forked from TencentOS/TencentOS-kernel
3 Commits
| Author | SHA1 | Message | Date |
|---|---|---|---|
Alexander Lobakin
|
33d9cd7a01 |
net: dsa: add GRO support via gro_cells
commit e131a5634830047923c694b4ce0c3b31745ff01b upstream. gro_cells lib is used by different encapsulating netdevices, such as geneve, macsec, vxlan etc. to speed up decapsulated traffic processing. CPU tag is a sort of "encapsulation", and we can use the same mechs to greatly improve overall DSA performance. skbs are passed to the GRO layer after removing CPU tags, so we don't need any new packet offload types as it was firstly proposed by me in the first GRO-over-DSA variant [1]. The size of struct gro_cells is sizeof(void *), so hot struct dsa_slave_priv becomes only 4/8 bytes bigger, and all critical fields remain in one 32-byte cacheline. The other positive side effect is that drivers for network devices that can be shipped as CPU ports of DSA-driven switches can now use napi_gro_frags() to pass skbs to kernel. Packets built that way are completely non-linear and are likely being dropped without GRO. This was tested on to-be-mainlined-soon Ethernet driver that uses napi_gro_frags(), and the overall performance was on par with the variant from [1], sometimes even better due to minimal overhead. net.core.gro_normal_batch tuning may help to push it to the limit on particular setups and platforms. iperf3 IPoE VLAN NAT TCP forwarding (port1.218 -> port0) setup on 1.2 GHz MIPS board: 5.7-rc2 baseline: [ID] Interval Transfer Bitrate Retr [ 5] 0.00-120.01 sec 9.00 GBytes 644 Mbits/sec 413 sender [ 5] 0.00-120.00 sec 8.99 GBytes 644 Mbits/sec receiver Iface RX packets TX packets eth0 7097731 7097702 port0 426050 6671829 port1 6671681 425862 port1.218 6671677 425851 With this patch: [ID] Interval Transfer Bitrate Retr [ 5] 0.00-120.01 sec 12.2 GBytes 870 Mbits/sec 122 sender [ 5] 0.00-120.00 sec 12.2 GBytes 870 Mbits/sec receiver Iface RX packets TX packets eth0 9474792 9474777 port0 455200 353288 port1 9019592 455035 port1.218 353144 455024 v2: - Add some performance examples in the commit message; - No functional changes. [1] https://lore.kernel.org/netdev/20191230143028.27313-1-alobakin@dlink.ru/ Signed-off-by: Alexander Lobakin <bloodyreaper@yandex.ru> Signed-off-by: David S. Miller <davem@davemloft.net> Cc: Pali Rohár <pali@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> |
|
|
Vladimir Oltean
|
17ce884842 |
net: dsa: declare lockless TX feature for slave ports
commit 2b86cb8299765688c5119fd18d5f436716c81010 upstream.
Be there a platform with the following layout:
Regular NIC
|
+----> DSA master for switch port
|
+----> DSA master for another switch port
After changing DSA back to static lockdep class keys in commit
1a33e10e4a95 ("net: partially revert dynamic lockdep key changes"), this
kernel splat can be seen:
[ 13.361198] ============================================
[ 13.366524] WARNING: possible recursive locking detected
[ 13.371851] 5.7.0-rc4-02121-gc32a05ecd7af-dirty #988 Not tainted
[ 13.377874] --------------------------------------------
[ 13.383201] swapper/0/0 is trying to acquire lock:
[ 13.388004] ffff0000668ff298 (&dsa_slave_netdev_xmit_lock_key){+.-.}-{2:2}, at: __dev_queue_xmit+0x84c/0xbe0
[ 13.397879]
[ 13.397879] but task is already holding lock:
[ 13.403727] ffff0000661a1698 (&dsa_slave_netdev_xmit_lock_key){+.-.}-{2:2}, at: __dev_queue_xmit+0x84c/0xbe0
[ 13.413593]
[ 13.413593] other info that might help us debug this:
[ 13.420140] Possible unsafe locking scenario:
[ 13.420140]
[ 13.426075] CPU0
[ 13.428523] ----
[ 13.430969] lock(&dsa_slave_netdev_xmit_lock_key);
[ 13.435946] lock(&dsa_slave_netdev_xmit_lock_key);
[ 13.440924]
[ 13.440924] *** DEADLOCK ***
[ 13.440924]
[ 13.446860] May be due to missing lock nesting notation
[ 13.446860]
[ 13.453668] 6 locks held by swapper/0/0:
[ 13.457598] #0: ffff800010003de0 ((&idev->mc_ifc_timer)){+.-.}-{0:0}, at: call_timer_fn+0x0/0x400
[ 13.466593] #1: ffffd4d3fb478700 (rcu_read_lock){....}-{1:2}, at: mld_sendpack+0x0/0x560
[ 13.474803] #2: ffffd4d3fb478728 (rcu_read_lock_bh){....}-{1:2}, at: ip6_finish_output2+0x64/0xb10
[ 13.483886] #3: ffffd4d3fb478728 (rcu_read_lock_bh){....}-{1:2}, at: __dev_queue_xmit+0x6c/0xbe0
[ 13.492793] #4: ffff0000661a1698 (&dsa_slave_netdev_xmit_lock_key){+.-.}-{2:2}, at: __dev_queue_xmit+0x84c/0xbe0
[ 13.503094] #5: ffffd4d3fb478728 (rcu_read_lock_bh){....}-{1:2}, at: __dev_queue_xmit+0x6c/0xbe0
[ 13.512000]
[ 13.512000] stack backtrace:
