[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: xen-swiotlb issue when NVMe driver is enabled in Dom0 on ARM


  • To: Stefano Stabellini <sstabellini@xxxxxxxxxx>
  • From: Rahul Singh <Rahul.Singh@xxxxxxx>
  • Date: Thu, 14 Apr 2022 17:44:24 +0000
  • Accept-language: en-US
  • Arc-authentication-results: i=1; mx.microsoft.com 1; spf=pass smtp.mailfrom=arm.com; dmarc=pass action=none header.from=arm.com; dkim=pass header.d=arm.com; arc=none
  • Arc-message-signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=microsoft.com; s=arcselector9901; h=From:Date:Subject:Message-ID:Content-Type:MIME-Version:X-MS-Exchange-AntiSpam-MessageData-ChunkCount:X-MS-Exchange-AntiSpam-MessageData-0:X-MS-Exchange-AntiSpam-MessageData-1; bh=rbQpKhI45P+wbcg885XGYSReiuO27OvDqIZBFs70wAw=; b=KzFGfP6HjrHxOrC2QvXjqEtUrNc/tMdbzmzpr+ZhMAANr2swM76hwJrjGIgQYYaDKsnbdFZO5LTRQZ/+5jBOg17eJA9+FSZEsHiRgoNPIch+kjvR5BNf4Nv9YXzuLccP/Wr9yTG8SaNCeA6rCJmYe+h17KC1V/YpMYrLLGaIq4M3tuF+dbf0UyMk5FjTtXw6SSwp06R6yeBRdY2jD38TB8MlFvJ7usLJZ911YQ5GwFuVkX238LMVJMHZ+aulzt5Lu0cE7YxW2HABX+hb0AQO12ngjJTL2SDKUcoQ4YXbzQnIXNVQ3xgctCfjYXDcVh9g6ysx7mnlVCd3N4jMpXduhw==
  • Arc-seal: i=1; a=rsa-sha256; s=arcselector9901; d=microsoft.com; cv=none; b=OEQl0xQSo699SdBi20U3MdGQUKd5jYeQq/GJthL5QlLbblc+bA4Au22PPFQD9wWz7r+YherYkb77Lbyp1yeSmeTqOVYhmc0Sys7MnQLtleTBppqljKq9Uxu60JqxflVaicCzo/rOuXJCbjiDA98bGONYuPIfXHDeDeBkt2tLi+cosDS9S3WvNYjUDYAqh45e6xHSqM36Y2H8izMdLlGoYpKg39O5CN+db95VInqht+B4fxX7nZ26w/wZqoewcii9x950cnx4aeNdf7Ws14k06HiKnGkMQOBN6CCxhPU1RJI9iQB6rq5s1yT68mXlq3KSNDRQ7WTHGUpmrt4eZSivgw==
  • Authentication-results-original: dkim=none (message not signed) header.d=none;dmarc=none action=none header.from=arm.com;
  • Cc: xen-devel <xen-devel@xxxxxxxxxxxxxxxxxxxx>, Bertrand Marquis <Bertrand.Marquis@xxxxxxx>, Julien Grall <julien@xxxxxxx>, Jan Beulich <jbeulich@xxxxxxxx>, "jgross@xxxxxxxx" <jgross@xxxxxxxx>, "boris.ostrovsky@xxxxxxxxxx" <boris.ostrovsky@xxxxxxxxxx>
  • Delivery-date: Thu, 14 Apr 2022 18:10:09 +0000
  • List-id: Xen developer discussion <xen-devel.lists.xenproject.org>
  • Nodisclaimer: true
  • Original-authentication-results: dkim=none (message not signed) header.d=none;dmarc=none action=none header.from=arm.com;
  • Thread-index: AQHYTzcUggEUWxM2006iot1qnYxlIKzuW0oAgAFU1AA=
  • Thread-topic: xen-swiotlb issue when NVMe driver is enabled in Dom0 on ARM

