On Wed, Feb 22, 2017 at 11:17 PM, Qi Zhang <qi.z.zh...@intel.com> wrote:
> --- a/lib/librte_eal/linuxapp/eal/eal_interrupts.c
> +++ b/lib/librte_eal/linuxapp/eal/eal_interrupts.c
> @@ -278,29 +278,6 @@ vfio_disable_msi(const struct rte_intr_handle
> *intr_handle) {
> return ret;
> }
>
> -static int
> -get_max_intr(const struct rte_intr_handle *intr_handle)
> -{
> - struct rte_intr_source *src;
> -
> - TAILQ_FOREACH(src, &intr_sources, next) {
> - if (src->intr_handle.fd != intr_handle->fd)
> - continue;
> -
> - if (src->intr_handle.max_intr < intr_handle->max_intr)
> - src->intr_handle.max_intr = intr_handle->max_intr;
> - if (!src->intr_handle.max_intr)
> - src->intr_handle.max_intr = 1;
> - else if (src->intr_handle.max_intr > RTE_MAX_RXTX_INTR_VEC_ID)
> - src->intr_handle.max_intr
> - = RTE_MAX_RXTX_INTR_VEC_ID + 1;
> -
> - return src->intr_handle.max_intr;
> - }
> -
> - return -1;
> -}
> -
> /* enable MSI-X interrupts */
> static int
> vfio_enable_msix(const struct rte_intr_handle *intr_handle) {
> @@ -313,15 +290,10 @@ vfio_enable_msix(const struct rte_intr_handle
> *intr_handle) {
>
> irq_set = (struct vfio_irq_set *) irq_set_buf;
> irq_set->argsz = len;
> -
> - ret = get_max_intr(intr_handle);
> - if (ret < 0) {
> - RTE_LOG(ERR, EAL, "Invalid number of MSI-X irqs for fd %d\n",
> - intr_handle->fd);
> - return -1;
> - }
> -
> - irq_set->count = ret;
> + /* 0 < irq_set->count < RTE_MAX_RXTX_INTR_VEC_ID + 1 */
> + irq_set->count = intr_handle->max_intr ?
> + (intr_handle->max_intr > RTE_MAX_RXTX_INTR_VEC_ID + 1 ?
> + RTE_MAX_RXTX_INTR_VEC_ID + 1 : intr_handle->max_intr) : 1;
> irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD |
> VFIO_IRQ_SET_ACTION_TRIGGER;
> irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
> irq_set->start = 0;
The changes (to not change the interrupt handle in intr_sources) seems
to be unrelated to the API changes. Can you split this into two
commits, please?
Also I'm not 100% convinced that the low-level drivers should make
extensive use of the ethdev's const intr_handle. I believe that the
ethdev ops might be incomplete in that sense that it would be better
to add an rx_intr_ctl() operation to the ethdev ops structure and pass
control the the low-level device.
