Re: [Xen-devel] [PATCH RESEND 3/4] qdisk, hw/block/xen_disk: Perform grant copy instead of grant map.

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

On Tue, May 31, 2016 at 06:44:57AM +0200, Paulina Szubarczyk wrote:
> Grant copy operation is divided into two phases different for
> 'read' and 'write' operation.
> 
> For a 'read' operation the flow is as follow:
>     1. allocate local buffers for all the segments contained in
>        a request.
>     2. fill the request io vectors with the buffers' addresses
>     3. invoke read operation by qemu device
>     4. in the completition call grant copy
>     5. free the buffers
> 
> Function 'ioreq_read_init' implements 1. and 2. step. It is called
> instead of 'ioreq_map' in 'ioreq_runio_qemu_aio'. Then the function
> 'ioreq_runio_qemu_aio' continues withouth changes performing step 3.
> Steps 4. and 5. are called in the callback function
> 'qemu_aio_complete'. The ioreq_read' function is implemented for
> step 4 which calls the new function 'xc_gnttab_copy_grant' presented
> in the other part of the patch.
> 
> For a 'write' operation steps 4. happens before step 2.. First data
> are copied from calling guest domains and then qemu operates on
> them.
> ---
>  hw/block/xen_disk.c | 185 
> ++++++++++++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 185 insertions(+)
> 
> diff --git a/hw/block/xen_disk.c b/hw/block/xen_disk.c
> index 3b7882e..43cd9c9 100644
> --- a/hw/block/xen_disk.c
> +++ b/hw/block/xen_disk.c
> @@ -284,6 +284,154 @@ err:
>      return -1;
>  }
>  
> +
> +static void* get_buffer(void) {
> +    void *buf;
> +
> +    buf = mmap(NULL, 1 << XC_PAGE_SHIFT, PROT_READ | PROT_WRITE, 
> +               MAP_SHARED | MAP_ANONYMOUS, -1, 0);

Please use XC_PAGE_SIZE instead of XC_PAGE_SHIFT, it's more clear.

I would also suggest using xc_memalign instead of mmap, to make sure the 
resulting buffer is page aligned. mmap makes no guarantes that the resulting 
address will be page-aligned.

Roger.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel