[RFC] PCI: skip pci host bridge as early as possible
Lucas Stach
dev at lynxeye.de
Tue Jun 16 12:21:36 PDT 2015
Am Dienstag, den 16.06.2015, 22:22 +0300 schrieb Antony Pavlov:
> On Tue, 16 Jun 2015 21:04:47 +0200
> Lucas Stach <dev at lynxeye.de> wrote:
>
> > Hi Antony,
> >
> > I don't think ignoring the host bridge is the right thing to do in the
> > first place and unfortunately I'm unable to reproduce your problem.
>
> What is your qemu version?
>
QEMU emulator version 2.1.3 (qemu-2.1.3-7.fc21)
>
> Anyway linux PCI initialization code differ from barebox PCI initialization code:
> linux code tries to revert PBAR contents after determening PBAR sizes.
>
> Here is the patch that adds linux PCI setup behaviour to barebox:
>
Does this fix your problem?
Linux puts back the previous content as it tries to keep as much as
possible from a potentially existing firmware configuration. It doesn't
make too much sense in a bootloader that is essentially the firmware
itself. But if it fixes the problem you are seeing we might still
include something like that.
> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> index 7953c55..b60bd5b 100644
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -141,9 +141,12 @@ static void setup_device(struct pci_dev *dev, int max_bar)
>
> for (bar = 0; bar < max_bar; bar++) {
> resource_size_t last_addr;
> + u32 l;
>
> + pci_read_config_dword(dev, PCI_BASE_ADDRESS_0 + bar * 4, &l);
> pci_write_config_dword(dev, PCI_BASE_ADDRESS_0 + bar * 4, 0xfffffffe);
> pci_read_config_dword(dev, PCI_BASE_ADDRESS_0 + bar * 4, &mask);
> + pci_write_config_dword(dev, PCI_BASE_ADDRESS_0 + bar * 4, l);
>
> if (mask == 0 || mask == 0xffffffff) {
> pr_debug("pbar%d set bad mask\n", bar);
>
>
>
> > Am Dienstag, den 09.06.2015, 22:51 +0300 schrieb Antony Pavlov:
> > > The barebox' pci_scan_bus() function first tries to setup GT64120
> > > host bridge BARs but fails:
> > >
> > > barebox:/ dmesg
> > > pci: pci_scan_bus for bus 0
> > > pci: last_io = 0x18000000, last_mem = 0x10000000, last_mem_pref = 0x00000000
> > > pci: class = 00000600, hdr_type = 00000000
> > > pci: 00:00 [11ab:4620]
> > > pci: pbar0: mask=00000008 NP memory 0 bytes
> > > pci: pbar1: mask=01000008 NP memory -16777216 bytes
> > > pci: pbar2: mask=1c000000 NP memory -469762048 bytes
> > > pci: BAR does not fit within bus np-mem res
> > > pci: skip pci host bridge
> > >
> > I can reproduce up until here. But the messages suggest that barebox is
> > in fact skipping the BAR configuration, so this isn't really a problem.
> >
> > This might be improved by checking for invalid BAR sizes before trying
> > to fit them.
> >
> > > so PCI host bridge and all discovered devices are misconfigured, e.g.
> > > miitool can't find rtl8139 emulated NIC tranceiver:
> > >
> > > rtl8139_eth pci-10ec:8139.0: rtl8139 (rev 20) at 90: 0200 (base=8f000000)
> > > rtl8139_eth pci-10ec:8139.0: enabling bus mastering
> > > ...
> > > barebox:/ miitool
> > > miibus0: registered phy as /dev/phy0
> > > phy0: eth0: No MII transceiver present!.
> > >
> >
> > Works for me:
> >
> > barebox:/ miitool
> > miibus0: registered phy as /dev/phy0
> > phy0: eth0: no autonegotiation, 100baseTx-FD flow-control, link ok
> >
> > > Here is linux-3.19 GT64120 host bridge discovery log:
> > >
> > > PCI host bridge to bus 0000:00
> > > pci_bus 0000:00: root bus resource [mem 0x10000000-0x17ffffff]
> > > pci_bus 0000:00: root bus resource [io 0x1000-0x1fffff]
> > > pci_bus 0000:00: No busn resource found for root bus, will use [bus 00-ff]
> > > pci 0000:00:00.0: [Firmware Bug]: reg 0x14: invalid BAR (can't size)
> > > pci 0000:00:00.0: [Firmware Bug]: reg 0x18: invalid BAR (can't size)
> > > pci 0000:00:00.0: [Firmware Bug]: reg 0x1c: invalid BAR (can't size)
> > > pci 0000:00:00.0: [Firmware Bug]: reg 0x20: invalid BAR (can't size)
> > > pci 0000:00:00.0: [Firmware Bug]: reg 0x24: invalid BAR (can't size)
> > >
> > > The appropriate lspci -v log:
> > >
> > > 00:00.0 Host bridge: Marvell Technology Group Ltd. GT-64120/64120A/64121A System Controller (rev 10)
> > > Subsystem: Red Hat, Inc Device 1100
> > > Control: I/O- Mem- BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
> > > Status: Cap- 66MHz- UDF- FastB2B+ ParErr- DEVSEL=medium >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
> > > Interrupt: pin A routed to IRQ 0
> > > Subsystem: Red Hat, Inc Device 1100
> > > Flags: medium devsel
> > > Memory at <unassigned> (32-bit, prefetchable) [disabled]
> > > Memory at <ignored> (32-bit, prefetchable) [disabled]
> > > Memory at <ignored> (32-bit, non-prefetchable) [disabled]
> > > Memory at <ignored> (32-bit, non-prefetchable) [disabled]
> > > Memory at <ignored> (32-bit, non-prefetchable) [disabled]
> > > I/O ports at <ignored> [disabled]
> > >
> > > It looks like qemu-system-mips does not fully emulates GT64120 host
> > > bridge BARs so skipping host bridge BAR's setup fixes qemu-malta
> > > problem. But can we skip host bridge BAR's setup for other targets?
> > >
> > > Signed-off-by: Antony Pavlov <antonynpavlov at gmail.com>
> > > ---
> > > drivers/pci/pci.c | 10 +++++-----
> > > 1 file changed, 5 insertions(+), 5 deletions(-)
> > >
> > > diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> > > index 3a0e7a5..7953c55 100644
> > > --- a/drivers/pci/pci.c
> > > +++ b/drivers/pci/pci.c
> > > @@ -331,6 +331,11 @@ unsigned int pci_scan_bus(struct pci_bus *bus)
> > > pr_debug("%02x:%02x [%04x:%04x]\n", bus->number, dev->devfn,
> > > dev->vendor, dev->device);
> > >
> > > + if (class == PCI_CLASS_BRIDGE_HOST) {
> > > + pr_debug("skip pci host bridge\n");
> > > + continue;
> > > + }
> > > +
> > > switch (hdr_type & 0x7f) {
> > > case PCI_HEADER_TYPE_NORMAL:
> > > if (class == PCI_CLASS_BRIDGE_PCI)
> > > @@ -391,11 +396,6 @@ unsigned int pci_scan_bus(struct pci_bus *bus)
> > > bus->number, dev->devfn, dev->vendor, dev->device, class, hdr_type);
> > > continue;
> > > }
> > > -
> > > - if (class == PCI_CLASS_BRIDGE_HOST) {
> > > - pr_debug("skip pci host bridge\n");
> > > - continue;
> > > - }
> > > }
> > >
> > > /*
> >
> >
>
>
More information about the barebox
mailing list