Re: [OpenWrt-Devel] [RFC/RFT PATCH] ath9k: implement kthread entropy collection for AR5008 and AR9002 PHYs.
Hi again, Felix, On Mon, 10 Feb 2020 at 15:46, Rui Salvaterra wrote: > > As an aside, this WZR-HP-AG300H is running as an AP, at a public > institution, with 12 STAs connected, at the moment (8 at 2,4 GHz, 4 at > 5 GHz), without any reported issues. So, since Tuesday, I've been running cat /dev/random > /dev/null on this router to stress the ADC entropy collection as much as possible. Everything's working just fine, though I'm seeing this in /sys/kernel/debug/ieee80211/phy?/ath9k/reset: phy0 (2.4 GHz): Baseband Hang: 0 Baseband Watchdog: 0 Fatal HW Error: 0 TX HW error: 0 Transmit timeout: 0 TX Path Hang: 0 PLL RX Hang: 0 MAC Hang: 1 Stuck Beacon: 76 MCI Reset: 0 Calibration error: 0 Tx DMA stop error: 69 Rx DMA stop error: 0 phy1 (5 GHz): Baseband Hang: 0 Baseband Watchdog: 0 Fatal HW Error: 0 TX HW error: 0 Transmit timeout: 0 TX Path Hang: 0 PLL RX Hang: 0 MAC Hang: 1 Stuck Beacon: 1 MCI Reset: 0 Calibration error: 0 Tx DMA stop error: 0 Rx DMA stop error: 0 I suspect these errors might be related to the DMA stop sequence (which you actually fixed/mitigated [1], as I seem to recall having this issue on my TL-WDR3600), not the entropy collection. In any case, I'm going to test an image which reverses the stop sequence also on AR9002, to see if this has any effect. Note that /sys/kernel/debug/ieee80211/phy?/ath9k/queues are completely clean on both PHYs: phy0 (2.4 GHz): (VO): qnum: 3 qdepth: 0 ampdu-depth: 0 pending: 0 (VI): qnum: 2 qdepth: 0 ampdu-depth: 0 pending: 0 (BE): qnum: 1 qdepth: 0 ampdu-depth: 0 pending: 0 (BK): qnum: 0 qdepth: 0 ampdu-depth: 0 pending: 0 (CAB): qnum: 8 qdepth: 0 ampdu-depth: 0 pending: 0 phy1 (5 GHz): (VO): qnum: 3 qdepth: 0 ampdu-depth: 0 pending: 0 (VI): qnum: 2 qdepth: 0 ampdu-depth: 0 pending: 0 (BE): qnum: 1 qdepth: 0 ampdu-depth: 0 pending: 0 (BK): qnum: 0 qdepth: 0 ampdu-depth: 0 pending: 0 (CAB): qnum: 8 qdepth: 0 ampdu-depth: 0 pending: 0 Thanks, Rui [1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/drivers/net/wireless/ath/ath9k?id=300f77c08ded96d33f492aaa02549103852f0c12 ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [OpenWrt-Devel] [RFC/RFT PATCH] ath9k: implement kthread entropy collection for AR5008 and AR9002 PHYs.
Hi again, Felix, On Mon, 10 Feb 2020 at 13:37, Felix Fietkau wrote: > The issues that were reported years ago were mainly stuck beacons, > increase in hardware resets, connection stability issues. Thanks for the info, I'll keep an eye out for those. The dmesg is clean, I haven't enabled any debugging, but I'll do so for further testing. As an aside, this WZR-HP-AG300H is running as an AP, at a public institution, with 12 STAs connected, at the moment (8 at 2,4 GHz, 4 at 5 GHz), without any reported issues. Thanks, Rui ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [OpenWrt-Devel] [RFC/RFT PATCH] ath9k: implement kthread entropy collection for AR5008 and AR9002 PHYs.
