From mboxrd@z Thu Jan 1 00:00:00 1970 Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.remote-ip=100.103.45.18 ARC-Seal: i=1; d=subspace.kernel.org; s=arc20261015; a=rsa-sha256; cv=none; t=1795303609; b=Y4WtifwdMqS+Vpqq4k3KlFQ7neyCYipB0kSuAJPXsJqAOdFi0hW9xeOIKF/5pqjg/iBB 2BrJJOh/HLIna9x+z5aFJ0EUx8PcqUhcuYI5eqazZiqoUar66sVa6zR+uPeXHVoOD9YcW 6zUYWEwGKQQBWu60S2/Uv7DfPmHKboMXLpTckL7O5lfbMwhOhj6LXdJ41bsXRcxWxInGv n6CSoLqM01KoXO4cEfChzo/CAcTsbk+6aGevwV2qOp5PHgkQ4vyaN2GIaGZ45/F1ifDIc x9yqyTc64pKTCD08mBcQj6vid5PrnfLT4mCr6ySNqcNteLSyI+jlSer0XbFAEUPwysg== ARC-Message-Signature: i=1; d=subspace.kernel.org; s=arc20261015; a=rsa-sha256; c=relaxed/relaxed; t=1795303609; h=Received:DKIM-Signature:From:To:Cc:Subject:Date:Message-ID:X-Mailer: In-Reply-To:References:MIME-Version:Content-Transfer-Encoding; bh=uioZ55CJ2GHe6TiiERn3Q6GU4DCpV+IM+fvb4tz39pA=; b=DEG9HiqPcadjGB/uET04OJmVniQsUANq+scxNIIf6W5GmJoycp2FPj0FCgUAynWsDk6f eX295xvW8WEIWZwBM2p+pnotr/U8yyCgMH6tLg/050g5io3l/E4guH+/7dhwBQTVyBiUA yWlfrEmf8bU6+brGNnPpsEZIaNznMK2IZHOmv+/OQvwxfl7ctPcnP44DJVzT5YMuwi5m6 01ISu4use7rPDSSoaLNgsLkERD/8yq6JToHKGesdHtOFgllcIC3wIx9j8UnY0ct42Lg5O t5Q3MX1AXHzRPC9ncwbHCSkIgm3JywEYK4NaxcAigHB7RpXH42eEuYk+CrTwyFQvPxw== ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dkim=pass header.d=kernel.org header.i=@kernel.org header.a=rsa-sha256 header.s=k20260515 header.b=XbHPKRY5; arc=none smtp.remote-ip=100.103.45.18 Received: from smtp.kernel.org (quasi.space.kernel.org [100.103.45.18]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature ECDSA (prime256v1)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 61BAF606A339; Sat, 21 Nov 2026 23:26:49 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.a=rsa-sha256 header.s=k20260515 header.b=XbHPKRY5 Received: by smtp.kernel.org (Postfix) with ESMTPSA id 607751F000D4; Sat, 21 Nov 2026 23:26:48 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=kernel.org; s=k20260515; t=1795303609; bh=uioZ55CJ2GHe6TiiERn3Q6GU4DCpV+IM+fvb4tz39pA=; h=From:To:Cc:Subject:Date:In-Reply-To:References; b=XbHPKRY5WsIx3t8S+1s5NS/T06LZA9GvIXh7SJ368t62lMYFNvFa62BSVEde1md6l Q6fPcCERMSviZFU/Twdh6qNKVrLse62C1H1rz5cX1k4HmIwzBKngMEqAaRHs4yG9dx 1PtCeWKbqYjyd9r6eYWiLLg0A6dEEUHkEN4GRPA+QfhECz6xd71HkU1xbbLjBSZyTy 0Y6XgsaOCJ8X/1JcxcGdj+viclM55VeMeUTFIp3YETHP0TVdpOHS6qJm60ezby369W MStHZ4lP9Oj2VW4E/x5u9hYKVeK67Cz4F8tNQkNu+xr6XYYy27Kr2ZgcZLvpMWpTJW ihqY8Dgza/srw== From: Jakub Kicinski To: ansuelsmth@gmail.com Cc: Jakub Kicinski , andrew+netdev@lunn.ch, davem@davemloft.net, edumazet@kernel.org, pabeni@redhat.com, robh@kernel.org, krzk+dt@kernel.org, conor+dt@kernel.org, horms@kernel.org, corbet@lwn.net, skhan@linuxfoundation.org, rdunlap@infradead.org, lorenzo@kernel.org, hkallweit1@gmail.com, linux@armlinux.org.uk, maxime.chevallier@bootlin.com, p.zabel@pengutronix.de, nathan@kernel.org, ndesaulniers@google.com, morbo@google.com, justinstitt@google.com, rmk+kernel@armlinux.org.uk, netdev@vger.kernel.org, devicetree@vger.kernel.org, linux-kernel@vger.kernel.org, linux-doc@vger.kernel.org, linux-arm-kernel@lists.infradead.org, linux-mediatek@lists.infradead.org, llvm@lists.linux.dev, zhengping.zhang@airoha.com Subject: Re: [PATCH net-next v22 07/15] net: phylink: add capabilities helpers and track modified advertising Date: Sat, 21 Nov 2026 15:26:47 -0800 Message-ID: <20261121232647.1365377-1-kuba@kernel.org> X-Mailer: git-send-email 2.56.1 In-Reply-To: <20261117123956.11981-8-ansuelsmth@gmail.com> References: <20261117123956.11981-8-ansuelsmth@gmail.com> X-Mailing-List: devicetree@vger.kernel.org List-Id: List-Post: List-Help: List-Subscribe: List-Unsubscribe: List-Owner: Precedence: list MIME-Version: 1.0 Content-Transfer-Encoding: 8bit This is an AI-generated review of your patch. The human sending this email has considered the AI review valid, or at least plausible. --- net: phylink: add capabilities helpers and track modified advertising This prepares for PCS release and late attach. It adds phylink_update_capabilities() to recompute the supported and advertised link modes, and it tracks the full advertisement and a user-disabled