Skip to contentSkip to conversation

virtio_pci_modern: replace msleep(1) with usleep_range(1000, 2000)

PATCH In review Mirza Ishan Beg 2 participants 4 messages Aug 16 – Aug 17 virtio core

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 virtio_pci_modern: replace msleep(1) with usleep_range(1000, 2000)

Replace msleep(1) with usleep_range(1000, 2000) in the queue reset loops.
msleep(1) relies on jiffies and can sleep up to 4-10 ms depending on HZ, whereas usleep_range provides predictable microsecond-level delay.

---
 drivers/virtio/virtio_pci_modern_dev.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/virtio/virtio_pci_modern_dev.c +2 -2
index 413a8c353463..8e74a813ca97 100644
@@ -535,10 +535,10 @@ void vp_modern_set_queue_reset(struct virtio_pci_modern_device *mdev, u16 index)
 	vp_iowrite16(1, &cfg->queue_reset);
 
 	while (vp_ioread16(&cfg->queue_reset))
-		msleep(1);
+		usleep_range(1000, 2000);
 
 	while (vp_ioread16(&cfg->cfg.queue_enable))
-		msleep(1);
+		usleep_range(1000, 2000);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_queue_reset);
 
Signature · 1 line
2.47.3
#1 lore raw
Quoted 6 lines · Mirza Ishan Beg

Replace msleep(1) with usleep_range(1000, 2000) in the queue reset loops.
msleep(1) relies on jiffies and can sleep up to 4-10 ms depending on HZ, whereas usleep_range provides predictable microsecond-level delay.

why do we care?

Quoted 23 lines
---
 drivers/virtio/virtio_pci_modern_dev.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/virtio/virtio_pci_modern_dev.c +2 -2
index 413a8c353463..8e74a813ca97 100644
@@ -535,10 +535,10 @@ void vp_modern_set_queue_reset(struct virtio_pci_modern_device *mdev, u16 index)
 	vp_iowrite16(1, &cfg->queue_reset);
 
 	while (vp_ioread16(&cfg->queue_reset))
-		msleep(1);
+		usleep_range(1000, 2000);
 
 	while (vp_ioread16(&cfg->cfg.queue_enable))
-		msleep(1);
+		usleep_range(1000, 2000);
 }
 EXPORT_SYMBOL_GPL(vp_modern_set_queue_reset);
 
Signature · 1 line
2.47.3
#2 lore raw
Quoted 1 line · Michael S. Tsirkin

why do we care?

Hi Michael,

The motivation originally came from a checkpatch.pl warning, but looking directly at Documentation/timers/timers-howto.rst, there is a practical penalty here.

The documentation explicitly notes for the 1-20ms range: "msleep(1~20) may not do what the caller intends, and will often sleep longer (~20 ms actual sleep for any value given in the 1~20ms range)."

In these queue reset loops, using msleep(1) means each iteration may be extended to 10-20ms depending on HZ and timer slack. The document recommends using usleep_range() to avoid this overhead and provide the intended precision.

Let me know if you'd like this explanation folded into a v2 commit message.

--
Mirza

#3 lore raw
Quoted 8 lines · Mirza Ishan Beg
Quoted 1 line · Michael S. Tsirkin

why do we care?

Hi Michael,

The motivation originally came from a checkpatch.pl warning, but looking directly at Documentation/timers/timers-howto.rst, there is a practical penalty here.

Quoted 8 lines

The documentation explicitly notes for the 1-20ms range: "msleep(1~20) may not do what the caller intends, and will often sleep longer (~20 ms actual sleep for any value given in the 1~20ms range)."

In these queue reset loops, using msleep(1) means each iteration may be extended to 10-20ms depending on HZ and timer slack. The document recommends using usleep_range() to avoid this overhead and provide the intended precision.

Let me know if you'd like this explanation folded into a v2 commit message.

--
Mirza

I don't think we care anyway. We probably just want fsleep?