net: if_arp: add ARPHRD_PUREIP type
Subsystems (7)
PATCH 1/4 net: if_arp: add ARPHRD_PUREIP type
This patch add the definition of ARPHRD_PUREIP which can for example be used by mobile ccmni device as device type. ARPHRD_PUREIP means that this device doesn't need kernel to generate ipv6 link-local address in any addr_gen_mode.
--- include/uapi/linux/if_arp.h | 1 + 1 file changed, 1 insertion(+)
include/uapi/linux/if_arp.h +1 -0
@@ -61,6 +61,7 @@#define ARPHRD_DDCMP 517/* Digital's DDCMP protocol */ #define ARPHRD_RAWHDLC 518/* Raw HDLC */ #define ARPHRD_RAWIP 519/* Raw IP */ +#define ARPHRD_PUREIP 520/* Pure IP */ #define ARPHRD_TUNNEL 768/* IPIP tunnel */ #define ARPHRD_TUNNEL6 769/* IP6IP6 tunnel */
Signature · 1 line
2.18.0
PATCH 2/4 net: ipv6: don't generate link local address on PUREIP device
PUREIP device such as ccmni does not need kernel to generate ipv6 link-local address in any addr_gen_mode, generally, it shall use the IPv6 Interface Identifier, as provided by the GGSN, to create its IPv6 link-ocal Unicast Address.
--- net/ipv6/addrconf.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
net/ipv6/addrconf.c +8 -1
@@ -3339,7 +3339,8 @@ static void addrconf_dev_config(struct net_device *dev) (dev->type != ARPHRD_IPGRE) && (dev->type != ARPHRD_TUNNEL) && (dev->type != ARPHRD_NONE) && - (dev->type != ARPHRD_RAWIP)) { + (dev->type != ARPHRD_RAWIP) && + (dev->type != ARPHRD_PUREIP)) {/* Alas, we support only Ethernet autoconfiguration. */ idev =__in6_dev_get (dev); if (!IS_ERR_OR_NULL(idev) && dev->flags & IFF_UP &&
@@ -3352,6 +3353,12 @@ static void addrconf_dev_config(struct net_device *dev)if (IS_ERR (idev))return ; +/* this device type doesn't need to generate +* link-local address in any addr_gen_mode +*/ +if (dev->type ==ARPHRD_PUREIP ) +return ; +/* this device type has no EUI support */ if (dev->type ==ARPHRD_NONE && idev->cnf.addr_gen_mode ==IN6_ADDR_GEN_MODE_EUI64 )
Signature · 1 line
2.18.0
PATCH 3/4 net: dev_is_mac_header_xmit() return false for ARPHRD_PUREIP
In reply to Rocco Yuethe incoming/outgoing packets on that pure ip interface do not have an ethernet header. So adding ARPHRD_PUREIP to the dev_is_mac_header_xmit() make the function return false, and __bpf_redirect() checks this boolean to determine whether to prefix an ethernet header.
--- include/linux/if_arp.h | 1 + 1 file changed, 1 insertion(+)
include/linux/if_arp.h +1 -0
@@ -51,6 +51,7 @@ static inline bool dev_is_mac_header_xmit(const struct net_device *dev)case ARPHRD_VOID :case ARPHRD_NONE :case ARPHRD_RAWIP : +case ARPHRD_PUREIP :return false;default :return true;
Signature · 1 line
2.18.0
PATCH 4/4 drivers: net: mediatek: initial implementation of ccmni
In reply to Rocco Yueccmni driver is used by all recent chipsets using MediaTek Inc. modems. It provides a bridge to shield hardware differences and connect kernel netdevice. Module provides virtual network devices which can be used to support different mobile networks, such as a default cellular internet,or IMS network.
--- .../device_drivers/cellular/ccmni/ccmni.rst | 151 +++++++++ MAINTAINERS | 9 + drivers/net/ethernet/mediatek/Kconfig | 5 + drivers/net/ethernet/mediatek/Makefile | 4 +- drivers/net/ethernet/mediatek/ccmni/Kconfig | 15 + drivers/net/ethernet/mediatek/ccmni/Makefile | 7 + drivers/net/ethernet/mediatek/ccmni/ccmni.c | 291 ++++++++++++++++++ drivers/net/ethernet/mediatek/ccmni/ccmni.h | 53 ++++ 8 files changed, 534 insertions(+), 1 deletion(-) create mode 100644 Documentation/networking/device_drivers/cellular/ccmni/ccmni.rst create mode 100644 drivers/net/ethernet/mediatek/ccmni/Kconfig create mode 100644 drivers/net/ethernet/mediatek/ccmni/Makefile create mode 100644 drivers/net/ethernet/mediatek/ccmni/ccmni.c create mode 100644 drivers/net/ethernet/mediatek/ccmni/ccmni.h
Documentation/networking/device_drivers/cellular/ccmni/ccmni.rst new file +151 -0
@@ -0,0 +1,151 @@ +.. SPDX-License-Identifier: GPL-2.0+ + +=================================================================== +Linux kernel driver for Cross Core Multi Network Interface (ccmni): +=================================================================== + + +1. Introduction +=============== + +ccmni driver is built on the top of AP-CCCI (Application Processor Cross +Core Communite Interface) to emulate the modem’s data networking service +as a network interface card. This driver is used by all recent chipsets +using MediaTek Inc. modems. + +This driver can support multiple tx/rx channels, each channel has its +own IP address, and ipv4 and ipv6 link-local addresses on the interface +is set through ioctl, which means that ccmni driver is a pure ip device +and it does not need the kernel to generate these ip addresses +automatically. + +Creating logical network devices (ccmni devices) can be used to handle +multiple private data networks (PDNs), such as a defaule cellular +internet, IP Multimedia Subsystem (IMS) network (VoWiFi/VoLTE), IMS +Emergency, Tethering, Multimedia Messaging Service (MMS), and so on. +In general, one ccmni device corresponds to one PDN. + +ccmni design +------------ + +- AT commands and responses between the Application Processor and Modem + Processor take place over the ccci tty serial port emulation. +- IP packets coming into the ccmni driver (from the Modem) needs to be + un-framed (via the Packet Framing Protocol) before it can be stuffed + into the Socket Buffer. +- IP Packets going out of the ccmni driver (to the Modem) needs to be + framed (via the Packet Framing Protocol) before it can be pushed into + the respective ccci tx buffer. +- The Packet Framing Protocol module contains algorithms to correctly + add or remove frames, going out to and coming into, the ccmni driver + respectively. +- Packets are passed to, and received from the Linux networking system, + via Socket Buffers. +- The Android Application Framework contains codes to setup the ccmni’s + parameters (such as netmask), and the routing table. + + +2. Architecture +=============== + + +------------------------+ +---------------------+ +user | MTK RilD | | network process | +space | (config/up/down ccmni) | | (send/recv packets) | + +------------------------+ +---------------------+ ++--------------------------------------------------------------------+ + +--------------------------------------------------+ + | socket | + +--------------------------------------------------+ + +---------------------+ + | TCP/IP stack | + +---------------------+ + +--------------------------------------------------+ + | net device layer | + +--------------------------------------------------+ + +--------+ +--------+ +--------+ +kernel | ccmni0 | | ccmni1 | ... | ccmnix | +space +--------+ +--------+ +--------+ + +-------------------------------------------------+ + | ccmni driver | + +-------------------------------------------------+ + +-------------------------------------------------+ + | AP-CCCI | + +-------------------------------------------------+ ++---------------------------------------------------------------------+ + +-------------------------------------------------+ + | +-------------------+ +-------------------+ | + | | DPMAIF hardware | | MD-CCCI | | + | +-------------------+ +-------------------+ | + | ... ... | + | Modem Processor | + +-------------------------------------------------+ + + +3. Driver information and notes +=============================== + +Data Connection Set Up +---------------------- + +The data framework will first SetupDataCall with passing ccmni index, +and then RilD will activate PDN connection and get CID (Connection ID). + +Next, RilD will creat ccmni socket to use ioctl to configure ccmni up, +and then ccmni_open() will be called. + +In addition, since ccmni is a pureip device, RilD needs to use ioctl to +configure the ipv4/ipv6-link-local address for ccmni after it is up. + +Alternatively, you can use the ip command as follows:: + ip link set up dev ccmni<x> + ip addr add a.b.c.d dev ccmni<x> + ip -6 addr add fe80::1/64 dev ccmni<x> + +Data Connection Set Down +------------------------ + +The data service implements a method to tear down the data connection, +after RilD deactivate the PDN connection, RilD will down the specific +interface of ccmnix through ioctl SIOCSIFFLAGS, and then ccmni_close() +will be called. After that, if any network process (such as browser) wants +to write data to ccmni socket, TCP/IP stack will return an error to +this socket. + +Data Transmit +------------- + +In the uplink direction, when there is data to be transmitted to the cellular +network, ccmni_start_xmit() will be called by the Linux networking system. + +main operations in ccmni_start_xmit(): +- the datagram to be transmitted is housed in the Socket Buffer. +- check if the datagram is within limits (i.e. 1500 bytes) acceptable by the + Modem. If the datagram exceeds limit, the datagram will be dropped, and free + the Socket Buffer. +- check if the AP-CCCI TX buffer is busy, or do not have enough space for this + datagram. If it is busy, or the free space is too small, ccmni_start_xmit() + return NETDEV_TX_BUSY and ask the Linux netdevice to stop the tx queue. + +To handle outcoming datagrams to the Modem, ccmni register a callback function +for AP-CCCI driver. ccmni_hif_hook() means ccci can implement specific egress +function to send these packets to the specific hardware. + +Data Receive +------------ + +In the downlink direction, DPMAIF (Data Path MD AP Interface) hardware sends +packets and messages with channel id matching these packets to AP-CCCI driver. + +To handle incoming datagrams from the Modem, ccmni register a callback +function for the AP-CCCI driver. ccmni_rx_push() responsible for extracting +the incoming packets from the ccci rx buffer, and updating skb. Once ready, +ccmni signal to the Linux networking system to take out Socket Buffer +(via netif_rx() / netif_rx_ni()). + + +Support +======= + +If an issue is identified by published source code and supported adapter on +the supported kernel, please email the specific information about the issue +to [email protected] and [email protected]
MAINTAINERS +9 -0
@@ -11612,6 +11612,15 @@ F: Documentation/devicetree/bindings/usb/mediatek,* F: drivers/usb/host/xhci-mtk* F: drivers/usb/mtu3/ +MEDIATEK CCMNI DRIVER +M: Rocco Yue <[email protected]> +M: Chao Song <[email protected]> +M: Zhuoliang Zhang <[email protected]> +L: [email protected] +S: Maintained +F: Documentation/networking/device_drivers/cellular/mediatek/ccmni.rst +F: drivers/net/ethernet/mediatek/ccmni/ + MEGACHIPS STDPXXXX-GE-B850V3-FW LVDS/DP++ BRIDGES M: Peter Senna Tschudin <[email protected]> M: Martin Donnelly <[email protected]>
drivers/net/ethernet/mediatek/Kconfig +5 -0
@@ -1,4 +1,7 @@ # SPDX-License-Identifier: GPL-2.0-only +# +# MediaTek network device configuration +# config NET_VENDOR_MEDIATEK bool "MediaTek devices" depends on ARCH_MEDIATEK || SOC_MT7621 || SOC_MT7620
@@ -24,4 +27,6 @@ config NET_MEDIATEK_STAR_EMAC This driver supports the ethernet MAC IP first used on MediaTek MT85** SoCs. +source "drivers/net/ethernet/mediatek/ccmni/Kconfig" + endif #NET_VENDOR_MEDIATEK
drivers/net/ethernet/mediatek/Makefile +3 -1
@@ -1,8 +1,10 @@ # SPDX-License-Identifier: GPL-2.0-only # -# Makefile for the Mediatek SoCs built-in ethernet macs +# Makefile for the Mediatek network device drivers. # obj-$(CONFIG_NET_MEDIATEK_SOC) += mtk_eth.o mtk_eth-y := mtk_eth_soc.o mtk_sgmii.o mtk_eth_path.o mtk_ppe.o mtk_ppe_debugfs.o mtk_ppe_offload.o obj-$(CONFIG_NET_MEDIATEK_STAR_EMAC) += mtk_star_emac.o + +obj-$(CONFIG_MTK_NET_CCMNI) += ccmni/
drivers/net/ethernet/mediatek/ccmni/Kconfig new file +15 -0
@@ -0,0 +1,15 @@ +# SPDX-License-Identifier: GPL-2.0 +# +# MediaTek CCMNI driver +# + +menuconfig MTK_NET_CCMNI + tristate "MediaTek CCMNI driver" + default n + help + Cross Core Multi Network Interface: + If you select this, you will enable the CCMNI module which is used + to shield hardware differences and communicate with kernel netdevice + and be used for handling outgoing/incoimg mobile data. Module provides + virtual network devices which can be used to support different mobile + networks.
drivers/net/ethernet/mediatek/ccmni/Makefile new file +7 -0
@@ -0,0 +1,7 @@ +# SPDX-License-Identifier: GPL-2.0 +# +# Makefile for the ccmni module +# +ccflags-y += -I$(srctree)/drivers/net/ethernet/mediatek/ccmni/ +obj-$(CONFIG_MTK_NET_CCMNI) += ccmni.o +
drivers/net/ethernet/mediatek/ccmni/ccmni.c new file +291 -0
291 lines not on this page · raw message
drivers/net/ethernet/mediatek/ccmni/ccmni.h new file +53 -0
53 lines not on this page · raw message
Signature · 1 line
2.18.0
Re: [PATCH 4/4] drivers: net: mediatek: initial implementation of ccmni
Quoted 1 line · Rocco Yue
+EXPORT_SYMBOL(ccmni_rx_push);
Why are you exporting symbols that are not used by anyone in this patch series? That doesn't feel right. Who needs this?
Quoted 1 line
+EXPORT_SYMBOL(ccmni_hif_hook);
Same with this, who calls this?
