soc: silabs: siwx91x: Improve performance #98060

ragurram26 · 2025-10-22T09:34:08Z

To enhance throughput performance on the SiWx91x series, certain configurations need to be enabled in the SiWx91x.

jerome-pouiller · 2025-10-22T10:25:31Z

drivers/wifi/siwx91x/Kconfig.siwx91x

+	help
+	  Enable this option to improve the throughput performance of the SiWx91x
+	  series. This feature provides enhanced data transfer capabilities and
+	  improved network performance.


Why anyone wouldn't want to enable this option? If there is not reason to not enable it, maybe we don't need a config parameter.

(BTW, this symbol is orphan. You can definitely remove it then).

This description is just a serie a paraphrase of the title. You have to guide the user in his choice: why he should want to not enable it.

Typically, it would probably makes sense to warn will break support for WPA3 and MPF (so so generate some security weakness).

It would be nice to include a couple of number if you have ones.

soc/silabs/silabs_siwx91x/siwg917/siwx91x_nwp.c

jerome-pouiller · 2025-10-22T10:38:42Z

soc/silabs/silabs_siwx91x/siwg917/siwx91x_nwp.c

+					   SL_SI91X_CUSTOM_FEAT_EXTENSION_VALID)
+					: (SL_SI91X_CUSTOM_FEAT_EXTENSION_VALID |
+					   SL_SI91X_CUSTOM_FEAT_ASYNC_CONNECTION_STATUS |
+					   SL_SI91X_CUSTOM_FEAT_RTC_FROM_HOST),


Do you need the condition? Why not just

.custom_feature_bit_map = SL_SI91X_CUSTOM_FEAT_EXTENSION_VALID | SL_SI91X_CUSTOM_FEAT_SOC_CLK_CONFIG_160MHZ;

My question is still relevant.

soc/silabs/silabs_siwx91x/siwg917/siwx91x_nwp.c

jerome-pouiller · 2025-10-23T09:16:26Z

drivers/wifi/siwx91x/Kconfig.siwx91x

+	help
+	  Enable this option to improve the throughput performance of the SiWx91x
+	  series. This feature provides enhanced data transfer capabilities and
+	  improved network performance.


This description is just a serie a paraphrase of the title. You have to guide the user in his choice: why he should want to not enable it.

Typically, it would probably makes sense to warn will break support for WPA3 and MPF (so so generate some security weakness).

It would be nice to include a couple of number if you have ones.

jerome-pouiller · 2025-10-23T09:22:00Z

soc/silabs/silabs_siwx91x/siwg917/siwx91x_nwp.c

 		SL_SI91X_EXT_FEAT_FRONT_END_SWITCH_PINS_ULP_GPIO_4_5_0;
+	if (!IS_ENABLED(CONFIG_WIFI_SIWX91X_HIGH_THROUGHPUT)) {
+		boot_config->ext_custom_feature_bit_map |= SL_SI91X_EXT_FEAT_IEEE_80211W;
+	}


Can you confirm the rest of the driver still behave properly in the various MFP scenario (I assume sl_wifi_set_mfp() will return an error and everything is fine).

BTW, I think disabling 11w should be a separated parameters (I don't think many user want it). Do you know what is the performance benefit of this parameter alone?

jerome-pouiller · 2025-10-23T09:27:15Z

soc/silabs/silabs_siwx91x/siwg917/siwx91x_nwp.c

+					   SL_SI91X_CUSTOM_FEAT_EXTENSION_VALID)
+					: (SL_SI91X_CUSTOM_FEAT_EXTENSION_VALID |
+					   SL_SI91X_CUSTOM_FEAT_ASYNC_CONNECTION_STATUS |
+					   SL_SI91X_CUSTOM_FEAT_RTC_FROM_HOST),


My question is still relevant.