On 2020-02-10 10:44, Rui Salvaterra wrote: > On Fri, 31 Jan 2020 at 08:12, Felix Fietkau wrote: >> >> For at least AR5008 and AR9002, but probably also for AR9003 I would >> like to keep the behavior of collecting entropy only once at driver >> initialization. >> Last time I worked on this I noticed that on several chips, sampling >> entropy during normal operation caused stability issues that were hard >> to pin down but quite noticeable. >> I think the benefit of continuous entropy collection is simply not worth >> the extra cost of potential stability issues and debugging time. >> >> - Felix > > Hi again, Felix, > > FWIW, this patch has just survived a whole weekend of rngtest < > /dev/random on a Buffalo WZR-HP-AG300H (dual AR9002). Output follows: > > root@AP157427:~# rngtest < /dev/random > rngtest 6.6 > Copyright (c) 2004 by Henrique de Moraes Holschuh > This is free software; see the source for copying conditions. There > is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A > PARTICULAR PURPOSE. > > rngtest: starting FIPS tests... > ^Crngtest: bits received from input: 224241980032 > rngtest: FIPS 140-2 successes: 11203058 > rngtest: FIPS 140-2 failures: 9041 > rngtest: FIPS 140-2(2001-10-10) Monobit: 1165 > rngtest: FIPS 140-2(2001-10-10) Poker: 1175 > rngtest: FIPS 140-2(2001-10-10) Runs: 3340 > rngtest: FIPS 140-2(2001-10-10) Long run: 3405 > rngtest: FIPS 140-2(2001-10-10) Continuous run: 0 > rngtest: input channel speed: (min=165.459; avg=1025.932; > max=1093.147)Kibits/s > rngtest: FIPS tests speed: (min=503.707; avg=14584.165; max=14920.741)Kibits/s > rngtest: Program run time: 228508694816 microseconds The issues that were reported years ago were mainly stuck beacons, increase in hardware resets, connection stability issues. - Felix ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [OpenWrt-Devel] [RFC/RFT PATCH] ath9k: implement kthread entropy collection for AR5008 and AR9002 PHYs.
On Fri, 31 Jan 2020 at 08:12, Felix Fietkau wrote: > > For at least AR5008 and AR9002, but probably also for AR9003 I would > like to keep the behavior of collecting entropy only once at driver > initialization. > Last time I worked on this I noticed that on several chips, sampling > entropy during normal operation caused stability issues that were hard > to pin down but quite noticeable. > I think the benefit of continuous entropy collection is simply not worth > the extra cost of potential stability issues and debugging time. > > - Felix Hi again, Felix, FWIW, this patch has just survived a whole weekend of rngtest < /dev/random on a Buffalo WZR-HP-AG300H (dual AR9002). Output follows: root@AP157427:~# rngtest < /dev/random rngtest 6.6 Copyright (c) 2004 by Henrique de Moraes Holschuh This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. rngtest: starting FIPS tests... ^Crngtest: bits received from input: 224241980032 rngtest: FIPS 140-2 successes: 11203058 rngtest: FIPS 140-2 failures: 9041 rngtest: FIPS 140-2(2001-10-10) Monobit: 1165 rngtest: FIPS 140-2(2001-10-10) Poker: 1175 rngtest: FIPS 140-2(2001-10-10) Runs: 3340 rngtest: FIPS 140-2(2001-10-10) Long run: 3405 rngtest: FIPS 140-2(2001-10-10) Continuous run: 0 rngtest: input channel speed: (min=165.459; avg=1025.932; max=1093.147)Kibits/s rngtest: FIPS tests speed: (min=503.707; avg=14584.165; max=14920.741)Kibits/s rngtest: Program run time: 228508694816 microseconds Best regards, Rui ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [OpenWrt-Devel] [RFC/RFT PATCH] ath9k: implement kthread entropy collection for AR5008 and AR9002 PHYs.
On Fri, 31 Jan 2020 at 09:06, Rui Salvaterra wrote: > > Like I wrote in the forum, I don't have (yet) AR5008 hardware to test. And I just ordered an AR5BXB72* for testing purposes. I can configure it as AP on my Turris Omnia. * https://www.aliexpress.com/item/32356022951.html ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [OpenWrt-Devel] [RFC/RFT PATCH] ath9k: implement kthread entropy collection for AR5008 and AR9002 PHYs.
