Re: Pinebook pro IOMMU enabled crashes
Jesper Schmitz Mouridsen <[email protected]>
| Newsgroups | gmane.os.freebsd.devel.arm |
|---|---|
| Message-ID | <[email protected]> |
On 30.09.2021 18.39, Jesper Schmitz Mouridsen wrote:
>>>>> main-n249225-f673cc5edac3-dirty
>>>>> nvme0: <Generic NVMe Device> at device 0.0 on pci1
>>>>> Fatal data abort:
>>>>> x0: 0
>>>>> x1: 1000
>>>>> x2: 10040
>>>>> x3: 2000
>>>>> x4: 1
>>>>> x5: ffff00009a7a0168
>>>>> x6: 1d00000000000000
>>>>> x7: 10000000000000
>>>>> x8: 1168
>>>>> x9: 1
>>>>> x10: 0
>>>>> x11: ffff000000f35140
>>>>> x12: ffff000000f350c0
>>>>> x13: 1
>>>>> x14: 10000
>>>>> x15: 1
>>>>> x16: 10000
>>>>> x17: ffff000000f350dc
>>>>> x18: ffff00000110d180
>>>>> x19: ffff00000110d248
>>>>> x20: 0
>>>>> x21: ffff00009a79f000
>>>>> x22: ffffa000010b0a00
>>>>> x23: ffffa000010a2880
>>>>> x24: ffffa0000116da00
>>>>> x25: ffff000000b4fd78
>>>>> x26: ffffa0000116db00
>>>>> x27: ffff000000e83000
>>>>> x28: 7
>>>>> x29: ffff00000110d190
>>>>> sp: ffff00000110d180
>>>>> lr: ffff00000077520c
>>>>> elr: ffff0000007a03ac
>>>>> spsr: 200000c5
>>>>> far: 0
>>>>> esr: 96000004
>>>>> panic: vm_fault failed: ffff0000007a03ac error 1
>>>>> cpuid = 0
>>>>> time = 1
>>>>> KDB: stack backtrace:
>>>>> db_trace_self() at db_trace_self
>>>>> db_trace_self_wrapper() at db_trace_self_wrapper+0x30
>>>>> vpanic() at vpanic+0x184
>>>>> panic() at panic+0x44
>>>>> data_abort() at data_abort+0x23c
>>>>> handle_el1h_sync() at handle_el1h_sync+0x78
>>>>> --- exception, esr 0x96000004
>>>>> iommu_map_msi() at iommu_map_msi+0x20
>>>>> gicv3_iommu_init() at gicv3_iommu_init+0x4c
>>>>> intr_alloc_msix() at intr_alloc_msix+0x13c
>>>>> rk_pcie_alloc_msix() at rk_pcie_alloc_msix+0xfc
>>>>> pci_alloc_msix_method() at pci_alloc_msix_method+0x1a8
>>>>> nvme_pci_attach() at nvme_pci_attach+0x378
>>>>> device_attach() at device_attach+0x400
>>>>> device_probe_and_attach() at device_probe_and_attach+0x7c
>>>>> bus_generic_attach() at bus_generic_attach+0x18
>>>>> pci_attach() at pci_attach+0xe8
>>>>> device_attach() at device_attach+0x400
>>>>> device_probe_and_attach() at device_probe_and_attach+0x7c
>>>>> bus_generic_attach() at bus_generic_attach+0x18
>>>>> device_attach() at device_attach+0x400
>>>>> device_probe_and_attach() at device_probe_and_attach+0x7c
>>>>> bus_generic_attach() at bus_generic_attach+0x18
>>>>> pci_attach() at pci_attach+0xe8
>>>>> device_attach() at device_attach+0x400
>>>>> device_probe_and_attach() at device_probe_and_attach+0x7c
>>>>> bus_generic_attach() at bus_generic_attach+0x18
>>>>> rk_pcie_attach() at rk_pcie_attach+0x14cc
>>>>> device_attach() at device_attach+0x400
>>>>> device_probe_and_attach() at device_probe_and_attach+0x7c
>>>>> bus_generic_new_pass() at bus_generic_new_pass+0xf8
>>>>> bus_generic_new_pass() at bus_generic_new_pass+0xa8
>>>>> bus_generic_new_pass() at bus_generic_new_pass+0xa8
>>>>> bus_set_pass() at bus_set_pass+0x4c
>>>>> mi_startup() at mi_startup+0x12c
>>>>> virtdone() at virtdone+0x6c
>>>>>
>>>> That's an old commit. Did you see this panic only recently or ?
>>>>
>>>
>>
>>
>> Even on stable/13-n247374-9faebc1e664d-dirty
>>
>> I get the same backtrace when IOMMU is enabled and the nvme is attached.
