[PATCH] ethdev: allow using devices with shortened PCI addresses

[PATCH] ethdev: allow using devices with shortened PCI addresses

[Bruce Richardson]

When using a secondary process, e.g. pdump, not all PCI devices will be
probed correctly if the primary process has allowlisted devices using
the shortened syntax, omitting the initial "0000:" domain. This is due
to the strcmp failing to match a full address with a shortened one.

Fix this issue by comparing the device names/addresses normally and then
having a special case second comparison for devices which start with
"0000:".

Fixes: d948f596fee2 ("ethdev: fix port data mismatched in multiple process model")
Cc: stable@dpdk.org

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 lib/ethdev/ethdev_driver.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/lib/ethdev/ethdev_driver.c b/lib/ethdev/ethdev_driver.c
index 9afef06431..e5590076c1 100644
--- a/lib/ethdev/ethdev_driver.c
+++ b/lib/ethdev/ethdev_driver.c
@@ -156,6 +156,7 @@ rte_eth_dev_attach_secondary(const char *name)
 	uint16_t i;
 	struct rte_eth_dev *eth_dev = NULL;
 
+
 	/* Synchronize port attachment to primary port creation and release. */
 	rte_spinlock_lock(rte_mcfg_ethdev_get_lock());
 
@@ -165,6 +166,11 @@ rte_eth_dev_attach_secondary(const char *name)
 	for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
 		if (strcmp(eth_dev_shared_data->data[i].name, name) == 0)
 			break;
+
+		/* for PCI devices, allow specifying name without a leading "0000:" */
+		if (strncmp(name, "0000:", 5) == 0 && /* is PCI address */
+				strcmp(eth_dev_shared_data->data[i].name, &name[5]) == 0)
+			break;
 	}
 	if (i == RTE_MAX_ETHPORTS) {
 		RTE_ETHDEV_LOG_LINE(ERR,
-- 
2.43.0

Re: [PATCH] ethdev: allow using devices with shortened PCI addresses

[Thomas Monjalon]

19/11/2024 16:57, Bruce Richardson:
> When using a secondary process, e.g. pdump, not all PCI devices will be
> probed correctly if the primary process has allowlisted devices using
> the shortened syntax, omitting the initial "0000:" domain. This is due
> to the strcmp failing to match a full address with a shortened one.
> 
> Fix this issue by comparing the device names/addresses normally and then
> having a special case second comparison for devices which start with
> "0000:".
> 
> Fixes: d948f596fee2 ("ethdev: fix port data mismatched in multiple process model")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
> ---
>  lib/ethdev/ethdev_driver.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/lib/ethdev/ethdev_driver.c b/lib/ethdev/ethdev_driver.c
> index 9afef06431..e5590076c1 100644
> --- a/lib/ethdev/ethdev_driver.c
> +++ b/lib/ethdev/ethdev_driver.c
> +		/* for PCI devices, allow specifying name without a leading "0000:" */
> +		if (strncmp(name, "0000:", 5) == 0 && /* is PCI address */
> +				strcmp(eth_dev_shared_data->data[i].name, &name[5]) == 0)
> +			break;

I'm afraid it is not enough in all cases.
Shani is preparing a patch which uses the same PCI address formatting
in all cases.

Re: [PATCH] ethdev: allow using devices with shortened PCI addresses

[Bruce Richardson]

On Tue, Nov 19, 2024 at 05:22:21PM +0100, Thomas Monjalon wrote:
> 19/11/2024 16:57, Bruce Richardson:
> > When using a secondary process, e.g. pdump, not all PCI devices will be
> > probed correctly if the primary process has allowlisted devices using
> > the shortened syntax, omitting the initial "0000:" domain. This is due
> > to the strcmp failing to match a full address with a shortened one.
> > 
> > Fix this issue by comparing the device names/addresses normally and then
> > having a special case second comparison for devices which start with
> > "0000:".
> > 
> > Fixes: d948f596fee2 ("ethdev: fix port data mismatched in multiple process model")
> > Cc: stable@dpdk.org
> > 
> > Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
> > ---
> >  lib/ethdev/ethdev_driver.c | 6 ++++++
> >  1 file changed, 6 insertions(+)
> > 
> > diff --git a/lib/ethdev/ethdev_driver.c b/lib/ethdev/ethdev_driver.c
> > index 9afef06431..e5590076c1 100644
> > --- a/lib/ethdev/ethdev_driver.c
> > +++ b/lib/ethdev/ethdev_driver.c
> > +		/* for PCI devices, allow specifying name without a leading "0000:" */
> > +		if (strncmp(name, "0000:", 5) == 0 && /* is PCI address */
> > +				strcmp(eth_dev_shared_data->data[i].name, &name[5]) == 0)
> > +			break;
> 
> I'm afraid it is not enough in all cases.
> Shani is preparing a patch which uses the same PCI address formatting
> in all cases.
> 
Great. I'll test that patch when it gets submitted.

/Bruce.