Hi, Felix, On Fri, 31 Jan 2020 at 08:12, Felix Fietkau wrote: > For at least AR5008 and AR9002, but probably also for AR9003 I would > like to keep the behavior of collecting entropy only once at driver > initialization. Well, you could have told me this before I started working on it, but I guess you don't read the developer section of the forum, even when you're mentioned. ;) (What's the best communication channel for OpenWrt development? There are far too many options, I think.) FWIW, my Archer C6 v2 has been running OpenWrt 19.07 with my entropy patch for days without any problems (1 x AR9003, though). I'll be testing soon on a TL-WDR3600 (2 x AR9003) and a WNDR3700 v1 (2 x AR9002). Like I wrote in the forum, I don't have (yet) AR5008 hardware to test. > Last time I worked on this I noticed that on several chips, sampling > entropy during normal operation caused stability issues that were hard > to pin down but quite noticeable. Maybe it's not a good idea to sample from a low power state? :) The old code didn't call ath9k_ps_wakeup before collecting (and ath9k_ps_restore afterwards), it's possible that the stability problems are related to the power state at collection time. The current mainline code is from 2015, and it's nothing like the one in the OpenWrt patch (from 2013). The entropy collection is not continuous at all, it's done on a kthread which only wakes up when the available entropy drops below a certain threshold. I only extended the mainline code to support the previous PHYs. ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [OpenWrt-Devel] [RFC/RFT PATCH] ath9k: implement kthread entropy collection for AR5008 and AR9002 PHYs.
On 2020-01-30 21:03, Rui Salvaterra wrote: > The mainline ath9k driver is able to provide a hardware random number > generator by collecting radio noise from the PHY ADC (using a kthread > to fill up the entropy pool as needed). Nevertheless, this feature has > only been implemented for the more recent AR9003 PHYs. > > Meanwhile, OpenWrt has been carrying a patch to provide entropy from the > ADC for the three supported PHYs for a long time, but this patch only > collects entropy once per existing PHY, at the driver initialisation > time. > > This patch enables kconfig support for this feature and updates the > OpenWrt patch, in order to add ADC entropy collection support to both > AR5008 and AR9002 PHYs. > > Signed-off-by: Rui Salvaterra For at least AR5008 and AR9002, but probably also for AR9003 I would like to keep the behavior of collecting entropy only once at driver initialization. Last time I worked on this I noticed that on several chips, sampling entropy during normal operation caused stability issues that were hard to pin down but quite noticeable. I think the benefit of continuous entropy collection is simply not worth the extra cost of potential stability issues and debugging time. - Felix ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
[OpenWrt-Devel] [RFC/RFT PATCH] ath9k: implement kthread entropy collection for AR5008 and AR9002 PHYs.
The mainline ath9k driver is able to provide a hardware random number generator by collecting radio noise from the PHY ADC (using a kthread to fill up the entropy pool as needed). Nevertheless, this feature has only been implemented for the more recent AR9003 PHYs. Meanwhile, OpenWrt has been carrying a patch to provide entropy from the ADC for the three supported PHYs for a long time, but this patch only collects entropy once per existing PHY, at the driver initialisation time. This patch enables kconfig support for this feature and updates the OpenWrt patch, in order to add ADC entropy collection support to both AR5008 and AR9002 PHYs. Signed-off-by: Rui