Observed issues in amd/iommu_v2pgmode_test.py test

[AMD-Muthukumar-S]

Describe the bug

• We are aware that this test currently checks for Level 5 paging, even though our system only supports up to Level 4 paging.
• Please add support for Level 4 paging in this test.

Environment

• Platform: Bare Metal (BM) host with Nutanix AHV

[root@NTNX-TITANITE-34B6-A ~]# cat /etc/os-release
NAME="Nutanix AHV"
VERSION="10.3 (enif)"
VERSION_ID="10.3"
VERSION_CODENAME="enif"
ID="ahv"
ID_LIKE="rhel fedora"
PRETTY_NAME="AHV 10.3 (enif)"
ANSI_COLOR="0;35"
HOME_URL=https://www.nutanix.com/
SUPPORT_URL=https://portal.nutanix.com/
VENDOR_NAME="Nutanix"
NUTANIX_AHV_BUILD_ID="422"
NUTANIX_AHV_GIT_HASH="a70725a92"
[root@NTNX-TITANITE-34B6-A ~]# uname -r
6.6.52-10.3s5c15r5.el8.x86_64
[root@NTNX-TITANITE-34B6-A ~]#

Steps to reproduce

• Please check the dmesg log.

My system supports 4 Level page table:

[root@NTNX-TITANITE-34B6-A ~]# dmesg | grep -i "V2 page table enabled"
[    2.308726] AMD-Vi: V2 page table enabled (Paging mode : 4 level)
[root@NTNX-TITANITE-34B6-A ~]#
[root@NTNX-TITANITE-34B6-A ~]# grep "address sizes" /proc/cpuinfo
address sizes   : 46 bits physical, 57 bits virtual
address sizes   : 46 bits physical, 57 bits virtual
address sizes   : 46 bits physical, 57 bits virtual

Expected behavior
self.bits_to_pgmode = {'57': '5', '48': '4', '39': '3', '30': '2', '21': '1'}

Debug print lines added in the test:

def test(self):
        '''
        Test if host page table mode matches with iommu v2 page table mode
        '''
        if check_dmesg('V2 page table enabled'):
            if (cpu.cpu_has_flags(["la57"]) and self.check_kernelconf_5lvl()):
                if check_v2pgtbl_mode("5"):  Dynamic function call  cpu.get_va_bits()
                    self.log.info("Host page table mode (5lvl) match with IOMMU V2 Page mode")
                else:
                    self.fail("Host page table mode (5lvl) does not match with IOMMU V2 Paging mode")
            else:
                self.log.info(f"VA={cpu.get_va_bits()}")  ==> Time stamp= 06:29:12,258
                self.log.info(f"Value1: {self.bits_to_pgmode[cpu.get_va_bits()]}") ==> Time stamp= 06:29:12,260
                self.log.info(f"Value2: {check_v2pgtbl_mode(self.bits_to_pgmode[cpu.get_va_bits()])}") ==> Time stamp= 06:29:12,325

                if check_v2pgtbl_mode(self.bits_to_pgmode[cpu.get_va_bits()]):
                    self.log.info("Host page table mode match with IOMMU V2 Page mode")
                else:
                    self.fail("Host page table mode does does not match with IOMMU V2 Paging mode")
        else:
            self.cancel("IOMMU is in v1 page table")

From Debug log: /root/elves/results/job-2025-07-30T06.29-2bc066f/test-results/1-_root_elves_tests_avocado-misc-tests_io_iommu_amd_iommu_v2pgmode_test.py_IommuPageTable.test/debug.log

