On Mon, Dec 20, 2021 at 03:40:17PM +0100, Peter Zijlstra wrote:
> On Fri, Dec 17, 2021 at 04:19:20PM +0100, Greg KH wrote:
> > On Fri, Dec 17, 2021 at 02:55:07PM +0100, Peter Zijlstra wrote:
> > > On Fri, Dec 17, 2021 at 01:01:43PM +0200, Mathias Nyman wrote:
> > > > I can reproduce this.
> > > > Looks like problems started when driver converted to readl_poll_timeout_atomic() in:
> > > >
> > > > 796eed4b2342 usb: early: convert to readl_poll_timeout_atomic()
> > >
> > > I can confirm, reverting that solves the boot hang, things aren't quite
> > > working for me though.
> > >
> > > > Seems to hang when read_poll_timeout_atomic() calls ktime_* functions.
> > > > Maybe it's too early for ktime.
> > >
> > > It certainly is, using ktime for delay loops sounds daft to me anyhow.
> >
> > It was a "find a pattern and replace it with a function call" type of
> > cleanup series. It's obviously wrong, I will go revert it now.
>
> 796eed4b2342 is definitely wrong, but instead of a straight up revert,
> perhaps do something like this ?
>
> ---
> Subject: usb: early: Revert from readl_poll_timeout_atomic()
>
> Reverts commit 796eed4b2342 ("usb: early: convert to
> readl_poll_timeout_atomic()") and puts in a comment to avoid the same
> happening again.
>
> Specifically that commit is wrong because readl_poll_timeout_atomic()
> relies on ktime() working while this earlyprintk driver should be usable
> long before that.
>
> Fixes: 796eed4b2342 ("usb: early: convert to readl_poll_timeout_atomic()")
> Debugged-by: Mathias Nyman <mathias.nyman@xxxxxxxxxxxxxxx>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> ---
> drivers/usb/early/xhci-dbc.c | 18 ++++++++++++++----
> 1 file changed, 14 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/usb/early/xhci-dbc.c b/drivers/usb/early/xhci-dbc.c
> index 933d77ad0a64..ff279a830653 100644
> --- a/drivers/usb/early/xhci-dbc.c
> +++ b/drivers/usb/early/xhci-dbc.c
> @@ -14,7 +14,6 @@
> #include <linux/pci_ids.h>
> #include <linux/memblock.h>
> #include <linux/io.h>
> -#include <linux/iopoll.h>
> #include <asm/pci-direct.h>
> #include <asm/fixmap.h>
> #include <linux/bcd.h>
> @@ -136,9 +135,20 @@ static int handshake(void __iomem *ptr, u32 mask, u32 done, int wait, int delay)
> {
> u32 result;
>
> - return readl_poll_timeout_atomic(ptr, result,
> - ((result & mask) == done),
> - delay, wait);
> + /*
> + * This must not be readl_poll_timeout_atomic(), as this is used
> + * *early*, before ktime lives.
> + */
> + do {
> + result = readl(ptr);
> + result &= mask;
> + if (result == done)
> + return 0;
> + udelay(delay);
> + wait -= delay;
> + } while (wait > 0);
> +
> + return -ETIMEDOUT;
> }
>
> static void __init xdbc_bios_handoff(void)
Please see c4d936efa46d ("Revert "usb: early: convert to
readl_poll_timeout_atomic()"") in Linus's tree. I already added a
comment much like yours. If that's not sufficient, I'll be glad to
re-word it.
thanks,
greg k-h
