Hi!
On Wed 2019-05-15 12:55:33, Greg Kroah-Hartman wrote:
[ Upstream commit a8fd48b50deaa20808bbf0f6685f6f1acba6a64c ]
Preemption disabled at: [<ffff000008cabd54>] dev_set_rx_mode+0x1c/0x38 Call trace: [<ffff00000808a5c0>] dump_backtrace+0x0/0x3d0 [<ffff00000808a9a4>] show_stack+0x14/0x20 [<ffff000008e6c0c0>] dump_stack+0xac/0xe4 [<ffff0000080fe76c>] ___might_sleep+0x164/0x238 [<ffff0000080fe890>] __might_sleep+0x50/0x88 [<ffff0000082261e4>] kmem_cache_alloc+0x17c/0x1d0 [<ffff000000ea0ae8>] ocelot_set_rx_mode+0x108/0x188
[mscc_ocelot_common]
[<ffff000008cabcf0>] __dev_set_rx_mode+0x58/0xa0 [<ffff000008cabd5c>] dev_set_rx_mode+0x24/0x38
Fixes: a556c76adc05 ("net: mscc: Add initial Ocelot switch support")
Is it right fix? Warning is gone, but now allocation is more likely to fail, causing mc_add() to fail under memory pressure.
So far this contributes to fixing a kernel hang issue, seen occasionally when the switch interfaces were brought up. Other than that I would look into improving this code. It looks suboptimal at least. Do we really need to allocate whole struct netdev_hw_addr elements? Can the allocation size be reduced? What about pre-allocating enough room for ha elements outside the atomic context (set_rx_mode() in this case)?
Pre-allocating the elements sounds like a obvious solution, yes.
Best regards, Pavel