Messages in this thread | | | From | Bjorn Helgaas <> | Date | Tue, 9 Dec 2014 11:21:11 -0700 | Subject | Re: [PATCH] PCI: Clear bridge MEM_64 flag if one child does not support it |
| |
On Tue, Dec 9, 2014 at 12:56 AM, Wei Yang <weiyang@linux.vnet.ibm.com> wrote:
> As you mentioned in another thread, "5b28541552ef is taking the wrong > approach". (http://www.spinics.net/lists/linux-pci/msg37374.html) Maybe I > don't catch it clearly. Put a 32bit prefetchable resource in a 32bit > non-prefetchable bridge window is a bad idea?
A 32-bit prefetchable resource *can* be put in a 32-bit non-prefetchable window, but the device won't perform as well as it would if the resource were in a prefetchable window.
What I object to is the fact that we put a 32-bit prefetchable resource in the non-prefetchable window and leave the 64-bit prefetchable window unused. This gives up performance for no benefit.
> But in my mind, if the bridge > prefetchable window is 64bit, we can't put a 32bit prefetchable resource in > it.
If the window is programmed to be above 4GB, of course we can't put a 32-bit resource in it. My point is that if the bridge *supports* a 64-bit prefetchable window, we can decide where to place it. If we put the window below 4GB, we can put a 32-bit prefetchable resource in it.
I think maybe you're thinking of "64-bit window" as "a window programmed to be above 4GB." I'm using "64-bit window" to mean "a window that supports 64-bit addressing," i.e., one where PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 are implemented. That's analogous to the way we talk about 64-bit BARs. A 64-bit BAR is still a 64-bit BAR even if it is currently programmed to be below 4GB.
Bjorn
| |