[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH 0/4] tcg: Add debug facilities for TCGv

From: Andreas Färber
Subject: Re: [Qemu-devel] [PATCH 0/4] tcg: Add debug facilities for TCGv
Date: Tue, 13 Dec 2011 13:43:04 +0100
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:8.0) Gecko/20111105 Thunderbird/8.0

Am 12.12.2011 16:58, schrieb Paul Brook:
>>> Trying to make a 32-bit target "64-bit safe" without actually
>>> implementing the 64-bit target is a complete waste of time.
>> That's where we disagree. I rather do things right from the start than
>> leaving the cleanup work to someone else later on.
>>> You've almost no chance of getting
>>> it right.  In some cases the correct answer will be to use 32-bit
>>> arithmetic, then sign/zero extend the result. In other cases the correct
>>> answer will be to perform word size arithmetic.  Blindly picking one
>>> just makes the bugs harder to find later.
>> This series picks nothing blindly. 
> Sure it does

No, start by reading the git summary. These four patches don't touch
target-* at all.

This is intentionally NOT some Coccinelle script running wild doing
refactorings. That's what I would call "blindly".

> Ther are three ways to resolve a mismatch:
> - Change everything to TCGv_i32.
> - Change everything to TCGv.
> - Add an explicit extension/truncation (compiles to no-op on 32-bit targets).
> There's no way of the developer of a 32-bit architecture to know

Again, that's where we disagree:

The whole point of TCGv and tl is to have variable-sized operations
scaling with target_long.

Thus, using them for fixed-size i32 or i64 operations is a semantic
error by definition. Whether or not an i64 target exists.

And I certainly don't want to knowingly introduce semantic errors in my
new code just so that at another time someone else can use that to
review a 64-bit port. That's just plain stupid. As the developer I must
know what semantics I am implementing for my target.

Note that the three choices are independent of this series, same holds
without. The difference is, my series offers a way to *flag* cases where
this has been ignored.

>> If you have a better proposal how to introduce the checks I want, please
>> let us hear it.
> I still don't understand how your additional restructions are ever useful.
> Please give an example of an actual error your checks will catch.

My stated requirement is that I want to detect ALL uses of TCGv_i32 with
tl functions and all uses of TCGv with i32 functions, be it an error or
a warning. Whether or not such consistency seems useful to you.

I have already given four examples to Peter, that you quoted previously.
Consider a uint32_t 8-bit status register on a 20-bit architecture - it
never scales to i64 so I know that TCGv/tl is definitely wrong!

Either point out something that's technically wrong with these patches
and I'll gladly fix it, or, again, propose a constructive solution.
Reappearing after a year and destructively objecting to patches is
something we've been through before.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]