Salvaterra --- package/kernel/mac80211/ath.mk| 7 + .../ath/543-ath9k_entropy_from_adc.patch | 216 ++ 2 files changed, 134 insertions(+), 89 deletions(-) diff --git a/package/kernel/mac80211/ath.mk b/package/kernel/mac80211/ath.mk index 788131b751..fb6e893494 100644 --- a/package/kernel/mac80211/ath.mk +++ b/package/kernel/mac80211/ath.mk @@ -8,6 +8,7 @@ PKG_CONFIG_DEPENDS += \ CONFIG_PACKAGE_ATH_SPECTRAL \ CONFIG_PACKAGE_ATH_DYNACK \ CONFIG_ATH9K_SUPPORT_PCOEM \ + CONFIG_ATH9K_HWRNG \ CONFIG_ATH9K_TX99 \ CONFIG_ATH10K_LEDS \ CONFIG_ATH10K_THERMAL \ @@ -45,6 +46,7 @@ config-$(CONFIG_TARGET_ipq40xx) += ATH10K_AHB config-$(CONFIG_PCI) += ATH9K_PCI config-$(CONFIG_ATH_USER_REGD) += ATH_USER_REGD config-$(CONFIG_ATH9K_SUPPORT_PCOEM) += ATH9K_PCOEM +config-$(CONFIG_ATH9K_HWRNG) += ATH9K_HWRNG config-$(CONFIG_ATH9K_TX99) += ATH9K_TX99 config-$(CONFIG_ATH9K_UBNTHSR) += ATH9K_UBNTHSR config-$(CONFIG_ATH10K_LEDS) += ATH10K_LEDS @@ -211,6 +213,11 @@ define KernelPackage/ath9k/config bool "Support chips used in PC OEM cards" depends on PACKAGE_kmod-ath9k + config ATH9K_HWRNG + bool "Use the PHY ADC as a hardware random number generator" + depends on PACKAGE_kmod-ath9k + select PACKAGE_kmod-random-core + config ATH9K_TX99 bool "Enable TX99 support (WARNING: testing only, breaks normal operation!)" depends on PACKAGE_kmod-ath9k diff --git a/package/kernel/mac80211/patches/ath/543-ath9k_entropy_from_adc.patch b/package/kernel/mac80211/patches/ath/543-ath9k_entropy_from_adc.patch index 64bd6cacfd..8b92fbcb0b 100644 --- a/package/kernel/mac80211/patches/ath/543-ath9k_entropy_from_adc.patch +++ b/package/kernel/mac80211/patches/ath/543-ath9k_entropy_from_adc.patch @@ -8,145 +8,113 @@ * @spectral_scan_config: set parameters for spectral scan and enable/disable it * @spectral_scan_trigger: trigger a spectral scan run * @spectral_scan_wait: wait for a spectral scan run to finish -@@ -744,6 +745,7 @@ struct ath_hw_ops { - struct ath_hw_antcomb_conf *antconf); - void (*antdiv_comb_conf_set)(struct ath_hw *ah, - struct ath_hw_antcomb_conf *antconf); -+ void (*get_adc_entropy)(struct ath_hw *ah, u8 *buf, size_t len); - void (*spectral_scan_config)(struct ath_hw *ah, -struct ath_spec_scan *param); - void (*spectral_scan_trigger)(struct ath_hw *ah); +@@ -756,6 +757,10 @@ struct ath_hw_ops { + #ifdef CPTCFG_ATH9K_BTCOEX_SUPPORT + void (*set_bt_ant_diversity)(struct ath_hw *hw, bool enable); + #endif ++ ++#ifdef CPTCFG_ATH9K_HWRNG ++ int (*get_adc_entropy)(struct ath_hw *ah, u32 *buf, const u32 buf_size, u32 *rng_last); ++#endif + }; + + struct ath_nf_limits { --- a/drivers/net/wireless/ath/ath9k/ar9003_phy.c +++ b/drivers/net/wireless/ath/ath9k/ar9003_phy.c -@@ -1927,6 +1927,26 @@ void ar9003_hw_init_rate_txpower(struct +@@ -1927,6 +1927,33 @@ void ar9003_hw_init_rate_txpower(struct } } -+static void ar9003_hw_get_adc_entropy(struct ath_hw *ah, u8 *buf, size_t len) ++static int __maybe_unused ++ar9003_hw_get_adc_entropy(struct ath_hw *ah, u32 *buf, const u32 buf_size, u32 *rng_last) +{ + int i, j; ++ u32 v1, v2, last = *rng_last; + + REG_RMW_FIELD(ah, AR_PHY_TEST, AR_PHY_TEST_BBB_OBS_SEL, 1); + REG_CLR_BIT(ah, AR_PHY_TEST, AR_PHY_TEST_RX_OBS_SEL_BIT5); + REG_RMW_FIELD(ah, AR_PHY_TEST_CTL_STATUS, AR_PHY_TEST_CTL_RX_OBS_SEL, 0); + -+ memset(buf, 0, len); -+ for (i = 0; i < len; i++) { -+ for (j = 0; j < 4; j++) { -+ u32 regval = REG_READ(ah, AR_PHY_TST_ADC); ++ for (i = 0, j = 0; i < buf_size; i++) { ++ v1 = REG_READ(ah, AR_PHY_TST_ADC) & 0x; ++ v2 = REG_READ(ah, AR_PHY_TST_ADC) & 0x; ++ ++ /* wait for data ready */ ++ if (v1 && v2 && last != v1 && v1 != v2 && v1 != 0x && ++ v2 != 0x) ++ buf[j++] = (v1 << 16) | v2; + -+ buf[i] <<= 2; -+ buf[i] |= (regval & 1) | ((regval & BIT(10)) >> 9); -