Re: [Qemu-devel] Functions to intercept Disk IO information?

[Fam Zheng <famz@redhat.com>]

On Thu, 10/22 14:18, QuQ Edsel wrote:
> Thank you for reply！
> But as I insert a printf message in dma_buf_rw, compile and boot up the VM,
> there are no messages printed even when I open/create/edit a .txt or .jpeg
> file ?
> If dma_buf_rw is called it is supposed to print messages in the monitor...
> On the other hand,
> the dma_blk_io called in ide_dma_ cb did happen when files are
> manipulated....

Yes, it was a mistake, I meant dma_blk_io, sorry for confusion. :) dma_buf_rw
only works for some device types, while dma_blk_io works more commonly.

> Would it be point to determine Disk IO and to intercept the information I
> need?
> Sorry I really lack knowledge in this field...
> Does the PIO you mentioned means NIC IO?

PIO is port I/O which doesn't use DMA.

Fam

> 
> Edsel
> 
> 2015-10-22 9:22 GMT+08:00 Fam Zheng <famz@redhat.com>:
> 
> > On Thu, 10/22 01:13, QuQ Edsel wrote:
> > > Hi,
> > > My friends and I were assigned a task to find out a point to insert a
> > > callback function to intercept Disk IO activities such as read/write a
> > .txt
> > > file. Our final goal is to generate a report for target process/file 's
> > > Disk IO activities. We have QEMU 2.3 with KVM enabled. We have been
> > looking
> > > for such a point for long...but not so capable of such a task.
> > > People who had this project last year done so with TEMU 1.0 (probably
> > QEMU
> > > 0.9), and the implementation point they had is dma_buf_rw(). The
> > > information in their report shows pid, timestamps, disk sector, buffer
> > size
> > > and write/read for a target file (I am not even sure if such information
> > is
> > > meaningful or useful)
> > > Currently I have tried to printf in functions such as dma_buf_rw /
> > > dma_blk_io / bdrv_aio_readv...etc. to see if they print out message when
> > I
> > > open/edit/save a .txt or .jpeg file. The first one just don't print at
> > all,
> > > and the second and third one print a lot after booting up the guest....
> > > I can see that the dma_blk_io function call in ide_dma_cb (core.c) may be
> > > related because it prints as I have activities on files. However it also
> > > prints sometime when I am not doing any thing... so I not that certain
> > > about it. (and I don't know if write/read activities invoke such
> > > function..)
> >
> > Yes, dma_buf_rw should handle all your I/O unless you're doing PIO. I
> > think the
> > other activities you see when you're not doing anything is from guest
> > system's
> > background tasks.
> >
> > Thanks,
> > Fam
> >
> > >
> > > Is there a correct /better point to intercept disk IO information
> > > ?(especially for activities such as read/write a .file)
> > > Or what should I do to clearly get the needed information from Disk IO
> > > functions?
> > > I would be so grateful to have the information.
> > > Thank you.
> > >
> > > The guest environment I have is 64bit Windows 7 with qcow2 image (not
> > sure
> > > if relative)
> > >
> >