[Qemu-devel] runtime configurable semihosting

[Qemu-devel] runtime configurable semihosting

[Liviu Ionescu]

> semihosting ... be runtime configurable on this, I think.

I took a quick look at the syntax of command options and monitor commands, and I would suggest the following:

- extend the option "-semihosting" with an optional "target=native|gdb|auto", default auto

- if necessary, add a monitor command "semihosting target=native|gdb|auto", default auto


any comments?

Liviu

Re: [Qemu-devel] runtime configurable semihosting

[Liviu Ionescu]

On 12 Nov 2014, at 21:17, Liviu Ionescu <ilg@livius.net> wrote:

> 
>> semihosting ... be runtime configurable on this, I think.
> 
> I took a quick look at the syntax of command options and monitor commands, and I would suggest the following:
> 
> - extend the option "-semihosting" with an optional "target=native|gdb|auto", default auto

unfortunately the parser is not able to detect a missing optional, and will always consume the next option, so this syntax is not available.

the only reliable solution was to add a new option

	-semihosting-target native|gdb|auto

which works as expected.


regards,

Liviu

Re: [Qemu-devel] runtime configurable semihosting

[Christopher Covington]

Hi Liviu,

On 11/12/2014 02:17 PM, Liviu Ionescu wrote:
> 
>> semihosting ... be runtime configurable on this, I think.
> 
> I took a quick look at the syntax of command options and monitor commands, and I would suggest the following:
> 
> - extend the option "-semihosting" with an optional "target=native|gdb|auto", default auto
> 
> - if necessary, add a monitor command "semihosting target=native|gdb|auto", default auto
> 
> any comments?

How do these targets change the existing behavior?

Thanks,
Chris

-- 
Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

Re: [Qemu-devel] runtime configurable semihosting

[Peter Maydell]

On 13 November 2014 17:48, Christopher Covington <cov@codeaurora.org> wrote:
> Hi Liviu,
>
> On 11/12/2014 02:17 PM, Liviu Ionescu wrote:
>>
>>> semihosting ... be runtime configurable on this, I think.
>>
>> I took a quick look at the syntax of command options and monitor commands, and I would suggest the following:
>>
>> - extend the option "-semihosting" with an optional "target=native|gdb|auto", default auto
>>
>> - if necessary, add a monitor command "semihosting target=native|gdb|auto", default auto
>>
>> any comments?
>
> How do these targets change the existing behavior?

Our current behaviour is "auto", which means "send semihosting
commands via gdb if and only if the gdbstub has a gdb connected
when the first semihosting command is made by the guest". "native"
would mean "handle semihosting ourselves even if there's a gdb
connected". I'm not sure how useful "gdb" is as distinct from
"auto" (what do you do with the semihosting requests if no
gdb is connected in this case?).

-- PMM

Re: [Qemu-devel] runtime configurable semihosting

[Liviu Ionescu]

On 13 Nov 2014, at 19:52, Peter Maydell <peter.maydell@linaro.org> wrote:

> ... I'm not sure how useful "gdb" is as distinct from
> "auto" (what do you do with the semihosting requests if no
> gdb is connected in this case?).

well, I also had some doubts about "gdb", but since I never used this configuration, I preferred to keep it, mainly for completeness.

I did some tests and apparently nothing wrong happens when setting "gdb" and no connection is available.

I'm about to prepare the patch, do you suggest to remove this case?

Liviu

Re: [Qemu-devel] runtime configurable semihosting

[Liviu Ionescu]

On 13 Nov 2014, at 23:23, Liviu Ionescu <ilg@livius.net> wrote:

> ... "gdb", but since I never used this configuration...

btw, can anyone suggest how to configure things to test this kind of 'semihosting forwarding'?

does the standard GDB support this, or a custom one is required?

from the source code I saw that some special packages are sent to GDB, but my arm-none-eabi-gdb silently ignores them. or am I missing something?


regards,

Liviu

Re: [Qemu-devel] runtime configurable semihosting

[Peter Maydell]

