Re: Panic on mlx5en.

On 2020-06-15 11:12, Santiago Martinez wrote:
> Hi everyone, while doing some tests for an MRSAS panic I hit another one 
> on mlx5en.
> 
> The device is a LenovoSR655 with 2xMellanox NIC.
> 
> 1 - Mellanox CX-4 Lx 10/25GbE SFP28 2-port OCP Ethernet Adapter
> 2 - Mellanox CX-4 Lx 10/25GbE SFP28 2-port PCIe Ethernet Adapter
> 
> This only happens only with current, tried different snapshots and the 
> same in all.
> On 12.1 works without a problem.
> 
> Trace (please not that is OCR+ manual corrections):
> 
> Tracing pid 2288 tid 182178 td _at_8xfffffeB385fe6500
> kdb_enter() at kbd_inter+0x37/frame 0xfffffe0386030ba0
> vpanic() at vpanic+0x19e/frame 0xfffffe0386030bf0
> panic() at panic+0x43/frame 0xfffffe038630c50
> trap_fatal() at trap_fatal+0x387/frame 0xfffffe0386030cb0
> trap() at trap+0x80/frame Bxfffffe0386030dc0
> calltrap() at calltrap+0x80/frame Bxfffffed386830dc0
> --- trap 0x9, rip = 0xfffffff8275c060, rsp = 0xfffffe0386030e90, rbp = 
> 0xfffffe0386030e90 ---
> linux_root_RB_INSERT_COLOR() at linux_root_RB_INSERT COLOR+0x40/frame 
> 0xfffffe0386030f60
> give_pages() at give pages+0x163/frame 0xfffffe0386830f20
> mlx5_satisfy_startup_pages() at mlx5_satisfy_startup_pages+0x76/frame 
> 0xfffffe0386030f60
> mlx5_load_one () at mlx5_load_one+0x6b7/frame 0xfffffe0386031080
> init_one() at init_one+0x12d5/frame 0xfffffe03860310f0
> linux_pci_attach_device() at linux_pci_attach device+0x573/frame 
> 0xfffffe0386031150
> device_attach() at device_attach+0x3ca/frame 0xfffffe0386031190
> device_probe_and_attach() at device_probe_and_attach+0x70/frame 
> 0xfffffe03860311c0
> pci_driver_added() at pci_driver_added+0xf6/frame 0xfffffe0386031200
> devclass_driver_added() at devclass_driver_added+0x39/frame 
> 0xfffffe0386031240
> devclass_add_driver() at devclass_add_driver+0x147/frame 0xfffffe0386031280
> _linux_pci_register_driver() at _linux_pci_register_driver+0xc9/frame 
> 0xfffffe03860312a0
> 

Are you using the latest version of kernel & mlx5en as of today? There 
was a regression issue with the rbtree.h implementation which recently 
was fixed.

--HPS