Re: [Qemu-devel] [PATCH V5 0/7] pxb: fix 64-bit MMIO allocation

[Marcel Apfelbaum <marcel@redhat.com>]

On 07/26/2016 09:30 PM, Michael S. Tsirkin wrote:
> On Sun, Jul 17, 2016 at 07:53:07PM +0300, Marcel Apfelbaum wrote:
>>
>> v4 -> v5:
>>   Addressed the pull request issues: (Peter Maydell)
>>   See: https://lists.gnu.org/archive/html/qemu-devel/2016-07/msg00882.html
>>   - cland warning -> "hw/pci/pci.c:196:23: runtime error: shift exponent -1 is negative":
>>     The PCIe Root port was not initialized properly, the interrupt pin was left 0. This
>>     is a long standing issue exposed by the new test. (Patch 1/7)
>>   - 'make check' fails on 32-bit:
>>      Fix it by changing the ivshmem mem size from 4G
>>      to 1G, since 4G is not a valid value on 32-bit archs. (Patch 2/7)
>>      (4G is truncated to 0 on 32-bit systems)
>>   - Rebased on mst's pci branch.
>>   Since all the new changes are not related to the series, I kept the existing
>>   "Reviewed-by"/"Tested-by" signatures.
>

Hi Michael,

> Applied,

Thanks!

Can you please apply also '[PATCH V3] hw/virtio-pci: fix virtio behaviour'?
(https://lists.gnu.org/archive/html/qemu-devel/2016-07/msg04744.html)

I think it worth having it in QEMU 2.7.

  but dropped tests for now - from experience they are too likely
> to break on weird platforms.
> Pls post new tests after 2.7.
>

Sure, I'll do that.

> Also, I would really like to see options to speed tests up.
> How about running them under kvm?
> There are a couple of minor acpi differences when running with kvm,
> we'll need to white-list them in some way.
>

Sure, we can do kvm 'variant' maybe for for 2.8.

Thanks,
Marcel

>
>

>> v3 -> v4:
>>  Addressed Igor's comments (thanks for the productive review!)
>>  - Split pxb test patch (previously patch 3/3) into the test itself (patch 1/6) and the blobs (patch 6/6).
>>  - New patch declaring pxb/pxb-pxie as not hot-pluggable.
>>     - Note that it does not solve the DSDT issue, but it is a prerequisite for the next patch.
>>  - New patch solving the DSDT issue spotted by Igor.
>>      - Using V=1 DIFF=diff make check does make it easier to review the ACPI changes, thanks.
>>  - Patches 4 and 5 untouched (previously patches 1/3 and 2/3)
>>
>> v2 -> v3:
>>  - split original series "pci: better support for 64-bit MMIO allocation" into 2 series:
>>     - this is the first part dealing with correct 64-bit MMIO ACPI computation
>>     - the second one will include 64-bit MMIO reservation for PCI hotplug
>>  - Add pxb/pxb-pcie tests (Igor) - See diffs below (*)
>>  - Re-based on latest master.
>>
>> v1 -> v2:
>>  - resolved some styling issues (Laszlo)
>>  - rebase on latest master (Laszlo)
>>
>>
>>
>> 64-bit BARs allocations fix for devices behind PXBs/PXB-PCIEs.
>>
>> In build_crs() the calculation and merging of the ranges already happens
>> in 64-bit, but the entry boundaries are silently truncated to 32-bit in the
>> call to aml_dword_memory(). Fix it by handling the 64-bit MMIO ranges separately.
>>
>>
>> Thank you,
>> Marcel
>>
>> Marcel Apfelbaum (7):
>>   hw/pcie-root-port: Fix PCIe root port initialization
>>   tests/acpi: add pxb/pxb-pcie tests
>>   hw/pxb: declare pxb devices as not hot-pluggable
>>   hw/acpi: fix a DSDT table issue when a pxb is present.
>>   acpi: refactor pxb crs computation
>>   hw/apci: handle 64-bit MMIO regions correctly
>>   tests/acpi: Add pxb/pxb-pcie tests blobs
>>
>>  hw/i386/acpi-build.c                   | 131 ++++++++++++++++++++++++---------
>>  hw/pci-bridge/ioh3420.c                |   1 +
>>  hw/pci-bridge/pci_expander_bridge.c    |   2 +
>>  tests/acpi-test-data/pc/DSDT.pxb       | Bin 0 -> 6286 bytes
>>  tests/acpi-test-data/q35/DSDT.pxb_pcie | Bin 0 -> 9098 bytes
>>  tests/bios-tables-test.c               |  37 ++++++++++
>>  6 files changed, 135 insertions(+), 36 deletions(-)
>>  create mode 100644 tests/acpi-test-data/pc/DSDT.pxb
>>  create mode 100644 tests/acpi-test-data/q35/DSDT.pxb_pcie
>>
>> --
>> 2.4.3