[dpdk-dev] [PATCH v4] VFIO: Avoid to enable vfio while the module not loaded

[Burakov, Anatoly]

> When vfio module is not loaded when kernel support vfio feature, the
> routine still try to open the container to get file description.
> 
> This action is not safe, and of cause got error messages:
> 
> EAL: Detected 40 lcore(s)
> EAL:   unsupported IOMMU type!
> EAL: VFIO support could not be initialized
> EAL: Setting up memory...
> 
> This may make user confuse, this patch make it reasonable and much more
> soomth to user.
> 
> Signed-off-by: Michael Qiu <michael.qiu at intel.com>
> ---
> v4 --> v3:
> 	1. Remove RTE_LOG for params check
> 	2. Remove "vfio" module check as "vfio_iommu_type1"
> 	   loaded indecated "vfio" loaded
> 
> v3 --> v2:
>         1. Add error log in rte_eal_check_module()
>         2. Some code clean up.
> 
> v2 --> v1:
>         1. Move check_module() from rte_common.h to eal_private.h
>            and rename to rte_eal_check_module().
>            To make it linuxapp only.
>         2. Some code clean up.
> 
>  lib/librte_eal/common/eal_private.h        | 42
> ++++++++++++++++++++++++++++++
>  lib/librte_eal/linuxapp/eal/eal_pci_vfio.c | 29 ++++++++++++++++++---
>  2 files changed, 68 insertions(+), 3 deletions(-)
> 
> diff --git a/lib/librte_eal/common/eal_private.h
> b/lib/librte_eal/common/eal_private.h
> index 232fcec..e877a25 100644
> --- a/lib/librte_eal/common/eal_private.h
> +++ b/lib/librte_eal/common/eal_private.h
> @@ -35,6 +35,9 @@
>  #define _EAL_PRIVATE_H_
> 
>  #include <stdio.h>
> +#include <string.h>
> +#include <rte_log.h>
> +#include <errno.h>
> 
>  /**
>   * Initialize the memzone subsystem (private to eal).
> @@ -203,4 +206,43 @@ int rte_eal_alarm_init(void);
>   */
>  int rte_eal_dev_init(void);
> 
> +/**
> + * Function is to check if the kernel module(like, vfio,
> +vfio_iommu_type1,
> + * etc.) loaded.
> + *
> + * @param module_name
> + *	The module's name which need to be checked
> + *
> + * @return
> + * 	-1 means some error happens(NULL pointer or open failure)
> + * 	0  means the module not loaded
> + * 	1  means the module loaded
> + */
> +static inline int
> +rte_eal_check_module(const char *module_name) {
> +	char mod_name[30]; /* Any module names can be longer than 30
> bytes? */
> +	int ret = 0;
> +
> +	if (NULL == module_name)
> +		return -1;
> +
> +	FILE * fd = fopen("/proc/modules", "r");
> +	if (NULL == fd) {
> +		RTE_LOG(ERR, EAL, "Open /proc/modules failed!"
> +			" error %i (%s)\n", errno, strerror(errno));
> +		return -1;
> +	}
> +	while(!feof(fd)) {
> +		fscanf(fd, "%s %*[^\n]", mod_name);
> +		if(!strcmp(mod_name, module_name)) {
> +			ret = 1;
> +			break;
> +		}
> +	}
> +	fclose(fd);
> +
> +	return ret;
> +}
> +

Apologies for not bringing this up before, but do we really want the rte_eal_check_module inline in the header? I think it would be better to declare it in eal_private but move the definition into eal.c.

>  #endif /* _EAL_PRIVATE_H_ */
> diff --git a/lib/librte_eal/linuxapp/eal/eal_pci_vfio.c
> b/lib/librte_eal/linuxapp/eal/eal_pci_vfio.c
> index c1246e8..8c54d2a 100644
> --- a/lib/librte_eal/linuxapp/eal/eal_pci_vfio.c
> +++ b/lib/librte_eal/linuxapp/eal/eal_pci_vfio.c
> @@ -44,6 +44,7 @@
>  #include <rte_tailq.h>
>  #include <rte_eal_memconfig.h>
>  #include <rte_malloc.h>
> +#include <eal_private.h>
> 
>  #include "eal_filesystem.h"
>  #include "eal_pci_init.h"
> @@ -339,10 +340,15 @@ pci_vfio_get_container_fd(void)
>  		ret = ioctl(vfio_container_fd, VFIO_CHECK_EXTENSION,
> VFIO_TYPE1_IOMMU);
>  		if (ret != 1) {
>  			if (ret < 0)
> -				RTE_LOG(ERR, EAL, "  could not get IOMMU
> type, "
> -						"error %i (%s)\n", errno,
> strerror(errno));
> +				RTE_LOG(ERR, EAL, "  could not get IOMMU
> type,"
> +					" error %i (%s)\n", errno,
> +					strerror(errno));
>  			else
> -				RTE_LOG(ERR, EAL, "  unsupported IOMMU
> type!\n");
> +				/* Better to show the IOMMU type return
> from
> +				 * kernel for easy debug
> +				 */
> +				RTE_LOG(ERR, EAL, "  unsupported IOMMU
> type"
> +					" detected: %d in VFIO\n", ret);

I'm not sure this message is meaningful. That ioctl call can either -1, 0 or 1. We already handle 1 separately; -1 means an error; 0 means IOMMU type 1 is not supported. The return value will *not* indicate which IOMMU types *are* currently supported - it will only indicate that the IOMMU type you requested is not supported. So there's really no point in indicating the return value in case of ret 0 - it is best to just mention that requested IOMMU type support is not enabled in VFIO.

>  			close(vfio_container_fd);
>  			return -1;
>  		}
> @@ -783,11 +789,28 @@ pci_vfio_enable(void)  {
>  	/* initialize group list */
>  	int i;
> +	int module_vfio_type1;
> 
>  	for (i = 0; i < VFIO_MAX_GROUPS; i++) {
>  		vfio_cfg.vfio_groups[i].fd = -1;
>  		vfio_cfg.vfio_groups[i].group_no = -1;
>  	}
> +
> +	module_vfio_type1 = rte_eal_check_module("vfio_iommu_type1");
> +
> +	/* return error directly */
> +	if (module_vfio_type1 == -1) {
> +		RTE_LOG(INFO, EAL, "Could not get loaded module
> details!\n");
> +		return -1;
> +	}
> +
> +	/* return 0 if VFIO modules not loaded */
> +	if (module_vfio_type1 == 0) {
> +		RTE_LOG(INFO, EAL, "VFIO modules not all loaded,"
> +			" skip VFIO support ...\n");
> +		return 0;
> +	}
> +
>  	vfio_cfg.vfio_container_fd = pci_vfio_get_container_fd();
> 
>  	/* check if we have VFIO driver enabled */
> --
> 1.9.3