On Fri, Oct 02, 2026 at 10:57:05PM +0200, Kirill A. Korinsky wrote:

> On Fri, 02 Oct 2026 22:39:52 +0200,
> Mark Kettenis <mark.kettenis@xs4all.nl> wrote:
> > 
> > > Date: Fri, 02 Oct 2026 22:12:26 +0200
> > > From: Kirill A. Korinsky <kirill@korins.ky>
> > > 
> > > Marcus, Mark,
> > > 
> > > we had already increased register status polling bus wait in v 1.47, and my
> > > Honor needs increase a bit as well.
> > > 
> > > With 600ms maximal delay it survives suspend, at least it goes out from X11
> > > to console and I see that it detached USB and waits. It doesn't response on
> > > ping, but pressing any key reattach the USB, it goes back to X11 and
> > > responses to ping.
> > > 
> > > I had played with different values and printed threshold when it doesn't
> > > survive. On my device it needs 59 or 60 retries usually to pass.
> > > 
> > > I think it is safe to increase and ask to include to 8.0, and probably we
> > > can put here something larger than 60. For example 75.
> > > 
> > > I read the code near by and that looks quite safe.
> > > 
> > > Thoughts? Ok?
> > 
> > So our driver seems to wait for 500us.  But Linux uses a timeout of
> > 500ms as far as I can tell.  U-Boot even uses a full second (1000ms).
> > 
> > So I think we should raise retry to 50000.

And FreeBSD seems to use 500ms.

> Like that?

There is no defined wait time for UFSHCI_REG_UICCMD_CMDOP_DME_LINKSTARTUP                                                                                                                       
in the specs what I have checked.  The change adapts with other drivers,                                                                                                                        
and has no harm to existing device behavior, so I am OK to increase it                                                                                                                          
to 500ms.                                                                                                                                                                                       
                                                                                                                                                                                                
One small note for the future; currently ufshci_is_poll() is only used                                                                                                                          
for polling the UFSHCI_REG_IS_UCCS register status.  If we should use                                                                                                                           
it in the future for polling other register status', we eventually                                                                                                                              
should pass the wait time as an argument to ufshci_is_poll().  But for                                                                                                                          
now it's fine to increase it globally.
 
> Index: sys/dev/ic/ufshci.c
> ===================================================================
> RCS file: /home/cvs/src/sys/dev/ic/ufshci.c,v
> diff -u -p -r1.49 ufshci.c
> --- sys/dev/ic/ufshci.c	6 Sep 2026 19:31:26 -0000	1.49
> +++ sys/dev/ic/ufshci.c	2 Oct 2026 20:56:18 -0000
> @@ -311,7 +311,7 @@ int
>  ufshci_is_poll(struct ufshci_softc *sc, uint32_t type)
>  {
>  	uint32_t status;
> -	int i, retry = 50;
> +	int i, retry = 50000;
>  
>  	for (i = 0; i < retry; i++) {
>  		status = UFSHCI_READ_4(sc, UFSHCI_REG_IS);
> 
> 
> -- 
> wbr, Kirill
>