mask in struct phylink. It also saves the PHY advertisement before restricting it, and stops the SFP path from modifying pl->sfp_support in place. > diff --git a/drivers/net/phy/phylink.c b/drivers/net/phy/phylink.c > index 88280866c19a2..921c227a5b3c0 100644 > --- a/drivers/net/phy/phylink.c > +++ b/drivers/net/phy/phylink.c [ ... ] > @@ -2120,6 +2233,142 @@ static void phylink_update_pause_state(struct phylink *pl, int pause_state) [ ... ] > +static void phylink_update_capabilities(struct phylink *pl, > + struct phylink_link_state *config) > +{ > + if (pl->phydev) { > + lockdep_assert_held(&pl->phydev_mutex); > + lockdep_assert_held(&pl->phydev->lock); > + } > + lockdep_assert_held(&pl->state_mutex); > + > + if (pl->phydev) { > + phylink_phy_update_capabilities(pl, pl->phydev, > + pl->phy_state.interface); > + return; > + } > + > + if (pl->cfg_link_an_mode == MLO_AN_FIXED) { > + phylink_parse_fixedlink_capabilities(pl, config); > + goto exit; > + } > + > + if (pl->cfg_link_an_mode == MLO_AN_INBAND) { > + phylink_parse_inband_capabilities(pl, config); > + goto exit; > + } [Severity: Medium] Can the SFP branch below ever run for an SFP cage described with managed = "in-band-status"? phylink_sfp_set_config() only updates req_link_an_mode and link_config.interface. cfg_link_an_mode therefore stays MLO_AN_INBAND while an optical module is in use. This check then sends the recompute to phylink_parse_inband_capabilities(). That function zeroes pl->supported and rebuilds it from the generic interface caps, not from pl->sfp_support. For 10GBASE-R this gives every 10G link mode plus Autoneg instead of the modes read from the module's EEPROM. It also brings back the Autoneg bit that phylink_sfp_config_optical() removes when the only interface is INBAND_NONE. Some module interfaces are not in the in-band switch, for example PHY_INTERFACE_MODE_100BASEX. For those, phylink_parse_inband_capabilities() returns after pl->supported has already been cut down to MII, Autoneg, Pause and Asym_Pause: default: phylink_err(pl, "incorrect link mode %s for in-band status\n", phy_modes(config->interface)); return -EINVAL; The return value is ignored here, so pl->supported is left with no speed modes. The old code restricted the result with linkmode_and(pl->supported, pl->supported, pl->sfp_support) whatever the AN mode was. At this commit, phylink_update_mac_pause_capabilities() reaches this path. Later in the series pcs_provider_notify() does too, and the branch order there is unchanged. > + > + /* Revalidate the SFP module modes against every available interface. */ > + if (pl->sfp_bus && !pl->phydev && config->interface && > + !phy_interface_empty(pl->sfp_interfaces)) { > + __ETHTOOL_DECLARE_LINK_MODE_MASK(supported); > + > + linkmode_copy(supported, pl->sfp_support); > + linkmode_copy(config->advertising, pl->sfp_support); > + if (!__phylink_validate_mask(pl, NULL, supported, config, > + pl->sfp_interfaces)) { > + phylink_sfp_update_capabilities(pl, supported, config, > + config->interface); > + linkmode_copy(pl->supported, supported); > + linkmode_copy(pl->advertising, config->advertising); > + } [Severity: Medium] Should this also narrow the result to the active link_config.interface? __phylink_validate_mask() returns the union across all of pl->sfp_interfaces. phylink_sfp_update_capabilities() only adjusts the Autoneg bit. Other paths do narrow. phylink_sfp_config_optical() validates against the chosen interface before phylink_sfp_set_config() stores the advertisement. The old phylink_update_mac_pause_capabilities() did the same through __phylink_validate(pl, pl->supported, config). After a recompute, pl->advertising and link_config.advertising can contain both 1000baseX_Full and 10000baseSR_Full while the link runs 10GBASE-R. phylink_ethtool_ksettings_get() then reports modes that the current interface can't run. phylink_ethtool_ksettings_set() also computes user_disabled_advertising against that wider mask. > + > + goto exit; > + } [ ... ] > @@ -2146,20 +2396,16 @@ void phylink_update_mac_pause_capabilities(struct phylink *pl, unsigned