Re: [PATCH v2] PCI: Coalesce contiguous regions for host bridges

[Guenter Roeck <linux@roeck-us.net>]

Hi,

On Thu, Apr 01, 2021 at 09:12:52PM +0800, Kai-Heng Feng wrote:
> Built-in graphics on HP EliteDesk 805 G6 doesn't work because graphics
> can't get the BAR it needs:
> [    0.611504] pci_bus 0000:00: root bus resource [mem 0x10020200000-0x100303fffff window]
> [    0.611505] pci_bus 0000:00: root bus resource [mem 0x10030400000-0x100401fffff window]
> ...
> [    0.638083] pci 0000:00:08.1:   bridge window [mem 0xd2000000-0xd23fffff]
> [    0.638086] pci 0000:00:08.1:   bridge window [mem 0x10030000000-0x100401fffff 64bit pref]
> [    0.962086] pci 0000:00:08.1: can't claim BAR 15 [mem 0x10030000000-0x100401fffff 64bit pref]: no compatible bridge window
> [    0.962086] pci 0000:00:08.1: [mem 0x10030000000-0x100401fffff 64bit pref] clipped to [mem 0x10030000000-0x100303fffff 64bit pref]
> [    0.962086] pci 0000:00:08.1:   bridge window [mem 0x10030000000-0x100303fffff 64bit pref]
> [    0.962086] pci 0000:07:00.0: can't claim BAR 0 [mem 0x10030000000-0x1003fffffff 64bit pref]: no compatible bridge window
> [    0.962086] pci 0000:07:00.0: can't claim BAR 2 [mem 0x10040000000-0x100401fffff 64bit pref]: no compatible bridge window
> 
> However, the root bus has two contiguous regions that can contain the
> child resource requested.
> 
> Bjorn Helgaas pointed out that we can simply coalesce contiguous regions
> for host bridges, since host bridge don't have _SRS. So do that
> accordingly to make child resource can be contained. This change makes
> the graphics works on the system in question.
> 
> Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=212013
> Suggested-by: Bjorn Helgaas <bhelgaas@google.com>
> Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>

With this patch in place, I can no longer boot the ppc:sam460ex
qemu emulation from nvme. I see the following boot error:

nvme nvme0: Device not ready; aborting initialisation, CSTS=0x0
nvme nvme0: Removing after probe failure status: -19

A key difference seems to be swapped region addresses:

ok:

PCI host bridge to bus 0002:00^M
pci_bus 0002:00: root bus resource [io  0x0000-0xffff]
pci_bus 0002:00: root bus resource [mem 0xd80000000-0xdffffffff] (bus address [0x80000000-0xffffffff])
pci_bus 0002:00: root bus resource [mem 0xc0ee00000-0xc0eefffff] (bus address [0x00000000-0x000fffff])

bad:

PCI host bridge to bus 0002:00^M
pci_bus 0002:00: root bus resource [io  0x0000-0xffff]
pci_bus 0002:00: root bus resource [mem 0xc0ee00000-0xc0eefffff] (bus address [0x00000000-0x000fffff])
pci_bus 0002:00: root bus resource [mem 0xd80000000-0xdffffffff] (bus address [0x80000000-0xffffffff])

and then bar address assignments are swapped/changed.

ok:

pci 0002:00:06.0: BAR 0: assigned [mem 0xd80000000-0xd83ffffff]^M
pci 0002:00:06.0: BAR 1: assigned [mem 0xd84000000-0xd841fffff]^M
pci 0002:00:02.0: BAR 0: assigned [mem 0xd84200000-0xd84203fff 64bit]^M
pci 0002:00:01.0: BAR 5: assigned [mem 0xd84204000-0xd842041ff]^M
pci 0002:00:03.0: BAR 0: assigned [io  0x1000-0x107f]^M
pci 0002:00:03.0: BAR 1: assigned [mem 0xd84204200-0xd8420427f]^M
pci 0002:00:01.0: BAR 4: assigned [io  0x1080-0x108f]^M
pci 0002:00:01.0: BAR 0: assigned [io  0x1090-0x1097]^M
pci 0002:00:01.0: BAR 2: assigned [io  0x1098-0x109f]^M
pci 0002:00:01.0: BAR 1: assigned [io  0x10a0-0x10a3]^M
pci 0002:00:01.0: BAR 3: assigned [io  0x10a4-0x10a7]^M
pci_bus 0002:00: resource 4 [io  0x0000-0xffff]^M
pci_bus 0002:00: resource 5 [mem 0xd80000000-0xdffffffff]^M
pci_bus 0002:00: resource 6 [mem 0xc0ee00000-0xc0eefffff]^M