On 13 November 2014 21:23, Liviu Ionescu <ilg@livius.net> wrote:
>
> On 13 Nov 2014, at 19:52, Peter Maydell <peter.maydell@linaro.org> wrote:
>
>> ... I'm not sure how useful "gdb" is as distinct from
>> "auto" (what do you do with the semihosting requests if no
>> gdb is connected in this case?).
>
> well, I also had some doubts about "gdb", but since I never used
> this configuration, I preferred to keep it, mainly for completeness.
>
> I did some tests and apparently nothing wrong happens when setting
> "gdb" and no connection is available.
>
> I'm about to prepare the patch, do you suggest to remove this case?


...I'm confused, I think. I thought 'gdb' was an option you added?
Can you explain what the three options you're proposing do, and
which of them is the behaviour we have today?

thanks
-- PMM

Re: [Qemu-devel] runtime configurable semihosting

[Liviu Ionescu]

On 13 Nov 2014, at 23:38, Peter Maydell <peter.maydell@linaro.org> wrote:

> On 13 November 2014 21:23, Liviu Ionescu <ilg@livius.net> wrote:

> ...I'm confused, I think. I thought 'gdb' was an option you added?

that's correct, but only part of the whole image, in fact I added "-semihosting-target auto|native|gdb"; however my patch only changes the destination of the semihosting calls, not the actual behaviour.

> Can you explain what the three options you're proposing do, and
> which of them is the behaviour we have today?

the behaviour is exactly the one you presented, 'auto' is the actual behaviour, 'native' is via qemu, 'gdb' is via... gbd. :-)

---

there were many details that required fixing, hopefully I'll figure out how to send the patch soon.



regards,

Liviu

p.s. I'm a bit puzzled by your patch submission procedure (the Eclipse CDT team uses Gerrit, and things are more or less automated), manually preparing patches and sending them to a mailing list seems quite tedious...

Re: [Qemu-devel] runtime configurable semihosting

[Peter Maydell]

On 13 November 2014 21:51, Liviu Ionescu <ilg@livius.net> wrote:
>
> On 13 Nov 2014, at 23:38, Peter Maydell <peter.maydell@linaro.org> wrote:
>
>> On 13 November 2014 21:23, Liviu Ionescu <ilg@livius.net> wrote:
>
>> ...I'm confused, I think. I thought 'gdb' was an option you added?
>
> that's correct, but only part of the whole image, in fact I added "-semihosting-target auto|native|gdb"; however my patch only changes the destination of the semihosting calls, not the actual behaviour.
>
>> Can you explain what the three options you're proposing do, and
>> which of them is the behaviour we have today?
>
> the behaviour is exactly the one you presented, 'auto' is the actual
> behaviour, 'native' is via qemu, 'gdb' is via... gbd. :-)

Right, but what does it mean to say 'gdb' that's different from
'auto' ? What happens in that case if you don't have a gdb
attached?

> p.s. I'm a bit puzzled by your patch submission procedure
> (the Eclipse CDT team uses Gerrit, and things are more or
> less automated), manually preparing patches and sending them
> to a mailing list seems quite tedious...

I use a flow like:
 git format-patch -o ~/my-sent-patches HEAD^
 # check that the ~/my-sent-patches/0001-whatever.patch looks ok
 git send-email --to=qemu-devel@nongnu.org ~/my-sent-patches/0001-whatever.patch

but that does require you to have set up git send-email
so it knows how to send emails.
(You can probably do it in one command but I happen to have
my main dev machine and the machine I send email from being
different, so I copy the patchfiles around in the middle.)

-- PMM

Re: [Qemu-devel] runtime configurable semihosting

[Liviu Ionescu]

On 14 Nov 2014, at 00:16, Peter Maydell <peter.maydell@linaro.org> wrote:

> Right, but what does it mean to say 'gdb' that's different from
> 'auto' ?

auto sends either to gdb or qemu, gdb always sends to gdb, native always to qemu.

