[PATCH] nvme-pci: add adaptive interrupt polling

[changfengnan]

> From: "Christoph Hellwig"<hch at lst.de>
> Date:  Wed, Aug 19, 2026, 13:25
> Subject:  Re: [PATCH] nvme-pci: add adaptive interrupt polling
> To: "changfengnan"<changfengnan at bytedance.com>
> Cc: "Keith Busch"<kbusch at kernel.org>, "Jens Axboe"<axboe at kernel.dk>, "Christoph Hellwig"<hch at lst.de>, "Sagi Grimberg"<sagi at grimberg.me>, <linux-nvme at lists.infradead.org>, <linux-kernel at vger.kernel.org>, <jun1.zeng at intel.com>, "Guzebing"<guzebing at bytedance.com>
> On Tue, Aug 18, 2026 at 11:53:55AM +0800, changfengnan wrote:
> > 1. Regarding the NVME_ADAPTIVE_POLL_PERIOD_NS parameter, is it necessary to make it a configurable setting? Perhaps Gen6 SSDs require a smaller value? I currently do not have a testing environment for this, but the default value of 10 us appears to be sufficient for Gen4 and Gen5. 
> 
> The most important thing to consider in the short term is not the fastest
> possible devices, but the worst.  I.e. consider what this does on really
> crappy older cusomer drives, as that is probably the biggest install base
> using the Linux nvme driver.
That makes sense, we also need to see how it performs with consumer-grade
NVMe drives. I only have an old Samsung 970 NVMe here, so I don’t have
much of a testing environment for consumer-grade NVMe drives. 
I hope someone can help me test it out :)

> 
> > 2. Should MSI be supported? I do not have a similar environment, so I am unsure how it would perform in such scenarios. 
> 
> Or legacy interrupts :)  I don't think we have to support them as long
> as we cleanly disable the feature.
> 
> > 3. Should it be enabled by default? Although current tests show only a minor performance hit, I still lean toward disabling it by default and waiting to gather more test feedback.
> 
> I think it should eventually enabled by default, but I'd also be happy
> to merge it fist and then flip the bit later.
>