Quoted 1 line
+++
b/drivers/net/ethernet/mediatek/ccmni/ccmni.h
Why do you have a .h file for a single .c file? that shouldn't be needed.
thanks,
greg k-h
kernel test robot
build botRe: [PATCH 4/4] drivers: net: mediatek: initial implementation of ccmni
In reply to Rocco YueI love your patch! Perhaps something to improve:
Hi Rocco,
I love your patch! Perhaps something to improve:
[auto build test WARNING on net-next/master]
[also build test WARNING on net/master linus/master sparc-next/master v5.13-rc7 next-20210623] [If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]
url: https://github.com/0day-ci/linux/commits/Rocco-Yue/net-if_arp-add-ARPHRD_PUREIP-type/20210623-195120 base: https://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git 38f75922a6905b010f597fc70dbb5db28398728e config: arm64-allyesconfig (attached as .config) compiler: aarch64-linux-gcc (GCC) 9.3.0
reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# https://github.com/0day-ci/linux/commit/ed82e2a7a56fce99884075bbf84bc76195971ddc
git remote add linux-review https://github.com/0day-ci/linux
git fetch --no-tags linux-review Rocco-Yue/net-if_arp-add-ARPHRD_PUREIP-type/20210623-195120 git checkout ed82e2a7a56fce99884075bbf84bc76195971ddc
# save the attached .config to linux build tree
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-9.3.0 make.cross ARCH=arm64
If you fix the issue, kindly add following tag as appropriate
All warnings (new ones prefixed by >>):
drivers/net/ethernet/mediatek/ccmni/ccmni.c: In function 'ccmni_exit':
>> drivers/net/ethernet/mediatek/ccmni/ccmni.c:190:26: warning: variable 'ctlb' set but not used [-Wunused-but-set-variable]
190 | struct ccmni_ctl_block *ctlb = NULL;
| ^~~~vim +/ctlb +190 drivers/net/ethernet/mediatek/ccmni/ccmni.c
187
188 static void __exit ccmni_exit(void)
189 {
> 190 struct ccmni_ctl_block *ctlb = NULL;
191 struct ccmni_inst *ccmni = NULL;
192 int i;
193
194 ctlb = s_ccmni_ctlb;
195 if (!s_ccmni_ctlb)
196 return;
197 for (i = 0; i < s_ccmni_ctlb->max_num; i++) {
198 ccmni = s_ccmni_ctlb->ccmni_inst[i];
199 if (ccmni) {
200 unregister_netdev(ccmni->dev);
201 s_ccmni_ctlb->ccmni_inst[i] = NULL;
202 }
203 }
204 kfree(s_ccmni_ctlb);
205 s_ccmni_ctlb = NULL;
206 }
207
---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org
Re: [PATCH 4/4] drivers: net: mediatek: initial implementation of ccmni
In reply to Rocco Yuedrivers/net/ethernet/mediatek/ccmni/ccmni.c line 30
Quoted 11 lines · Rocco Yue
drivers/net/ethernet/mediatek/ccmni/ccmni.c new file +11
-0+static int ccmni_open (struct net_device *ccmni_dev) +{ +struct ccmni_inst *ccmni =netdev_priv (ccmni_dev); + +netif_tx_start_all_queues (ccmni_dev); +netif_carrier_on (ccmni_dev); + +if (atomic_inc_return (&ccmni->usage) >1 ) { +atomic_dec (&ccmni->usage); +netdev_err (ccmni_dev,"dev already open\n" ); +return -EINVAL ;
You only check this _AFTER_ starting up? If so, why even check a count at all? Why does it matter as it's not keeping anything from working here.
Quoted 57 lines
drivers/net/ethernet/mediatek/ccmni/ccmni.c new file +57
-0+ } + +return 0 ; +} + +static int ccmni_close (struct net_device *ccmni_dev) +{ +struct ccmni_inst *ccmni =netdev_priv (ccmni_dev); + +atomic_dec (&ccmni->usage); +netif_tx_disable (ccmni_dev); + +return 0 ; +} + +static netdev_tx_t +ccmni_start_xmit (struct sk_buff *skb,struct net_device *ccmni_dev) +{ +struct ccmni_inst *ccmni =NULL ; + +if (unlikely (!ccmni_hook_ready)) + goto tx_ok; + +if (!skb || !ccmni_dev) + goto tx_ok; + + ccmni =netdev_priv (ccmni_dev); + +/* some process can modify ccmni_dev->mtu */ +if (skb->len > ccmni_dev->mtu) { +netdev_err (ccmni_dev,"xmit fail: len(0x%x) > MTU(0x%x, 0x%x)" , + skb->len,CCMNI_MTU , ccmni_dev->mtu); + goto tx_ok; + } + +/* hardware driver send packet will return a negative value +* ask the Linux netdevice to stop the tx queue +*/ +if ((s_ccmni_ctlb->xmit_pkt (ccmni->index, skb,0 )) <0 ) +return NETDEV_TX_BUSY ; + +return NETDEV_TX_OK ; +tx_ok: +dev_kfree_skb (skb); + ccmni_dev->stats.tx_dropped++; +return NETDEV_TX_OK ; +} + +static int ccmni_change_mtu (struct net_device *ccmni_dev,int new_mtu) +{ +if (new_mtu <0 || new_mtu >CCMNI_MTU ) +return -EINVAL ; + +if (unlikely (!ccmni_dev)) +return -EINVAL ; + + ccmni_dev->mtu = new_mtu;
drivers/net/ethernet/mediatek/ccmni/ccmni.c line 97
Quoted 10 lines
drivers/net/ethernet/mediatek/ccmni/ccmni.c new file +10
-0+return 0 ; +} + +static void ccmni_tx_timeout (struct net_device *ccmni_dev,unsigned int txqueue) +{ +struct ccmni_inst *ccmni =netdev_priv (ccmni_dev); + + ccmni_dev->stats.tx_errors++; +if (atomic_read (&ccmni->usage) >0 ) +netif_tx_wake_all_queues (ccmni_dev);
Why does it matter what the reference count is? What happens if it drops _RIGHT_ after testing for it?
Anytime you do an atomic_read() call, it's almost always a sign that the logic is not correct.
Again, why have this reference count at all? What is it protecting?
drivers/net/ethernet/mediatek/ccmni/ccmni.c line 212
Quoted 5 lines
drivers/net/ethernet/mediatek/ccmni/ccmni.c new file +5
-0+/* exposed API +* receive incoming datagrams from the Modem and push them to the +* kernel networking system +*/ +int ccmni_rx_push (unsigned int ccmni_idx,struct sk_buff *skb)
Ah, so this driver doesn't really do anything on its own, as there is no modem driver for it.
So without a modem driver, it will never be used? Please submit the modem driver at the same time, otherwise it's impossible to review this correctly.
thanks,
greg k-h
On Wed, 2021-06-23 at 19:31 +0200, Greg KH wrote:
Quoted 7 lines · Rocco Yue
Quoted 7 lines
drivers/net/ethernet/mediatek/ccmni/ccmni.c new file +7
-0+static int ccmni_open (struct net_device *ccmni_dev) +{ +struct ccmni_inst *ccmni =netdev_priv (ccmni_dev); + +netif_tx_start_all_queues (ccmni_dev); +netif_carrier_on (ccmni_dev); +
Quoted 9 lines
drivers/net/ethernet/mediatek/ccmni/ccmni.c line 30
Quoted 4 lines
drivers/net/ethernet/mediatek/ccmni/ccmni.c new file +4
-0+if (atomic_inc_return (&ccmni->usage) >1 ) { +atomic_dec (&ccmni->usage); +netdev_err (ccmni_dev,"dev already open\n" ); +return -EINVAL ;You only check this _AFTER_ starting up? If so, why even check a count at all? Why does it matter as it's not keeping anything from working here.
Thanks for your review.
Looking back at this code block, it does have some ploblems, ccmni->usage hasn't been used to protect some resources or do some specific things in the current code, I will delete them.
Quoted 72 lines
Quoted 58 lines
drivers/net/ethernet/mediatek/ccmni/ccmni.c new file +57
-0+ } + +return 0 ; +} + +static int ccmni_close (struct net_device *ccmni_dev) +{ +struct ccmni_inst *ccmni =netdev_priv (ccmni_dev); + +atomic_dec (&ccmni->usage); +netif_tx_disable (ccmni_dev); + +return 0 ; +} + +static netdev_tx_t +ccmni_start_xmit (struct sk_buff *skb,struct net_device *ccmni_dev) +{ +struct ccmni_inst *ccmni =NULL ; + +if (unlikely (!ccmni_hook_ready)) + goto tx_ok; + +if (!skb || !ccmni_dev) + goto tx_ok; + + ccmni =netdev_priv (ccmni_dev); + +/* some process can modify ccmni_dev->mtu */ +if (skb->len > ccmni_dev->mtu) { +netdev_err (ccmni_dev,"xmit fail: len(0x%x) > MTU(0x%x, 0x%x)" , + skb->len,CCMNI_MTU , ccmni_dev->mtu); + goto tx_ok; + } + +/* hardware driver send packet will return a negative value +* ask the Linux netdevice to stop the tx queue +*/ +if ((s_ccmni_ctlb->xmit_pkt (ccmni->index, skb,0 )) <0 ) +return NETDEV_TX_BUSY ; + +return NETDEV_TX_OK ; +tx_ok: +dev_kfree_skb (skb); + ccmni_dev->stats.tx_dropped++; +return NETDEV_TX_OK ; +} + +static int ccmni_change_mtu (struct net_device *ccmni_dev,int new_mtu) +{ +if (new_mtu <0 || new_mtu >CCMNI_MTU ) +return -EINVAL ; + +if (unlikely (!ccmni_dev)) +return -EINVAL ; + + ccmni_dev->mtu = new_mtu;drivers/net/ethernet/mediatek/ccmni/ccmni.c line 97
Quoted 10 lines
drivers/net/ethernet/mediatek/ccmni/ccmni.c new file +10
-0+return 0 ; +} + +static void ccmni_tx_timeout (struct net_device *ccmni_dev,unsigned int txqueue) +{ +struct ccmni_inst *ccmni =netdev_priv (ccmni_dev); + + ccmni_dev->stats.tx_errors++; +if (atomic_read (&ccmni->usage) >0 ) +netif_tx_wake_all_queues (ccmni_dev);Why does it matter what the reference count is? What happens if it drops _RIGHT_ after testing for it?
Quoted 5 lines
Anytime you do an
atomic_read()call, it's almost always a sign that the logic is not correct.Again, why have this reference count at all? What is it protecting?
The jedgment of ccmni->usage here is to ensure that the ccmnix interface is already up when do wake up tx queue behavior.
Then I re-read the kernel code, I think my previous ider should be wrong. the reason is that before calling ccmni_tx_timeout(), it will check whether the dev exist or not, for example, it will be checked in dev_watchdog().
I can delete this code.
Quoted 9 lines
drivers/net/ethernet/mediatek/ccmni/ccmni.c line 212
Quoted 5 lines
drivers/net/ethernet/mediatek/ccmni/ccmni.c new file +5
-0+/* exposed API +* receive incoming datagrams from the Modem and push them to the +* kernel networking system +*/ +int ccmni_rx_push (unsigned int ccmni_idx,struct sk_buff *skb)Ah, so this driver doesn't really do anything on its own, as there is no modem driver for it.
Quoted 4 lines
So without a modem driver, it will never be used? Please submit the modem driver at the same time, otherwise it's impossible to review this correctly.
without MTK ap ccci driver (modem driver), ccmni_rx_push() and ccmni_hif_hook() are not be used.
Both of them are exported as symbols because MTK ap ccci driver will be complied to the ccci.ko file.
In current codes, I implementated the basic functionality of ccmni, such as open, close, xmit packet, rcv packet. And my original intention was that I can gradually complete some of the more functions of ccmni on this basis, such as sw-gro, napi, or meet the requirement of high throughput performance.
In addition, the code of MTK's modem driver is a bit complicated, because this part has more than 30,000 lines of code and contains more than 10 modules. We are completeing the upload of this huge code step by step. Our original intention was to upload the ccmni driver that directly interacts with the kernel first, and then complete the code from ccmni to the bottom layer one by one from top to bottom. We expect the completion period to be about 1 year.
Quoted 4 lines
+++
b/drivers/net/ethernet/mediatek/ccmni/ccmni.hWhy do you have a .h file for a single .c file? that shouldn't be needed.
I add a .h file to facilitate subsequent code expansion. If it's not appropriate to do this here, I can add the content of .h into .c file.
Thanks,
Rocco
Quoted 14 lines · Rocco Yue
Quoted 9 lines
drivers/net/ethernet/mediatek/ccmni/ccmni.c line 212
Quoted 5 lines
drivers/net/ethernet/mediatek/ccmni/ccmni.c new file +5
-0+/* exposed API +* receive incoming datagrams from the Modem and push them to the +* kernel networking system +*/ +int ccmni_rx_push (unsigned int ccmni_idx,struct sk_buff *skb)Ah, so this driver doesn't really do anything on its own, as there is no modem driver for it.
Quoted 4 lines
So without a modem driver, it will never be used? Please submit the modem driver at the same time, otherwise it's impossible to review this correctly.
Quoted 5 lines
without MTK ap ccci driver (modem driver),
ccmni_rx_push()andccmni_hif_hook()are not be used.Both of them are exported as symbols because MTK ap ccci driver will be complied to the ccci.ko file.
But I do not see any code in this series that use these symbols. We can not have exports that no one uses. Please add the driver to this patch series when you resend it.
Quoted 7 lines
In addition, the code of MTK's modem driver is a bit complicated, because this part has more than 30,000 lines of code and contains more than 10 modules. We are completeing the upload of this huge code step by step. Our original intention was to upload the ccmni driver that directly interacts with the kernel first, and then complete the code from ccmni to the bottom layer one by one from top to bottom. We expect the completion period to be about 1 year.
Again, we can not add code to the kernel that is not used, sorry. That would not make any sense, would you want to maintain such a thing?
And 30k of code seems a bit excesive for a modem driver. Vendors find
that when they submit code for inclusion in the kernel tree, in the end, they end up 1/3 the original size, so 10k is reasonable.
I can also take any drivers today into the drivers/staging/ tree, and you can do the cleanups there as well as getting help from others.
1 year seems like a long time to do "cleanup", good luck!
Quoted 8 lines
Quoted 4 lines
+++
b/drivers/net/ethernet/mediatek/ccmni/ccmni.hWhy do you have a .h file for a single .c file? that shouldn't be needed.
I add a .h file to facilitate subsequent code expansion. If it's not appropriate to do this here, I can add the content of .h into .c file.
If nothing other than a single .c file needs it, put it into that .c file please.
thanks,
greg k-h
Re: [PATCH 4/4] drivers: net: mediatek: initial implementation of ccmni
On Thu, 2021-06-24 at 14:23 +0200, Greg KH wrote:
Quoted 8 lines · Rocco Yue
Quoted 6 lines
without MTK ap ccci driver (modem driver),
ccmni_rx_push()andccmni_hif_hook()are not be used.Both of them are exported as symbols because MTK ap ccci driver will be compiled to the ccci.ko file.
But I do not see any code in this series that use these symbols. We can
will delete these symbols.
Quoted 3 lines
not have exports that no one uses. Please add the driver to this patch series when you resend it.
I've just took a look at what the Linux staging tree is. It looks like a good choice for the current ccmni driver.
honstly, If I simply upload the relevant driver code B that calls A (e.g. ccmni_rx_push), there is still a lack of code to call B. This seems to be a continuty problem, unless all drivers codes are uploaded (e.g. power on modem, get hardware status, complete tx/rx flow).
Quoted 11 lines
Quoted 7 lines
In addition, the code of MTK's modem driver is a bit complicated, because this part has more than 30,000 lines of code and contains more than 10 modules. We are completeing the upload of this huge code step by step. Our original intention was to upload the ccmni driver that directly interacts with the kernel first, and then complete the code from ccmni to the bottom layer one by one from top to bottom. We expect the completion period to be about 1 year.
Again, we can not add code to the kernel that is not used, sorry. That would not make any sense, would you want to maintain such a thing?
Quoted 9 lines
And 30k of code seems a bit excesive for a modem driver. Vendors find
that when they submit code for inclusion in the kernel tree, in the end, they end up 1/3 the original size, so 10k is reasonable.I can also take any drivers today into the
drivers/staging/tree, and you can do the cleanups there as well as getting help from others.1 year seems like a long time to do "cleanup", good luck!
Thanks~
Can I resend patch set as follows:
(1) supplement the details of pureip for patch 1/4;
(2) the document of ccmni.rst still live in the Documentation/...
(3) modify ccmni and move it into the drivers/staging/...
Quoted 11 lines
Quoted 8 lines
Quoted 4 lines
+++
b/drivers/net/ethernet/mediatek/ccmni/ccmni.hWhy do you have a .h file for a single .c file? that shouldn't be needed.
I add a .h file to facilitate subsequent code expansion. If it's not appropriate to do this here, I can add the content of .h into .c file.
If nothing other than a single .c file needs it, put it into that .c file please.
will do.
Thanks,
Rocco
Re: [PATCH 4/4] drivers: net: mediatek: initial implementation of ccmni
Quoted 17 lines · Rocco Yue
On Thu, 2021-06-24 at 14:23 +0200, Greg KH wrote:
Quoted 8 lines · Rocco Yue
Quoted 6 lines
without MTK ap ccci driver (modem driver),
ccmni_rx_push()andccmni_hif_hook()are not be used.Both of them are exported as symbols because MTK ap ccci driver will be compiled to the ccci.ko file.
But I do not see any code in this series that use these symbols. We can
will delete these symbols.
Quoted 3 lines
not have exports that no one uses. Please add the driver to this patch series when you resend it.
Quoted 7 lines
I've just took a look at what the Linux staging tree is. It looks like a good choice for the current ccmni driver.
honstly, If I simply upload the relevant driver code B that calls A (e.g.
ccmni_rx_push), there is still a lack of code to call B. This seems to be a continuty problem, unless all drivers codes are uploaded (e.g. power on modem, get hardware status, complete tx/rx flow).
Great, send it all! Why is it different modules, it's only for one chunk of hardware, no need to split it up into tiny pieces. That way only causes it to be more code overall.
Quoted 21 lines
Quoted 11 lines
Quoted 7 lines
In addition, the code of MTK's modem driver is a bit complicated, because this part has more than 30,000 lines of code and contains more than 10 modules. We are completeing the upload of this huge code step by step. Our original intention was to upload the ccmni driver that directly interacts with the kernel first, and then complete the code from ccmni to the bottom layer one by one from top to bottom. We expect the completion period to be about 1 year.
Again, we can not add code to the kernel that is not used, sorry. That would not make any sense, would you want to maintain such a thing?
Quoted 9 lines
And 30k of code seems a bit excesive for a modem driver. Vendors find
that when they submit code for inclusion in the kernel tree, in the end, they end up 1/3 the original size, so 10k is reasonable.I can also take any drivers today into the
drivers/staging/tree, and you can do the cleanups there as well as getting help from others.1 year seems like a long time to do "cleanup", good luck!
Quoted 6 lines
Thanks~
Can I resend patch set as follows:
(1) supplement the details of pureip for patch 1/4;
(2) the document ofccmni.rststill live in theDocumentation/...
(3) modify ccmni and move it into thedrivers/staging/...
for drivers/staging/ the code needs to be "self contained" in that it does not require adding anything outside of the directory for it.
If you still require this core networking change, that needs to be accepted first by the networking developers and maintainers.
thanks,
greg k-h
Re: [PATCH 4/4] drivers: net: mediatek: initial implementation of ccmni
On Thu, 2021-06-24 at 18:51 +0200, Greg KH wrote:
Quoted 30 lines · Rocco Yue
Quoted 7 lines · Greg KH
Quoted 4 lines · Rocco Yue
not have exports that no one uses. Please add the driver to this patch series when you resend it.
Quoted 7 lines
I've just took a look at what the Linux staging tree is. It looks like a good choice for the current ccmni driver.
honstly, If I simply upload the relevant driver code B that calls A (e.g.
ccmni_rx_push), there is still a lack of code to call B. This seems to be a continuty problem, unless all drivers codes are uploaded (e.g. power on modem, get hardware status, complete tx/rx flow).Great, send it all! Why is it different modules, it's only for one chunk of hardware, no need to split it up into tiny pieces. That way only causes it to be more code overall.
Quoted 7 lines
Thanks~
Can I resend patch set as follows:
(1) supplement the details of pureip for patch 1/4;
(2) the document ofccmni.rststill live in theDocumentation/...
(3) modify ccmni and move it into thedrivers/staging/...for
drivers/staging/the code needs to be "self contained" in that it does not require adding anything outside of the directory for it.
Quoted 7 lines
If you still require this core networking change, that needs to be accepted first by the networking developers and maintainers.
thanks,
greg k-h
Hi Greg,
I am grateful for your help.
Both ccmni change and networking changes are needed, because as far as I know, usually a device type should have at least one device to use it, and pureip is what the ccmni driver needs, so I uploaded the networking change and ccmni driver together;
Since MTK’s modem driver has a large amount of code and strong code coupling, it takes some time to clean up them. At this stage, it may be difficult to upstream all the codes together.
During this period, even if ccmni is incomplete, can I put the ccmni driver initial code in the driver/staging first ? After that, we will gradually implement more functions of ccmni in the staging tree, and we can also gradually sort out and clean up modem driver in the staging.
In addition, due to the requirements of GKI 2.0, if ccmni device uses RAWIP or NONE, it will hit ipv6 issue; and if ccmni uses a device type other than PUREIP/RAWIP/NONE, there will be tethering ebpf offload or clat ebpf offload can not work problems.
I hope PUREIP and ccmni can be accepted by the Linux community.
Thanks,
Rocco
Re: [PATCH 4/4] drivers: net: mediatek: initial implementation of ccmni
Quoted 44 lines · Rocco Yue
On Thu, 2021-06-24 at 18:51 +0200, Greg KH wrote:
Quoted 30 lines · Rocco Yue
Quoted 7 lines · Greg KH
Quoted 4 lines · Rocco Yue
not have exports that no one uses. Please add the driver to this patch series when you resend it.
Quoted 7 lines
I've just took a look at what the Linux staging tree is. It looks like a good choice for the current ccmni driver.
honstly, If I simply upload the relevant driver code B that calls A (e.g.
ccmni_rx_push), there is still a lack of code to call B. This seems to be a continuty problem, unless all drivers codes are uploaded (e.g. power on modem, get hardware status, complete tx/rx flow).Great, send it all! Why is it different modules, it's only for one chunk of hardware, no need to split it up into tiny pieces. That way only causes it to be more code overall.
Quoted 7 lines
Thanks~
Can I resend patch set as follows:
(1) supplement the details of pureip for patch 1/4;
(2) the document ofccmni.rststill live in theDocumentation/...
(3) modify ccmni and move it into thedrivers/staging/...for
drivers/staging/the code needs to be "self contained" in that it does not require adding anything outside of the directory for it.Quoted 7 lines
If you still require this core networking change, that needs to be accepted first by the networking developers and maintainers.
thanks,
greg k-h
Hi Greg,
I am grateful for your help.
Quoted 8 lines
Both ccmni change and networking changes are needed, because as far as I know, usually a device type should have at least one device to use it, and pureip is what the ccmni driver needs, so I uploaded the networking change and ccmni driver together;
Since MTK’s modem driver has a large amount of code and strong code coupling, it takes some time to clean up them. At this stage, it may be difficult to upstream all the codes together.
Why? Just dump the whole thing in a drivers/staging/mtk/ directory structure and all should be fine.
Quoted 4 lines
During this period, even if ccmni is incomplete, can I put the ccmni driver initial code in the driver/staging first ? After that, we will gradually implement more functions of ccmni in the staging tree, and we can also gradually sort out and clean up modem driver in the staging.
I do not know, let's see the code first. But we can not add frameworks with no in-kernel users, as that does not make any sense at all.
Quoted 1 line
In addition, due to the requirements of GKI 2.0,
That is a Google requirement, not a kernel.org requirement. Please work with Google if you have questions/issues about that, there is NOTHING we can do about that here in the community for obvious reasons.
Quoted 6 lines
if ccmni device
uses RAWIP or NONE, it will hit ipv6 issue; and if ccmni uses a device type other than PUREIP/RAWIP/NONE, there will be tethering ebpf offload or clat ebpf offload can not work problems.I hope PUREIP and ccmni can be accepted by the Linux community.
As I stated before you need to have an in-kernel user for us to be able to accept frameworks and functions into the tree. Otherwise Linux would quickly become unmanagable and unmaintainable. Would you want to try to maintain code with no in-tree users? What would you do if you were in our position?
But for networking flags like this that go into userspace, I do not know what the maintainers of the networking stack require, so that really is up to them, not me.
thanks,
greg k-h
Quoted 15 lines · Rocco Yue
This patch add the definition of
ARPHRD_PUREIPwhich can for example be used by mobile ccmni device as device type.ARPHRD_PUREIPmeans that this device doesn't need kernel to generate ipv6 link-local address in anyaddr_gen_mode.Signed-off-by Rocco Yue <[email protected]>--- include/uapi/linux/if_arp.h | 1 + 1 file changed, 1 insertion(+)include/uapi/linux/if_arp.h +0
-0
include/uapi/linux/if_arp.h line 64
Quoted 4 lines
include/uapi/linux/if_arp.h +1
-0@@ -61,6 +61,7 @@#define ARPHRD_DDCMP 517/* Digital's DDCMP protocol */ #define ARPHRD_RAWHDLC 518/* Raw HDLC */ #define ARPHRD_RAWIP 519/* Raw IP */ +#define ARPHRD_PUREIP 520/* Pure IP */
In looking at the patches, what differs "PUREIP" from "RAWIP"? It seems to be the same to me. If they are different, where is that documented?
thanks,
greg k-h
On Wed, 2021-06-23 at 19:19 +0200, Greg KH wrote:
Quoted 10 lines · Rocco Yue
Quoted 10 lines
This patch add the definition of
ARPHRD_PUREIPwhich can for example be used by mobile ccmni device as device type.ARPHRD_PUREIPmeans that this device doesn't need kernel to generate ipv6 link-local address in anyaddr_gen_mode.Signed-off-by Rocco Yue <[email protected]>--- include/uapi/linux/if_arp.h | 1 + 1 file changed, 1 insertion(+)
Quoted 11 lines
include/uapi/linux/if_arp.h line 64
Quoted 9 lines
include/uapi/linux/if_arp.h +1
-0@@ -61,6 +61,7 @@#define ARPHRD_DDCMP 517/* Digital's DDCMP protocol */ #define ARPHRD_RAWHDLC 518/* Raw HDLC */ #define ARPHRD_RAWIP 519/* Raw IP */ +#define ARPHRD_PUREIP 520/* Pure IP */ In looking at the patches, what differs "PUREIP" from "RAWIP"? It seems
Thanks for your review.
The difference between RAWIP and PUREIP is that they generate IPv6 link-local address and IPv6 global address in different ways.
RAWIP:
~~~~~~
In the ipv6_generate_eui64() function, using RAWIP will always return 0, which will cause the kernel to automatically generate an IPv6 link-local address in EUI64 format and an IPv6 global address in EUI64 format.
PUREIP:
~~~~~~~
After this patch set, when using PUREIP, kernel doesn't generate IPv6 link-local address regardless of which IN6_ADDR_GEN_MODE is used.
@@ static void addrconf_dev_config(struct net_device *dev)
+ if (dev->type == ARPHRD_PUREIP)
+ return;
And after recving RA message, kernel iterates over the link-local address
that exists for the interface and uses the low 64bits of the link-local
address to generate the IPv6 global address.
The general process is as follows:ndisc_router_discovery() -> addrconf_prefix_rcv() -> ipv6_generate_eui64() -> ipv6_inherit_eui64()
Quoted 5 lines
to be the same to me. If they are different, where is that documented?
thanks,
greg k-h
I tried to find corresponding documents about other device types, but I am sorry I didn't find it. If it is needed, I am willing to provide.
Thanks,
Rocco
Quoted 19 lines · Rocco Yue
The difference between RAWIP and PUREIP is that they generate IPv6 link-local address and IPv6 global address in different ways.
RAWIP:
~~~~~~
In theipv6_generate_eui64()function, using RAWIP will always return 0, which will cause the kernel to automatically generate an IPv6 link-local address in EUI64 format and an IPv6 global address in EUI64 format.PUREIP:
~~~~~~~
After this patch set, when using PUREIP, kernel doesn't generate IPv6 link-local address regardless of whichIN6_ADDR_GEN_MODEis used.@@ static void
addrconf_dev_config(struct net_device *dev)(quoted diff) +2
-0+ if (dev->type == ARPHRD_PUREIP) + return;
Quoted 6 lines
And after recving RA message, kernel iterates over the link-local address
that exists for the interface and uses the low 64bits of the link-local
address to generate the IPv6 global address.
The general process is as follows:ndisc_router_discovery()->addrconf_prefix_rcv()->ipv6_generate_eui64()->ipv6_inherit_eui64()
please add that to the commit message.
On Wed, 2021-06-23 at 23:15 -0600, David Ahern wrote:
Quoted 27 lines · Rocco Yue
Quoted 19 lines
The difference between RAWIP and PUREIP is that they generate IPv6 link-local address and IPv6 global address in different ways.
RAWIP:
~~~~~~
In theipv6_generate_eui64()function, using RAWIP will always return 0, which will cause the kernel to automatically generate an IPv6 link-local address in EUI64 format and an IPv6 global address in EUI64 format.PUREIP:
~~~~~~~
After this patch set, when using PUREIP, kernel doesn't generate IPv6 link-local address regardless of whichIN6_ADDR_GEN_MODEis used.@@ static void
addrconf_dev_config(struct net_device *dev)(quoted diff) +2
-0+ if (dev->type == ARPHRD_PUREIP) + return;Quoted 7 lines
And after recving RA message, kernel iterates over the link-local address that exists for the interface and uses the low 64bits of the link-local address to generate the IPv6 global address.
The general process is as follows:ndisc_router_discovery()->addrconf_prefix_rcv()->ipv6_generate_eui64()->ipv6_inherit_eui64()
Quoted 1 line
please add that to the commit message.
Thanks for your reminding, will do.
Thanks,
Rocco
Quoted 44 lines · Rocco Yue
On Wed, 2021-06-23 at 19:19 +0200, Greg KH wrote:
Quoted 10 lines · Rocco Yue
Quoted 10 lines
This patch add the definition of
ARPHRD_PUREIPwhich can for example be used by mobile ccmni device as device type.ARPHRD_PUREIPmeans that this device doesn't need kernel to generate ipv6 link-local address in anyaddr_gen_mode.Signed-off-by Rocco Yue <[email protected]>--- include/uapi/linux/if_arp.h | 1 + 1 file changed, 1 insertion(+)Quoted 11 lines
include/uapi/linux/if_arp.h line 64
Quoted 9 lines
include/uapi/linux/if_arp.h +1
-0@@ -61,6 +61,7 @@#define ARPHRD_DDCMP 517/* Digital's DDCMP protocol */ #define ARPHRD_RAWHDLC 518/* Raw HDLC */ #define ARPHRD_RAWIP 519/* Raw IP */ +#define ARPHRD_PUREIP 520/* Pure IP */ In looking at the patches, what differs "PUREIP" from "RAWIP"? It seems
Thanks for your review.
The difference between RAWIP and PUREIP is that they generate IPv6 link-local address and IPv6 global address in different ways.
RAWIP:
~~~~~~
In theipv6_generate_eui64()function, using RAWIP will always return 0, which will cause the kernel to automatically generate an IPv6 link-local address in EUI64 format and an IPv6 global address in EUI64 format.PUREIP:
~~~~~~~
After this patch set, when using PUREIP, kernel doesn't generate IPv6 link-local address regardless of whichIN6_ADDR_GEN_MODEis used.@@ static void
addrconf_dev_config(struct net_device *dev)(quoted diff) +2
-0+ if (dev->type == ARPHRD_PUREIP) + return;
Quoted 5 lines
And after recving RA message, kernel iterates over the link-local address
that exists for the interface and uses the low 64bits of the link-local
address to generate the IPv6 global address.
The general process is as follows:ndisc_router_discovery()->addrconf_prefix_rcv()->ipv6_generate_eui64()->ipv6_inherit_eui64()
Thanks for the explaination, why is this hardware somehow "special" in this way that this has never been needed before?
thanks,
greg k-h
Quoted 8 lines · Greg KH
Thanks for the explaination, why is this hardware somehow "special" in this way that this has never been needed before?
thanks,
greg k-h
Before kernel-4.18, RAWIP was the same as PUREIP, neither of them automatically generates an IPv6 link-local address, and the way to generate an IPv6 global address is the same.
After kernel-4.18 (include 4.18 version), the behavior of RAWIP had changed due to the following patch:
@@ static int ipv6_generate_eui64(u8 *eui, struct net_device *dev)
+ case ARPHRD_RAWIP:
+ return addrconf_ifid_rawip(eui, dev);
}
return -1;
}
the reason why the kernel doesn't need to generate the link-local address automatically is as follows:
In the 3GPP 29.061, here is some description as follows:
"in order to avoid any conflict between the link-local address of MS and that of the GGSN, the Interface-Identifier used by the MS to build its link-local address shall be assigned by the GGSN. The GGSN ensures the uniqueness of this Interface-Identifier. Then MT shall then enforce the use of this Interface-Identifier by the TE"
In other words, in the cellular network, GGSN determines whether to reply to the Router Solicitation message of UE by identifying the low 64bits of UE interface's ipv6 link-local address.
When using a new kernel and RAWIP, kernel will generate an EUI64 format ipv6 link-local address, and if the device uses this address to send RS, GGSN will not reply RA message.
Therefore, in that background, we came up with PUREIP to make kernel doesn't generate a ipv6 link-local address in any address generate mode.
Thanks,
Rocco
Quoted 37 lines · Rocco Yue
Quoted 8 lines · Greg KH
Thanks for the explaination, why is this hardware somehow "special" in this way that this has never been needed before?
thanks,
greg k-h
Before kernel-4.18, RAWIP was the same as PUREIP, neither of them automatically generates an IPv6 link-local address, and the way to generate an IPv6 global address is the same.
After kernel-4.18 (include 4.18 version), the behavior of RAWIP had changed due to the following patch:
@@ static intipv6_generate_eui64(u8 *eui, struct net_device *dev)(quoted diff) +2
-0+ case ARPHRD_RAWIP: + return addrconf_ifid_rawip(eui, dev); } return -1;}
the reason why the kernel doesn't need to generate the link-local address automatically is as follows:
In the 3GPP 29.061, here is some description as follows:
"in order to avoid any conflict between the link-local address of MS and that of the GGSN, the Interface-Identifier used by the MS to build its link-local address shall be assigned by the GGSN. The GGSN ensures the uniqueness of this Interface-Identifier. Then MT shall then enforce the use of this Interface-Identifier by the TE"In other words, in the cellular network, GGSN determines whether to reply to the Router Solicitation message of UE by identifying the low 64bits of UE interface's ipv6 link-local address.
Quoted 7 lines
When using a new kernel and RAWIP, kernel will generate an EUI64 format ipv6 link-local address, and if the device uses this address to send RS, GGSN will not reply RA message.
Therefore, in that background, we came up with PUREIP to make kernel doesn't generate a ipv6 link-local address in any address generate mode.
Thanks for the better description. That should go into the changelog text somewhere so that others know what is going on here with this new option.
And are these user-visable flags documented in a man page or something else somewhere? If not, how does userspace know about them?
thanks,
greg k-h
On Thu, 2021-06-24 at 11:04 +0200, Greg KH wrote:
Quoted 41 lines · Rocco Yue
Quoted 41 lines · Greg KH
Quoted 8 lines
Thanks for the explaination, why is this hardware somehow "special" in this way that this has never been needed before?
thanks,
greg k-h
Before kernel-4.18, RAWIP was the same as PUREIP, neither of them automatically generates an IPv6 link-local address, and the way to generate an IPv6 global address is the same.
After kernel-4.18 (include 4.18 version), the behavior of RAWIP had changed due to the following patch:
@@ static intipv6_generate_eui64(u8 *eui, struct net_device *dev)(quoted diff) +2
-0+ case ARPHRD_RAWIP: + return addrconf_ifid_rawip(eui, dev); } return -1;}
the reason why the kernel doesn't need to generate the link-local address automatically is as follows:
In the 3GPP 29.061, here is some description as follows:
"in order to avoid any conflict between the link-local address of MS and that of the GGSN, the Interface-Identifier used by the MS to build its link-local address shall be assigned by the GGSN. The GGSN ensures the uniqueness of this Interface-Identifier. Then MT shall then enforce the use of this Interface-Identifier by the TE"In other words, in the cellular network, GGSN determines whether to reply to the Router Solicitation message of UE by identifying the low 64bits of UE interface's ipv6 link-local address.
When using a new kernel and RAWIP, kernel will generate an EUI64 format ipv6 link-local address, and if the device uses this address to send RS, GGSN will not reply RA message.
Quoted 8 lines
Quoted 3 lines
Therefore, in that background, we came up with PUREIP to make kernel doesn't generate a ipv6 link-local address in any address generate mode.
Thanks for the better description. That should go into the changelog text somewhere so that others know what is going on here with this new option.
Does changelog mean adding these details to the commit message ? I am willing do it.
Quoted 3 lines
And are these user-visable flags documented in a man page or something else somewhere? If not, how does userspace know about them?
There are mappings of these device types value in the libc:
"/bionic/libc/kernel/uapi/linux/if_arp.h". userspace can get it from here.
But I also failed to find a man page or a description of these device types.
Thanks,
Rocco
Quoted 52 lines · Rocco Yue
On Thu, 2021-06-24 at 11:04 +0200, Greg KH wrote:
Quoted 41 lines · Rocco Yue
Quoted 41 lines · Greg KH
Quoted 8 lines
Thanks for the explaination, why is this hardware somehow "special" in this way that this has never been needed before?
thanks,
greg k-h
Before kernel-4.18, RAWIP was the same as PUREIP, neither of them automatically generates an IPv6 link-local address, and the way to generate an IPv6 global address is the same.
After kernel-4.18 (include 4.18 version), the behavior of RAWIP had changed due to the following patch:
@@ static intipv6_generate_eui64(u8 *eui, struct net_device *dev)(quoted diff) +2
-0+ case ARPHRD_RAWIP: + return addrconf_ifid_rawip(eui, dev); } return -1;}
the reason why the kernel doesn't need to generate the link-local address automatically is as follows:
In the 3GPP 29.061, here is some description as follows:
"in order to avoid any conflict between the link-local address of MS and that of the GGSN, the Interface-Identifier used by the MS to build its link-local address shall be assigned by the GGSN. The GGSN ensures the uniqueness of this Interface-Identifier. Then MT shall then enforce the use of this Interface-Identifier by the TE"In other words, in the cellular network, GGSN determines whether to reply to the Router Solicitation message of UE by identifying the low 64bits of UE interface's ipv6 link-local address.
When using a new kernel and RAWIP, kernel will generate an EUI64 format ipv6 link-local address, and if the device uses this address to send RS, GGSN will not reply RA message.
Quoted 8 lines
Quoted 3 lines
Therefore, in that background, we came up with PUREIP to make kernel doesn't generate a ipv6 link-local address in any address generate mode.
Thanks for the better description. That should go into the changelog text somewhere so that others know what is going on here with this new option.
Quoted 1 line
Does changelog mean adding these details to the commit message ?
Yes please.
Quoted 7 lines
Quoted 3 lines
And are these user-visable flags documented in a man page or something else somewhere? If not, how does userspace know about them?
There are mappings of these device types value in the libc:
"/bionic/libc/kernel/uapi/linux/if_arp.h". userspace can get it from here.
Yes, they will show up in a libc definition, but where is it documented in text form what the flag does?
thanks,
greg k-h
Quoted 4 lines
Does changelog mean adding these details to the commit message ?
Yes please.
will do.
Quoted 10 lines
Quoted 7 lines
Quoted 3 lines
And are these user-visable flags documented in a man page or something else somewhere? If not, how does userspace know about them?
There are mappings of these device types value in the libc:
"/bionic/libc/kernel/uapi/linux/if_arp.h". userspace can get it from here.Yes, they will show up in a libc definition, but where is it documented in text form what the flag does?
Judging from the changes of ARPHRD_xxx submitted before, I am sorry I could not find the corresponding doucuments to describe their respective behaviors in details. Perhaps the best way to understand their behaviors is to read the code directly.
Thanks,
Rocco
Quoted 6 lines · Rocco Yue
Quoted 5 lines · Greg KH
Thanks for the explaination, why is this hardware somehow "special" in
this way that this has never been needed before?
Quoted 8 lines
Quoted 4 lines
thanks,
greg k-h
Before kernel-4.18, RAWIP was the same as PUREIP, neither of them automatically generates an IPv6 link-local address, and the way to generate an IPv6 global address is the same.
This distinction seems confusing from a kernel standpoint if it only changes how v6 IIDs are determined. Do we really need something that's also reflected to userspace (in struct ifinfomsg -> ifi_type) if the kernel is handling the behavior that's different? Why should userspace care?
I'm also curious why this isn't an issue for the ipa/rmnet (Qualcomm) modem drivers. There's probably a good reason, but would be good to know what that is from Alex Elder or Loic or Bjorn...
Dan
Quoted 35 lines
After kernel-4.18 (include 4.18 version), the behavior of RAWIP had changed due to the following patch:
@@ static intipv6_generate_eui64(u8 *eui, struct net_device *dev)+ caseARPHRD_RAWIP:
+ returnaddrconf_ifid_rawip(eui, dev);
}
return -1;
}the reason why the kernel doesn't need to generate the link-local address automatically is as follows:
In the 3GPP 29.061, here is some description as follows:
"in order to avoid any conflict between the link-local address of MS and that of the GGSN, the Interface-Identifier used by the MS to build its link-local address shall be assigned by the GGSN. The GGSN ensures the uniqueness of this Interface-Identifier. Then MT shall then enforce the use of this Interface-Identifier by the TE"In other words, in the cellular network, GGSN determines whether to reply to the Router Solicitation message of UE by identifying the low 64bits of UE interface's ipv6 link-local address.
When using a new kernel and RAWIP, kernel will generate an EUI64 format ipv6 link-local address, and if the device uses this address to send RS, GGSN will not reply RA message.
Therefore, in that background, we came up with PUREIP to make kernel doesn't generate a ipv6 link-local address in any address generate mode.
Thanks,
Rocco
On Thu, 2021-06-24 at 11:14 -0500, Dan Williams wrote:
Quoted 8 lines · Rocco Yue
Quoted 5 lines · Greg KH
Before kernel-4.18, RAWIP was the same as PUREIP, neither of them automatically generates an IPv6 link-local address, and the way to generate an IPv6 global address is the same.
This distinction seems confusing from a kernel standpoint if it only changes how v6 IIDs are determined. Do we really need something that's
Hi Dan,
Thanks for your comment,
In the cellular network, v6 IID is important, If the device use the link-local address formed by the incorrect IID to send RS message to the network, based on 3GPP, GGSN will not reply solicited RA message. It will lead to the device can get ipv6 address prefix and ipv6 route.
Maybe the table below is a little bit clearer
three device type: ARPHRD_RAWIP , ARPHRD_PUREIP, ARPHRD_NONE
three mode: IN6_ADDR_GEN_MODE_EUI64 , IN6_ADDR_GEN_MODE_NONE, IN6_ADDR_GEN_MODE_STABLE_PRIVACY
ipv6 link-local address generate behavior in the kernel:
+---------+-------------------+---------------------+----------------+ | | MODE_EUI64 | MODE_STABLE_PRIVACY | MODE_NONE | +---------+-------------------+---------------------+----------------+ | RAWIP | fe80::(eui64-id) | fe80::(privacy-id) | no address gen | +---------+-------------------+---------------------+----------------+ | PUREIP | no address gen | no address gen | no address gen | +---------+-------------------+---------------------+----------------+ | NONE | fe80::(random-id) | fe80::(privacy-id) | no address gen | +---------+-------------------+---------------------+----------------+
ipv6 global address generate behavior in the kernel:
+---------+-------------------+---------------------+-------------------+ | | MODE_EUI64 | MODE_STABLE_PRIVACY | MODE_NONE | +---------+-------------------+---------------------+-------------------+ | RAWIP | prefix+(eui64-id) | prefix+(privacy-id) | prefix+(eui64-id) | +---------+-------------------+---------------------+-------------------+ | PUREIP | prefix+(GGSN-id) | prefix+(privacy-id) | prefix+(GGSN-id) | +---------+-------------------+---------------------+-------------------+ | NONE | prefix+(random-id)| prefix+(privacy-id) | prefix+(random-id)| +---------+-------------------+---------------------+-------------------+
Quoted 4 lines
also reflected to userspace (in struct ifinfomsg ->
ifi_type) if the kernel is handling the behavior that's different? Why should userspace care?
In my opinion, userspace program cares about it because the kernel behaves differently for different device types.
userspace can get the device type of the interface through ioctl, such as the following code weblink:
https://cs.android.com/android/platform/superproject/+/master:system/netd/server/OffloadUtils.cpp;drc=master;l=41?q=ARPHRD_RAWIP&ss=android%2Fplatform%2Fsuperproject&start=11
Quoted 5 lines
I'm also curious why this isn't an issue for the ipa/rmnet (Qualcomm) modem drivers. There's probably a good reason, but would be good to know what that is from Alex Elder or Loic or Bjorn...
Dan
MediaTek and Qualcomm has different hardware or modem design. For the MediaTek platform, device send the RS message that generated by the kernel to the GGSN.
Thanks,
Rocco