bad:

pci 0002:00:06.0: BAR 0: assigned [mem 0xd80000000-0xd83ffffff]^M
pci 0002:00:06.0: BAR 1: assigned [mem 0xd84000000-0xd841fffff]^M
pci 0002:00:02.0: BAR 0: assigned [mem 0xc0ee00000-0xc0ee03fff 64bit]^M
pci 0002:00:01.0: BAR 5: assigned [mem 0xc0ee04000-0xc0ee041ff]^M
pci 0002:00:03.0: BAR 0: assigned [io  0x1000-0x107f]^M
pci 0002:00:03.0: BAR 1: assigned [mem 0xc0ee04200-0xc0ee0427f]^M
pci 0002:00:01.0: BAR 4: assigned [io  0x1080-0x108f]^M
pci 0002:00:01.0: BAR 0: assigned [io  0x1090-0x1097]^M
pci 0002:00:01.0: BAR 2: assigned [io  0x1098-0x109f]^M
pci 0002:00:01.0: BAR 1: assigned [io  0x10a0-0x10a3]^M
pci 0002:00:01.0: BAR 3: assigned [io  0x10a4-0x10a7]^M
pci_bus 0002:00: resource 4 [io  0x0000-0xffff]^M
pci_bus 0002:00: resource 5 [mem 0xc0ee00000-0xc0eefffff]^M
pci_bus 0002:00: resource 6 [mem 0xd80000000-0xdffffffff]^M

Reverting this patch fixes the problem.

Guenter

---
bisect log:

# bad: [f55966571d5eb2876a11e48e798b4592fa1ffbb7] Merge tag 'drm-next-2021-07-08-1' of git://anongit.freedesktop.org/drm/drm
# good: [e9f1cbc0c4114880090c7a578117d3b9cf184ad4] Merge tag 'acpi-5.14-rc1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
git bisect start 'f55966571d5e' 'e9f1cbc0c411'
# bad: [b0dfd9af28b60d7ec42c359ae84c1ba97e093100] Merge tag 'clk-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux
git bisect bad b0dfd9af28b60d7ec42c359ae84c1ba97e093100
# bad: [364a716bd73e9846d3118a43f600f8f517658b38] Merge branch 'pci/host/intel-gw'
git bisect bad 364a716bd73e9846d3118a43f600f8f517658b38
# good: [c9fb9042c98df94197a1ba4cf14a77c8053b0fae] Merge branch 'pci/p2pdma'
git bisect good c9fb9042c98df94197a1ba4cf14a77c8053b0fae
# bad: [7132700067f234d37c234e5d711bb49ea06d2352] Merge branch 'pci/sysfs'
git bisect bad 7132700067f234d37c234e5d711bb49ea06d2352
# bad: [131e4f76c9ae9636046bf04d19d43af0e4ae9807] Merge branch 'pci/resource'
git bisect bad 131e4f76c9ae9636046bf04d19d43af0e4ae9807
# good: [411e2a43d210e98730713acf6d01dcf823ee35e3] PCI: Work around Huawei Intelligent NIC VF FLR erratum
git bisect good 411e2a43d210e98730713acf6d01dcf823ee35e3
# good: [e92605b0a0cdafb6c37b9d1ad24fe1cf8280eeb6] Merge branch 'pci/pm'
git bisect good e92605b0a0cdafb6c37b9d1ad24fe1cf8280eeb6
# bad: [65db04053efea3f3e412a7e0cc599962999c96b4] PCI: Coalesce host bridge contiguous apertures
git bisect bad 65db04053efea3f3e412a7e0cc599962999c96b4
# first bad commit: [65db04053efea3f3e412a7e0cc599962999c96b4] PCI: Coalesce host bridge contiguous apertures