On Sun, 22 Dec 2002 11:38:14 +0100 (CET), Jimi wrote:
>I want to discuss the design of FindDevice/FindClass APIs.
>
>I have to options. First use a zero-based index to get multiple matches e.g.
>FindDevice (0x1000, -1, FIND_STRAIGHT, 0) would get the 1st device
>FindDevice (0x1000, -1, FIND_STRAIGHT, 1) would get the 2nd device and so on.
>
>Or via LastMatch HWDeviceID which would give a slight performance increase, because i
>could directly
>match the HWDeviceID first (this would also work via internal cache) and then search
>from that position w/o
>going through all Base-Block-Area-Segments e.g.
>HWDeviceID = FindDevice(0x1000, -1, FIND_STRAIGHT, 0) would get the 1st device
>NextHWDeviceID = FindDevice(0x1000, -1, FIND_STRAIGHT, HWDeviceID) would get the 2nd.
>
>What do you think?
While the first one resembles the BIOS PCI functions I like the second
approach more.
>I also think that the Base-Block access APIs (for searches described by Dani) should
>copy out the
>Base-Blocks to a caller supplied buffer. This needs to get done imho, because the
>device could get detached
>and the Base-Block could get overwritten with a different one, so a pointer wouldn't
>work. The base-block is
>only 64bytes, so this should be fine.
Copy out to caller.
>The other option would be to offer 2 separate APIs. One that would only
>allow access to fixed HWDevices and giving back a pointer. The other one for hotplug
>devices that would
>copy over. Also one could remove hotplug device support for this, because the
>attachment messages already
>contain a pointer to the base-block (which is okay, because the pointer is true at
>least till hotplug notification
>is done, even if the hotplug device gets unplugged during that notification).
>
>Getting the full detection area from Ring-3 (application!) will get somewhat tricky.
>The current method is
>opening gconfig$ and reading till EOF. Now, that the detection area is not linear
>anymore, I will have to
>transfer single base-blocks only and remember the last Base-Block position (aka
>Base-Block Index) that got
>read out, so if a device detach occurs during read-out, the data won't get mangled
>up. The read-out
>detection area will be linear again that way and built up like the original one.
You'll have to keep the read state per open file handle!
Ciao
Dani
-----------
To unsubscribe yourself from this list, send the following message
to [EMAIL PROTECTED]
unsubscribe acpi-os2
end
