Re: [dpdk-dev] [PATCH v2 09/10] igbuio: show irq mode in sysfs

["Carew, Alan" <alan.carew@intel.com>]

> -----Original Message-----
> From: Stephen Hemminger [mailto:stephen@networkplumber.org]
> Sent: Wednesday, June 11, 2014 9:08 PM
> To: Carew, Alan
> Cc: Neil Horman; dev@dpdk.org
> Subject: Re: [dpdk-dev] [PATCH v2 09/10] igbuio: show irq mode in sysfs
> 
> This is what I am testing, along with 10 other virtio patches.
> 
> Subject: virtio: check for using msix interrupt
> 
> Fix how the device driver detects MSI-X vs INTX mode.
> Look in sysfs to find if MSI-X is enabled.
> 
> Suggested-by: Neil Horman <nhorman@tuxdriver.com>
> Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
> 
> 
> --- a/lib/librte_pmd_virtio/virtio_ethdev.c
> +++ b/lib/librte_pmd_virtio/virtio_ethdev.c
> @@ -709,6 +709,28 @@
>  	return 0;
>  }
> 
> +/*
> + * Detect if using INTX or MSI-X by looking for:
> + *  /sys/bus/pci/devices/<D:B:D.F>/msi_irqs/
> + * if directory exists, must be using msi-x
> + */
> +static int
> +has_msix(const struct rte_pci_addr *loc)
> +{
> +	DIR *d;
> +	char dirname[PATH_MAX];
> +
> +	rte_snprintf(dirname, sizeof(dirname),
> +		     SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/msi_irqs",
> +		     loc->domain, loc->bus, loc->devid, loc->function);
> +
> +	d = opendir(dirname);
> +	if (d)
> +		closedir(d);
> +
> +	return (d != NULL);
> +}
> +
>  static int get_uio_dev(struct rte_pci_addr *loc, char *buf, unsigned int buflen)
>  {
>  	unsigned int uio_num;
> @@ -872,6 +894,8 @@
>  		PMD_INIT_LOG(DEBUG,
>  			     "PCI Port IO found start=0x%lx with size=0x%lx\n",
>  			     start, size);
> +
> +		hw->use_msix = has_msix(&pci_dev->addr);
>  	}
>  #endif
>  	hw->io_base = (uint32_t)(uintptr_t)pci_dev->mem_resource[0].addr;
> --- a/lib/librte_pmd_virtio/virtio_pci.h
> +++ b/lib/librte_pmd_virtio/virtio_pci.h
> @@ -177,6 +177,7 @@
>  	uint16_t    subsystem_device_id;
>  	uint16_t    subsystem_vendor_id;
>  	uint8_t     revision_id;
> +	uint8_t	    use_msix;
>  	uint8_t     mac_addr[ETHER_ADDR_LEN];
>  	int         adapter_stopped;
>  };
> @@ -194,13 +195,11 @@
>  	uint16_t   max_virtqueue_pairs;
>  } __attribute__((packed));
> 
> -/* Value indicated in device config */
> -#define VIRTIO_PCI_FLAG_MSIX  0x0020
>  /*
>   * The remaining space is defined by each driver as the per-driver
>   * configuration space.
>   */
> -#define VIRTIO_PCI_CONFIG(hw) (((hw)->guest_features &
> VIRTIO_PCI_FLAG_MSIX) ? 24 : 20)
> +#define VIRTIO_PCI_CONFIG(hw) (((hw)->use_msix) ? 24 : 20)
> 
>  /*
>   * How many bits to shift physical queue address written to QUEUE_PFN.

Hi Stephen,

The mechanism is fine, however I would be against OS specific code in abstracted drivers. If "has_msix" was a helper function in eal (and suitably renamed) we could then add a FreeBSD equivalent implementation. A less favourable solution would be conditional compilation(RTE_EXEC_ENV_LINUXAPP/BSDPAPP) around the above function.

Thanks,
Alan