| Message ID | 20240321134911.120091-1-tushar.vyavahare@intel.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Selftests/xsk: Test with maximum and minimum HW ring size configurations | expand |
On Thu, 21 Mar 2024 at 15:05, Tushar Vyavahare <tushar.vyavahare@intel.com> wrote: > > Please find enclosed a patch set that introduces enhancements and new test > cases to the selftests/xsk framework. These test the robustness and > reliability of AF_XDP across both minimal and maximal ring size > configurations. > > While running these tests, a bug [1] was identified when the batch size is > roughly the same as the NIC ring size. This has now been addressed by > Maciej's fix. > > [1] https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net.git/commit/?id=913eda2b08cc49d31f382579e2be34c2709eb789 > > Patch Summary: > > 1. This commit syncs the ethtool.h header file between the kernel source > tree and the tools directory to maintain consistency. > > 2: Modifies the BATCH_SIZE from a constant to a variable, batch_size, to > support dynamic modification at runtime for testing different hardware > ring sizes. > > 3: Implements a function, get_hw_ring_size, to retrieve the current > maximum interface size and store this information in the > ethtool_ringparam structure. > > 4: Implements a new function, set_hw_ring_size, which allows for the > dynamic configuration of the ring size within an interface. > > 5: Introduce a new function, set_ring_size(), to manage asynchronous AF_XDP > socket closure. Make sure to retry the set_hw_ring_size function > multiple times, up to SOCK_RECONF_CTR, if it fails due to an active > AF_XDP socket. Immediately return an error for non-EBUSY errors. > > 6: Adds a new test case that puts the AF_XDP driver under stress by > configuring minimal hardware and software ring sizes, verifying its > functionality under constrained conditions. > > 7: Add a new test case that evaluates the maximum ring sizes for AF_XDP, > ensuring its reliability under maximum ring utilization. > > Testing Strategy: > > Check the system in extreme scenarios, such as maximum and minimum > configurations. This helps identify and fix any bugs that may occur. Thanks Tushar for this patch set. A year and a half ago, we had some bugs in this area in the ice driver, so these new tests are very welcome. Acked-by: Magnus Karlsson <magnus.karlsson@intel.com> > Tushar Vyavahare (7): > tools/include: copy ethtool.h to tools directory > selftests/xsk: make batch size variable > selftests/bpf: implement get_hw_ring_size function to retrieve current > and max interface size > selftests/bpf: implement set_hw_ring_size function to configure > interface ring size > selftests/xsk: introduce set_ring_size function with a retry mechanism > for handling AF_XDP socket closures > selftests/xsk: test AF_XDP functionality under minimal ring > configurations > selftests/xsk: add new test case for AF_XDP under max ring sizes > > Signed-off-by: Tushar Vyavahare <tushar.vyavahare@intel.com> > > --- > Changelog: > v1->v2 > - copy ethtool.h to tools directory [Stanislav] > - Use ethtool_ringparam directly for get_hw_ring_size() [Stanislav] > - get_hw_ring_size() and get_hw_ring_size() moved to network_helpers.c [Stanislav] > - return -errno to match the other cases where errors are < 0. [Stanislav] > - Cleaned up set_ring_size() function by removing unused variables and > refactoring logic for clarity. [Alexei] > - Implement a retry mechanism for the set_ring_size function to handle > the asynchronous nature of AF_XDP socket closure. [Magnus] > --- > > tools/include/uapi/linux/ethtool.h | 2229 ++++++++++++++++- > tools/testing/selftests/bpf/Makefile | 2 +- > tools/testing/selftests/bpf/network_helpers.c | 48 + > tools/testing/selftests/bpf/network_helpers.h | 5 + > tools/testing/selftests/bpf/xdp_hw_metadata.c | 14 - > tools/testing/selftests/bpf/xskxceiver.c | 123 +- > tools/testing/selftests/bpf/xskxceiver.h | 12 +- > 7 files changed, 2376 insertions(+), 57 deletions(-) > > -- > 2.34.1 > >
On Tue, Mar 26, 2024 at 12:24 AM Magnus Karlsson <magnus.karlsson@gmail.com> wrote: > > On Thu, 21 Mar 2024 at 15:05, Tushar Vyavahare > <tushar.vyavahare@intel.com> wrote: > > > > Please find enclosed a patch set that introduces enhancements and new test > > cases to the selftests/xsk framework. These test the robustness and > > reliability of AF_XDP across both minimal and maximal ring size > > configurations. > > > > While running these tests, a bug [1] was identified when the batch size is > > roughly the same as the NIC ring size. This has now been addressed by > > Maciej's fix. > > > > [1] https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net.git/commit/?id=913eda2b08cc49d31f382579e2be34c2709eb789 > > > > Patch Summary: > > > > 1. This commit syncs the ethtool.h header file between the kernel source > > tree and the tools directory to maintain consistency. > > > > 2: Modifies