Fix XDP documentation to a unified location for ICSSG and CPSW#642
Fix XDP documentation to a unified location for ICSSG and CPSW#642MeghanaMalladiTI wants to merge 2 commits into
Conversation
|
@danish-ti and @daolin can you please review this PR |
0e1fd27 to
c38b907
Compare
c38b907 to
eedbe25
Compare
eedbe25 to
5480f52
Compare
5480f52 to
5ca55f7
Compare
|
New warnings found with rstcheck: |
61e0e52 to
bf8615a
Compare
bf8615a to
7433a5e
Compare
StaticRocket
left a comment
There was a problem hiding this comment.
Ah, section depth still looks weird.
https://github.com/TexasInstruments/processor-sdk-doc/blob/master/CONTRIBUTING.md#headings--sections
…CSSG and CPSW - Add new centralized XDP documentation at Kernel/Kernel_Drivers/Network/XDP.rst - Add kernel_xdp reference label that can be used throughout the docs - Update PRU_ICSSG_XDP.rst to redirect to the consolidated XDP docs - Update features supported in PRU_ICSSG_Ethernet.rst - Update CPSW-Ethernet.rst to reference kernel_xdp instead of pru_icssg_xdp - Add XDP.rst to AM64X device-specific table of contents - Add XDP.rst to main Kernel Drivers toctree The new XDP documentation covers: - XDP introduction and concepts - Use cases and kernel configuration - AF_XDP sockets and zero-copy mode - Testing with xdp-tools (xdp-bench, xdp-trafficgen) - Performance comparison for CPSW and ICSSG drivers Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com> Signed-off-by: Meghana Malladi <m-malladi@ti.com>
Add XDP documentation support to the following SoC platforms by including Network/XDP.rst in their device-specific table of contents: - AM62AX (am62axx_sk-fs) - AM62PX (am62pxx_sk-fs) - AM62X (am62xx_sk-fs) - J7200 (j7200-evm) - J721E (j721e-idk-gw) - J722S (j722s_evm-fs) Note: AM64X (am64xx-evm, am64xx-hsevm) already has XDP support from the initial commit. This ensures that users of these platforms can access the consolidated XDP documentation covering Native XDP, Generic XDP, and Zero-copy modes. Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com> Signed-off-by: Meghana Malladi <m-malladi@ti.com>
Ahh, I used claude to address this commit last time when you pointed it out. Should have verified the changes without blindly pushing the changes :) |
7433a5e to
b51b552
Compare
|
@danish-ti , @dao-qiu please review. |
| * - rxdrop | ||
| - 322 | ||
| - 491 | ||
| - 845 |
There was a problem hiding this comment.
I didn't extensively test this but on some latest test with all AM6x platforms and all configurations but I found that on AM62Px tests on SDK 11.2 the actual kps rate was was much less.
For documentation purposes I can approve this for now but later I might create JIRA if I indeed see kps rate is poor on latest SDK
There was a problem hiding this comment.
Update: A couple addition points to address before approving
- Can the specific XDP application and options used to obtain these performance numbers be added? For example, if for txonly benchmark "xdpsock -i eth0 -q0 -t -x --tx-dmac ff:ff:ff:ff:ff:ff -s64 -N -z" was used then that should be indicated in the docs
- Are these performance numbers the same between CPSW and PRU-ICSSG Ethernet? If not, then there should be a separate section for CPSW and another section for PRU-ICSSG Ethernet.
- If xdpsock was used to obtain these performance numbers, I think some documentation on xdpsock should be included in the documentation here otherwise the benchmarks should be done with xdp-tools since that already has a section in the documentation here (I see that xdp-tools here is the new utility tool to use so if that is the case, it would be inconsistent to show performance numbers using xdpsock)
- If xdpsock is used to obtain the performance numbers, then xdpsock utility tool should also be built into the SDK release so customers can also evaluate easily on their systems
dao-qiu
left a comment
There was a problem hiding this comment.
Please see my comment on XDP zero copy benchmark:
I didn't extensively test this but on some latest test with all AM6x platforms and all configurations but I found that on AM62Px tests on SDK 11.2 the actual kps rate was was much less.
For documentation purposes I can approve this for now but later I might create JIRA if I indeed see kps rate is poor on latest SDK
| * - rxdrop | ||
| - 322 | ||
| - 491 | ||
| - 845 |
There was a problem hiding this comment.
Update: A couple addition points to address before approving
- Can the specific XDP application and options used to obtain these performance numbers be added? For example, if for txonly benchmark "xdpsock -i eth0 -q0 -t -x --tx-dmac ff:ff:ff:ff:ff:ff -s64 -N -z" was used then that should be indicated in the docs
- Are these performance numbers the same between CPSW and PRU-ICSSG Ethernet? If not, then there should be a separate section for CPSW and another section for PRU-ICSSG Ethernet.
- If xdpsock was used to obtain these performance numbers, I think some documentation on xdpsock should be included in the documentation here otherwise the benchmarks should be done with xdp-tools since that already has a section in the documentation here (I see that xdp-tools here is the new utility tool to use so if that is the case, it would be inconsistent to show performance numbers using xdpsock)
- If xdpsock is used to obtain the performance numbers, then xdpsock utility tool should also be built into the SDK release so customers can also evaluate easily on their systems
Add new centralized XDP documentation at Kernel/Kernel_Drivers/Network/XDP.rst for CPSW and ICSSG