Re: [PATCH] net/gve: add passive device-requested reset

[Stephen Hemminger <stephen@networkplumber.org>]

On Fri, 27 Mar 2026 00:20:38 +0000
"Jasper Tran O'Leary" <jtranoleary@google.com> wrote:

> In addition to application-initiated resets, during normal operation,
> the device can indicate the need to reset by writing a value to a device
> status register. This patch introduces an alarm that polls the
> status of that register and informs the application of the need to
> reset. The application is responsible for registering a callback that
> will execute in the event that the alarm discovers that the device
> requests a reset.
> 
> Signed-off-by: Jasper Tran O'Leary <jtranoleary@google.com>
> Reviewed-by: Joshua Washington <joshwash@google.com>

Makes sense, AI analysis had one item worth noting.
Not sure where documentation related (if any) should go.

The alarm callback calls rte_eth_dev_callback_process() synchronously,
which means the application's reset handler runs in the alarm thread
context. The documentation example shows calling rte_eth_dev_reset()
directly in the callback — this works (rte_eal_alarm_cancel from within
a callback is allowed), but a production application would likely want
to set a flag and handle the reset from the main thread. Worth a note
in the doc example.
The polling interval is 1 second vs 20 seconds in the kernel gve driver.
Not wrong, but more aggressive — is there a reason for the shorter interval?
Otherwise looks good.

Applied to next-net