soc/silabs/silabs_siwx91x/siwg917/siwx91x_nwp.c

jerome-pouiller · 2025-11-03T15:34:01Z

soc/silabs/silabs_siwx91x/siwg917/siwx91x_nwp.c

 		}
+		.ta_pool = {.tx_ratio_in_buffer_pool     = 1,
+			    .rx_ratio_in_buffer_pool     = 1,
+			    .global_ratio_in_buffer_pool = 1}


New line after opening brace:

.ta_pool = { .tx_ratio_in_buffer_pool = 1, .rx_ratio_in_buffer_pool = 1, .global_ratio_in_buffer_pool = 1, }

I have not been to understand how these ratio work despite the documentation in sl_wifi_device.h. If all the values are identical, every buffer type has one third of the memory? Then, how the new values are different from the previous ones?
Or every buffer type get x/10 of the memory? Then, how it worked when these values were 0?

For the record: when this structure was filled with zeros, the NWP only reserved a small fixed numbers (~10) of RX/TX buffers. With the new code, the number of buffers is a ratio (1/3 Tx, 1/3 Rx) of the available RAM.

soc/silabs/silabs_siwx91x/siwg917/siwx91x_nwp.c

Martinhoff-maker · 2025-11-04T08:30:54Z

dts/arm/silabs/siwg917.dtsi

 		 * to NWP, better are the WiFi and BLE performances.
 		 */
-		reg = <0x00000400 DT_SIZE_K(255)>;
+		reg = <0x00000400 DT_SIZE_K(319)>;


If I understand right, you give less memory to NWP and you got better perf ?

The original values has been computed to offload socket/TCP/IP to the NWP. However, since the Zephyr is taking care about these layers, the NWP is less solicited than initially expected. So, it makes sense to allocate more memory to the M4. Probably the user would want to allocate more memory to the NWP if he enables CONFIG_WIFI_SILABS_SIWX91X_NET_STACK_OFFLOAD.

@ragurram26, maye the sentence "Less memory is allocated to Zephyr, more memory is allocated to NWP, better are the WiFi and BLE performances." is not accurate anymore?

@ragurram26 don't you believe the comment should be updated?

soc/silabs/silabs_siwx91x/siwg917/siwx91x_nwp.c

jerome-pouiller · 2025-11-05T16:04:58Z

dts/arm/silabs/siwg917.dtsi

 		 * to NWP, better are the WiFi and BLE performances.
 		 */
-		reg = <0x00000400 DT_SIZE_K(255)>;
+		reg = <0x00000400 DT_SIZE_K(319)>;


The original values has been computed to offload socket/TCP/IP to the NWP. However, since the Zephyr is taking care about these layers, the NWP is less solicited than initially expected. So, it makes sense to allocate more memory to the M4. Probably the user would want to allocate more memory to the NWP if he enables CONFIG_WIFI_SILABS_SIWX91X_NET_STACK_OFFLOAD.

@ragurram26, maye the sentence "Less memory is allocated to Zephyr, more memory is allocated to NWP, better are the WiFi and BLE performances." is not accurate anymore?

jhedberg · 2025-11-24T10:31:22Z

dts/bindings/net/wireless/silabs,siwx91x-nwp.yaml

      - deep-sleep-with-ram-retention
    required: true
+
+  soc-120mhz-clk:


Is this a standard naming for this property? If it's specific to Silabs, perhaps it should have some appropriate prefix? (e.g. silabs,)

It seems to me like a clock-frequency property would be better, potentially with an enum to limit the selection to allowed values.

BTW, I am a bit confused by the name. What soc means here? Is it clock the NWP clock? If this parameter is not set what is the frequency of the NWP?

SL_SI91X_CUSTOM_FEAT_SOC_CLK_CONFIG_160MHZ also exists. Why we don't use 160MHz?

I assume 160MHZ is exclusive wit 120MHZ, so a boolean is probably not the right type.

