On Tue, Mar 8, 2011 at 5:52 AM, Gunasekaran Dharman <guna.s...@gmail.com> wrote: > Iam using QEMU emulated ARM Realview board with Linux as guest OS for my > application development. This virtual machine is running on x86 Linux host. > My application is facing performance issue as memcpy in guest machine is > taking lot of time compared to the same memcpy in host machine. > To improve the performance, Iam planning to add new device (Virtual DMA) in > QEMU which will implement memcpy operation in host for guest application. > The memcpy function in guest will be replaced with ioctl to instruct the > Virtual DMA to do memcpy in host for the guest application. > Iam planning to work on this. As Iam new to QEMU, before starting the work, > I would like to know whether this is a possibility. > Will there be any problem in accessing guest buffer addresses in QEMU? Is > this approach right? > Please let me know your views about this.
The same effort could be spent on improving TCG performance in your ARM -> x86 case. That might make other non-memcpy code faster too. Stefan