Skip to contentSkip to conversation

floppy: flush pending work before releasing resources on init failure

PATCH In review Karl Mehltretter 2 participants 4 messages Jul 22 – Sep 12

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 floppy: flush pending work before releasing resources on init failure

do_floppy_init() probes each FDC with user_reset_fdc(), which queues a redo_fd_request() work item on floppy_wq that may still be pending when the probe finishes. When no controller is found, the error path calls floppy_release_irq_and_dma() to free the IRQ and DMA channel, and only later destroys the workqueue. destroy_workqueue() drains the queued work last, so it can run after the IRQ and DMA are already gone:

floppy0: no floppy controllers found
work still pending

The flush used to be here. Commit 070ad7e793dc ("floppy: convert to delayed work and single-thread wq") renamed the label out_flush_work to out_release_dma and dropped it. Restore it so the work drains before the IRQ and DMA are released, as floppy_module_exit() already does.

Fixes 070ad7e793dc ("floppy: convert to delayed work and single-thread wq")
Assisted-by Claude:claude-opus-4-8

---

Noticed while running mainline on an emulated Acorn RiscPC, a custom personal QEMU machine. It has no floppy controller and rpc_defconfig sets CONFIG_BLK_DEV_FD=y, so every boot takes the no-controller error path. No disk is registered there, so the drained work takes the empty-request fast path and just unlocks the FDC, leaving only the spurious message.

Verified with 100 QEMU boots per platform on v7.2-rc4, no controller present. "work still pending" appeared on an unfixed kernel vs with this patch:

Acorn RiscPC, rpc_defconfig: 100/100 -> 0/100
x86_64 defconfig, qemu -M q35: 6/100 -> 0/100

With a controller present on qemu -M pc, reading and writing a 1.44M floppy image still works.

 drivers/block/floppy.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/block/floppy.c +2 -0
index f04397b8e381..6cfd114d27df 100644
@@ -4744,6 +4744,8 @@ static int __init do_floppy_init(void)
 		}
 	}
 out_release_dma:
+	/* Drain the queued redo_fd_request() before releasing IRQ and DMA. */
+	flush_workqueue(floppy_wq);
 	if (atomic_read(&usage_count))
 		floppy_release_irq_and_dma();
 out_unreg_driver:
Signature · 1 line
2.53.0
maintainer Reviewed-by on v1 1/1
#1 lore raw

Hello,

Thank you for the patch.

Quoted 10 lines · Karl Mehltretter

do_floppy_init() probes each FDC with user_reset_fdc(), which queues a redo_fd_request() work item on floppy_wq that may still be pending when the probe finishes. When no controller is found, the error path calls floppy_release_irq_and_dma() to free the IRQ and DMA channel, and only later destroys the workqueue. destroy_workqueue() drains the queued work last, so it can run after the IRQ and DMA are already gone:

floppy0: no floppy controllers found
work still pending

Quoted 8 lines

The flush used to be here. Commit 070ad7e793dc ("floppy: convert to delayed work and single-thread wq") renamed the label out_flush_work to out_release_dma and dropped it. Restore it so the work drains before the IRQ and DMA are released, as floppy_module_exit() already does.

Fixes 070ad7e793dc ("floppy: convert to delayed work and single-thread wq")
Assisted-by Claude:claude-opus-4-8
Quoted 27 lines

---

Noticed while running mainline on an emulated Acorn RiscPC, a custom personal QEMU machine. It has no floppy controller and rpc_defconfig sets CONFIG_BLK_DEV_FD=y, so every boot takes the no-controller error path. No disk is registered there, so the drained work takes the empty-request fast path and just unlocks the FDC, leaving only the spurious message.

Verified with 100 QEMU boots per platform on v7.2-rc4, no controller present. "work still pending" appeared on an unfixed kernel vs with this patch:

Acorn RiscPC, rpc_defconfig: 100/100 -> 0/100
x86_64 defconfig, qemu -M q35: 6/100 -> 0/100

With a controller present on qemu -M pc, reading and writing a 1.44M floppy image still works.

 drivers/block/floppy.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/block/floppy.c +0 -0
index f04397b8e381..6cfd114d27df 100644

drivers/block/floppy.c static int __init do_floppy_init(void) · line 4751

Quoted 8 lines
drivers/block/floppy.c +2 -0
@@ -4744,6 +4744,8 @@ static int __init do_floppy_init(void)
 		}
 	}
 out_release_dma:
+	/* Drain the queued redo_fd_request() before releasing IRQ and DMA. */
+	flush_workqueue(floppy_wq);
 	if (atomic_read(&usage_count))
 		floppy_release_irq_and_dma();
 out_unreg_driver:

Thanks,
Denis

#2 lore raw
In reply to Karl Mehltretter
Quoted 8 lines · Karl Mehltretter

The flush used to be here. Commit 070ad7e793dc ("floppy: convert to delayed work and single-thread wq") renamed the label out_flush_work to out_release_dma and dropped it. Restore it so the work drains before the IRQ and DMA are released, as floppy_module_exit() already does.

Fixes 070ad7e793dc ("floppy: convert to delayed work and single-thread wq")
Assisted-by Claude:claude-opus-4-8

Friendly ping on this patch. It already has a Reviewed-by.

Thanks,
Karl

#3 lore raw

Hello,

Quoted 6 lines · Karl Mehltretter
Quoted 5 lines · Karl Mehltretter

The flush used to be here. Commit 070ad7e793dc ("floppy: convert to delayed work and single-thread wq") renamed the label out_flush_work to out_release_dma and dropped it. Restore it so the work drains before the IRQ and DMA are released, as floppy_module_exit() already does.

Quoted 6 lines
Quoted 3 lines
Fixes 070ad7e793dc ("floppy: convert to delayed work and single-thread wq")
Assisted-by Claude:claude-opus-4-8

Friendly ping on this patch. It already has a Reviewed-by.

Thank you for the reminder. I took you patch
https://github.com/evdenis/linux-floppy/commit/acf6cbd113f0cc68dc42e2ad7a83dd77ae2def04
and will send merge request to Jens with rest of the changes shortly after.

Thanks,
Denis