[stdlog] 2025-07-30 06:29:12,258 avocado.test iommu_v2pgmode_t L0122 INFO | VA=57
[stdlog] 2025-07-30 06:29:12,260 avocado.test iommu_v2pgmode_t L0123 INFO | Value1: 5
[stdlog] 2025-07-30 06:29:12,261 avocado.utils.process process          L0662 INFO | Running 'dmesg | grep -i "V2 page table enabled (Paging mode : 5 level)"'
[stdlog] 2025-07-30 06:29:12,261 avocado.utils.process process          L0711 INFO | Command dmesg | grep -i "V2 page table enabled (Paging mode : 5 level)" running on a thread
[stdlog] 2025-07-30 06:29:12,276 avocado.utils.process process          L0719 INFO | Command 'dmesg | grep -i "V2 page table enabled (Paging mode : 5 level)"' finished with 1 after 0.014709414s
[stdlog] 2025-07-30 06:29:12,276 avocado.utils.process process          L0662 INFO | Running 'journalctl -k -b | grep -i "V2 page table enabled (Paging mode : 5 level)"'
[stdlog] 2025-07-30 06:29:12,277 avocado.utils.process process          L0711 INFO | Command journalctl -k -b | grep -i "V2 page table enabled (Paging mode : 5 level)" running on a thread
[stdlog] 2025-07-30 06:29:12,325 avocado.utils.process process          L0719 INFO | Command 'journalctl -k -b | grep -i "V2 page table enabled (Paging mode : 5 level)"' finished with 1 after 0.048514932s
[stdlog] 2025-07-30 06:29:12,325 avocado.test iommu_v2pgmode_t L0124 INFO | Value2: False

Investigation:
https://github.com/AMDESE/avocado-misc-tests/blob/AMD_elves/io/iommu/amd/iommu_v2pgmode_test.py#L122

Root Cause:

• This failure is correct as per the current code, but it is actually a false fail.
• check_dmesg is applying the mode as "5" from this function: https://github.com/avocado-framework/avocado/blob/master/avocado/utils/cpu.py#L183

Code Snippet:

def get_va_bits():
    """
    Check for VA address bit size in /proc/cpuinfo

    :return: VA address bit size
    :rtype: str
    """
    cpu_info = genio.read_file("/proc/cpuinfo")
    for line in cpu_info.splitlines():
        if "address sizes" in line:
            return line.split()[-3].strip()    is "57"
    return ""            

• cat proc/cpuinfo Returns:

address sizes   : 46 bits physical, 57 bits virtual

Code Snippet:

def check_v2pgtbl_mode(mode):
    '''
    Check if v2 page table is enabled in "mode" level of paging
    '''
    return check_dmesg(f'V2 page table enabled (Paging mode : {mode} level)')

Causing point:
In Above Function:

• check_dmesg(f'V2 page table enabled (Paging mode : {mode} level)')
• Here expected value should be "4" (Physical address size) based on actual IOMMU log, but we are incorrectly comparing it with “5” (derived from Virtual address size)

Current behavior
In the test:

• check_v2pgtbl_mode(self.bits_to_pgmode[cpu.get_va_bits()])  returned False
• The final test result is FAIL: Host page table mode does does not match with IOMMU V2 Paging mode (0.21 s)

System information

• Platform: Bare Metal (BM)

• OS: Nutanix AOS/AHV

• Kernel: 6.6.52

• Elves tag/branch/version: elves_05_2025 branch]

• Hardware platform: AMD EPYC 9575F 64-Core Processor - Titanite

Additional information

• Debug log:
  Dmesg output:

[root@NTNX-TITANITE-34B6-A ~]# dmesg | grep -i "V2 page table enabled"
[    2.308726] AMD-Vi: V2 page table enabled (Paging mode : 4 level)

Console Log:

