"swiotlb: coherent allocation failed" dmesg spam with linux 4.15.0-9.10

Bug #1749202 reported by Seth Forshee on 2018-02-13
This bug affects 1 person
Affects Status Importance Assigned to Milestone
linux (Ubuntu)
Seth Forshee

Bug Description

I'm seeing this appear fairly regularly in dmesg:

[23630.791701] nouveau 0000:02:00.0: swiotlb buffer is full (sz: 2097152 bytes)
[23630.791703] swiotlb: coherent allocation failed for device 0000:02:00.0 size=2097152
[23630.791706] CPU: 12 PID: 2609 Comm: gnome-shell Not tainted 4.15.0-9-generic #10-Ubuntu
[23630.791709] Hardware name: Intel Corp. GRANTLEY/GRANTLEY, BIOS GRNDCRB1.86B.0032.R02.1405090848 05/09/2014
[23630.791710] Call Trace:
[23630.791718] dump_stack+0x63/0x8b
[23630.791723] swiotlb_alloc_coherent+0xe8/0x160
[23630.791728] x86_swiotlb_alloc_coherent+0x43/0x50
[23630.791734] ttm_dma_pool_get_pages+0x1f1/0x5b0 [ttm]
[23630.791737] ttm_dma_populate+0x249/0x330 [ttm]
[23630.791770] nouveau_ttm_tt_populate+0x156/0x1f0 [nouveau]
[23630.791773] ttm_tt_bind+0x2e/0x60 [ttm]
[23630.791776] ttm_bo_handle_move_mem+0x589/0x5c0 [ttm]
[23630.791779] ? ttm_bo_mem_space+0x39d/0x470 [ttm]
[23630.791782] ttm_bo_validate+0x144/0x160 [ttm]
[23630.791784] ttm_bo_init_reserved+0x393/0x430 [ttm]
[23630.791787] ttm_bo_init+0x2d/0x90 [ttm]
[23630.791808] ? nouveau_bo_invalidate_caches+0x10/0x10 [nouveau]
[23630.791827] nouveau_bo_new+0x40c/0x580 [nouveau]
[23630.791847] ? nouveau_bo_invalidate_caches+0x10/0x10 [nouveau]
[23630.791866] nouveau_gem_new+0x60/0x130 [nouveau]
[23630.791883] nouveau_gem_ioctl_new+0x59/0xe0 [nouveau]
[23630.791899] ? nouveau_gem_new+0x130/0x130 [nouveau]
[23630.791912] drm_ioctl_kernel+0x5f/0xb0 [drm]
[23630.791919] drm_ioctl+0x31b/0x3d0 [drm]
[23630.791936] ? nouveau_gem_new+0x130/0x130 [nouveau]
[23630.791954] nouveau_drm_ioctl+0x72/0xc0 [nouveau]
[23630.791957] do_vfs_ioctl+0xa8/0x630
[23630.791961] ? __sys_recvmsg+0x80/0x90
[23630.791964] SyS_ioctl+0x79/0x90
[23630.791967] do_syscall_64+0x76/0x130
[23630.791971] entry_SYSCALL_64_after_hwframe+0x21/0x86
[23630.791972] RIP: 0033:0x7f26f5357ea7
[23630.791973] RSP: 002b:00007ffe50c20ee8 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
[23630.791975] RAX: ffffffffffffffda RBX: 00005570b6b1b490 RCX: 00007f26f5357ea7
[23630.791975] RDX: 00007ffe50c20f40 RSI: 00000000c0306480 RDI: 000000000000000b
[23630.791976] RBP: 00007ffe50c20f40 R08: 0000000000000004 R09: 00005570b6b1b490
[23630.791977] R10: 0000000000000030 R11: 0000000000000246 R12: 00000000c0306480
[23630.791977] R13: 000000000000000b R14: 00005570b6ad5b88 R15: 00005570b0ad0690

This seems to be due to an attempt to allocate using huge pages, which has a fallback to allocating normal pages. There are patches in 4.16-rc1 to suppress this spam; backport them to bionic.

