Date: Mon, 2 Dec 2024 17:41:34 +0800 From: Zhenlei Huang <zlei@FreeBSD.org> To: Warner Losh <imp@freebsd.org> Cc: stable@freebsd.org, Edward Tomasz Napierala <trasz@FreeBSD.org> Subject: MFC fixes for uninitialized kernel stack variables in sys/cam or do direct fix for pvscsi driver Message-ID: <0DDE1B66-B794-472D-A901-54FA2FF1E853@FreeBSD.org>
next in thread | raw e-mail | index | archive | help
Hi Warner,
Recently I upgraded some ESXi vms from 13.3 to 13.4 and noticed weird =
report for sas speed.
The boot console has the following,
```
da0 at pvscsi0 bus 0 scbus2 target 0 lun 0
da0: <VMware Virtual disk 2.0> Fixed Direct Access SPC-4 SCSI device
da0: 4294967.295MB/s transfers
```
But camcontrol report the correct value,
```
# camcontrol inquiry da0 -R
pass1: 750.000MB/s transfers, Command Queueing Enabled
```
The `4294967.295MB` is actually 0xffff_ffff or -1 but I do not see any =
logic set those values.
Finally I managed to get the stack trace,
```
_scsi_announce_periph
scsi_announce_periph_sbuf
xpt_announce_periph_sbuf
dadone_proberc
xpt_done_process
xpt_done_td
fork_exit
fork_trampoline
```
and noticed that the last param `cts` of `_scsi_announce_periph(struct =
cam_periph *periph, u_int *speed, u_int *freq, struct ccb_trans_settings =
*cts)`
is from kernel stack and is not properly initialized, latter I found =
some commits related to this,
076686fe0703 cam: make sure to clear CCBs allocated on the stack
ec5325dbca62 cam: make sure to clear even more CCBs allocated on the =
stack
0f206cc91279 cam: add missing zeroing of a stack-allocated CCB.
616a676a0535 cam: clear stack-allocated CCB in the target layer
I applied them to stable/13, rebuild and reboot, now the speed of da0 is =
reported correctly. I also tried to patch the pvscsi driver with few =
lines and
it also works as intended.
```
--- a/sys/dev/vmware/pvscsi/pvscsi.c
+++ b/sys/dev/vmware/pvscsi/pvscsi.c
@@ -1444,6 +1444,10 @@ pvscsi_action(struct cam_sim *sim, union ccb =
*ccb)
cts->proto_specific.scsi.flags =3D =
CTS_SCSI_FLAGS_TAG_ENB;
cts->proto_specific.scsi.valid =3D CTS_SCSI_VALID_TQ;
=20
+ /* Prefer connection speed over sas port speed */
+ /* cts->xport_specific.sas.bitrate =3D 0; */
+ cts->xport_specific.sas.valid =3D 0;
+
ccb_h->status =3D CAM_REQ_CMP;
xpt_done(ccb);
```
Things come clear and I know why this weird speed happens, now it is =
time to decide how to fix it.
Fixing the consumer of cam, aka pvscsi driver, is quite simple and =
promising. I did a quick search it appears other consumers set =
`cts->xport_specific.sas.valid` correctly. It does not convince me as =
I'm quite new to cam subsystem.
Which one do you prefer, MFC commits to stable/13, or do direct fix for =
pvscsi driver to stable/13 ?
Best regards,
Zhenlei
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?0DDE1B66-B794-472D-A901-54FA2FF1E853>