PCI/PM: Reduce D3hot delay with usleep_range()
PCIe r6.0, sec 5.9, requires a 10ms delay between programming a device to
change to or from D3hot and the time the device is next accessed (unless
Readiness Notifications are used).
The 10ms value (PCI_PM_D3HOT_WAIT) doesn't appear directly here because
some chipsets require 120ms for devices *below* them (pci_pm_d3hot_delay)
and some devices require more or less than 10ms (dev->d3hot_delay).
But msleep(10) typically waits about *20*ms, which is more than we need.
Switch to usleep_range() to improve the delay accuracy.
Based on a commit from Sajid in the Pixel 6 kernel tree [1]. On a Pixel 6,
the 10ms delay for the Exynos PCIe device delayed for an average of 19ms.
Switching to usleep_range() decreased the resume time by about 9ms.
[1] 18a8cad68d
[bhelgaas commit log, add timers-howto.rst link]
Link: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/timers/timers-howto.rst?id=v5.19#n73
Link: https://lore.kernel.org/r/20220921212735.2131588-1-willmcvicker@google.com
Signed-off-by: Sajid Dalvi <sdalvi@google.com>
Signed-off-by: Will McVicker <willmcvicker@google.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Matthias Kaehlcke <mka@chromium.org>
This commit is contained in:
parent
4c00cba122
commit
3e347969a5
@ -66,13 +66,15 @@ struct pci_pme_device {
|
||||
|
||||
static void pci_dev_d3_sleep(struct pci_dev *dev)
|
||||
{
|
||||
unsigned int delay = dev->d3hot_delay;
|
||||
unsigned int delay_ms = max(dev->d3hot_delay, pci_pm_d3hot_delay);
|
||||
unsigned int upper;
|
||||
|
||||
if (delay < pci_pm_d3hot_delay)
|
||||
delay = pci_pm_d3hot_delay;
|
||||
|
||||
if (delay)
|
||||
msleep(delay);
|
||||
if (delay_ms) {
|
||||
/* Use a 20% upper bound, 1ms minimum */
|
||||
upper = max(DIV_ROUND_CLOSEST(delay_ms, 5), 1U);
|
||||
usleep_range(delay_ms * USEC_PER_MSEC,
|
||||
(delay_ms + upper) * USEC_PER_MSEC);
|
||||
}
|
||||
}
|
||||
|
||||
bool pci_reset_supported(struct pci_dev *dev)
|
||||
|
Loading…
Reference in New Issue
Block a user