On November 28, 2018 10:24:04 AM GMT+01:00, Thierry Reding <thierry.reding@xxxxxxxxx> wrote: >On Thu, Nov 01, 2018 at 10:41:52AM +0200, Peter De Schrijver wrote: >> On Thu, Nov 01, 2018 at 02:52:29AM +0100, Marcel Ziswiler wrote: >> > From: Marcel Ziswiler <marcel.ziswiler@xxxxxxxxxxx> >> > >> > As UARTB and VFIR share their clock enable bit it is rather unwise >for >> > the kernel to turn off the VFIR one should that be unused (and >> > potentially vice versa but so far there anyway is no VFIR driver). >> > >> > Without this patch trying to use UARTB with the regular 8250 driver >> > will freeze as soon as ttyS1 is accessed after boot. Luckily, using >the >> > high-speed Tegra serial driver won't exhibit the issue as clocks >are >> > dynamically enabled/disabled on every access. >> > >> > This has been reproduced both on Apalis T30 as well as Apalis TK1 >but >> > may be an issue on all Tegra UARTB's which share the clock enable >with >> > VFIR. >> > >> >> Ah.. the correct fix for this is to initialize the enable_refcnt >based on the >> hw state. This is done in 9619dba8325fce098bbc9ee2911d1b0150fec0c9 >for >> periph gate clocks, but obviously also applies to normal periph >clocks. > >Hi Marcel, > >were you going to send a new version with the alternative fix as >suggested by Peter? Yes, sorry. Let me look at that now. >Thierry
