Dan Mick wrote: > Christopher Mi wrote: >> Dan Mick wrote: >>> Christopher Mi wrote: >>>> Dan Mick wrote: >>>>> How do these tools access PCI config space? >>>>> >>>> Hi Dan, >>>> >>>> Solaris doesn't have the corresponding file system in Linux, so >>>> it uses the method describe in the following section of PCI Local >>>> Bus Specification v2.2 to access PCI config space. >>>> 3.2.2.3.2. Software Generation of Configuration Transactions >>>> >>>> And because of the above reason, PCI Express config space can be >>>> accessed and we can't port it to SPARC architecture for Solaris. >>>> >>>> In the pciutils-3.0.3 source code, it uses the functions conf1_read >>>> (line 117 in lib/i386-ports.c) and conf1_write(line 144 in >>>> lib/i386-ports.c). >>>> >>>> Thanks, >>>> Chris >>> >>> That's really really bad; collisions in accessing PCI config space >>> can result, and I'm guessing that doesn't handle MMIO config space >>> access at all? >>> >>> scanpci and Xorg have recently moved to using libpciaccess for these >>> accesses, and I think pcitools should as well. >>> >> Hi Dan, >> >> If the collisions happen, it is the problem of pciutils. They should >> fix it and we integrate the new version >> into sfw gate. > > well, someone should fix it, and the fix should be integrated; I'm > suggesting that maybe that should happen at "port to Solaris" time, > which seems to be now. > Nothing stops the team integrating into Solaris from doing The Right > Thing. Hi Dan,
Yes, I agree that quality and stability are very important for Solaris. I'll investigate this problem. But I'm not the PCI guru. Could you please describe the problem accurately, or give me some bug IDs? And I think we can discuss it off-line. Thanks, Chris > >> I haven't investigate libpciaccess and scanpci yet. But I think if I >> use libpciaccess in pciutils, >> pciutils is not itself. Maybe I can't integrate it into sfw gate. >> >> Thanks, >> Chris >