[root@NTNX-TITANITE-34B6-A elves]# python3 ./avocado-setup.py --nrunner --run-tests avocado-misc-tests/io/iommu/amd/iommu_v2pgmode_test.py  --no-download
06:29:09 INFO    : Env Config: /root/elves/config/wrapper/env.conf
06:29:09 INFO    : No Run Config: /root/elves/config/wrapper/no_run_tests.conf
06:29:10 INFO    : Check for environment
06:29:10 INFO    : Creating temporary mux dir
06:29:10 INFO    :
06:29:10 INFO    : Running Host Tests Suite dynamic_test_suite_dynamic_test_suite
06:29:10 INFO    : Running: /usr/local/bin/avocado run   /root/elves/tests/avocado-misc-tests/io/iommu/amd/iommu_v2pgmode_test.py --force-job-id 2bc066fb7f6caa6c8ab719388c84cf9aaa4d7515                  --job-results-dir /root/elves/results
JOB ID     : 2bc066fb7f6caa6c8ab719388c84cf9aaa4d7515
JOB LOG    : /root/elves/results/job-2025-07-30T06.29-2bc066f/job.log
(1/1) /root/elves/tests/avocado-misc-tests/io/iommu/amd/iommu_v2pgmode_test.py:IommuPageTable.test: STARTED
(1/1) /root/elves/tests/avocado-misc-tests/io/iommu/amd/iommu_v2pgmode_test.py:IommuPageTable.test: FAIL: Host page table mode does does not match with IOMMU V2 Paging mode (0.21 s)
RESULTS    : PASS 0 | ERROR 0 | FAIL 1 | SKIP 0 | WARN 0 | INTERRUPT 0 | CANCEL 0
JOB HTML   : /root/elves/results/job-2025-07-30T06.29-2bc066f/results.html
JOB TIME   : 192.52 s

[AMD-Muthukumar-S]

Log:

1-_root_elves_tests_avocado-misc-tests_io_iommu_amd_iommu_v2pgmode_test.py_IommuPageTable.test.zip

[dhsrivas]

@AMD-Muthukumar-S Can you please provide the output of the below command?

dmesg | grep "Extended features"

Thanks

[AMD-Muthukumar-S]

Hi @dhsrivas

Please find the output:

[root@NTNX-TITANITE-34B6-A ~]# dmesg | grep -i "Extended features"
[    2.294241] AMD-Vi: Extended features (0xa5bf732fa2295afe, 0x53f): PPR X2APIC NX GT [5] IA GA PC GA_vAPIC SNP
[    2.294246] AMD-Vi: Extended features (0xa5bf732fa2295afe, 0x53f): PPR X2APIC NX GT [5] IA GA PC GA_vAPIC SNP
[    2.294249] AMD-Vi: Extended features (0xa5bf732fa2295afe, 0x53f): PPR X2APIC NX GT [5] IA GA PC GA_vAPIC SNP
[    2.294251] AMD-Vi: Extended features (0xa5bf732fa2295afe, 0x53f): PPR X2APIC NX GT [5] IA GA PC GA_vAPIC SNP
[    2.294252] AMD-Vi: Extended features (0xa5bf732fa2295afe, 0x53f): PPR X2APIC NX GT [5] IA GA PC GA_vAPIC SNP
[    2.294254] AMD-Vi: Extended features (0xa5bf732fa2295afe, 0x53f): PPR X2APIC NX GT [5] IA GA PC GA_vAPIC SNP
[    2.294256] AMD-Vi: Extended features (0xa5bf732fa2295afe, 0x53f): PPR X2APIC NX GT [5] IA GA PC GA_vAPIC SNP
[    2.294258] AMD-Vi: Extended features (0xa5bf732fa2295afe, 0x53f): PPR X2APIC NX GT [5] IA GA PC GA_vAPIC SNP
[root@NTNX-TITANITE-34B6-A ~]

[dhsrivas]

@AMD-Muthukumar-S Could you please test the patch below and check if it resolves the issue?

0001-io-iommu-amd-Use-CPU-la57-flag-and-IOMMU-feature-reg.patch

[dhsrivas]

Hi @AMD-Muthukumar-S Gentle reminder. Could you kindly test the latest patch and confirm if it fixes the issue?

v2-0001-io-iommu-amd-Use-CPU-la57-flag-and-IOMMU-feature-.patch

[AMD-Muthukumar-S]

Hi @dhsrivas

• I validated the patch(v2-0001-io-iommu-amd-Use-CPU-la57-flag-and-IOMMU-feature-.patch) on both Ubuntu 6.8 kernel and AHV 6.6 kernel with Turin CPUs.

• On Ubuntu, 5-level paging was detected and all tests passed successfully.

• On AHV, only 4-level paging is enabled. As a result, the 5-level paging test was skipped, and one test failed with a ./run-tests permission issue.

Request:
Currently, it’s not clearly visible in the test results which paging mode (4-level or 5-level) the system is actually using during execution.

