[DPDK/ethdev Bug 2031] cn10k: NPA_POOL_ERR_INT_RANGE on Pool 1 during mbuf pool creation in testpmd

[DPDK/ethdev Bug 2031] cn10k: NPA_POOL_ERR_INT_RANGE on Pool 1 during mbuf pool creation in testpmd

[bugzilla]

https://bugs.dpdk.org/show_bug.cgi?id=2031

            Bug ID: 2031
           Summary: cn10k: NPA_POOL_ERR_INT_RANGE on Pool 1 during mbuf
                    pool creation in testpmd
           Product: DPDK
           Version: 24.11
          Hardware: ARM
                OS: Linux
            Status: UNCONFIRMED
          Severity: critical
          Priority: Normal
         Component: ethdev
          Assignee: dev@dpdk.org
          Reporter: amiyaranjan.mohakud@gmail.com
  Target Milestone: ---

Created attachment 390
  --> https://bugs.dpdk.org/attachment.cgi?id=390&action=edit
testpmd-cn10k-mempool-issue.txt

Environment:
------------

  - SoC: Marvell Octeon CN10K (cn10kb_a0)
  - Kernel: 6.12.68
  - DPDK: 24.11.2
  - IOVA mode: PA
  - VFIO: no-iommu mode
  - Hugepages: 1024 x 2MB (2048 MB total)
  - Mempool ops: cn10k_mempool_ops

Description:
--------------
testpmd creates mbuf pool mb_pool_0 (n=203456, size=2304) using
cn10k_mempool_ops. During pool populate, Pool 1 triggers NPA_POOL_ERR_INT_RANGE
repeatedly, followed by Err_intr=0x1000:

  testpmd: create a new mbuf pool <mb_pool_0>: n=203456, size=2304, socket=0
  testpmd: preferred mempool ops selected: cn10k_mempool_ops

  CNXK: npa_q_irq():150 queue_intr=0x1 qintx=1
  CNXK: npa_q_irq():165 Pool=1 NPA_POOL_ERR_INT_RANGE
    (repeats 4 times during mbuf pool populate)

  Configuring Port 0 (socket 0)
  CNXK: npa_err_irq():18 Err_intr=0x1000
  CNXK: npa_err_irq():18 Err_intr=0x1000

  Pool 1 context at time of error:
  ptr_start:    0x24808d200
  ptr_end:      0x267200000
  buf_size:     20 (units of 128 bytes = 2560 bytes)
  buf_offset:   1
  count:        decreasing across dumps (0x31aa6 → 0x31a91 → 0x31a79 → 0)
  op_pc:        increasing (0x1a → 0x2f → 0x47 → 0x31ac0)
  stack_pages:  0 (remains zero across all dumps)

Notable observations:
----------------------
  - stack_pages remains 0 throughout, even as op_pc increases — suggesting
buffers are being freed but the stack is not being updated
  - count decreases to 0 by the final dump, indicating the pool is fully
drained
  - The address range [0x24808d200, 0x267200000) spans ~512 MB, but the mempool
requests ~203K buffers of 2304 bytes (~447 MB)


Expected Result
-------------------
No NPA pool range errors. Pool 1 should initialize and populate without
triggering NPA_POOL_ERR_INT_RANGE or npa_err_irq interrupts.

-- 
You are receiving this mail because:
You are the assignee for the bug.

[DPDK/ethdev Bug 2031] cn10k: NPA_POOL_ERR_INT_RANGE on Pool 1 during mbuf pool creation in testpmd

[bugzilla]

https://bugs.dpdk.org/show_bug.cgi?id=2031

Ashwin Sekhar T K (asekhar@marvell.com) changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |asekhar@marvell.com
             Status|CONFIRMED                   |RESOLVED
         Resolution|---                         |INVALID

--- Comment #5 from Ashwin Sekhar T K (asekhar@marvell.com) ---
PCI_BUS:   Expecting 'VA' IOVA mode but current mode is 'PA', not initializing
EAL: Failed to probe device 0002:01:01.7
PCI_BUS:   Expecting 'VA' IOVA mode but current mode is 'PA', not initializing
EAL: Failed to probe device 0002:01:02.0
PCI_BUS:   Expecting 'VA' IOVA mode but current mode is 'PA', not initializing
EAL: Failed to probe device 0002:02:00.0
PCI_BUS:   Expecting 'VA' IOVA mode but current mode is 'PA', not initializing
EAL: Failed to probe device 0002:03:00.0
PCI_BUS:   Expecting 'VA' IOVA mode but current mode is 'PA', not initializing

These prints should be already coming. No further checks are required.

@Amiya, 
It is not a supported mode. This means any address that you are configuring
(whether it be contexts or stacks or pointers) will run into all sorts of
issues.

You may print the pointers that you are trying to free to understand why the
Range error happened. Mostly probably there is a difference in the range
configured and the pointers being freed.

-- 
You are receiving this mail because:
You are the assignee for the bug.