From mboxrd@z Thu Jan 1 00:00:00 1970 From: Claude Code Review Bot To: dri-devel-reviews@example.com Subject: Claude review: drm/imagination: Synchronize interrupts before suspending the GPU Date: Mon, 25 May 2026 18:34:26 +1000 Message-ID: In-Reply-To: <20260522-sync-irqs-6-12-v1-1-b0ecc9675078@imgtec.com> References: <20260522-sync-irqs-6-12-v1-1-b0ecc9675078@imgtec.com> X-Mailer: Claude Code Patch Reviewer Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable MIME-Version: 1.0 Overall Series Review Subject: drm/imagination: Synchronize interrupts before suspending the GPU Author: Alessio Belle Patches: 2 Reviewed: 2026-05-25T18:34:26.620059 --- This is a single-patch backport targeting the 6.12 stable kernel. It fixes = a real race condition where the runtime PM suspend callback can power off t= he GPU while the threaded IRQ handler is still running on another CPU, caus= ing a synchronous external abort when the handler tries to read GPU registe= rs. The fix is straightforward: call `synchronize_irq()` after requesting f= irmware power-off but before calling `pvr_fw_stop()`, ensuring any in-fligh= t IRQ handler has completed. The approach is a sensible minimal backport. The upstream commit (2d7f05cdd= f4c) on drm-next went further =E2=80=94 it uses `disable_irq()` / `enable_i= rq()` pairs in the suspend/resume paths (and adds a corresponding `pvr_powe= r_fw_enable` signature change with `rpm_resume`). This 6.12 backport intent= ionally takes only the synchronization part without the full disable/enable= refactoring, which is a reasonable choice for a stable backport: it fixes = the race with minimal code churn and avoids the need to also modify the res= ume path. **Verdict: The patch looks correct and suitable for stable.** One minor observation and one potential concern are noted below. --- Generated by Claude Code Patch Reviewer