For example, on AHV:
CPUID shows: LA57 = true → CPU supports 5-level paging.
dmesg shows: AMD-Vi: V2 page table enabled (Paging mode : 4 level) → kernel is running in 4-level mode.

Log Snippet:

[root@NTNX-TITANITE1-A ~]# cpuid -1 -l 0x7 -s 0 | grep -i la57
      LA57: 57-bit addrs & 5-level paging      = true
[root@NTNX-TITANITE1-A ~]#  dmesg | grep -i "V2 page table enabled"
[    3.934676] AMD-Vi: V2 page table enabled (Paging mode : 4 level)
[root@NTNX-TITANITE1-A ~]#

Since dmesg can be rotated or cleared, end users may not always have this context when a test is skipped.
Would it be possible to enhance the test output to explicitly log both the CPU capability (from CPUID) and the active paging mode (from kernel/dmesg)? That way, the reason for skipping (e.g., "CPU supports 5-level, but kernel is running in 4-level mode") is clear to customers.

Logs:

Ubuntu with Turin:

Kernel and Paging Mode Logs:

root@amd:~/elves# dmesg | grep -i "V2 page table enabled"
[    1.485328] AMD-Vi: V2 page table enabled (Paging mode : 5 level)

root@amd:~/elves# uname -a
Linux amd 6.8.0-79-generic #79-Ubuntu SMP PREEMPT_DYNAMIC Tue Aug 12 14:42:46 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
root@amd:~/elves#

Git Log:

root@amd:~/elves/tests/avocado-misc-tests# git log --oneline
9aad74e1 (HEAD -> AMD_elves) io/iommu/amd: Use CPU la57 flag and IOMMU feature register for 5-level paging support
a5e14c26 (origin/AMD_elves) Remove 5lvl kernel config check test
9f9b3dd7 Remove check added for more than 255 CPUs and added correct link for msr-tools src
3e0808e2 Fix handling bus lock trap failed case

Execution Log:

04:31:20 INFO    : Running Host Tests Suite AMD_elves_page_table
04:31:20 INFO    : Running: /usr/local/bin/avocado run  /root/elves/tests/avocado-misc-tests/memory/page_table.py --force-job-id 57b59c57d4134f980e33329cdc0e60232070ecc5                  --job-results-dir /root/elves/results   --max-parallel-tasks=1
JOB ID     : 57b59c57d4134f980e33329cdc0e60232070ecc5
JOB LOG    : /root/elves/results/job-2025-09-15T04.31-57b59c5/job.log
 (1/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_CPUID: STARTED
 (1/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_CPUID: PASS (1.09 s)
 (2/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_detect_5lvl: STARTED
 (2/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_detect_5lvl: PASS (0.35 s)
 (3/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_pg_table_tests: STARTED
 (3/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_pg_table_tests: PASS (0.96 s)
RESULTS    : PASS 3 | ERROR 0 | FAIL 0 | SKIP 0 | WARN 0 | INTERRUPT 0 | CANCEL 0
JOB HTML   : /root/elves/results/job-2025-09-15T04.31-57b59c5/results.html
JOB TIME   : 9.63 s

AHV with Turin:

Kernel and Paging Mode Logs:

[root@NTNX-TITANITE1-A ~]# dmesg | grep -i "V2 page table enabled"
[    3.934676] AMD-Vi: V2 page table enabled (Paging mode : 4 level)

[root@NTNX-TITANITE1-A ~]# uname -a
Linux NTNX-TITANITE1-A 6.6.52-10.3.0.1s4c7r5.el8.x86_64 #1 SMP Tue Jul  8 12:00:00 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
[root@NTNX-TITANITE1-A ~]#

Git Log:

[root@NTNX-TITANITE1-A avocado-misc-tests]# git log --oneline
3b57dd73 (HEAD -> AMD_elves) io/iommu/amd: Use CPU la57 flag and IOMMU feature register for 5-level paging support
a5e14c26 (origin/AMD_elves) Remove 5lvl kernel config check test
9f9b3dd7 Remove check added for more than 255 CPUs and added correct link for msr-tools src
3e0808e2 Fix handling bus lock trap failed case