Hi Stefano,

> On 13 Apr 2022, at 10:24 pm, Stefano Stabellini <sstabellini@xxxxxxxxxx> wrote:
>
> On Wed, 13 Apr 2022, Rahul Singh wrote:
>> Hello All,
>>
>> We are trying to boot the Xen 4.15.1 and dom0 Linux Kernel (5.10.27-ampere-lts-standard from [1] ) on Ampere Altra / AVA Developer Platform
>> [2] with ACPI.
>>
>> NVMe storage is connected to PCIe. Native Linux kernel boot fine and also I am able to detect and access NVMe storage.
>> However, during XEN boot when NVME driver is requesting the DMA buffer we are observing the Oops with XEN.
>
> Hi Rahul,
>
> Thanks for the bug report. More comments below.
>
>
>
>> Please find the attached detail logs for Xen and dom0 booting.
>>
>> Snip from logs:
>> (XEN) d0v0: vGICR: SGI: unhandled word write 0x000000ffffffff to ICACTIVER0
>> [  0.000000] Booting Linux on physical CPU 0x0000000000 [0x413fd0c1]
>> [  0.000000] Linux version 5.10.27-ampere-lts-standard (oe-user@oe-host) (aarch64-poky-linux-gcc (GCC) 11.2.0, GNU ld (GNU Binutils)
>> 2.37.20210721) #1 SMP PREEMPT Sat Sep 18 06:01:59 UTC 2021
>> [  0.000000] Xen XEN_VERSION.XEN_SUBVERSION support found
>> [  0.000000] efi: EFI v2.50 by Xen
>> [  0.000000] efi: ACPI 2.0=0x807f66cece8
>> [  0.000000] ACPI: Early table checksum verification disabled
>> [  0.000000] ACPI: RSDP 0x00000807F66CECE8 000024 (v02 Ampere)
>> [  0.000000] ACPI: XSDT 0x00000807F66CEC38 0000AC (v01 Ampere Altra  00000002 AMP. 01000013)
>> [  0.000000] ACPI: FACP 0x00000807F66CE000 000114 (v06 Ampere Altra  00000002 AMP. 01000013)
>> [  0.000000] ACPI: DSDT 0x00000807F8DB0018 02C19E (v02 Ampere Jade  00000001 INTL 20201217)
>> [  0.000000] ACPI: BERT 0x00000807FA0DFF98 000030 (v01 Ampere Altra  00000002 AMP. 01000013)
>> [  0.000000] ACPI: DBG2 0x00000807FA0DFA98 00005C (v00 Ampere Altra  00000002 AMP. 01000013)
>> [  0.000000] ACPI: GTDT 0x00000807FA0DE998 000110 (v03 Ampere Altra  00000002 AMP. 01000013)
>> [  0.000000] ACPI: SPCR 0x00000807FA0DFE18 000050 (v02 Ampere Altra  00000002 AMP. 01000013)
>> [  0.000000] ACPI: EINJ 0x00000807FA0DF598 000150 (v01 Ampere Altra  00000001 INTL 20201217)
>> [  0.000000] ACPI: HEST 0x00000807FA0DEB18 0001F4 (v01 Ampere Altra  00000001 INTL 20201217)
>> [  0.000000] ACPI: SSDT 0x00000807FA0DFA18 00002D (v02 Ampere Altra  00000001 INTL 20201217)
>> [  0.000000] ACPI: TPM2 0x00000807FA0DFD18 00004C (v04 Ampere Altra  00000002 AMP. 01000013)
>> [  0.000000] ACPI: MCFG 0x00000807FA0DF718 00007C (v01 Ampere Altra  00000001 AMP. 01000013)
>> [  0.000000] ACPI: IORT 0x00000807FA0DEF18 0003DC (v00 Ampere Altra  00000002 AMP. 01000013)
>> [  0.000000] ACPI: APIC 0x00000807F66CE118 000AF4 (v05 Ampere Altra  00000002 AMP. 01000013)
>> [  0.000000] ACPI: PPTT 0x00000807FA0D8618 004520 (v02 Ampere Altra  00000002 AMP. 01000013)
>> [  0.000000] ACPI: SLIT 0x00000807FA0DFD98 00002D (v01 Ampere Altra  00000002 AMP. 01000013)
>> [  0.000000] ACPI: SRAT 0x00000807FA0DCE18 000370 (v03 Ampere Altra  00000002 AMP. 01000013)
>> [  0.000000] ACPI: PCCT 0x00000807FA0DE318 000576 (v02 Ampere Altra  00000002 AMP. 01000013)
>> [  0.000000] ACPI: STAO 0x00000807F66CEC10 000025 (v01 Ampere Altra  00000002 AMP. 01000013)
>> [  0.000000] ACPI: SPCR: console: pl011,mmio32,0x100002600000,115200
>> [  0.000000] ACPI: SRAT: Node 0 PXM 0 [mem 0x88300000-0x883fffff]
>> [  0.000000] ACPI: SRAT: Node 0 PXM 0 [mem 0x90000000-0xffffffff]
>> [  0.000000] ACPI: SRAT: Node 0 PXM 0 [mem 0x80000000000-0x8007fffffff]
>> [  0.000000] ACPI: SRAT: Node 0 PXM 0 [mem 0x80100000000-0x807ffffffff]
>> [  0.000000] NUMA: NODE_DATA [mem 0x8079fbf5e00-0x8079fbf7fff]
>> [  0.000000] Zone ranges:
>> [  0.000000]  DMA  [mem 0x0000000098000000-0x00000000ffffffff]
>> [  0.000000]  DMA32  empty
>> [  0.000000]  Normal  [mem 0x0000000100000000-0x00000807fa0dffff]
>> [  0.000000] Movable zone start for each node
>> [  0.000000] Early memory node ranges
>> ….
>>
>> [  0.000000] Dentry cache hash table entries: 262144 (order: 9, 2097152 bytes, linear)
>> [  0.000000] Inode-cache hash table entries: 131072 (order: 8, 1048576 bytes, linear)
>> [  0.000000] mem auto-init: stack:off, heap alloc:off, heap free:off
>> [  0.000000] software IO TLB: mapped [mem 0x00000000f4000000-0x00000000f8000000] (64MB)
>> [  0.000000] Memory: 1929152K/2097412K available (13568K kernel code, 1996K rwdata, 3476K rodata, 4160K init, 822K bss, 168260K reserved,
