[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH] ehci: avoid string arguments in trace events
From: |
Stefan Hajnoczi |
Subject: |
Re: [Qemu-devel] [PATCH] ehci: avoid string arguments in trace events |
Date: |
Mon, 5 Sep 2011 14:08:03 +0100 |
On Mon, Sep 5, 2011 at 9:38 AM, Gerd Hoffmann <address@hidden> wrote:
> On 09/03/11 17:22, Stefan Hajnoczi wrote:
>>
>> String arguments are not supported by all trace backends. This patch
>> replaces existing string arguments in hw/usb-ehci.c either with
>> individual trace events that remain human-friendly or by printing raw
>> addresses when there is no alternative or downside to that.
>
> Printing raw addresses *is* a downside.
>
>> States and usbsts bits remain human-friendly since it is hard to
>> remember all of them. MMIO addresses are printed raw because they would
>> create many individual trace events and the addresses are usually easy
>> to remember when debugging.
>
> I find it hard to rememeber them. There is a reason why the code to print
> the names for the mmio addresses is there in the first place. I don't want
> to loose that.
>
> Can't we just fix the backends instead? Replacing debug fprintf with trace
> points isn't going to work if tracing can't handle strings.
I looked at the capabilities of the backends again and I think we
*can* allow strings. The simple trace backend does not support them
but it's possible to add that. I thought SystemTap doesn't either but
it turns out there is a userspace string copy function available.
Both stderr and ust are fine with strings.
Let's drop this patch. I will update the tracing documentation.
Stefan