[ 13.516369] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.7.0-rc4-02121-gc32a05ecd7af-dirty #988
[ 13.530421] Call trace:
[ 13.532871] dump_backtrace+0x0/0x1d8
[ 13.536539] show_stack+0x24/0x30
[ 13.539862] dump_stack+0xe8/0x150
[ 13.543271] __lock_acquire+0x1030/0x1678
[ 13.547290] lock_acquire+0xf8/0x458
[ 13.550873] _raw_spin_lock+0x44/0x58
[ 13.554543] __dev_queue_xmit+0x84c/0xbe0
[ 13.558562] dev_queue_xmit+0x24/0x30
[ 13.562232] dsa_slave_xmit+0xe0/0x128
[ 13.565988] dev_hard_start_xmit+0xf4/0x448
[ 13.570182] __dev_queue_xmit+0x808/0xbe0
[ 13.574200] dev_queue_xmit+0x24/0x30
[ 13.577869] neigh_resolve_output+0x15c/0x220
[ 13.582237] ip6_finish_output2+0x244/0xb10
[ 13.586430] __ip6_finish_output+0x1dc/0x298
[ 13.590709] ip6_output+0x84/0x358
[ 13.594116] mld_sendpack+0x2bc/0x560
[ 13.597786] mld_ifc_timer_expire+0x210/0x390
[ 13.602153] call_timer_fn+0xcc/0x400
[ 13.605822] run_timer_softirq+0x588/0x6e0
[ 13.609927] __do_softirq+0x118/0x590
[ 13.613597] irq_exit+0x13c/0x148
[ 13.616918] __handle_domain_irq+0x6c/0xc0
[ 13.621023] gic_handle_irq+0x6c/0x160
[ 13.624779] el1_irq+0xbc/0x180
[ 13.627927] cpuidle_enter_state+0xb4/0x4d0
[ 13.632120] cpuidle_enter+0x3c/0x50
[ 13.635703] call_cpuidle+0x44/0x78
[ 13.639199] do_idle+0x228/0x2c8
[ 13.642433] cpu_startup_entry+0x2c/0x48
[ 13.646363] rest_init+0x1ac/0x280
[ 13.649773] arch_call_rest_init+0x14/0x1c
[ 13.653878] start_kernel+0x490/0x4bc
Lockdep keys themselves were added in commit ab92d68fc22f ("net: core:
add generic lockdep keys"), and it's very likely that this splat existed
since then, but I have no real way to check, since this stacked platform
wasn't supported by mainline back then.
>From Taehee's own words:
This patch was considered that all stackable devices have LLTX flag.
But the dsa doesn't have LLTX, so this splat happened.
After this patch, dsa shares the same lockdep class key.
On the nested dsa interface architecture, which you illustrated,
the same lockdep class key will be used in __dev_queue_xmit() because
dsa doesn't have LLTX.
So that lockdep detects deadlock because the same lockdep class key is
used recursively although actually the different locks are used.
There are some ways to fix this problem.
1. using NETIF_F_LLTX flag.
If possible, using the LLTX flag is a very clear way for it.
But I'm so sorry I don't know whether the dsa could have LLTX or not.
2. using dynamic lockdep again.
It means that each interface uses a separate lockdep class key.
So, lockdep will not detect recursive locking.
But this way has a problem that it could consume lockdep class key
too many.
Currently, lockdep can have 8192 lockdep class keys.
- you can see this number with the following command.
cat /proc/lockdep_stats
lock-classes: 1251 [max: 8192]
...
The [max: 8192] means that the maximum number of lockdep class keys.
If too many lockdep class keys are registered, lockdep stops to work.
So, using a dynamic(separated) lockdep class key should be considered
carefully.
In addition, updating lockdep class key routine might have to be existing.
(lockdep_register_key(), lockdep_set_class(), lockdep_unregister_key())
3. Using lockdep subclass.
A lockdep class key could have 8 subclasses.
The different subclass is considered different locks by lockdep
infrastructure.
But "lock-classes" is not counted by subclasses.
So, it could avoid stopping lockdep infrastructure by an overflow of
lockdep class keys.
This approach should also have an updating lockdep class key routine.
(lockdep_set_subclass())
4. Using nonvalidate lockdep class key.
The lockdep infrastructure supports nonvalidate lockdep class key type.
It means this lockdep is not validated by lockdep infrastructure.
So, the splat will not happen but lockdep couldn't detect real deadlock
case because lockdep really doesn't validate it.
I think this should be used for really special cases.
(lockdep_set_novalidate_class())
Further discussion here:
https://patchwork.ozlabs.org/project/netdev/patch/20200503052220.4536-2-xiyou.wangcong@gmail.com/
There appears to be no negative side-effect to declaring lockless TX for
the DSA virtual interfaces, which means they handle their own locking.
So that's what we do to make the splat go away.
Patch tested in a wide variety of cases: unicast, multicast, PTP, etc.
Fixes: ab92d68fc22f ("net: core: add generic lockdep keys")
Suggested-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
|
|
|
Kaixu Xia
|
590eaf1fec |
Init Repo base on linux 5.4.32 long term, and add base tlinux kernel interfaces.
Signed-off-by: Kaixu Xia <kaixuxia@tencent.com> |