>> 0K cma-reserved)
>> [  0.000000] SLUB: HWalign=64, Order=0-3, MinObjects=0, CPUs=32, Nodes=1
>> [  0.000000] ftrace: allocating 41306 entries in 162 pages
>> ….
>>
>> ….
>> [  12.599484] loop: module loaded
>> [  12.603160] nvme nvme0: pci function 0005:04:00.0
>> [  12.608129] igb: Intel(R) Gigabit Ethernet Network Driver
>> [  12.613495] igb: Copyright (c) 2007-2014 Intel Corporation.
>> [  12.613636] nvme nvme0: missing or invalid SUBNQN field.
>> [  12.625941] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000008
>> [  12.634726] Mem abort info:
>> [  12.637520]  ESR = 0x96000044
>> [  12.640646]  EC = 0x25: DABT (current EL), IL = 32 bits
>> [  12.646055]  SET = 0, FnV = 0
>> [  12.649153]  EA = 0, S1PTW = 0
>> [  12.652365] Data abort info:
>> [  12.655314]  ISV = 0, ISS = 0x00000044
>> [  12.659231]  CM = 0, WnR = 1
>> [  12.662260] [0000000000000008] user address but active_mm is swapper
>> [  12.668724] Internal error: Oops: 96000044 [#1] PREEMPT SMP
>> [  12.674358] Modules linked in:
>> [  12.677455] CPU: 0 PID: 7 Comm: kworker/u64:0 Tainted: G  W  5.10.27-ampere-lts-standard #1
>> [  12.687083] Workqueue: nvme-reset-wq nvme_reset_work
>> [  12.692059] pstate: 60c00085 (nZCv daIf +PAN +UAO -TCO BTYPE=--)
>> [  12.698149] pc : steal_suitable_fallback+0x138/0x2f0
>> [  12.703170] lr : steal_suitable_fallback+0x1bc/0x2f0
>> [  12.708203] sp : ffff80001196b820
>> [  12.711569] x29: ffff80001196b820 x28: 0000000000000000
>> [  12.716975] x27: 0000000000000000 x26: ffff8000114dbcb0
>> [  12.722357] x25: fffffdffffe00000 x24: 0000000000000001
>> [  12.727740] x23: 0000000000000000 x22: fffffe201bf60000
>> [  12.733120] x21: ffff08071fbf6980 x20: 0000000000000901
>> [  12.738502] x19: 0000000000080000 x18: ffffffffffffffff
>> [  12.743884] x17: 0000000000000000 x16: 0000000000000012
>> [  12.749266] x15: ffff08070508c683 x14: 0000000000000058
>> [  12.754648] x13: 00000000000000c0 x12: 0000000000000000
>> [  12.760030] x11: 0000000000000400 x10: 000000000000000c
>> [  12.765412] x9 : ffff800010039d58 x8 : 0000000020000000
>> [  12.770794] x7 : 0000000000000018 x6 : ffff800011750890
>> [  12.776176] x5 : ffff800011750878 x4 : 0000000000000000
>> [  12.781558] x3 : 0000000000000000 x2 : 0000000000000000
>> [  12.786940] x1 : 0000000000000200 x0 : 0000000000000000
>> [  12.792322] Call trace:
>> [  12.794806]  steal_suitable_fallback+0x138/0x2f0
>> [  12.799520]  get_page_from_freelist+0xe30/0x12a0
>> [  12.804207]  __alloc_pages_nodemask+0x148/0xe00
>> [  12.808809]  __dma_direct_alloc_pages+0xa4/0x1d0
>> [  12.813496]  dma_direct_alloc+0x1d8/0x340
>> [  12.817571]  xen_swiotlb_alloc_coherent+0x68/0x370
>> [  12.822439]  dma_alloc_attrs+0xe8/0xf0
>> [  12.826246]  nvme_reset_work+0x1030/0x1520
>> [  12.830417]  process_one_work+0x1dc/0x4bc
>> [  12.834495]  worker_thread+0x144/0x470
>> [  12.838313]  kthread+0x14c/0x160
>> [  12.841604]  ret_from_fork+0x10/0x38
>> [  12.845255] Code: a94082c4 d37ef463 cb3c4063 8b3c4042 (f9000480)
>> [  12.851447] ---[ end trace f68728a0d3053b72 ]---
>> [  12.856117] note: kworker/u64:0[7] exited with preempt_count
>
> xen_swiotlb_alloc_coherent calls dma_direct_alloc which fails for the
> device.
>
> Without swiotlb_xen, dma_alloc_attrs would do:
>
>        if (dma_alloc_direct(dev, ops))
>                cpu_addr = dma_direct_alloc(dev, size, dma_handle, flag, attrs);
>        else if (ops->alloc)
>                cpu_addr = ops->alloc(dev, size, dma_handle, flag, attrs);
>
> dma_alloc_direct is the very same call that fails when called from
> xen_swiotlb_alloc_coherent. So it must be that the path meant to be
> taken is ops->alloc instead.
>
> ops->alloc is the dma_ops function to allocate a coherent buffer.
> When swiotlb-xen is enabled, it points to xen_swiotlb_alloc_coherent.
> Often dma_ops is NULL when Xen is disabled.
>
> I think the most likely explanation is that the nvme device require a
> specific dma_ops. When swiotlb-xen is enabled, it gets overwritten by
> arch/arm64/mm/dma-mapping.c:arch_setup_dma_ops.
>
> Can you double-check this theory by adding a few printks in
> arch/arm64/mm/dma-mapping.c:arch_setup_dma_ops, to check the dma_ops
> used when !CONFIG_XEN, and also in kernel/dma/mapping.c:dma_alloc_attrs
> to check the code path taken when !CONFIG_XEN?