the BATCH_SIZE from a constant to a variable, batch_size, to > > support dynamic modification at runtime for testing different hardware > > ring sizes. > > > > 3: Implements a function, get_hw_ring_size, to retrieve the current > > maximum interface size and store this information in the > > ethtool_ringparam structure. > > > > 4: Implements a new function, set_hw_ring_size, which allows for the > > dynamic configuration of the ring size within an interface. > > > > 5: Introduce a new function, set_ring_size(), to manage asynchronous AF_XDP > > socket closure. Make sure to retry the set_hw_ring_size function > > multiple times, up to SOCK_RECONF_CTR, if it fails due to an active > > AF_XDP socket. Immediately return an error for non-EBUSY errors. > > > > 6: Adds a new test case that puts the AF_XDP driver under stress by > > configuring minimal hardware and software ring sizes, verifying its > > functionality under constrained conditions. > > > > 7: Add a new test case that evaluates the maximum ring sizes for AF_XDP, > > ensuring its reliability under maximum ring utilization. > > > > Testing Strategy: > > > > Check the system in extreme scenarios, such as maximum and minimum > > configurations. This helps identify and fix any bugs that may occur. > > Thanks Tushar for this patch set. A year and a half ago, we had some > bugs in this area in the ice driver, so these new tests are very > welcome. > > Acked-by: Magnus Karlsson <magnus.karlsson@intel.com> Agree, but CI doesn't like the set. Tushar, pls keep the Ack when you respin.
Please find enclosed a patch set that introduces enhancements and new test cases to the selftests/xsk framework. These test the robustness and reliability of AF_XDP across both minimal and maximal ring size configurations. While running these tests, a bug [1] was identified when the batch size is roughly the same as the NIC ring size. This has now been addressed by Maciej's fix. [1] https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net.git/commit/?id=913eda2b08cc49d31f382579e2be34c2709eb789 Patch Summary: 1. This commit syncs the ethtool.h header file between the kernel source tree and the tools directory to maintain consistency. 2: Modifies the BATCH_SIZE from a constant to a variable, batch_size, to support dynamic modification at runtime for testing different hardware ring sizes. 3: Implements a function, get_hw_ring_size, to retrieve the current maximum interface size and store this information in the ethtool_ringparam structure. 4: Implements a new function, set_hw_ring_size, which allows for the dynamic configuration of the ring size within an interface. 5: Introduce a new function, set_ring_size(), to manage asynchronous AF_XDP socket closure. Make sure to retry the set_hw_ring_size function multiple times, up to SOCK_RECONF_CTR, if it fails due to an active AF_XDP socket. Immediately return an error for non-EBUSY errors. 6: Adds a new test case that puts the AF_XDP driver under stress by configuring minimal hardware and software ring sizes, verifying its functionality under constrained conditions. 7: Add a new test case that evaluates the maximum ring sizes for AF_XDP, ensuring its reliability under maximum ring utilization. Testing Strategy: Check the system in extreme scenarios, such as maximum and minimum configurations. This helps identify and fix any bugs that may occur. Tushar Vyavahare (7): tools/include: copy ethtool.h to tools directory selftests/xsk: make batch size variable selftests/bpf: implement get_hw_ring_size function to retrieve current and max interface size selftests/bpf: implement set_hw_ring_size function to configure interface ring size selftests/xsk: introduce set_ring_size function with a retry mechanism for handling AF_XDP socket closures selftests/xsk: test AF_XDP functionality under minimal ring configurations selftests/xsk: add new test case for AF_XDP under max ring sizes Signed-off-by: Tushar Vyavahare <tushar.vyavahare@intel.com> --- Changelog: v1->v2 - copy ethtool.h to tools directory [Stanislav] - Use ethtool_ringparam directly for get_hw_ring_size() [Stanislav] - get_hw_ring_size() and get_hw_ring_size() moved to network_helpers.c [Stanislav] - return -errno to match the other cases where errors are < 0. [Stanislav] - Cleaned up set_ring_size() function by removing unused variables and refactoring logic for clarity. [Alexei] - Implement a retry mechanism for the set_ring_size function to handle the asynchronous nature of AF_XDP socket closure. [Magnus] --- tools/include/uapi/linux/ethtool.h | 2229 ++++++++++++++++- tools/testing/selftests/bpf/Makefile | 2 +- tools/testing/selftests/bpf/network_helpers.c | 48 + tools/testing/selftests/bpf/network_helpers.h | 5 + tools/testing/selftests/bpf/xdp_hw_metadata.c | 14 - tools/testing/selftests/bpf/xskxceiver.c | 123 +- tools/testing/selftests/bpf/xskxceiver.h | 12 +- 7 files changed, 2376 insertions(+), 57 deletions(-)