2020-12-15 13:23:40

by Chen Yu

[permalink] [raw]
Subject: [PATCH] xhci: Introduce max wait timeout in xhci_handshake()

The time to finish a xhci_handshake() is platform specific
and sometimes during suspend resume test the followng
errors were encountered:
[53455.418330] ACPI: Waking up from system sleep state S4
[66838.490856] xhci_hcd 0000:00:14.0: xHCI dying, ignoring interrupt.
Shouldn't IRQs be disabled?
After changing the poll time granularity from 1 usec to 20 usec in
xhci_handshake() this issue was not reproduced. While tuning on the
poll time granularity might be painful on different platforms, it is
applicable to introduce a module parameter to allow the xhci driver to wait
for at max 16 ms.

Reported-by: "Muchowski, MaciejX" <[email protected]>
Signed-off-by: Chen Yu <[email protected]>
---
drivers/usb/host/xhci.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/usb/host/xhci.c b/drivers/usb/host/xhci.c
index d4a8d0efbbc4..b8be9f3cc987 100644
--- a/drivers/usb/host/xhci.c
+++ b/drivers/usb/host/xhci.c
@@ -38,6 +38,10 @@ static unsigned long long quirks;
module_param(quirks, ullong, S_IRUGO);
MODULE_PARM_DESC(quirks, "Bit flags for quirks to be enabled as default");

+static int wait_handshake;
+module_param(wait_handshake, int, 0644);
+MODULE_PARM_DESC(wait_handshake, "Force wait for completion of handshake");
+
static bool td_on_ring(struct xhci_td *td, struct xhci_ring *ring)
{
struct xhci_segment *seg = ring->first_seg;
@@ -74,7 +78,7 @@ int xhci_handshake(void __iomem *ptr, u32 mask, u32 done, int usec)
ret = readl_poll_timeout_atomic(ptr, result,
(result & mask) == done ||
result == U32_MAX,
- 1, usec);
+ 1, wait_handshake ? XHCI_MAX_HALT_USEC : usec);
if (result == U32_MAX) /* card removed */
return -ENODEV;

--
2.17.1


2020-12-15 13:41:26

by Greg Kroah-Hartman

[permalink] [raw]
Subject: Re: [PATCH] xhci: Introduce max wait timeout in xhci_handshake()

On Tue, Dec 15, 2020 at 09:22:40PM +0800, Chen Yu wrote:
> The time to finish a xhci_handshake() is platform specific
> and sometimes during suspend resume test the followng
> errors were encountered:
> [53455.418330] ACPI: Waking up from system sleep state S4
> [66838.490856] xhci_hcd 0000:00:14.0: xHCI dying, ignoring interrupt.
> Shouldn't IRQs be disabled?
> After changing the poll time granularity from 1 usec to 20 usec in
> xhci_handshake() this issue was not reproduced. While tuning on the
> poll time granularity might be painful on different platforms, it is
> applicable to introduce a module parameter to allow the xhci driver to wait
> for at max 16 ms.
>
> Reported-by: "Muchowski, MaciejX" <[email protected]>

I doubt the "X" is part of this person's name, please just spell it out
without the "," please.

> Signed-off-by: Chen Yu <[email protected]>
> ---
> drivers/usb/host/xhci.c | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/usb/host/xhci.c b/drivers/usb/host/xhci.c
> index d4a8d0efbbc4..b8be9f3cc987 100644
> --- a/drivers/usb/host/xhci.c
> +++ b/drivers/usb/host/xhci.c
> @@ -38,6 +38,10 @@ static unsigned long long quirks;
> module_param(quirks, ullong, S_IRUGO);
> MODULE_PARM_DESC(quirks, "Bit flags for quirks to be enabled as default");
>
> +static int wait_handshake;
> +module_param(wait_handshake, int, 0644);
> +MODULE_PARM_DESC(wait_handshake, "Force wait for completion of handshake");

This is not the 1990's, we are not adding new module parameters that no
one will know how to change.

Make this dynamic, and per-device, and work properly instead. This can
not handle multiple controllers in the system at all :(

thanks,

greg k-h