I checked the code path when !CONFID_XEN and SMMU is not enabled in linux, code path is:
        dma_alloc_attrs() -> dma_alloc_direct()
and there is no issue in booting and allocating a coherent buffer.

When !CONFIG_XEN and SMMU is enabled code path is:
        dma_alloc_attrs()-> ops->alloc -> iommu_dma_alloc()
and there is no issue in booting and allocating a coherent buffer in this case also.

When Xen is enabled code path is:
        dma_alloc_attrs()-> ops->alloc() -> xen_swiotlb_alloc_coherent() -> dma_alloc_direct()
in this case allocating the coherent buffer failed.

Only difference I see is the size argument for dma_alloc_direct() in Xen case but I don’t think
that is causing the issue there is something else causing the issue maybe from
where the coherent memory is allocated.
       
        /* Convert the size to actually allocated. */
        size = 1UL << (order + XEN_PAGE_SHIFT);


Please find the attached detailed logs for each case.




Regards,
Rahul

Attachment: native_linux_boot_with_smmu_debug.log
Description: native_linux_boot_with_smmu_debug.log

Attachment: native_linux_boot_without_smmu_debug.log
Description: native_linux_boot_without_smmu_debug.log

Attachment: xen_linux_boot_debug.log
Description: xen_linux_boot_debug.log

Attachment: native_linux_boot.log
Description: native_linux_boot.log


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.