On 10/15/2026 1:50 PM, Thomas Monjalon wrote:
15/10/2026 13:21, Burakov, Anatoly:
On 10/15/2026 1:18 PM, David Marchand wrote:
On Thu, 15 Oct 2026 at 12:41, Burakov, Anatoly
<[email protected]> wrote:
On 10/15/2026 11:13 AM, Thomas Monjalon wrote:
15/10/2026 10:53, Burakov, Anatoly:
On 10/14/2026 10:34 PM, Thomas Monjalon wrote:
Testpmd commands should use the driver's public API
rather than access its private structures directly.
Note: ixgbe_testpmd.c should be considered as part of testpmd.
ixgbe_testpmd.c includes ixgbe_ethdev.h only to check whether bypass
operations are available through the private adapter structure.
Rely on the public bypass API to check availability
and handle -ENOTSUP in the commands.
Remove ixgbe_ethdev.h, ethdev_driver.h
and the now unnecessary RTE_LIBRTE_IXGBE_BYPASS definition.
Signed-off-by: Thomas Monjalon <[email protected]>
---
Reviewed-by: Anatoly Burakov <[email protected]>
I forgot to say it is mostly written by AI.
I didn't review it myself.
Please double check.
It doesn't look wrong to me, testpmd code should not be using
driver-internal things anyway except through pmd-specific API, which is
what this patch does.
On the principle, this looks good indeed.
My fault for just moving the code around, without considering the used APIs.
And there are probably similar issues in some other drivers _testpmd.c
files (at least net/ice seems to include ice_ethdev.h).
My guess is this doesn't blow up on ice because it is not using the
graph helper yet. I'll see about fixing our other drivers.
Please could you make it as a v2 of this series?
Sure. patches already submitted but I'll respin with this one included.
(I also took the liberty of checking if other drivers have this problem,
and no, only ours do)
--
Thanks,
Anatoly
