[PATCH v6 1/2] net: brcm: netXtreme driver

Pali Rohár pali at kernel.org
Fri Nov 5 22:51:49 CET 2021


On Friday 05 November 2021 14:41:44 Roman Bacik wrote:
> On Fri, Nov 5, 2021 at 2:29 PM Pali Rohár <pali at kernel.org> wrote:
> >
> > On Friday 05 November 2021 14:19:46 Roman Bacik wrote:
> > > Hi Pali,
> > >
> > > On Fri, Nov 5, 2021 at 2:09 PM Pali Rohár <pali at kernel.org> wrote:
> > > >
> > > > On Friday 05 November 2021 12:54:24 Roman Bacik wrote:
> > > > > > > + pci_read_word16(bp->pdev,
> > > > > > > +                 PCI_SUBSYSTEM_VENDOR_ID,
> > > > > > > +                 &bp->subsystem_vendor);
> > > > > > > + pci_read_word16(bp->pdev, PCI_SUBSYSTEM_ID, &bp-
> > > > > > >subsystem_device);
> > > > > > > + pci_read_word16(bp->pdev, PCI_COMMAND, &bp->cmd_reg);
> > > > > > > + pci_read_byte(bp->pdev, PCICFG_ME_REGISTER, &bp->pf_num);
> > > > > >
> > > > > > PCICFG_ME_REGISTER looks like an error as there is no such PCI config
> > > > > > space macro. What you are trying to read into pf_num? Currently I do not
> > > > > > know what "pf" abbreviation could mean.
> > > > >
> > > > > PF stands for physical function and pf_num is the number of physical
> > > > > functions configured.
> > > > > The macro is defined in bnxt.h:
> > > > > #define PCICFG_ME_REGISTER                      0x98
> > > >
> > > > pci_read_byte() reads from PCI(e) config space, which is standardized.
> > > > Therefore only standard macro constants from include/pci.h should be
> > > > used. Standard PCI config header is 64 byte long and after that is
> > > > linked list of capabilities. Order of capabilities is not defined.
> > > > Extended capabilities from linked list should be located by macro
> > > > constants PCI_CAP_ID_*.
> > > >
> > > > So above register is part of some extended capability. Correctly it
> > > > should be used some function to locate starting offset of that extended
> > > > capability based on PCI_CAP_ID_* (see pci.h file for these functions)
> > > > and then access that register as offset + PCI_* constant (which defined
> > > > as relative to the start of extended capability). In case standard macro
> > > > for this constant in pci.h is missing, it is a good idea to define it,
> > > > or copy it from linux header file pci_regs.h (to have consistent naming
> > > > of macros).
> > > >
> > > > Could you provide output of 'lspci -nn -vv' from linux for this card?
> > > > Or 'pci display.b ?.?.? 0 0x1000' dump from U-Boot?
> > > > This could help me to under what kind of register that 0x98 is.
> > > >
> > > > I can write this part of code, no problem, just I need to see layout of
> > > > config space of that card.
> > >
> > > Here it is:
> > >
> > > u-boot> pci display.b ?.?.? 0 1000
> >
> > Hello! '?.?.?' is placeholder for bus-device-function PCI address of
> > card. So should replace it by correct address at which is that card.
> > (What you have sent is dump of PCIe Root Port, probably from 0.0.0)
> 
> Oh right, I thought it would print all. Here is Linux output:
> 
> 0008:00:00.0 PCI bridge [0604]: Broadcom Limited Device [14e4:d750]
> 0008:01:00.0 Ethernet controller [0200]: Broadcom Limited Device [14e4:d804]
> 0008:01:00.1 Ethernet controller [0200]: Broadcom Limited Device [14e4:d804]
> 0008:01:00.2 Ethernet controller [0200]: Broadcom Limited Device [14e4:d804]
> 0008:01:00.3 Ethernet controller [0200]: Broadcom Limited Device [14e4:d804]

Hmmm? There is no ethernet controller with id 0x14e4:0x16F0 for which
you have wrote this U-Boot driver.


More information about the U-Boot mailing list