Re: [Qemu-devel] [PATCH 1/5] target-tile: Firstly add to qemu with minimized features

[Chris Metcalf <cmetcalf@ezchip.com>]

On 2/13/2015 10:08 PM, Chen Gang S wrote:
> On 2/14/15 08:26, Peter Maydell wrote:
>> On 13 February 2015 at 23:50, Chen Gang S <gang.chen@sunrus.com.cn> wrote:
>>> On 2/13/15 05:15, Chen Gang S wrote:
>>>> It almost likes a template for adding an architecture target.
>>>>
>>>> Signed-off-by: Chen Gang <gang.chen.5i5j@gmail.com>
>>>> ---
>>>>   configure                             |   7 ++
>>>>   default-configs/tilegx-linux-user.mak |   1 +
>>>>   target-tile/Makefile.objs             |   1 +
>>>>   target-tile/cpu-qom.h                 |  72 +++++++++++++++
>>>>   target-tile/cpu.c                     | 159 ++++++++++++++++++++++++++++++++++
>>>>   target-tile/cpu.h                     |  84 ++++++++++++++++++
>>>>   target-tile/helper.h                  |   0
>>>>   target-tile/translate.c               |  54 ++++++++++++
>>>>   8 files changed, 378 insertions(+)
>>>>   create mode 100644 default-configs/tilegx-linux-user.mak
>>>>   create mode 100644 target-tile/Makefile.objs
>>>>   create mode 100644 target-tile/cpu-qom.h
>>>>   create mode 100644 target-tile/cpu.c
>>>>   create mode 100644 target-tile/cpu.h
>>>>   create mode 100644 target-tile/helper.h
>>>>   create mode 100644 target-tile/translate.c
>>>>
>>>> diff --git a/configure b/configure
>>>> index 7ba4bcb..ffb4d56 100755
>>>> --- a/configure
>>>> +++ b/configure
>>>> @@ -5191,6 +5191,9 @@ case "$target_name" in
>>>>     s390x)
>>>>       gdb_xml_files="s390x-core64.xml s390-acr.xml s390-fpr.xml"
>>>>     ;;
>>>> +  tilegx)
>>>> +    TARGET_ARCH=tile
>>>> +  ;;
>>> Oh, it needs to use "TARGET_ARCH=tilegx" instead of "TARGET_ARCH=tile",
>>> or can not put "CONFIG_TILE_DIS=y" to config-all-disas.mak, below (when
>>> I am adding "disas/tilegx.c", I meet it).
>> You need to make up your mind about what your architecture
>> is called: is it "tile" or "tilegx"? Be consistent...
>>
> Oh, after think of, for me, I shall follow the other architectures has
> done (e.g. i386).
>
>   - Still use target-tile directory, since 'tile' is the architecture
>     name.
>
>   - 'tilegx' is one kind of cpu of tile, so I need consider about it just
>     like another architectures have done (e.g. i386).

I recommend pretty strongly that you use "tilegx" consistently for qemu.  There is enough commonality in the Linux arch code that I chose "tile" to support both tilepro and tilegx.  But elsewhere, we use "tilegx" vs "tilepro", for example in binutils and gcc.  I would argue "qemu" is much more like binutils or gcc.  Also, note that while x86_64 can run i386 binaries, tilegx can not run tilepro binaries.

>> You should only add this bit to configure in the patchset
>> where you add a Tile disassembler.
> Yeah, I shall add disassembler firstly, before generate tcg code:
>
>   - binutils tilegx disassembler is easy to be integrated into qemu.
>
>   - I shall reuse part of the tilegx disassembler code for generating tcg
>     code.
>
>   - For me, dumping disassemble code in qemu is much helpful for testing.
>
> At present, I have integrated binutils tilegx disassembler into qemu, it
> is passed building, and I shall give a test for it tonight. If no any
> issues, I plan to send related patch too.

The license issue is tricky with qemu.  The tilegx support was only returned to the community in release 2.22, and binutils has been GPL v3 since 2.18 or so.  So you can't use that code.  On the other hand, there is disassembly code in the kernel that we released under the kernel's GPL v2; see arch/tile/kernel/tile-desc_64.c and the matching include file.  So maybe there is enough there for you to use to implement the disassembly code.

I'm not sure whether Tilera can simply re-release the tilegx-specific stuff from binutils as a separate tarball with GPL v2 licensing.  Hopefully we can avoid having to figure that out. :-)

-- 
Chris Metcalf, EZChip Semiconductor
http://www.ezchip.com