M. Lureau, Thank you for your reply ! I had misunderstood some informations indicating that the device would create an entry in the /dev/shm of virtual machines. The lspci command assured me that the device is properly accessible in my virtual machine.
"Note that ivshmem is not well regarded by the qemu community, what are you trying to accomplish?" Ivshmem breaks one of the most important virtualization's principal : the isolation of virtual machines. I know some people don't like it. I'm working in the GPGPU context. We are working (again) on sharing available GPUs of a physical machine between the virtual machines running on that machine. To do so, we created a direct access shared buffer : data are written and read directly in a ring buffer without any kind of copy. That reduces transfer times but it requires memory pages to be shared between communicating processes. I used to work with Xen but Xen was incompatible with NVIDIA GPUs when I began working on this project. Therefore I went for Qemu as it's compatible with GPGPU on NVIDIA ! Ivshmem is a danger for public clusters, but it could bring speed-up for private infrastructures ! Best regards, Sebastien Fremal