Skip to contentSkip to conversation

nvme interrupt counters got reset after suspend/resume

DISCUSSION David Wang 2 participants 3 messages Mar 19, 2025 NVMe

lore
Share & tools
lore

David Wang

author
#0 lore raw

Hi,

I noticed that on my system, the counters from /proc/interrupts for nvme got reset after I suspend and then resume the system. For example, before systemctl suspend:

$ cat /proc/interrupts 
            CPU0       CPU1       CPU2       CPU3       CPU4       CPU5       CPU6       CPU7
...
  38:          0          0          0         22          0          0          0          0 IR-PCI-MSIX-0000:25:00.0    0-edge      nvme0q0
  39:          0          0          0          0          0          0          0          0 IR-PCI-MSIX-0000:22:00.0    0-edge      enp34s0
  40:         78          0          0          0          0          0          0          0 IR-PCI-MSIX-0000:25:00.0    1-edge      nvme0q1
  41:          0        147          0          0          0          0          0          0 IR-PCI-MSIX-0000:25:00.0    2-edge      nvme0q2
  42:          0          0         13          0          0          0          0          0 IR-PCI-MSIX-0000:25:00.0    3-edge      nvme0q3
  43:          0          0          0          9          0          0          0          0 IR-PCI-MSIX-0000:25:00.0    4-edge      nvme0q4
  44:          0          0          0          0         87          0          0          0 IR-PCI-MSIX-0000:25:00.0    5-edge      nvme0q5
  45:          0          0          0          0          0        112          0          0 IR-PCI-MSIX-0000:25:00.0    6-edge      nvme0q6
  46:          0          0          0          0          0          0         12          0 IR-PCI-MSIX-0000:25:00.0    7-edge      nvme0q7
  47:          0          0          0          0          0          0          0          8 IR-PCI-MSIX-0000:25:00.0    8-edge      nvme0q8

And right after resume the system, values are reset to 0s:

$ cat /proc/interrupts 
            CPU0       CPU1       CPU2       CPU3       CPU4       CPU5       CPU6       CPU7
...
  38:          0          0          0         21          0          0          0          0 IR-PCI-MSIX-0000:25:00.0    0-edge      nvme0q0
  39:          0          0          0          0          0          0          0          0 IR-PCI-MSIX-0000:22:00.0    0-edge      enp34s0
  40:          0          0          0          0          0          0          0          0 IR-PCI-MSIX-0000:25:00.0    1-edge      nvme0q1
  41:          0          0          0          0          0          0          0          0 IR-PCI-MSIX-0000:25:00.0    2-edge      nvme0q2
  42:          0          0          0          0          0          0          0          0 IR-PCI-MSIX-0000:25:00.0    3-edge      nvme0q3
  43:          0          0          0          0          0          0          0          0 IR-PCI-MSIX-0000:25:00.0    4-edge      nvme0q4
  44:          0          0          0          0          0          0          0          0 IR-PCI-MSIX-0000:25:00.0    5-edge      nvme0q5
  45:          0          0          0          0          0          0          0          0 IR-PCI-MSIX-0000:25:00.0    6-edge      nvme0q6
  46:          0          0          0          0          0          0          0          0 IR-PCI-MSIX-0000:25:00.0    7-edge      nvme0q7
  47:          0          0          0          0          0          0          0          0 IR-PCI-MSIX-0000:25:00.0    8-edge      nvme0q8

This does not happend to counters for other interrupts on my system.
Not sure whether this is designed explicitly this way, or happended to be this way to make suspend/resume work, or there is some bug behind this.
Just bring this up for discussion since I failed to find any discussion about it.

Thanks
David.

maintainer
#1 lore raw
Quoted 4 lines · David Wang

This does not happend to counters for other interrupts on my system.
Not sure whether this is designed explicitly this way, or happended to be this way to make suspend/resume work, or there is some bug behind this.

The nvme driver completely brings down the interrupts on suspend and creates new ones on resume. They might have the same numbers and labels on the other side of the resume, but they're not from the same original allocation that tracks these stats. Hope that helps.

author
#2 lore raw

Hi,

Quoted 10 lines · Keith Busch
Quoted 4 lines · David Wang

This does not happend to counters for other interrupts on my system.
Not sure whether this is designed explicitly this way, or happended to be this way to make suspend/resume work, or there is some bug behind this.

The nvme driver completely brings down the interrupts on suspend and creates new ones on resume. They might have the same numbers and labels on the other side of the resume, but they're not from the same original allocation that tracks these stats. Hope that helps.

Copy that~! And thanks for clearing things out.

David