Skip to content

hw/nvme: add irqfd support

Klaus Jensen qemu-devel 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v3 Latest2 patches 2 repliesIn review

Patches in v3 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/2hw/nvme: support irq(de)assertion with eventfdhw/nvme/ctrl.c hw/nvme/nvme.h+106 -170 replies
2/2hw/nvme: use KVM irqfd when availablehw/nvme/ctrl.c hw/nvme/nvme.h hw/nvme/trace-events+149 -20 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v3 yet.

Read the v3 thread

Cover letter [PATCH v3 0/2] hw/nvme: add irqfd support

From: Klaus Jensen <[email protected]>

This is a re-spin of Jinhao's irqfd support series that fixes msix vector masking/unmasking to work correctly.

I kept being bugged out about that msi route not getting updated, so I hit the code into submission with a hammer.

I finally noticed the core issue:

1. The vector notifiers was never set because msix is not enabled at the point where nvme_init_pci() is called.

Move this call to nvme_start_ctrl().

Since the unmask callback was suddenly getting called now, another fix was needed:

2. Call kvm_irqchip_add_irqfd_notifier_gsi() in the unmask handler - not in nvme_init_irq_notifier(). The vectors may potentially be masked/unmasked and shall cause a pair of add_irqfd and remove_irqfd calls. Removing it from nvme_init_irq_notifier() makes sure we do not try to double add.

Now it does what it is supposed to; no hacks required :)

Jinhao Fan (2):
hw/nvme: support irq(de)assertion with eventfd
hw/nvme: use KVM irqfd when available

 hw/nvme/ctrl.c       | 261 ++++++++++++++++++++++++++++++++++++++++---
 hw/nvme/nvme.h       |   6 +
 hw/nvme/trace-events |   3 +
 3 files changed, 253 insertions(+), 17 deletions(-)
Signature · 1 line
2.37.2