CVE References

Seth Forshee (sforshee) on 2018-02-13
Changed in linux (Ubuntu):
status: In Progress → Fix Committed
Launchpad Janitor (janitor) wrote :
Download full text (3.3 KiB)

This bug was fixed in the package linux - 4.15.0-10.11

linux (4.15.0-10.11) bionic; urgency=medium

  * linux: 4.15.0-10.11 -proposed tracker (LP: #1749250)

  * "swiotlb: coherent allocation failed" dmesg spam with linux 4.15.0-9.10
    (LP: #1749202)
    - swiotlb: suppress warning when __GFP_NOWARN is set
    - drm/ttm: specify DMA_ATTR_NO_WARN for huge page pools

  * linux-tools: perf incorrectly linking libbfd (LP: #1748922)
    - SAUCE: tools -- add ability to disable libbfd
    - [Packaging] correct disablement of libbfd

  * [Artful] Realtek ALC225: 2 secs noise when a headset plugged in
    (LP: #1744058)
    - ALSA: hda/realtek - update ALC225 depop optimize

  * [Artful] Support headset mode for DELL WYSE (LP: #1723913)
    - SAUCE: ALSA: hda/realtek - Add support headset mode for DELL WYSE

  * headset mic can't be detected on two Dell machines (LP: #1748807)
    - ALSA: hda/realtek - Support headset mode for ALC215/ALC285/ALC289
    - ALSA: hda - Fix headset mic detection problem for two Dell machines

  * Bionic update to v4.15.3 stable release (LP: #1749191)
    - ip6mr: fix stale iterator
    - net: igmp: add a missing rcu locking section
    - qlcnic: fix deadlock bug
    - qmi_wwan: Add support for Quectel EP06
    - r8169: fix RTL8168EP take too long to complete driver initialization.
    - tcp: release sk_frag.page in tcp_disconnect
    - vhost_net: stop device during reset owner
    - ipv6: addrconf: break critical section in addrconf_verify_rtnl()
    - ipv6: change route cache aging logic
    - Revert "defer call to mem_cgroup_sk_alloc()"
    - net: ipv6: send unsolicited NA after DAD
    - rocker: fix possible null pointer dereference in
    - tcp_bbr: fix pacing_gain to always be unity when using lt_bw
    - cls_u32: add missing RCU annotation.
    - ipv6: Fix SO_REUSEPORT UDP socket with implicit sk_ipv6only
    - soreuseport: fix mem leak in reuseport_add_sock()
    - net_sched: get rid of rcu_barrier() in tcf_block_put_ext()
    - net: sched: fix use-after-free in tcf_block_put_ext
    - media: mtk-vcodec: add missing MODULE_LICENSE/DESCRIPTION
    - media: soc_camera: soc_scale_crop: add missing
    - media: tegra-cec: add missing MODULE_DESCRIPTION/AUTHOR/LICENSE
    - gpio: uniphier: fix mismatch between license text and MODULE_LICENSE
    - crypto: tcrypt - fix S/G table for test_aead_speed()
    - Linux 4.15.3

  * bnx2x_attn_int_deasserted3:4323 MC assert! (LP: #1715519) //
    - net: create skb_gso_validate_mac_len()
    - bnx2x: disable GSO where gso_size is too big for hardware

  * ethtool -p fails to light NIC LED on HiSilicon D05 systems (LP: #1748567)
    - net: hns: add ACPI mode support for ethtool -p

  * CVE-2017-5715 (Spectre v2 Intel)
    - [Packaging] retpoline files must be sorted
    - [Packaging] pull in retpoline files

  * [Feature] PXE boot with Intel Omni-Path (LP: #1712031)
    - d-i: Add hfi1 to nic-modules

  * CVE-2017-5715 (Spectre v2 retpoline)
    - [Packaging] retpoline -- add call site validation
    - [Config] disable retpoline checks for first upload

  * Do not dup...


Changed in linux (Ubuntu):
status: Fix Committed → Fix Released
To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers