- 06 Feb, 2024 11 commits
-
-
Claudiu Beznea authored
ravb_set_gti() was computing the value of GTI based on the reference clock rate and then applied it to register. This was done on the driver's probe function. In order to implement runtime PM for all IP variants (as some IP variants switches to reset mode (and thus the registers content is lost) when module standby is configured through clock APIs) the GTI setup was split in 2 parts: one computing the value of the GTI register (done in the driver's probe function) and one applying the computed value to register (done in the driver's ndo_open API). Signed-off-by:
Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com> Reviewed-by:
Sergey Shtylyov <s.shtylyov@omp.ru> Signed-off-by:
Paolo Abeni <pabeni@redhat.com>
-
Claudiu Beznea authored
The runtime PM implementation will disable clocks at the end of ravb_probe(). As some IP variants switch to reset mode as a result of setting module standby through clock disable APIs, to implement runtime PM the resource parsing and requesting are moved in the probe function and IP settings are moved in the open function. This is done because at the end of the probe some IP variants will switch anyway to reset mode and the registers content is lost. Also keeping only register settings operations in the ravb_open()/ravb_close() functions will make them faster. Commit moves IRQ requests to ravb_probe() to have all the IRQs ready when the interface is open. As now getting/requesting IRQs is done in a single place there is no need to keep intermediary data (like ravb_rx_irqs[] and ravb_tx_irqs[] arrays or IRQs in struct ravb_private). In order to avoid accessing the IP registers while the IP is runtime suspended (e.g. in the timeframe b/w the probe requests shared IRQs and IP clocks are enabled) in the interrupt handlers were introduced pm_runtime_active() checks. The device runtime PM usage counter has been incremented to avoid disabling the device's clocks while the check is in progress (if any). This is a preparatory change to add runtime PM support for all IP variants. Signed-off-by:
-
Claudiu Beznea authored
Reference clock could be or not be part of the power domain. If it is part of the power domain, the power domain takes care of properly setting it. In case it is not part of the power domain and full runtime PM support is available in driver the clock will not be propertly disabled/enabled at runtime. For this, keep the prepare/unprepare operations in the driver's probe()/remove() functions and move the enable/disable in runtime PM functions. By doing this, the previous ravb_runtime_nop() function was renamed ravb_runtime_suspend() and the comment was removed. A proper runtime PM resume function was added (ravb_runtime_resume()). The current driver still don't need to make any register settings on runtime suspend/resume (as expressed in the removed comment) because, currently, pm_runtime_put_sync() is called on the driver remove function. This will be changed in the next commits (that extends the runtime PM support) such that proper register settings (along with runtime resume/suspend) will be done on ravb_open()/ravb_close(). Along with it, the other clock request operations were moved close to reference clock request and prepare to have all the clock requests specific code grouped together. Signed-off-by:
-
Claudiu Beznea authored
RZ/G3S can go to deep sleep states where power to most of the SoC parts is off. When resuming from such a state, the Ethernet controller needs to be reinitialized. De-asserting the reset signal for it should also be done. Thus, add reset assert/de-assert on suspend/resume functions. On the resume function, the de-assert was not reverted in case of failures to give the user a chance to restore the interface (e.g., bringing down/up the interface) in case suspend/resume failed. Signed-off-by:
-
Claudiu Beznea authored
Use tabs instead of spaces in the ravb_set_rate_gbeth() function. This aligns with the coding style requirements. Signed-off-by:
-
Claudiu Beznea authored
SET_SYSTEM_SLEEP_PM_OPS() and SET_RUNTIME_PM_OPS() are deprecated now and require __maybe_unused protection against unused function warnings. The usage of pm_ptr() and SYSTEM_SLEEP_PM_OPS()/RUNTIME_PM_OPS() allows the compiler to see the functions, thus suppressing the warning. Thus drop the __maybe_unused markings. Signed-off-by:
Geert Uytterhoeven <geert+renesas@glider.be> Signed-off-by:
-
Claudiu Beznea authored
On the RZ/G3S SoC the reset controller is mandatory for the IP to work. The device tree binding documentation for the ravb driver specifies that the resets are mandatory. Based on this, make the resets mandatory also in driver for all ravb devices. Signed-off-by:
-
Claudiu Beznea authored
ravb_rzv2m_hw_info::gptp_ref_clk is enabled only for RZ/V2M. RZ/V2M is an ARM64-based device which selects power domains by default and CONFIG_PM. The RZ/V2M Ethernet DT node has proper power-domain binding available in device tree from the commit that added the Ethernet node. (4872ca1f ("arm64: dts: renesas: r9a09g011: Add ethernet nodes")). Power domain support was available in the rzg2l-cpg.c driver when the Ethernet DT node has been enabled in RZ/V2M device tree. (ef3c613c ("clk: renesas: Add CPG core wrapper for RZ/G2L SoC")). Thus, remove the explicit clock enable for gptp_clk (and treat it as the other clocks are treated) as it is not needed and removing it doesn't break the ABI according to the above explanations. By removing the enable/disable operation from the driver we can add runtime PM support (which operates on clocks) w/o the need to handle the gptp_clk in the Ethernet driver functions like ravb_runtime_nop(). PM domain does all that is needed. Signed-off-by:
-
Claudiu Beznea authored
ravb_poll() initial code used to interrogate the first descriptor of the RX queue in case gPTP is false to determine if ravb_rx() should be called. This is done for non-gPTP IPs. For gPTP IPs the driver PTP-specific information was used to determine if receive function should be called. As every IP has its own receive function that interrogates the RX descriptors list in the same way the ravb_poll() was doing there is no need to double check this in ravb_poll(). Removing the code from ravb_poll() leads to a cleaner code. Signed-off-by:
-
Bo Liu authored
The maple tree register cache is based on a much more modern data structure than the rbtree cache and makes optimisation choices which are probably more appropriate for modern systems than those made by the rbtree cache. Signed-off-by:
Bo Liu <liubo03@inspur.com> Link: https://lore.kernel.org/r/20240202064336.39138-1-liubo03@inspur.comSigned-off-by:
-
Willem de Bruijn authored
This test is time sensitive. It may fail on virtual machines and for debug builds. Continue to run in these environments to get code coverage. But optionally suppress failure for timing errors (only). This is controlled with environment variable KSFT_MACHINE_SLOW. The test continues to return 0 (KSFT_PASS), rather than KSFT_XFAIL as previously discussed. Because making so_txtime.c return that and then making so_txtime.sh capture runs that pass that vs KSFT_FAIL and pass it on added a bunch of (fragile bash) boilerplate, while the result is interpreted the same as KSFT_PASS anyway. Signed-off-by:
Willem de Bruijn <willemb@google.com> Link: https://lore.kernel.org/r/20240201162130.2278240-1-willemdebruijn.kernel@gmail.comSigned-off-by:
-
- 05 Feb, 2024 15 commits
-
-
David Wei authored
Add a Makefile for netdevsim selftests and add selftests path to MAINTAINERS Signed-off-by:
David Wei <dw@davidwei.uk> Link: https://lore.kernel.org/r/20240130214620.3722189-5-dw@davidwei.ukSigned-off-by:
Jakub Kicinski <kuba@kernel.org>
-
David S. Miller authored
Vladimir Oltean says: ==================== Fixups for qca8k ds->user_mii_bus cleanup The series "ds->user_mii_bus cleanup (part 1)" from the last development cycle: https://patchwork.kernel.org/project/netdevbpf/cover/20240104140037.374166-1-vladimir.oltean@nxp.com/ had some review comments I didn't have the time to address at the time. One from Alvin and one from Luiz. They can reasonably be treated as improvements for v6.9. ==================== Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Vladimir Oltean authored
It was pointed out during the review [1] of commit 68e1010c ("net: dsa: qca8k: put MDIO bus OF node on qca8k_mdio_register() failure") that the rest of the qca8k driver uses "int ret" rather than "int err". Make everything consistent in that regard, not only qca8k_mdio_register(), but also qca8k_setup_mdio_bus(). [1] https://lore.kernel.org/netdev/qyl2w3ownx5q7363kqxib52j5htar4y6pkn7gen27rj45xr4on@pvy5agi6o2te/Suggested-by:
Alvin Šipraga <alsi@bang-olufsen.dk> Signed-off-by:
Vladimir Oltean <vladimir.oltean@nxp.com> Reviewed-by:
Andrew Lunn <andrew@lunn.ch> Reviewed-by:
Christian Marangi <ansuelsmth@gmail.com> Signed-off-by:
-
Vladimir Oltean authored
It was pointed out during the review [1] of commit e66bf63a ("net: dsa: qca8k: skip MDIO bus creation if its OF node has status = "disabled"") that we now leak a reference to the "mdio" OF node if it is disabled. This is only a concern when using dynamic OF as far as I can tell (like probing on an overlay), since OF nodes are never freed in the regular case. Additionally, I'm unaware of any actual device trees (in production or elsewhere) which have status = "disabled" for the MDIO OF node. So handling this as a simple enhancement. [1] https://lore.kernel.org/netdev/CAJq09z4--Ug+3FAmp=EimQ8HTQYOWOuVon-PUMGB5a1N=RPv4g@mail.gmail.com/Suggested-by:
Luiz Angelo Daros de Luca <luizluca@gmail.com> Signed-off-by:
-
Vladimir Oltean authored
These got misaligned after commit 6ca80638 ("net: dsa: Use conduit and user terms"). Signed-off-by:
Florian Fainelli <florian.fainelli@broadcom.com> Signed-off-by:
-
Breno Leitao authored
commit 1c870c63 ("net: fill in MODULE_DESCRIPTION()s for ocelot") got a suggestion from Vladimir Oltean after it had landed in net-next. Rewrite the module description according to Vladimir's suggestion. Fixes: 1c870c63 ("net: fill in MODULE_DESCRIPTION()s for ocelot") Suggested-by:
Breno Leitao <leitao@debian.org> Reviewed-by:
-
Breno Leitao authored
lib/test_blackhole_dev.c sets a variable that is never read, causing this following building warning: lib/test_blackhole_dev.c:32:17: warning: variable 'ethh' set but not used [-Wunused-but-set-variable] Remove the variable struct ethhdr *ethh, which is unused. Fixes: 509e56b3 ("blackhole_dev: add a selftest") Signed-off-by:
Jiri Pirko <jiri@nvidia.com> Signed-off-by:
-
David S. Miller authored
Matthieu Baerts says: ==================== mptcp: annotate lockless access This is a series of 5 patches from Paolo to annotate lockless access. The MPTCP locking schema is already quite complex. We need to clarify it and make the lockless access already there consistent, or later changes will be even harder to follow and understand. This series goes through all the msk fields accessed in the RX and TX path and makes the lockless annotation consistent with the in-use locking schema. As a bonus, this should fix data races eventually found by fuzzers -- even if we haven't seen many such reports so far. Patch 1/5 hints we could remove "local_key" and "remote_key" from the subflow context, and always use the ones from the msk socket, possibly reducing the context memory usage. That change is left over as a possible follow-up. ==================== Signed-off-by:
Matthieu Baerts (NGI0) <matttbe@kernel.org> Signed-off-by:
-
Paolo Abeni authored
The following MPTCP socket fields: - can_ack - fully_established - rcv_data_fin - snd_data_fin_enable - rcv_fastclose - use_64bit_ack are accessed without any lock, add the appropriate annotation. The schema is safe as each field can change its value at most once in the whole mptcp socket life cycle. Signed-off-by:
Mat Martineau <martineau@kernel.org> Signed-off-by:
-
Paolo Abeni authored
The token field is manipulated under the msk socket lock and accessed lockless in a few spots, add proper ONCE annotation Signed-off-by:
-
Paolo Abeni authored
The following fields: - ack_seq - snd_una - wnd_end - rmem_fwd_alloc are protected by the data lock end accessed lockless in a few spots. Ensure ONCE annotation for write (under such lock) and for lockless read. Signed-off-by:
-
Paolo Abeni authored
The mptcp-level TX path info (write_seq, bytes_sent, snd_nxt) are under the msk socket lock protection, and are accessed lockless in a few spots. Always mark the write operations with WRITE_ONCE, read operations outside the lock with READ_ONCE and drop the annotation for read under such lock. To simplify the annotations move mptcp_pending_data_fin_ack() from __mptcp_data_acked() to __mptcp_clean_una(), under the msk socket lock, where such call would belong. Signed-off-by:
-
Paolo Abeni authored
Both the local and the remote key follow the same locking schema, put in place the proper ONCE accessors. Signed-off-by:
-
Gerhard Engleder authored
Similar chunk of code is used in tsnep_rx_poll_zc() and tsnep_rx_reopen_xsk() to maintain the RX XDP_RING_NEED_WAKEUP flag. Consolidate the code to common helper function. Suggested-by:
Gerhard Engleder <gerhard@engleder-embedded.com> Reviewed-by:
Simon Horman <horms@kernel.org> Signed-off-by:
-
Eric Dumazet authored
We can change sctp_sk() to propagate its argument const qualifier, thanks to container_of_const(). Signed-off-by:
Eric Dumazet <edumazet@google.com> Cc: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com> Cc: Xin Long <lucien.xin@gmail.com> Acked-by:
Xin Long <lucien.xin@gmail.com> Signed-off-by:
-
- 04 Feb, 2024 6 commits
-
-
Eric Dumazet authored
We can use a global dev_unreg_count counter instead of a per netns one. As a bonus we can factorize the changes done on it for bulk device removals. Signed-off-by:
-
Yunjian Wang authored
Implement the tun .get_channels functionality. This feature is necessary for some tools, such as libxdp, which need to retrieve the queue count. Signed-off-by:
Yunjian Wang <wangyunjian@huawei.com> Reviewed-by:
Jason Wang <jasowang@redhat.com> Signed-off-by:
-
Yunjian Wang authored
This fixes the following code style problem: - WARNING: please, no spaces at the start of a line - CHECK: Please use a blank line after function/struct/union/enum declarations Signed-off-by: -
Heiner Kallweit authored
This adds support for the RTL8126A found on Asus z790 Maximus Formula. It was successfully tested w/o the firmware at 1000Mbps. Firmware file has been provided by Realtek and submitted to linux-firmware. 2.5G and 5G modes are untested. Signed-off-by:
Heiner Kallweit <hkallweit1@gmail.com> Signed-off-by:
-
Jakub Kicinski authored
Paolo points out that ifconfig is legacy and we should not use it. Signed-off-by:
-
Horatiu Vultur authored
By default lan8841's 1588 clock frequency is 125MHz. But when adjusting the frequency, it is using the 1PPM format of the lan8814. Which is the wrong format as lan8814 has a 1588 clock frequency of 250MHz. So then for each 1PPM adjustment would adjust less than expected. Therefore fix this by using the correct 1PPM format for lan8841. Signed-off-by:
Horatiu Vultur <horatiu.vultur@microchip.com> Reviewed-by:
-
- 03 Feb, 2024 8 commits
-
-
David S. Miller authored
Christian Marangi says: ==================== net: phy: qcom: qca808x: fixup qca808x LED This is a bit embarassing and totally my fault so sorry for that! While reworking the patch to phy_modify API, it was done a logic error and made the brightness_set function broken. It wasn't notice in last revisions test as the testing method was to verify if hw control was correctly working. Noticing this problem also made me notice an additional problem with the polarity. The introduced patch made the polarity configurable but I forgot to add the required code to enable Active High by default. (the PHY sets active low by default) This wasn't notice with hw control testing as the LED blink on traffic and polarity problem are not notice. It might be worth discussing if needed a change in implementation where the polarity function is always called but I think it's better this way where specific PHY apply fixup with the help of priv struct and on the config_init phase. ==================== Signed-off-by: -
Christian Marangi authored
qca808x PHY provide support for the led_polarity_set OP to configure and apply the active-low property but on PHY reset, the Active High bit is not set resulting in the LED driven as active-low. To fix this, check if active-low is not set in DT and enable Active High polarity by default to restore correct funcionality of the LED. Fixes: 7196062b ("net: phy: at803x: add LED support for qca808x") Signed-off-by:
-
Christian Marangi authored
In switching to using phy_modify_mmd and a more short version of the LED ON/OFF condition in later revision, it was made a logic error where value ? QCA808X_LED_FORCE_ON : QCA808X_LED_FORCE_OFF is always true as value is always OR with QCA808X_LED_FORCE_EN due to missing () resulting in the testing condition being QCA808X_LED_FORCE_EN | value. Add the () to apply the correct condition and restore correct functionality of the brightness ON/OFF. Fixes: 7196062b ("net: phy: at803x: add LED support for qca808x") Signed-off-by:
-
Jakub Kicinski authored
Jakub Kicinski says: ==================== tools: ynl: auto-gen for all genetlink families The code gen has caught up with all features required in genetlink families in Linux 6.8 already. We have also stopped committing auto- -generated user space code to the tree. Instead of listing all the families in the Makefile search the spec directory, and generate code for everything that's not legacy netlink. ==================== Link: https://lore.kernel.org/r/20240202004926.447803-1-kuba@kernel.orgSigned-off-by:
-
Jakub Kicinski authored
Instead of listing the genetlink families that we want to codegen for, always codegen for everyone. We can add an opt-out later but it seems like most families are not causing any issues, and yet folks forget to add them to the Makefile. Reviewed-by:
-
Jakub Kicinski authored
Add ovs_flow, ovs_vport and ovs_datapath to the families supported in C. ovs-flow has some circular nesting which is fun to deal with, but the necessary support has been added already in the previous release cycle. Add a sample that proves that dealing with fixed headers does actually work correctly. Reviewed-by:
-
Jakub Kicinski authored
The DPLL and mptcp_pm families are pretty clean, and YNL C codegen supports them fully with no changes. Add them to user space codegen so that C samples can be written, and we know immediately if changes to these families require YNL codegen work. Reviewed-by:
-
Eric Dumazet authored
Both addrconf_verify_work() and addrconf_dad_work() acquire rtnl, there is no point trying to have one thread per cpu. Signed-off-by:
David Ahern <dsahern@kernel.org> Link: https://lore.kernel.org/r/20240201173031.3654257-1-edumazet@google.comSigned-off-by:
-
