On Sun, Mar 24, 2019 at 03:29:04PM +0100, Thomas Gleixner wrote: >Find a mostly untested patch which implements this below. I just booted it >in a 64bit guest and it did not explode. > >It removes 55 lines of code instead of adding 35 and reduces the binary >size by 408 bytes on 64bit and 128 bytes on 32bit. > >Note, it's a combo of changes (including your patch 1/6) and needs to be >split up. It would be nice if you have time to split it up into separate >patches, add proper changelogs and test the heck out of it on both 32 and >64 bit. If you don't have time, please let me know.
I tried to test mem hotplug with this applied. On x86_64, this looks good. It split ranges well. While I found some problem on testing mem hotplug on x86_32. I start up a guest memory and trying to plug 1G memory. The original memory split range looks good: [ 0.004260] ywtest: [mem 0x00000000-0x00100000] [ 0.004261] ywtest: [mem 0x00000000-0x000fffff] page size 4k [ 0.004268] ywtest: [mem 0x37200000-0x37400000] [ 0.004269] ywtest: [mem 0x37200000-0x373fffff] page size 2M [ 0.004271] ywtest: [mem 0x20000000-0x37200000] [ 0.004272] ywtest: [mem 0x20000000-0x371fffff] page size 2M [ 0.004280] ywtest: [mem 0x00100000-0x20000000] [ 0.004281] ywtest: [mem 0x00100000-0x001fffff] page size 4k [ 0.004281] ywtest: [mem 0x00200000-0x1fffffff] page size 2M [ 0.004292] ywtest: [mem 0x37400000-0x375fe000] [ 0.004293] ywtest: [mem 0x37400000-0x375fdfff] page size 4k While I can't online the new added memory device. And the new memory device's range is out of 4G, which is a little bit strange. > >Thanks, > > tglx > -- Wei Yang Help you, Help me