Description of problem:

On three identical systems (Dell Optiplex 3620) with Intel i7-6770 CPUs, all of them equipped with M.2 NVME SSD's (Western Digital Black SSD model WDS256G1X0C-00ENX0 with B35900WD firmware), so that faulty hardware is almost assuredly out of the equation, starting 'sudo fstrim -va' manually or waiting for the 'fstrim.timer' to trigger (on monday nights) crashes the system with:

[cut]
Feb 25 00:00:49 ronin systemd[1]: Starting Discard unused blocks...
Feb 25 00:00:49 ronin kernel: DMAR: DRHD: handling fault status reg 3
Feb 25 00:00:49 ronin kernel: DMAR: [DMA Read] Request device [02:00.0] fault addr 0 [fault reason 06] PTE Read access is not set
Feb 25 00:01:20 ronin kernel: nvme nvme0: I/O 697 QID 5 timeout, aborting
Feb 25 00:01:26 ronin kernel: nvme nvme0: I/O 446 QID 2 timeout, aborting
Feb 25 00:01:38 ronin kernel: nvme nvme0: I/O 634 QID 4 timeout, aborting
Feb 25 00:01:38 ronin kernel: nvme nvme0: I/O 635 QID 4 timeout, aborting
Feb 25 00:01:38 ronin kernel: nvme nvme0: I/O 636 QID 4 timeout, aborting
Feb 25 00:01:50 ronin kernel: nvme nvme0: I/O 697 QID 5 timeout, reset controller
Feb 25 00:02:20 ronin kernel: nvme nvme0: I/O 4 QID 0 timeout, reset controller
Feb 25 00:03:13 ronin kernel: nvme nvme0: Device not ready; aborting reset
Feb 25 00:03:13 ronin kernel: nvme nvme0: Abort status: 0x7
Feb 25 00:03:13 ronin kernel: nvme nvme0: Abort status: 0x7
Feb 25 00:03:13 ronin kernel: nvme nvme0: Abort status: 0x7
Feb 25 00:03:13 ronin kernel: nvme nvme0: Abort status: 0x7
Feb 25 00:03:13 ronin kernel: nvme nvme0: Abort status: 0x7
Feb 25 00:03:36 ronin kernel: nvme nvme0: Device not ready; aborting reset
Feb 25 00:03:36 ronin kernel: nvme nvme0: Removing after probe failure status: -19
Feb 25 00:03:36 ronin kernel: print_req_error: I/O error, dev nvme0n1, sector 157472648
Feb 25 00:03:36 ronin kernel: print_req_error: I/O error, dev nvme0n1, sector 1050704
[/cut]

and more badness...

Version-Release number of selected component (if applicable):

fstrim from util-linux 2.32.1

How reproducible:

90 over 100 trials (every now and then it works, then retrying immediately after triggers the crash).

Steps to Reproduce:
1. any of the PCs with the configuration mentioned above boots;
2. SSH to console;
3. 'sudo fstrim -va'

Actual results:

console stays silent for a while, then dies with 'fstrim: /: FITRIM ioctl failed: Input/output error' and every attempt to issue any command results in 'input/output error', hard reset necessary.

Expected results:

[cut]
[fsimula@ronin ~]$ sudo fstrim -va
/opt: 0 B (0 bytes) trimmed
/boot: 148,4 MiB (155566080 bytes) trimmed
/: 644,1 MiB (675352576 bytes) trimmed
[/cut]

Additional info:

'dmesg' always outputs the following messages:
- DMAR: DRHD: handling fault status reg 3
- DMAR: [DMA Read] Request device [02:00.0] fault addr 0 [fault reason 06] PTE Read access is not set

more or less simultaneously with the crash, [02:00.0] being the PCI address of the NVME drive:

[cut]
[fsimula@ronin ~]$ sudo lspci -vs02:00.0
02:00.0 Non-Volatile memory controller: Sandisk Corp WD Black NVMe SSD (prog-if 02 [NVM Express])
	Subsystem: Marvell Technology Group Ltd. Device 1093
	Flags: bus master, fast devsel, latency 0, IRQ 16, NUMA node 0
	Memory at ef000000 (64-bit, non-prefetchable) [size=16K]
	Capabilities: [40] Power Management version 3
	Capabilities: [70] Express Endpoint, MSI 00
	Capabilities: [b0] MSI-X: Enable+ Count=19 Masked-
	Capabilities: [100] Advanced Error Reporting
	Capabilities: [148] Device Serial Number 03-4d-ff-7a-99-88-77-66
	Capabilities: [158] Power Budgeting <?>
	Capabilities: [168] Alternative Routing-ID Interpretation (ARI)
	Capabilities: [178] Secondary PCI Express <?>
	Capabilities: [2b8] Latency Tolerance Reporting
	Capabilities: [2c0] L1 PM Substates
	Kernel driver in use: nvme
	Kernel modules: nvme
[/cut]

and after reading:
- "https://unix.stackexchange.com/questions/456628/nvme-fstrim-causing-crash-on-linux-disabling-with-systemctl-doesnt-help"
and
- "https://access.redhat.com/documentation/en-us/red_hat_virtualization/4.1/html/installation_guide/appe-configuring_a_hypervisor_host_for_pci_passthrough"

I made up my mind that the problem was somehow linked to IOMMU and VT-d.

I decided to change the 'intel_iommu=on" GRUB setting with 'intel_iommu=pt'; guess what, this seems to solve the crash.

Can someone clarify?