Dismissing my review until my question is answered

To enhance throughput performance on the SiWx91x series, Added some Siwx91x configurations by deafult. Signed-off-by: Rahul Gurram <[email protected]>

Added support for 120Mhz soc clock via DTS.Setting the clock to 120 MHz can improve throughput Signed-off-by: Rahul Gurram <[email protected]>

sonarqubecloud · 2025-11-25T05:22:23Z

Please retry analysis of this Pull-Request directly on SonarQube Cloud

zephyrbot added area: Wi-Fi Wi-Fi area: Boards/SoCs platform: Silabs Silicon Labs labels Oct 22, 2025

zephyrbot requested review from Kludentwo, MaochenWang1, Martinhoff-maker, asmellby, jerome-pouiller, jhedberg, jukkar, krish2718 and rlubos October 22, 2025 09:35

zephyrbot assigned jhedberg and asmellby Oct 22, 2025

ragurram26 force-pushed the Throughput_configurations branch from 7487afc to 6cf1b1d Compare October 22, 2025 10:22

jerome-pouiller reviewed Oct 22, 2025

View reviewed changes

ragurram26 force-pushed the Throughput_configurations branch 3 times, most recently from 92977e7 to 576dc52 Compare October 23, 2025 09:04

jerome-pouiller reviewed Oct 23, 2025

View reviewed changes

ragurram26 force-pushed the Throughput_configurations branch 7 times, most recently from 42c44e2 to f226c12 Compare November 3, 2025 15:22

ragurram26 changed the title ~~drivers: wifi: siwx91x: SiWx91x high performance configurations~~ To improve SiWx91x performance Nov 3, 2025

jerome-pouiller reviewed Nov 3, 2025

View reviewed changes

ragurram26 force-pushed the Throughput_configurations branch from f226c12 to db5eaa1 Compare November 4, 2025 05:13

ragurram26 force-pushed the Throughput_configurations branch from db5eaa1 to 937f705 Compare November 4, 2025 05:26

Martinhoff-maker reviewed Nov 4, 2025

View reviewed changes

jerome-pouiller reviewed Nov 5, 2025

View reviewed changes

jhedberg changed the title ~~To improve SiWx91x performance~~ soc: silabs: siwx91x: Improve performance Nov 6, 2025

zephyrbot added the area: Devicetree Bindings label Nov 12, 2025

ragurram26 force-pushed the Throughput_configurations branch 5 times, most recently from 3577df3 to 93eeeee Compare November 13, 2025 04:13

jerome-pouiller previously approved these changes Nov 24, 2025

View reviewed changes

jhedberg previously approved these changes Nov 24, 2025

View reviewed changes

jhedberg reviewed Nov 24, 2025

View reviewed changes

ragurram26 dismissed jerome-pouiller’s stale review via ea8b5f6 November 25, 2025 04:56

ragurram26 force-pushed the Throughput_configurations branch from 93eeeee to ea8b5f6 Compare November 25, 2025 04:56

ragurram26 added 2 commits November 25, 2025 10:26

drivers: wifi: siwx91x: To improve SiWx91x performance

5dddb23

To enhance throughput performance on the SiWx91x series, Added some Siwx91x configurations by deafult. Signed-off-by: Rahul Gurram <[email protected]>

soc: silabs: siwg917: Added 120Mhz soc clock features via DTS

26af633

Added support for 120Mhz soc clock via DTS.Setting the clock to 120 MHz can improve throughput Signed-off-by: Rahul Gurram <[email protected]>

ragurram26 force-pushed the Throughput_configurations branch from ea8b5f6 to 26af633 Compare November 25, 2025 04:56

soc: silabs: siwx91x: Improve performance #98060

Are you sure you want to change the base?

soc: silabs: siwx91x: Improve performance #98060

Conversation

ragurram26 commented Oct 22, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jerome-pouiller Nov 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jerome-pouiller Nov 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sonarqubecloud bot commented Nov 25, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

jerome-pouiller Nov 24, 2025 •

edited

Loading

jerome-pouiller Nov 24, 2025 •

edited

Loading