> What happens in that case if you don't have a gdb
> attached?

according to my tests, nothing.

but, as I already mentioned, I have no experience with forwarding semihosting to gdb, and I may be missing the point.

> I use a flow like: ...

thank you, I'm trying to figure out the details (Signed-oof-by, cover letter, etc)


Liviu

Re: [Qemu-devel] runtime configurable semihosting

[Peter Maydell]

On 13 November 2014 22:23, Liviu Ionescu <ilg@livius.net> wrote:
>
> On 14 Nov 2014, at 00:16, Peter Maydell <peter.maydell@linaro.org> wrote:
>
>> Right, but what does it mean to say 'gdb' that's different from
>> 'auto' ?
>
> auto sends either to gdb or qemu, gdb always sends to gdb, native always to qemu.
>
>> What happens in that case if you don't have a gdb
>> attached?
>
> according to my tests, nothing.

Oh, I see now. gdb_do_syscall() will return doing nothing if
gdbserver_state is NULL (so "gdb" will throw away semihosting
requests until a gdb is eventually connected, at which point
we start to honour them). Yes, that makes sense as a behaviour
(though 'auto' and 'native' are going to be the major uses).

>> I use a flow like: ...
>
> thank you, I'm trying to figure out the details (Signed-off-by,
> cover letter, etc)

The Signed-off-by is just a matter of putting that line in your
commit message. It's pretty much the only thing we absolutely
have to insist on, since it's a statement that you're happy
for the code to go in and have the legal right to contribute it.
Most other things we can fix up formatting details if needed.
You don't need a cover letter unless you're submitting a series
with more than one patch in it.

-- PMM

Re: [Qemu-devel] runtime configurable semihosting

[Peter Maydell]

On 13 November 2014 09:44, Liviu Ionescu <ilg@livius.net> wrote:
> Peter Maydell wrote:
>> I took a quick look at the syntax of command options and monitor
>> commands, and I would suggest the following:
>>
>> - extend the option "-semihosting" with an optional
>> "target=native|gdb|auto", default auto
>
> unfortunately the parser is not able to detect a missing optional, and
> will always consume the next option, so this syntax is not available.

Sorry, yes, you're right; I should have looked more carefully.
So we do need a new option; however I think it would be better
for the new option to be a general qemuopts option, like this:

 -semihosting-config target=[native|gdb|auto]

because that then gives us a place to put future semihosting
related config options, and also allows the config files used
by -readconfig/-writeconfig to change this setting. (We can
have -semihosting-config have an implied 'enable=true', so
you don't have to use the old "-semihosting" option if you're
specifying -semihosting-config.)

thanks
-- PMM

Re: [Qemu-devel] runtime configurable semihosting

[Liviu Ionescu]

On 18 Nov 2014, at 15:31, Peter Maydell <peter.maydell@linaro.org> wrote:

> Sorry, yes, you're right; I should have looked more carefully.
> So we do need a new option;

or we should fix the parser. if the current keyword is not in the list, it can be pushed back, and defaults used.

> -semihosting-config target=[native|gdb|auto]

ok, I'll implement this, and if you fix the parser we'll change it to the plain -semihosting.

> have -semihosting-config have an implied 'enable=true', so

ok

Liviu

Re: [Qemu-devel] runtime configurable semihosting

[Peter Maydell]

On 18 November 2014 16:46, Liviu Ionescu <ilg@livius.net> wrote:
>
> On 18 Nov 2014, at 15:31, Peter Maydell <peter.maydell@linaro.org> wrote:
>
>> Sorry, yes, you're right; I should have looked more carefully.
>> So we do need a new option;
>
> or we should fix the parser. if the current keyword is not in
> the list, it can be pushed back, and defaults used.

This is veering quite close to magic, really. It changes
the current handling of some valid command lines (admittedly
ones with implausible image names, but still).

>> -semihosting-config target=[native|gdb|auto]
>
> ok, I'll implement this, and if you fix the parser we'll change it to the plain -semihosting.

Thanks.

-- PMM