On Tue, Nov 7, 2017 at 5:26 PM, Ferruh Yigit <ferruh.yi...@intel.com> wrote:
> On 11/7/2017 12:47 PM, Chas Williams wrote: > > I will confess I haven't looked into the issue too hard since I have a > > workaround. My first guess is that there is something going on with the > IOMMU > > and quiescing a PCI pass-through device/function from the guest (since I > don't > > think the IOMMU is "visible" to the guest) seems iffy. > > > > Most devices have some sort of reset to put the device into a known > state for > > setup/configuration (or enable/disable for the DMA engines). If this is > done at > > .dev_close(), shouldn't that be as sufficient as resetting the function? > > This is for the cases DPDK app terminated unexpectedly, proper exit path > already > does cleanup. > Call a usermode helper from igb_uio that does an open/close on the device about to be released? > > > > > On Tue, Nov 7, 2017 at 1:49 PM, Ferruh Yigit <ferruh.yi...@intel.com > > <mailto:ferruh.yi...@intel.com>> wrote: > > > > On 11/7/2017 10:12 AM, Chas Williams wrote: > > > Environment: Dell PowerEdge R730, Intel Corporation 82599ES > 10-Gigabit > > SFI/SFP+ > > > Network Connection shared via PCI pass-through > > > Host: Debian 8 > > > Guest: Custom Debian 8 with DPDK application based on 17.11 > > > > > > When we shutdown the guest, the kernel panics with: > > > > > > [ 279.021818] Do you have a strange power saving mode enabled? > > > [ 279.021819] Dazed and confused, but trying to continue > > > [ 279.021847] {1}[Hardware Error]: Hardware error from APEI > Generic Hardware > > > Error Source: 3 > > > [ 279.021849] {1}[Hardware Error]: event severity: fatal > > > [ 279.021850] {1}[Hardware Error]: Error 0, type: fatal > > > [ 279.021851] {1}[Hardware Error]: section_type: PCIe error > > > [ 279.021852] {1}[Hardware Error]: port_type: 0, PCIe end point > > > [ 279.021853] {1}[Hardware Error]: version: 1.16 > > > [ 279.021854] {1}[Hardware Error]: command: 0x0507, status: > 0x4010 > > > [ 279.021855] {1}[Hardware Error]: device_id: 0000:03:00.0 > > > [ 279.021855] {1}[Hardware Error]: slot: 0 > > > [ 279.021856] {1}[Hardware Error]: secondary_bus: 0x00 > > > [ 279.021857] {1}[Hardware Error]: vendor_id: 0x8086, > device_id: 0x10fb > > > [ 279.021858] {1}[Hardware Error]: class_code: 000002 > > > [ 279.021859] Kernel panic - not syncing: Fatal hardware error! > > > [ 279.021977] sched: Unexpected reschedule of offline CPU#1! > > > [ 279.021984] ------------[ cut here ]------------ > > > [ 279.021992] WARNING: CPU: 43 PID: 2807 at > > > /build/linux-fHlJSJ/linux-4.12.6/arch/x86/kernel/smp.c:128 > > > native_smp_send_reschedule+0x34/0x40 > > > [ 279.021993] Modules linked in: vfio_pci vfio_virqfd > vfio_iommu_type1 vfio > > > openvswitch nf_conntrack_ipv6 nf_nat_ipv6 nf_conntrack_ipv4 > nf_defrag_ipv4 > > > nf_nat_ipv4 nf_defrag_ipv6 nf_nat nf_conntrack libcrc32c > crc32c_generic nfsd > > > nfs_aclr > > > pcsec_gss_krb5 auth_rpcgss nfsv4 dns_resolver nfs lockd grace > sunrpc > > fscache tun > > > intel_rapl sb_edac x86_pkg_temp_thermal intel_powerclamp coretemp > > kvm_intel kvm > > > irqbypass mgag200 ttm drm_kms_helper drm joydev crct10dif_pclmul > crc32_pclmu > > > l ghash_clmulni_intel i2c_algo_bit ipmi_si ipmi_devintf iTCO_wdt > intel_cstate > > > iTCO_vendor_support evdev intel_uncore mxm_wmi lpc_ich > ipmi_msghandler > > mfd_core > > > ioatdma intel_rapl_perf dcdbas pcspkr shpchp mei_me button wmi mei > > acpi_power_m > > > eter tpm_crb autofs4 ext4 crc16 jbd2 fscrypto mbcache sr_mod cdrom > sg > > > hid_generic usbhid hid sd_mod > > > [ 279.022044] crc32c_intel aesni_intel aes_x86_64 crypto_simd > cryptd > > > glue_helper ahci ehci_pci libahci ehci_hcd ixgbe libata > megaraid_sas > > usbcore dca > > > i40e usb_common ptp pps_core scsi_mod mdio > > > [ 279.022060] CPU: 43 PID: 2807 Comm: revalidator85 Not tainted > > 4.12.0-1-amd64 > > > #1 Debian 4.12.6-1 > > > [ 279.022061] Hardware name: Dell Inc. PowerEdge R730/0WCJNT, > BIOS 2.3.4 > > 11/08/2016 > > > [ 279.022062] task: ffff91d0473f7100 task.stack: ffffafef8f4a4000 > > > [ 279.022066] RIP: 0010:native_smp_send_reschedule+0x34/0x40 > > > [ 279.022067] RSP: 0018:ffffafef8f4a7c98 EFLAGS: 00010082 > > > [ 279.022069] RAX: 000000000000002e RBX: ffff91d059d24080 RCX: > > 0000000000000001 > > > [ 279.022070] RDX: 0000000000000000 RSI: 0000000000000002 RDI: > > 0000000000000046 > > > [ 279.022071] RBP: ffff91d04691d100 R08: 0000000000000000 R09: > > 000000000000002e > > > [ 279.022072] R10: ffffafef8f4a7c90 R11: 00000000001cbb78 R12: > > ffff91d85d21ae80 > > > [ 279.022073] R13: ffff91d059d24000 R14: 0000000000000002 R15: > > 0000000000000008 > > > [ 279.022075] FS: 00007f726affd700(0000) > GS:ffff91d85d740000(0000) > > > knlGS:0000000000000000 > > > [ 279.022076] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > > > [ 279.022077] CR2: 00007fd422a52c48 CR3: 000000042d90f000 CR4: > > 00000000003426e0 > > > [ 279.022078] DR0: 0000000000000000 DR1: 0000000000000000 DR2: > > 0000000000000000 > > > [ 279.022079] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: > > 0000000000000400 > > > [ 279.022080] Call Trace: > > > [ 279.022086] ? check_preempt_wakeup+0x181/0x220 > > > [ 279.022091] ? check_preempt_curr+0x74/0x80 > > > [ 279.022094] ? ttwu_do_wakeup+0x19/0x140 > > > [ 279.022098] ? try_to_wake_up+0x1b8/0x470 > > > [ 279.022101] ? wake_up_q+0x3f/0x70 > > > [ 279.022106] ? futex_wake+0x15a/0x170 > > > [ 279.022108] ? do_futex+0x2df/0xa90 > > > [ 279.022111] ? SyS_futex+0x7a/0x170 > > > [ 279.022113] ? SyS_read+0x76/0xc0 > > > [ 279.022118] ? system_call_fast_compare_end+0xc/0x97 > > > [ 279.022119] Code: a3 05 51 fb cc 00 73 15 48 8b 05 28 74 a3 00 > be fd 00 > > 00 00 > > > 48 8b 80 a0 00 00 00 ff e0 89 fe 48 c7 c7 88 5c de b6 e8 e2 c9 13 > 00 <0f> > > ff c3 > > > 66 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 8b 05 5d 00 > > > [ 279.022151] ---[ end trace eddc980dc8648163 ]--- > > > [ 279.454274] Kernel Offset: 0x35400000 from 0xffffffff81000000 > (relocation > > > range: 0xffffffff80000000-0xffffffffbfffffff) > > > > > > The test engineer says this doesn't happen if we use SRIOV (which > makes sense > > > since the device isn't directly shared between the guest and the > host). If I > > > remove the pci_reset_function() from igb_uio's .release, then all > is well. > > > > This was tougher than expected, so many unexpected behavior. Why > resetting > > pass-through device in guest cause a crash in the host? > > > > Finally, I will send a patch to remove the reset. Hopefully no more > surprises > > for release. > > > > Still there will remain two improvement in igb_uio for better > security, > > disabling device interrupt on exit and clear master on exit. > > > > > > > > > > > On Tue, Nov 7, 2017 at 8:02 AM, Thomas Monjalon < > tho...@monjalon.net <mailto:tho...@monjalon.net> > > > <mailto:tho...@monjalon.net <mailto:tho...@monjalon.net>>> wrote: > > > > > > 07/11/2017 12:50, Chas Williams: > > > > We still have an issue with this and PCI pass-through. If a > guest is > > > > restarted while using PCI pass-through and igb_uio issues a > > > > pci_reset_function(), this causes the host to crash. > > > > > > Please, could you better explain the exact scenario and the > cause of > > the crash? > > > Thanks > > > > > > > > > > > >
