On Wed, Oct 19, 2022 at 02:11:18PM +0100, Bruce Richardson wrote:
> For historical reasons, a number of net vdev drivers also add a driver
> alias using the "eth_" prefix. Since this is done on a per-driver basis,
> the use of the alias in inconsistent and is spread across multiple
> files. We can remove the per-driver aliases by just adding the alias
> automatically at the vdev bus level.
>
> Signed-off-by: Bruce Richardson <bruce.richard...@intel.com>
> ---
> drivers/bus/vdev/vdev.c | 6 ++++++
> 1 file changed, 6 insertions(+)
>
> diff --git a/drivers/bus/vdev/vdev.c b/drivers/bus/vdev/vdev.c
> index f5b43f1930..bfd7ce60c1 100644
> --- a/drivers/bus/vdev/vdev.c
> +++ b/drivers/bus/vdev/vdev.c
> @@ -54,6 +54,12 @@ static rte_spinlock_t vdev_custom_scan_lock =
> RTE_SPINLOCK_INITIALIZER;
> void
> rte_vdev_register(struct rte_vdev_driver *driver)
> {
> + /* For net driver vdevs, add an automatic alias using "eth" prefix */
> + if (strncmp(driver->driver.name, "net_", 4) == 0 &&
> driver->driver.alias == NULL) {
> + char *alias = strdup(driver->driver.name);
> + memcpy(alias, "eth_", 4);
> + driver->driver.alias = alias;
> + }
> TAILQ_INSERT_TAIL(&vdev_driver_list, driver, next);
> }
>
As a self-review comment, I realise this solution has got an issue that it
leaks memory if drivers are constantly being registered or unregistered. I
find it hard to see situations where this can occur, but it is a potential
issue.
A second solution that does not have this problem is to move the aliasing
to EAL, as below:
index fb5d0a293b..37b86914a0 100644
--- a/lib/eal/common/eal_common_devargs.c
+++ b/lib/eal/common/eal_common_devargs.c
@@ -226,9 +226,14 @@ rte_devargs_parse(struct rte_devargs *da, const char *dev)
da->name[i] = '\0';
if (bus == NULL) {
bus = rte_bus_find_by_device_name(da->name);
+ if (bus == NULL && strncmp(da->name, "eth_", 4) == 0) {
+ RTE_LOG(INFO, EAL, "failed to parse device
\"%s\"...\n", da->name);
+ memcpy(da->name, "net_", 4);
+ RTE_LOG(INFO, EAL, "... trying device \"%s\"\n",
da->name);
+ bus = rte_bus_find_by_device_name(da->name);
+ }
if (bus == NULL) {
RTE_LOG(ERR, EAL, "failed to parse device \"%s\"\n",
da->name);
While this doesn't have a memory freeing issue, it's downside is obviously
that we have further abstraction leakage, from the vdev bus for net drivers
all the way to EAL. Again, since the code delta is fairly small, this may
be acceptable, so opinions welcome.
/Bruce