long mac > if (old_pause == mac_pause) > return; > > + mutex_lock(&pl->phydev_mutex); > + phydev = pl->phydev; > + if (phydev) > + mutex_lock(&phydev->lock); > mutex_lock(&pl->state_mutex); > > pl->config->mac_capabilities &= ~(MAC_SYM_PAUSE | MAC_ASYM_PAUSE); > pl->config->mac_capabilities |= mac_pause; > > - phylink_set(pl->supported, Pause); > - phylink_set(pl->supported, Asym_Pause); > - > - if (pl->phydev) > - linkmode_and(pl->supported, pl->supported, pl->phydev->supported); > - else if (pl->sfp_bus) > - linkmode_and(pl->supported, pl->supported, pl->sfp_support); > - > - __phylink_validate(pl, pl->supported, config); > + phylink_update_capabilities(pl, config); [Severity: Medium] Does the PHY still renegotiate when the MAC loses pause support? With a PHY attached, this call goes through phylink_phy_update_capabilities(). That function validates pl->phy_advertising against MAC caps that no longer include pause, then writes the result straight into the PHY: /* Restrict the phy advertisement according to the MAC support. */ linkmode_copy(phy->advertising, config.advertising); It does not call phy_config_aneg() or phy_start_aneg(). By the time phylink_update_pause_state() calls phy_set_asym_pause(), the pause bits are already clear: drivers/net/phy/phy_device.c:phy_set_asym_pause() { linkmode_copy(oldadv, phydev->advertising); linkmode_set_pause(phydev->advertising, tx, rx); if (!linkmode_equal(oldadv, phydev->advertising) && phydev->autoneg) phy_start_aneg(phydev); } linkmode_equal() is therefore true, and aneg is not restarted. With AN enabled, manual_changed is false, so no resolve is triggered either. Before this patch, phydev->advertising still had the pause bits at this point. phy_set_asym_pause() saw the difference and restarted aneg. The kernel-doc still says the function "triggers renegotiation if needed". The in-tree caller rtl8169_change_mtu() passes caps = 0 when switching to a jumbo MTU. Won't the PHY keep advertising PAUSE_CAP/PAUSE_ASYM, and flow control stay active in jumbo mode, until something else renegotiates? Writing phydev->advertising directly also skips phylib's own bookkeeping, such as the adv_old mask used by phy_speed_up(). [ ... ] > @@ -2507,18 +2683,17 @@ static int phylink_bringup_phy(struct phylink *pl, struct phy_device *phy, > */ > phy_support_asym_pause(phy); > > - memset(&config, 0, sizeof(config)); > - linkmode_copy(supported, phy->supported); > - linkmode_copy(config.advertising, phy->advertising); > - config.interface = interface; > + mutex_lock(&pl->phydev_mutex); > + mutex_lock(&phy->lock); > + mutex_lock(&pl->state_mutex); > + > + linkmode_copy(pl->phy_advertising, phy->advertising); > > - ret = phylink_validate_phy(pl, phy, supported, &config); > + ret = phylink_phy_update_capabilities(pl, phy, interface); [Severity: Low] The PHY-derived state is now written in this locked section: pl->phy_advertising, pl->supported, pl->advertising, link_config.advertising and phy->advertising. pl->phydev is published only later, after all locks have been dropped and phy_eee_rx_clock_stop() has run: mutex_lock(&pl->phydev_mutex); mutex_lock(&phy->lock); mutex_lock(&pl->state_mutex); pl->phydev = phy; A phylink_update_capabilities() call that runs in that window sees pl->phydev == NULL. In PHY mode without an SFP it takes the generic branch: /* No media is attached yet. Recalculate the generic capabilities. */ linkmode_fill(pl->supported); Doesn't that throw away the PHY and MAC intersection that was just computed? phylink_update_pause_state() also skips updating pl->phy_advertising while pl->phydev is NULL. The old mac-pause path only set the pause bits and revalidated the existing pl->supported, so the PHY-derived mask mostly survived. The in-tree drivers that trigger a recompute (r8169 and airoha) connect the PHY under rtnl, so this doesn't look reachable in-tree today. It could still happen for a driver that calls phylink_connect_phy() without rtnl while the fwnode PCS notifier is registered. > if (ret) { [ ... ] > @@ -3411,6 +3579,7 @@ int phylink_ethtool_ksettings_set(struct