PERFORCE change 1205651 for review

[John-Mark Gurney]

John Baldwin wrote this message on Tue, Feb 03, 2015 at 15:37 -0500:
> On Monday, February 02, 2015 7:12:49 pm John-Mark Gurney wrote:
> > http://p4web.freebsd.org/@@1205651?ac=10
> > 
> > Change 1205651 by jmg at jmg_pciehp on 2015/02/03 00:12:18
> > 
> > 	Add inital work to make PCIe HotPlug work...  This has been tested
> > 	w/:
> > 	<inserted ExpressCard>
> > 	uart3: <MosChip MCS9922 PCIe to Peripheral Controller> port 0x3000-0x3007 
> at device 0.0 on pci2
> > 	uart4: <MosChip MCS9922 PCIe to Peripheral Controller> port 0x3008-0x300f 
> at device 0.1 on pci2
> > 	<removed ExpressCard>
> > 	uart3: detached
> > 	uart4: detached
> > 	<inserted ExpressCard>
> > 	uart3: <MosChip MCS9922 PCIe to Peripheral Controller> port 0x3000-0x3007 
> at device 0.0 on pci2
> > 	uart4: <MosChip MCS9922 PCIe to Peripheral Controller> port 0x3008-0x300f 
> at device 0.1 on pci2
> > 	
> > 	This work is based upon gavin's project branch, but fixed
> > 	to make some things work..
> > 	
> > 	Right now there is a nasty bit as we don't have a way for a
> > 	non-acpi driver to add a new PCI device, so the code hard codes
> > 	the extra space necessary for ACPI...  This will be fixed by
> > 	adding necessary bus functions to tell ACPI to rescan the bus,
> > 	etc...
> 
> Note that Ryan already had to solve this problem for SRIOV, and I think his 
> solution will work fine.  In particular, you can have a pci_if.m method that 
> the bus drivers implement that is something like 'pci_add_device()' to tell it 
> about the new device(s) it needs to rescan without forcing an entire bus 
> rescan (though a bus rescan might be useful for other reasons).

Cool.. I'll take a look at this to see how to integrate this..

I wrote this code last year, around July, and brought up to HEAD and
verified that it still worked... So, I know that there is lots of work
to do...

> > +/* Interesting values for PCIe Hotplug */
> > +struct pcicfg_hp {
> > +    struct task	hp_inttask;
> > +    struct callout_handle hp_dllhndl;
> > +    int		hp_cnt;		/* Giant locked */
> > +    uint32_t	hp_slotcap;	/* cache this */
> > +};
> 
> Do not use a callout_handle.  timeout() is about to be removed from the tree 
> (there is only one consumer left).  Use a struct callout instead.

Yeh, I was just looking at that for another reason..  I'll update it
shortly..

oh, btw, has it been anounced that timeout is being removed beyond
-arch or -current? i.e. was it marked deprecated in 10?  if so, isn't
that distruptive to third party code that might be using it?

The reason I ask is that there were some ciphers I was thinking of
removing from OpenCrypto, but if we can remove features w/o a full
branch release, then I'll remove the ciphers before 11...

Thanks for the review!

-- 
  John-Mark Gurney				Voice: +1 415 225 5579

     "All that I will do, has been done, All that I have, has not."