On Tue, 10 Feb 2015, enuhtac wrote:
...
> But unfortunately this does not affect the "ehci_sync_hc: tsleep() = 35"
> error.
> Strangely it seams that "35" means that the code has been compiled without
> the macro "_KERNEL" being defined (in this case "tsleep" is a macro that
> expands to "35").
Hmm? That message is from this code:
...
do {
/* ask for doorbell */
EOWRITE4(sc, EHCI_USBCMD, EOREAD4(sc, EHCI_USBCMD) |
EHCI_CMD_IAAD);
DPRINTFN(1,("ehci_sync_hc: cmd=0x%08x sts=0x%08x\n",
EOREAD4(sc, EHCI_USBCMD), EOREAD4(sc, EHCI_USBSTS)));
/* bell wait */
error = tsleep(&sc->sc_async_head, PZERO, "ehcidi", hz / 2);
DPRINTFN(1,("ehci_sync_hc: cmd=0x%08x sts=0x%08x\n",
EOREAD4(sc, EHCI_USBCMD), EOREAD4(sc, EHCI_USBSTS)));
} while (error && ++tries < 10);
splx(s);
/* release doorbell */
rw_exit_write(&sc->sc_doorbell_lock);
#ifdef DIAGNOSTIC
if (error)
printf("ehci_sync_hc: tsleep() = %d\n", error);
#endif
...
35 == EWOULDBLOCK, which according to tsleep(9) manpage means the tsleep
timed out. So, this function is repeatedly setting some hardware
registers to get a "doorbell" interrupt, then waiting .5 second for it. If
it doesn't get that after 10 tries, it gives up and prints that message
showing the actual failure of "sleep until woken up" call.
So, is the interrupt not happening? Does, but the interrupt handler
doesn't think it should do a wakeup? etc...
Philip