phylink *pl, > > if (pl->phydev) { > struct ethtool_link_ksettings phy_kset = *kset; > + int ret; > > linkmode_and(phy_kset.link_modes.advertising, > phy_kset.link_modes.advertising, [ ... ] > @@ -3434,7 +3603,15 @@ int phylink_ethtool_ksettings_set(struct phylink *pl, > * the presence of a PHY, this should not be changed as that > * should be determined from the media side advertisement. > */ > - return phy_ethtool_ksettings_set(pl->phydev, &phy_kset); > + ret = phy_ethtool_ksettings_set(pl->phydev, &phy_kset); > + if (!ret) { > + mutex_lock(&pl->state_mutex); > + linkmode_copy(pl->phy_advertising, > + phy_kset.link_modes.advertising); > + mutex_unlock(&pl->state_mutex); > + } [Severity: Medium] Does this keep the meaning documented for the new field? /* PHY advertisement before restricting it to MAC/PCS capabilities */ __ETHTOOL_DECLARE_LINK_MODE_MASK(phy_advertising); phy_kset.link_modes.advertising has already been ANDed with pl->supported above, and pl->supported is limited to MAC/PCS. It is also the raw request, not what phylib applied. phy_ethtool_ksettings_set() ANDs it with phydev->supported and sets the Autoneg bit from base.autoneg. Later, phylink_phy_update_capabilities() replays this copy and writes the result into phy->advertising: linkmode_copy(config.advertising, pl->phy_advertising); Can that restore an Autoneg bit while phydev->autoneg is AUTONEG_DISABLE? Modes outside pl->supported at the time of the ethtool call are also lost for good. After a PCS is re-attached they won't be advertised again. The non-PHY path avoids this because it tracks user-disabled modes. The copy also goes stale when phylib changes phydev->advertising itself, for example in phy_speed_down(). The next recompute then silently overwrites that change. > + > + return ret; > } [ ... ] > @@ -4022,19 +4201,34 @@ static void phylink_sfp_detach(void *upstream, struct sfp_bus *bus) [ ... ] > + for (i = 0; i < ARRAY_SIZE(phylink_sfp_interface_preference); i++) { > + interface = phylink_sfp_interface_preference[i]; > + if (!test_bit(interface, intf)) > + continue; > > - return interface; > + tmp_state = *state; > + tmp_state.interface = interface; > + linkmode_copy(tmp_supported, supported); > + if (phylink_validate(pl, tmp_supported, &tmp_state)) > + continue; > + > + /* Skip this preferred interface if user-disabled in advertising */ > + linkmode_andnot(tmp_state.advertising, tmp_state.advertising, > + pl->user_disabled_advertising); > + if (!phylink_is_empty_linkmode(tmp_state.advertising)) > + return interface; > + } > + > + return PHY_INTERFACE_MODE_NA; [Severity: Medium] Is pl->user_disabled_advertising ever cleared? Its only writer is phylink_update_user_disabled_advertising(), which keeps restrictions for modes that are absent from the current pl->advertising. phylink_sfp_module_remove() only zeroes sfp_interfaces and sfp_module_interfaces. An ethtool restriction made for one SFP module therefore applies to every module inserted later. For example, take a dual-rate 1G/10G module running 10GBASE-R. Running "ethtool -s ... advertise 1000baseX/Full" records 10000baseSR_Full and the pause bits as user-disabled. Then a 10G-only module is inserted. This loop validates 10GBASE-R, masks out the stale disabled modes, and is left with only Autoneg. phylink_is_empty_linkmode() returns true and PHY_INTERFACE_MODE_NA is returned. phylink_sfp_config_optical() then fails with "failed to select SFP interface". Before this patch, the interface was chosen by preference alone and the module worked. There is also a mismatch in phylink_sfp_set_config(). phylink_validate_pcs_inband_autoneg() checks the unmasked config.advertising, but the masked one is what gets programmed, and it may have Autoneg removed: linkmode_copy(pl->advertising, state->advertising); if (!pl->phydev) linkmode_andnot(state->advertising, pl->advertising, pl->user_disabled_advertising); The commit message only mentions keeping user rules across PCS removal and re-attach. Are these restrictions meant to persist across SFP module swaps and change how the interface is selected? [ ... ]