Hello, finally I tried kernel 5.3.0 and unfortunately the same effect: 1. Kernel 5.3.0, i40e driver packed into kernel, NIC X722: load eBPF program in "driver" mode cause kernel NULL pointer dereference with following stack trace:
[20783.656513] BUG: kernel NULL pointer dereference, address: 0000000000000000 [20783.656561] #PF: supervisor read access in kernel mode [20783.656589] #PF: error_code(0x0000) - not-present page [20783.656618] PGD 0 P4D 0 [20783.656636] Oops: 0000 [#1] SMP NOPTI [20783.656659] CPU: 15 PID: 5763 Comm: bpf_program_man Not tainted 5.3.0-19-generic #20-Ubuntu [20783.656703] Hardware name: Supermicro Super Server/X11SDV-12C-TP8F, BIOS 1.1 01/19/2019 [20783.656759] RIP: 0010:i40e_xdp+0xf9/0x1d0 [i40e] [20783.656786] Code: 0c 00 00 45 38 f7 0f 85 bf 00 00 00 66 41 83 bc 24 f6 0c 00 00 00 74 2b 31 c0 49 8b 94 24 90 0c 00 00 49 8b 8c 24 d0 0c 00 00 <48> 8b 14 c2 48 83 c0 01 48 89 4a 20 41 0f b7 94 24 f6 0c 00 00 39 [20783.656897] RSP: 0018:ffffa98c80c03880 EFLAGS: 00010246 [20783.656927] RAX: 0000000000000000 RBX: ffffffffc03ec601 RCX: ffffa98c802cb000 [20783.656965] RDX: 0000000000000000 RSI: 0000000000000092 RDI: ffff9b533fdd7440 [20783.657003] RBP: ffffa98c80c038b0 R08: 0000000000000610 R09: 0000000000000004 [20783.657040] R10: 0000000000000000 R11: 0000000000000001 R12: ffff9b53247c1000 [20783.657078] R13: 0000000000000000 R14: ffffa98c802cb001 R15: ffffffffc03c0e00 [20783.657116] FS: 00007f51b8009f80(0000) GS:ffff9b533fdc0000(0000) knlGS:0000000000000000 [20783.657160] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [20783.657191] CR2: 0000000000000000 CR3: 0000001009ff2002 CR4: 00000000007626e0 [20783.657228] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [20783.657266] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 [20783.657303] PKRU: 55555554 [20783.657320] Call Trace: [20783.657351] ? i40e_reconfig_rss_queues+0x140/0x140 [i40e] [20783.657385] dev_xdp_install+0x4e/0x70 [20783.657409] dev_change_xdp_fd+0xd9/0x240 [20783.657433] do_setlink+0x944/0x980 .... .... 2. Kernel 5.3.0, latest driver 2.10.19.30 from Intel page, NIC X722: eBPF program loads fine but attempt to use zero copy mode on AF_XDP socket ends with errno 95 - "operation not supported". Any idea how can try to identify the root cause and help the community to fix this problem? Or maybe way of using zero copy has changed in 2.9.x and 2.10.x versions and this is not a bug but feature? Previously I had similar problem with XVV710 NIC - 2.8.x version works fine but 2.9.x - "operation not supported" trying to use XDP. Best regards, Andriy Korud ----------------------------------------------------------------------------------------- FALCON V SYSTEMS S.A. (Joint-Stock Company) incorporated under the laws of Poland Registered offices at Krzemowa 6 Street, Gdynia, Poland (postal code: 81-577) Entered into the National Court Register under KRS no. 0000740920 Documents of the Company are filed at the District Court in Gdansk (postal code: 80-126) Tax ID no./EU VAT no.: PL 9571107871 Niniejsza wiadomość, wraz z załącznikami, jest przeznaczona wyłącznie dla wskazanego w niej odbiorcy (odbiorców) i może zawierać informacje poufne, tajemnicę przedsiębiorstwa, informacje zastrzeżone lub niepodlegające ujawnieniu zgodnie z właściwymi przepisami. Bezwzględnie zabrania się nieuprawnionego wykorzystywania, rozpowszechniania lub kopiowania niniejszej wiadomości, w tym załączników. Jeżeli otrzymali Państwo tę wiadomość omyłkowo, a nie są Państwo jej adresatem, prosimy o niezwłoczne zawiadomienie o tym fakcie nadawcy oraz o usunięcie wiadomości wraz z załącznikami z Państwa komputera. Dziękujemy. This e-mail message, including attachments, is confidential, is intended only for the named recipient(s) above and may contain information that is privileged, trade secrets, proprietary or exempt from disclosure under applicable law. The unauthorized use, dissemination, distribution or reproduction of this e-mail message, including attachments, is strictly prohibited. If you have received this message in error, or are not an intended recipient, please immediately notify the sender and delete this e-mail message, including attachments, from your computer. Thank you.-----------------------------------------------------------------------------------------________________________________________ From: Jeff Kirsher <jeffrey.t.kirs...@intel.com> Sent: Wednesday, September 18, 2019 18:44 To: Andriy Korud; e1000-de...@lists.sf.net Cc: Björn Töpel; magnus.karlsson Subject: Re: [E1000-devel] AF_XDP zero copy not supported in 2.9.21 driver On Wed, 2019-09-18 at 11:27 +0000, Andriy Korud wrote: > Hi, > > we are developing system heavy using eBPF programs and AF_XDP and > noticed something that looks like regression. > > > In my setup I have system with 2 NICs: XXV710 and X722, kernel 5.2.1. > > > Driver version 2.8.20-k (packed into kernel): > > XXV710 - works fine with AF_XDP in zero copy mode; > > X722 - crash on loading eBPF program (seems to be known bug AFAIK). > > > Driver version 2.9.21 (latest available from intel.com): > > XXV710 - "operation not supported" error when initializing AF_XDP > socket in zero copy mode; > > X722 - the same, "operation not supported" error when initializing > AF_XDP socket in zero copy mode. > > > And my final goal is to run X722 in AF_XDP zero copy which seems hard > to achieve now. > > > Any points, ideas, thing to check or debug? - any help is very > valuable. Could you try the recently released 5.3 kernel? There were some bug fixes in that kernel release for AF_XDP for both the i40e and ixgbe driver. _______________________________________________ E1000-devel mailing list E1000-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/e1000-devel To learn more about Intel Ethernet, visit https://forums.intel.com/s/topic/0TO0P00000018NbWAI/intel-ethernet