>>
>> pcib1: <PCI-PCI bridge> at device 0.0 on pci0
>> pcib0: failed to reserve resource for pcib1
>> pcib1: failed to allocate initial memory window: 0-0xfffff
>> pci1: <PCI bus> on pcib1
>> nvme0: <Generic NVMe Device> at device 0.0 on pci1
>> Fatal data abort:
>> x0: 0
>> x1: 1000
>> x2: 10040
>> x3: 2000
>> x4: 1
>> x5: ffff00009a99e160
>> x6: 1400000000000000
>> x7: 10000000000000
>> x8: 1160
>> x9: ffff000000cd7cc0
>> x10: 0
>> x11: ffff000000d89540
>> x12: ffff000000d894c0
>> x13: 1
>> x14: 10000
>> x15: 1
>> x16: 10000
>> x17: 0
>> x18: ffff000000f5c250
>> x19: ffff000000f5c318
>> x20: 0
>> x21: ffff00009a99d000
>> x22: ffffa00000f06200
>> x23: ffffa00000f49700
>> x24: ffffa00000f8f500
>> x25: ffff0000009b85f8
>> x26: ffffa00000f8f600
>> x27: ffff000000cd7000
>> x28: 7
>> x29: ffff000000f5c260
>> sp: ffff000000f5c250
>> lr: ffff0000006bf3dc
>> elr: ffff0000006e15d0
>> spsr: 600001c5
>> far: 0
>> esr: 96000004
>> panic: vm_fault failed: ffff0000006e15d0
>> cpuid = 0
>> time = 1
>> KDB: stack backtrace:
>> #0 0xffff00000047c304 at kdb_backtrace+0x60
>> #1 0xffff000000437fd8 at vpanic+0x184
>> #2 0xffff000000437e50 at panic+0x44
>> #3 0xffff0000006d692c at data_abort+0x204
>> #4 0xffff0000006bb874 at handle_el1h_sync+0x74
>> #5 0xffff0000006bf3d8 at gicv3_iommu_init+0x4c
>> #6 0xffff0000006bf3d8 at gicv3_iommu_init+0x4c
>> #7 0xffff0000006b1940 at intr_alloc_msix+0x110
>> #8 0xffff0000007860c0 at rk_pcie_alloc_msix+0xfc
>> #9 0xffff000000219bbc at pci_alloc_msix_method+0x1a8
>> #10 0xffff00000020ba64 at nvme_pci_attach+0x378
>> #11 0xffff00000046bd80 at device_attach+0x400
>> #12 0xffff00000046d14c at bus_generic_attach+0x4c
>> #13 0xffff000000221f30 at pci_attach+0xe0
>> #14 0xffff00000046bd80 at device_attach+0x400
>> #15 0xffff00000046d14c at bus_generic_attach+0x4c
>> #16 0xffff00000046bd80 at device_attach+0x400
>> #17 0xffff00000046d14c at bus_generic_attach+0x4c
>> Uptime: 1s
>>
>>
> git checkout 50cedfede3d21824ec6023324b3ad41a435e1815
> sys/arm64/arm64/gicv3_its.c and the problem goes away. The commit is one
> before
> Add IOMMU support to GICv3 Interrupt Translation Service (ITS) driver.
> (ba196aec7dad1b73a9a3b86a06259d5e81f16fad)
>
It turns out iommu_get_dev_ctx returns NULL for at least my nvme device.
(Kingston A2000 M.2 NVMe SSD)
So the below patch "fixes" it..
diff --git a/sys/arm64/arm64/gicv3_its.c b/sys/arm64/arm64/gicv3_its.c
index 1a0e7a79e76b..28e2bcf70a5d 100644
--- a/sys/arm64/arm64/gicv3_its.c
+++ b/sys/arm64/arm64/gicv3_its.c
@@ -316,7 +316,7 @@ static const struct {
static device_attach_t gicv3_its_attach;
static device_detach_t gicv3_its_detach;
-static pic_disable_intr_t gicv3_its_disable_intr;
+tatic pic_disable_intr_t gicv3_its_disable_intr;
static pic_enable_intr_t gicv3_its_enable_intr;
static pic_map_intr_t gicv3_its_map_intr;
static pic_setup_intr_t gicv3_its_setup_intr;
@@ -1473,6 +1473,10 @@ gicv3_iommu_init(device_t dev, device_t child,
struct iommu_domain **domain)
sc = device_get_softc(dev);
ctx = iommu_get_dev_ctx(child);
+
+ if(ctx == NULL)
+ return (ENXIO);
+
error = iommu_map_msi(ctx, PAGE_SIZE, GITS_TRANSLATER,
IOMMU_MAP_ENTRY_WRITE, IOMMU_MF_CANWAIT, &sc->ma);
*domain = iommu_get_ctx_domain(ctx);
$