[dpdk-dev] [PATCH 3/3] eal: fix multi-process probe failure handling

Zhang, Qi Z qi.z.zhang at intel.com
Thu Mar 14 08:15:53 CET 2019



> -----Original Message-----
> From: Thomas Monjalon [mailto:thomas at monjalon.net]
> Sent: Saturday, March 2, 2019 10:43 AM
> To: dev at dpdk.org
> Cc: Zhang, Qi Z <qi.z.zhang at intel.com>; stable at dpdk.org
> Subject: [PATCH 3/3] eal: fix multi-process probe failure handling
> 
> If probe fails in multi-process context, the device must removed in other
> processes for consistency. This is a rollback mechanism.
> However the rollback should not happen for devices which were already probed
> before the current probe transaction.
> 
> When probing an already probed device, the driver may reject with -EEXIST or
> update and succeed with code 0.
> In order to distinguish successful new probe from re-probe, in the function
> local_dev_probe(), the positive EEXIST code is returned for the latter case.
> 
> The functions rte_dev_probe() and __handle_secondary_request() can test for
> -EEXIST and +EEXIST, and skip rollback in such case.
> 
> Fixes: 244d5130719c ("eal: enable hotplug on multi-process")
> Fixes: ac9e4a17370f ("eal: support attach/detach shared device from secondary")
> Cc: qi.z.zhang at intel.com
> Cc: stable at dpdk.org
> 
> Signed-off-by: Thomas Monjalon <thomas at monjalon.net>
> ---
>  lib/librte_eal/common/eal_common_dev.c | 12 ++++++++++--
>  lib/librte_eal/common/eal_private.h    |  2 +-
>  lib/librte_eal/common/hotplug_mp.c     |  8 ++++++--
>  3 files changed, 17 insertions(+), 5 deletions(-)
> 
> diff --git a/lib/librte_eal/common/eal_common_dev.c
> b/lib/librte_eal/common/eal_common_dev.c
> index deaaea9345..2c7b1ab071 100644
> --- a/lib/librte_eal/common/eal_common_dev.c
> +++ b/lib/librte_eal/common/eal_common_dev.c
> @@ -132,6 +132,7 @@ local_dev_probe(const char *devargs, struct rte_device
> **new_dev)  {
>  	struct rte_device *dev;
>  	struct rte_devargs *da;
> +	bool already_probed;
>  	int ret;
> 
>  	*new_dev = NULL;
> @@ -171,12 +172,15 @@ local_dev_probe(const char *devargs, struct rte_device
> **new_dev)
>  	 * those devargs shouldn't be removed manually anymore.
>  	 */
> 
> +	already_probed = rte_dev_is_probed(dev);
>  	ret = dev->bus->plug(dev);
>  	if (ret && !rte_dev_is_probed(dev)) { /* if hasn't ever succeeded */
>  		RTE_LOG(ERR, EAL, "Driver cannot attach the device (%s)\n",
>  			dev->name);
>  		return ret;
>  	}
> +	if (ret == 0 && already_probed)
> +		ret = EEXIST; /* hint to avoid any rollback */

What if bus->plug return -EEXIST and rte_dev_is_probed return true? (See rte_pci_probe_one_driver)
You will not give hint here, but is this expected?

> 
>  	*new_dev = dev;
>  	return ret;
> @@ -194,6 +198,7 @@ rte_dev_probe(const char *devargs)  {
>  	struct eal_dev_mp_req req;
>  	struct rte_device *dev;
> +	bool already_probed;
>  	int ret;
> 
>  	memset(&req, 0, sizeof(req));
> @@ -221,8 +226,8 @@ rte_dev_probe(const char *devargs)
> 
>  	/* primary attach the new device itself. */
>  	ret = local_dev_probe(devargs, &dev);
> -
> -	if (ret != 0 && ret != -EEXIST) {
> +	already_probed = (ret == -EEXIST || ret == EEXIST);
> +	if (ret < 0 && !already_probed) {
>  		RTE_LOG(ERR, EAL,
>  			"Failed to attach device on primary process\n");
>  		return ret;
> @@ -250,6 +255,9 @@ rte_dev_probe(const char *devargs)
>  	return 0;
> 
>  rollback:
> +	if (already_probed)
> +		return ret; /* skip rollback */
> +
>  	req.t = EAL_DEV_REQ_TYPE_ATTACH_ROLLBACK;
> 
>  	/* primary send rollback request to secondary. */ diff --git
> a/lib/librte_eal/common/eal_private.h b/lib/librte_eal/common/eal_private.h
> index 798ede553b..a01d252930 100644
> --- a/lib/librte_eal/common/eal_private.h
> +++ b/lib/librte_eal/common/eal_private.h
> @@ -304,7 +304,7 @@ rte_devargs_layers_parse(struct rte_devargs *devargs,
>   * @param new_dev
>   *   new device be probed as output.
>   * @return
> - *   0 on success, negative on error.
> + *   >=0 on success (+EEXIST if already probed), negative on error.
>   */
>  int local_dev_probe(const char *devargs, struct rte_device **new_dev);
> 
> diff --git a/lib/librte_eal/common/hotplug_mp.c
> b/lib/librte_eal/common/hotplug_mp.c
> index 69e9a16d6a..9f8ef28a3b 100644
> --- a/lib/librte_eal/common/hotplug_mp.c
> +++ b/lib/librte_eal/common/hotplug_mp.c
> @@ -90,13 +90,15 @@ __handle_secondary_request(void *param)
>  	struct rte_devargs da;
>  	struct rte_device *dev;
>  	struct rte_bus *bus;
> +	bool already_probed = false;
>  	int ret = 0;
> 
>  	tmp_req = *req;
> 
>  	if (req->t == EAL_DEV_REQ_TYPE_ATTACH) {
>  		ret = local_dev_probe(req->devargs, &dev);
> -		if (ret != 0 && ret != -EEXIST) {
> +		already_probed = (ret == -EEXIST || ret == EEXIST);
> +		if (ret < 0 && !already_probed) {
>  			RTE_LOG(ERR, EAL, "Failed to hotplug add device on primary\n");
>  			goto finish;
>  		}
> @@ -159,7 +161,7 @@ __handle_secondary_request(void *param)
>  	goto finish;
> 
>  rollback:
> -	if (req->t == EAL_DEV_REQ_TYPE_ATTACH) {
> +	if (req->t == EAL_DEV_REQ_TYPE_ATTACH && !already_probed) {
>  		tmp_req.t = EAL_DEV_REQ_TYPE_ATTACH_ROLLBACK;
>  		eal_dev_hotplug_request_to_secondary(&tmp_req);
>  		local_dev_remove(dev);
> @@ -238,6 +240,8 @@ static void __handle_primary_request(void *param)
>  	case EAL_DEV_REQ_TYPE_ATTACH:
>  	case EAL_DEV_REQ_TYPE_DETACH_ROLLBACK:
>  		ret = local_dev_probe(req->devargs, &dev);
> +		if (ret > 0)
> +			ret = 0; /* return only errors */
>  		break;
>  	case EAL_DEV_REQ_TYPE_DETACH:
>  	case EAL_DEV_REQ_TYPE_ATTACH_ROLLBACK:
> --
> 2.20.1



More information about the dev mailing list