Remove unnecessary sema, use msleep instead

[howard0su_gmail.com]

Summary

Code Cleanup to remove sema

[honzhan_microsoft.com]

sys/dev/hyperv/storvsc/hv_storvsc_drv_freebsd.c

625

Which resource need to be protected by this lock?

[decui_microsoft.com]

sys/dev/hyperv/storvsc/hv_storvsc_drv_freebsd.c

625

It's to make the msleep() usages work.
msleep(xxx, &sc->hslock, xxx,...) assumes the lock has been required.

996

I assume you've checked it's OK to remove the M_ZERO.

However, I found the "retries" field of hv_storvsc_request may be used before proper initialization:

grep 'retries = ' dev/hyperv/storvsc/ -nr

dev/hyperv/storvsc/hv_storvsc_drv_freebsd.c:2027: reqp->retries = 0;

We can see it's only initialized once in storvsc_io_done(), but in this function, it's also tested before the initialization.

1007

Why is it removed? reqp->softc is used in several places.

[decui_microsoft.com]

sys/dev/hyperv/storvsc/hv_storvsc_drv_freebsd.c

625

typo.. "required" -> "acquired".

[howard0su_gmail.com]

sys/dev/hyperv/storvsc/hv_storvsc_drv_freebsd.c

625

Yes. This is due to avoid race condition between sleep and wakeup. In our case, it is not possible but WITNESS requires it anyhow.

996

This is actual a pre-allocate buf pool. every time, grab one from the pool, we will call bzero to initialize it. so it is not needed to zero it here.

1007

it will bzero when grab it from the pool. and set softc again.

[decui_microsoft.com]

LGTM

sys/dev/hyperv/storvsc/hv_storvsc_drv_freebsd.c

996

Got it -- I saw that in storvsc_action()'s case XPT_SCSI_IO.

1007

Got it.

[sepherosa_gmail.com]

Most parts no longer apply; and we will take different approach, i.e. vmbus_xact.