nvme: machine check when running nvme subsystem-reset /dev/nvme0 against direct attach via PCIE slot

Keith Busch kbusch at kernel.org
Tue Oct 29 09:42:21 PDT 2024


On Tue, Oct 29, 2024 at 12:07:26PM -0400, Laurence Oberman wrote:
> Finishing this thread off but have a final question. 
> Bottom line is certain server hardware sees the nvme reset command
> create a machine check for PCIE plugged NVME devices going back quite
> far in kernel versions,  and we panic.
> 
> As Keith had said, that nvme reset command is too much impact

Sure, it takes the PCIe link down, and handling for that, if at all, is
platform specific.
 
> There is a final simple question for M2 connected NVME devices. 
> Are these expected to auto-re-connect after an nvme reset is issued. 
> 
> The complaint is the following
> 
> nvme subsystem-reset /dev/nvme0 
> Device is disconnected as expected but requires the following to
> reconnect
> 
> echo 1 >  /sys/bus/pci/devices/0000:02:00.0/remove
> echo 1 > /sys/bus/pci/rescan
> 
> Then it is reconnected.

For platforms that don't support link detected hotplug, that sequence
should get the device back to usable.



More information about the Linux-nvme mailing list