, 'G': 1 << 30}
-
-        total_pages = 0
-        total_bytes = 0
-
-        hp_sizes = options.get('kernel', {}).get('memory', {}).get('hugepage_size', {})
-        for size_str, hp_config in hp_sizes.items():
-            pages = int(hp_config.get('hugepage_count', 0))
-            total_pages += pages
-            total_bytes += pages * int(size_str[:-1]) * unit_map[size_str[-1]]
+        total_pages, total_bytes = _get_total_hugepages_and_memory(
+            options.get('kernel', {}).get('memory', {})
+        )
 
         # Minimum recommended system values
         max_map_count_min = 65530  # ensures large workload compatibility