Execution Log:

04:36:56 INFO    : Running Host Tests Suite AMD_elves_page_table
04:36:56 INFO    : Running: /root/myenv/bin/avocado run  /root/elves/tests/avocado-misc-tests/memory/page_table.py --force-job-id 61e9e8e4e865ddf98c7115c29190b6b1ba807f64                  --job-results-dir /root/elves/results   --max-parallel-tasks=1
JOB ID     : 61e9e8e4e865ddf98c7115c29190b6b1ba807f64
JOB LOG    : /root/elves/results/job-2025-09-15T04.36-61e9e8e/job.log
 (1/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_CPUID: STARTED
 (1/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_CPUID: PASS (2.33 s)
 (2/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_detect_5lvl: STARTED
 (2/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_detect_5lvl: CANCEL: 5-Level page table is not supported on the CPU (1.06 s)
 (3/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_pg_table_tests: STARTED
 (3/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_pg_table_tests: ERROR: [Errno 13] Permission denied (./run-tests): './run-tests' (1.10 s)
RESULTS    : PASS 1 | ERROR 1 | FAIL 0 | SKIP 0 | WARN 0 | INTERRUPT 0 | CANCEL 1
JOB HTML   : /root/elves/results/job-2025-09-15T04.36-61e9e8e/results.html
JOB TIME   : 6.87 s

[dhsrivas]

@AMD-Muthukumar-S Just to clarify: the IOMMU and the CPU each maintain their own separate page tables — they are not the same.

The initial issue you reported was with avocado-misc-tests/io/iommu/amd/iommu_v2pgmode_test.py, for which I have already provided a fix. This test validates whether both the CPU and the IOMMU boot with the expected page table levels for their respective page tables.

On the other hand, avocado-misc-tests/memory/page_table.py is a completely different test. It checks whether 5-level CPU page tables are supported and enabled.

The error you’re currently seeing occurs only on AVH, not on Ubuntu:

(3/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_pg_table_tests: ERROR: [Errno 13] Permission denied (./run-tests): './run-tests' (1.10 s)

At present, ELVES does not support AVH. If you encounter the same issue on Ubuntu 24.04, please file a separate issue.

For now, please confirm whether the fix for avocado-misc-tests/io/iommu/amd/iommu_v2pgmode_test.py resolves the original failure.

Thanks
Dheeraj

[AMD-Muthukumar-S]

Hi @dhsrivas,

For now, please confirm whether the fix for avocado-misc-tests/io/iommu/amd/iommu_v2pgmode_test.py resolves the original failure.

Currently, it’s not clearly visible in the test results which paging mode (4-level or 5-level) the system is actually using during execution. If possible, please add the current paging mode information to the test output when it fails. Otherwise, I’m fine with merging your changes to the origin branch, since end users can still cross-check the lscpu flags to verify support.

(2/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_detect_5lvl: CANCEL: 5-Level page table is not supported on the CPU (1.06 s)

Please ignore the ./run-tests permission issue failure.

Thanks,
Muthu

[dhsrivas]

@AMD-Muthukumar-S

Hi @dhsrivas,

For now, please confirm whether the fix for avocado-misc-tests/io/iommu/amd/iommu_v2pgmode_test.py resolves the original failure.

Currently, it’s not clearly visible in the test results which paging mode (4-level or 5-level) the system is actually using during execution. If possible, please add the current paging mode information to the test output when it fails. Otherwise, I’m fine with merging your changes to the origin branch, since end users can still cross-check the lscpu flags to verify support.

Hi @AMD-Muthukumar-S Attached is the updated patch showing current vs. expected IOMMU page table mode on failure. Kindly review and validate; I will merge the fix once confirmed.

v3-0001-io-iommu-amd-Use-CPU-la57-flag-and-IOMMU-feature-.patch

Thanks
Dheeraj

(2/3) /root/elves/tests/avocado-misc-tests/memory/page_table.py:PageTable.test_detect_5lvl: CANCEL: 5-Level page table is not supported on the CPU (1.06 s)

Please ignore the ./run-tests permission issue failure.

Thanks, Muthu