re: kern/55485: bwfm @ sdmmc hangs, with some diagnosis

[matthew green <mrg%eterna.com.au@localhost>]

more data.  and some likely less broken behaviour fixes, if not
a full fix.

with some help from riastradh@ and jmcneill@ i've tracked down
that when the failure is occuring, the only thing looping is
an interrupt.  we appear to be in dwc_mmc_exec_command() all
the time, but kernhist logging shows that we re-enter this
function many times (one time, 1000 times in 60ms.)

initially, byt checking what bits were set in the
BWFM_SDPCMD_INTSTATUS register, we determined that there was
the SDPCMD_INTSTATUS_XMTDATA_AVAIL bit set, which is not used
by our driver.  then we noticed that the interrupt mask reg
was set to 0xffffffff, with commented version asking for 2
bits we are interested in.  this allowed my test case to work
for about 10x longer than average before failing, and the
failure mode was less hard-hang of bwfm itself.   eg, when
i ran 'ifconfig -a', it took 15 seconds, but it did complete,
and 'ifconfig bwfm down up' restored functionality.  (there
are 3x5s slow periods here, each accompanied by a 'checksum
error' message on the console.)

the patch below helps a lot, but there are still stability.


.mrg.


Index: if_bwfm_sdio.c
===================================================================
RCS file: /cvsroot/src/sys/dev/sdmmc/if_bwfm_sdio.c,v
retrieving revision 1.19
diff -p -u -r1.19 if_bwfm_sdio.c
--- if_bwfm_sdio.c	23 Jun 2020 10:09:33 -0000	1.19
+++ if_bwfm_sdio.c	18 Jul 2020 04:13:51 -0000
@@ -505,9 +505,12 @@ bwfm_sdio_attachhook(device_t self)
 		goto err;
 	}
 
-//	bwfm_sdio_dev_write(sc, SDPCMD_HOSTINTMASK,
-//	    SDPCMD_INTSTATUS_HMB_SW_MASK | SDPCMD_INTSTATUS_CHIPACTIVE);
+#if 1
+	bwfm_sdio_dev_write(sc, SDPCMD_HOSTINTMASK,
+	    SDPCMD_INTSTATUS_HMB_SW_MASK | SDPCMD_INTSTATUS_CHIPACTIVE);
+#else
 	bwfm_sdio_dev_write(sc, SDPCMD_HOSTINTMASK, 0xffffffff);
+#endif
 	bwfm_sdio_write_1(sc, BWFM_SDIO_WATERMARK, 8);
 
 	if (bwfm_chip_sr_capable(bwfm)) {