cvs commit: src/sys/dev/bce if_bce.c src/sys/dev/em if_em.c if_em.h src/sys/dev/mpt mpt.h mpt_pci.c

[M. Warner Losh]

In message: <455B963A.4050200 at samsco.org>
            Scott Long <scottl at samsco.org> writes:
: John Baldwin wrote:
: > On Wednesday 15 November 2006 16:51, Ruslan Ermilov wrote:
: >> On Thu, Nov 16, 2006 at 12:51:19AM +0300, Ruslan Ermilov wrote:
: >>> On Wed, Nov 15, 2006 at 08:04:57PM +0000, John Baldwin wrote:
: >>>> jhb         2006-11-15 20:04:57 UTC
: >>>>
: >>>>   FreeBSD src repository
: >>>>
: >>>>   Modified files:
: >>>>     sys/dev/bce          if_bce.c 
: >>>>     sys/dev/em           if_em.c if_em.h 
: >>>>     sys/dev/mpt          mpt.h mpt_pci.c 
: >>>>   Log:
: >>>>   Add MSI support to em(4), bce(4), and mpt(4).  For now, we only 
: > support
: >>>>   devices that support a maximum of 1 message, and we use that 1 message
: >>>>   instead of the INTx rid 0 IRQ with the same interrupt handler, etc.
: >>>>   
: >>>>   Revision  Changes    Path
: >>>>   1.19      +11 -3     src/sys/dev/bce/if_bce.c
: >>>>   1.164     +11 -2     src/sys/dev/em/if_em.c
: >>>>   1.56      +1 -0      src/sys/dev/em/if_em.h
: >>>>   1.31      +1 -0      src/sys/dev/mpt/mpt.h
: >>>>   1.39      +14 -1     src/sys/dev/mpt/mpt_pci.c
: >>>>
: >>> How will the "vmstat -i" output look like for MSI-enabled devices?
: >>>
: >> irqXXXX, where XXXX>=1024?
: > 
: > s/1024/256/
: > 
: 
: There is a problem here, though.  Newbus prints out the IRQ number after
: a successful device probe phase.  It has no knowledge of MSI at that 
: point, so it just prints out the traditional IRQ value.  At some point,
: this needs to be fixed.  Having the driver tell newbus about its MSI
: intentions in the probe routine is unrealistic, so there is no quick
: fix there.  Probably need to delay printing the device message until
: later in the attach routine, once the driver has set up all of the 
: resources.

I've been wanting to move the printing of the attach string from
post-probe, pre-attach to post-attach for some time now.  We would
then report the resources assigned to the device.  I'm not sure if I'd
print all the resources assigned, or only those the driver activates.
Both sides of the argument have merit, imho.  On the pro side,
resources are used, and printing them will help highlight conflicts.
On the con side, people think it clutters things up too much and might
lead to false expectations.

When I've mentioned this desire at various developer summits, I was
told basically "go for it, but only at freebsd X.0 since people have
dmesg parsers" by many people (maybe even including Scott).

Warner