Re: [PATCH v15 02/15] accel/rocket: take the completion register writes under job_lock

Next message: David Woodhouse: "Re: [PATCH v4 00/12] KVM: x86/xen: Don't drop event deliveries which race memslot updates"
Previous message: Usama Arif: "Re: [PATCH] x86/lib: Use REP MOVSB for large memcpy() on ERMS CPUs without FSRM"
In reply to: Jiaxing Hu: "[PATCH v15 02/15] accel/rocket: take the completion register writes under job_lock"
Next in thread: Jiaxing Hu: "[PATCH v15 01/15] accel/rocket: request the core clocks by name"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

From: Heiko Stübner

Date: Tue Nov 03 2026 - 17:50:14 EST

Am Sonntag, 1. November 2026, 09:22:54 Mitteleuropäische Normalzeit schrieb Jiaxing Hu:
> rocket_job_handle_irq() writes OPERATION_ENABLE and INTERRUPT_CLEAR before
> taking job_lock, while rocket_job_hw_submit() writes OPERATION_ENABLE under
> it, so a completion can write its zero after a submit has written its one
> and stop a task that has just started. Move both writes inside the existing
> scoped_guard().
>
> Fixes: 0810d5ad88a1 ("accel/rocket: Add job submission IOCTL")
> Signed-off-by: Jiaxing Hu <gahing@xxxxxxxxxxxxx>
> Tested-by: Igor Paunovic <royalnet026@xxxxxxxxx> # RK3588, three cores, induced reset, JOB_TIMEOUT_MS=2

Reviewed-by: Heiko Stuebner <heiko@xxxxxxxxx>