If you get an execution trace right before this happens that might shed some light on it. Tracking how the address that is being used is assembled by the cpu is a good start.
Nothing jumps out at me though, so I'm pretty confused why I don't see the problem and you do. Ali On Feb 16, 2012, at 4:57 PM, Pritha Ghoshal wrote: > >> Hi Pritha, >> I took a old kernel from when i published the original paper in 2009 >> (2.6.27) > and it seems to work with the e1000 NIC if I just make the following change: >> diff -r ef8630054b5e configs/common/FSConfig.py--- > a/configs/common/FSConfig.py Tue Feb 14 14:15:30 2012 -0500+++ > b/configs/common/FSConfig.py Thu Feb 16 11:28:32 2012 -0600 <at> <at> > -58,7 > +58,7 <at> <at> def makeLinuxAlphaSystem(mem_mode, mdesc = None): > IO_address_space_base = 0x80000000000 class BaseTsunami(Tsunami):- > ethernet = NSGigE(pci_bus=0, pci_dev=1, pci_func=0)+ ethernet = > IGbE_e1000(pci_bus=0, pci_dev=1, pci_func=0) ide = > IdeController(disks= > [Parent.disk0, Parent.disk2], pci_func=0, > pci_dev=0, > pci_bus=0) >> I don't know what kernel you're using but it's likely there is either an >> issue > with the configuration of it or perhaps something has broken in the alpha > branch. >> From an Alpha/Tsunami perspective, virtual addresses that start with ffffc >> map > to physical memory directly and addresses that start with ffffd map to the > i/o. > I'd have to look at the tsunami memory map documentation which isn't close at > hand to what 80000000008 could be, but it doesn't seem right. You could use > the > PCIDev Ethernet trace flags to understand what addresses the PCI devices are > getting assigned. >> Thanks, >> Ali >> > > Hi Ali, > > I had tried using the same modification in FSConfig.py, and even the kernel I > am > using in 2.6.27. Should I try to build the kernel again and check? > > Regarding the addresses, I used the flag BusAddrRanges, I am not able to see > any > of the address ranges mapped to the IOBus which starts from 0x80000000000. > The > closest one I can see is: > 0: drivesys.tsunami.fake_OROM: registering range: 0x800000a0000-0x60000 > 0: drivesys.iobus: Adding range 0x800000a0000 - 0x800000fffff for id 12 > All the rest seem to be starting with 0x801**** instead of 0x800****. > For membus this is the range: > 0: drivesys.membus: Adding range 0x80000000000 - 0xffffffffffffffff for > id > 2 > So there is one range of addresses which are not mapped to anywhere on IO > bus, > even though the > IO_address_space_base = 0x80000000000 > is set in FSConfig.py. > But the mapping of addresses do not change from NSGigE adapter mappings, and > there is no error in that case. > > I enabled Fetch flag to see the addresses being accessed before the error > condition, and in the e100 NIC, this is the faulting address: > 51061947500: drivesys.cpu: Fetch: PC:0xfffffc0000324800 > 51061947500: drivesys.cpu: Address is fffffd0000000008, Physical address > 80000000008 > > But in the case of NSGigE, the same address brings forward a different > virtual > address for the read: > 51379655000: drivesys.cpu: Fetch: PC:0xfffffc0000324800 > 51379655000: drivesys.cpu: Address is fffffd0009000018, Physical address > 80009000018 > > For the other cases, the sequence addresses are identical in case of NSGigE > and > IGbE_e1000 adapters. eg: > NSGigE: > 51690478500: drivesys.cpu: Fetch: PC:0xfffffc0000319ce4 > 51690478500: drivesys.cpu: Address is fffffc000085e208, Physical address > 85e208 > e1000: > 51061946500: drivesys.cpu: Fetch: PC:0xfffffc0000319ce4 > 51061946500: drivesys.cpu: Address is fffffc000085e208, Physical address > 85e208 > > > Could you give any pointer regarding where this faulty address is getting > generated for this particular case? > > Pritha > > > > > > _______________________________________________ > gem5-users mailing list > gem5-users@gem5.org > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users _______________________________________________ gem5-users mailing list gem5-users@gem5.org http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
