[Top][All Lists]

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

Re: [Qemu-devel] [very-WIP 3/4] slirp: VMStatify sbuf

From: Halil Pasic
Subject: Re: [Qemu-devel] [very-WIP 3/4] slirp: VMStatify sbuf
Date: Tue, 18 Oct 2016 12:40:03 +0200
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Thunderbird/45.4.0

On 10/17/2016 09:06 PM, Dr. David Alan Gilbert wrote:
>> OK, memory was allocated at #2
>> > It is a bit confusing though (for a novice like me) that we have a non 
>> > whose pointer is NULL after post_load.
> I don't think this pointer can be NULL; the sbreserve at #2 causes it to be
> allocated.
> But yes, it's a shame I can't use VMS_ALLOC here, but the sbreserve is not
> a trivial allocation function.

Sorry my fault, wanted to say pre_load and not post_load. The assumption
that backing memory is allocated either during device realization or in the
pre_load of the parent or in vmstate_base_addr seemed a reasonable one
(prior to the extra possibilities introduced by the patches of Jianjun).

>> > Now if I imagine the original stream were written in the following 
>> > sequence:
>> > vbuffer_length (sb_datalen), vbuffer_data (sb_data), offsets (sb_wptr, 
>> > sb_rptr)
>> > which seems completely valid to me then the context would not be sufficient
>> > to compute sb_wptr and sb_rptr because the lifetime of vbuffer_data and
>> > the tmp do not overlap.
> If that was the case you could still do it pretty easily.
> You'd have to add the sb_datalen and sb_data fields to the temporary
> and then move the VMSTATE_VBUFFER_UINT32 into the tmp so it would operate
> on the copied fields.

That basically means you expand the area affected by the tmp handling
so that you have all the info you need to do the computation/transformation
before the temporary is freed, right? Then the worst it can get is that you 
need to
transform the first field and for that you need the last field.

>> > I aware it's a trade-off between how long the temporary data lives and
>> > how complicated the dependencies get. Or am I getting something wrong?
> No, I think that's right.  The other option I thought of was a macro
> to allocate a temporary and then another to free it and then someway
> to tell macros in between that they should operate on the temporary
> rather than the main pointer; but then you'd have to be VERY careful
> to not allow yourself to access a temporary that's been freed.
> This structure means you can't make that mistake.

I have a couple of crazy half ideas of myself, I just do not feel
very comfortable sharing them, because right now I do not have the
capacity to explore them properly. Besides I don't know the code-base
well enough to say if this reasoning has some practical relevance
or is it 'rather academic'. Nevertheless I did not want to keep
the opinion to myself that this thing with the dependencies can
get rather convoluted under circumstances.


> Dave
>> > 
>> > Cheers,
>> > Halil
>> > 
>>>> > >> +        VMSTATE_END_OF_LIST()
>>>> > >> +    }
>>>> > >> +};
>> > [..]
>> > 

reply via email to

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