Re: [Patch net-next v2 0/9] net: dsa: microchip: add support for phylink mac config and link up

[Vladimir Oltean]

Hello,

On Tue, Aug 30, 2022 at 08:15:59AM +0000, Arun.Ramadoss@xxxxxxxxxxxxx wrote:
> On Tue, 2022-08-30 at 08:55 +0200, Oleksij Rempel wrote:
> > Hi Arun,
> > 
> > starting with this patch set I have following regression on ksz8873
> > switch. Can you please take a look at it:
> >  8<--- cut here ---
> >  Unable to handle kernel NULL pointer dereference at virtual address 00000005
> >  ksz8863-switch gpio-0:00: nonfatal error -34 setting MTU to 1500 on port 0
> > ...
> >  Modules linked in:
> >  CPU: 0 PID: 16 Comm: kworker/0:1 Not tainted 6.0.0-rc2-00436-
> > g3da285df1324 #74
> >  Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree)
> >  Workqueue: events_power_efficient phylink_resolve
> >  PC is at ksz_set_gbit+0x5c/0xa4
> >  LR is at arch_atomic_cmpxchg_relaxed+0x1c/0x38
> > ....
> >  Backtrace:
> >   ksz_set_gbit from ksz_phylink_mac_link_up+0x15c/0x1c8
> >   ksz_phylink_mac_link_up from dsa_port_phylink_mac_link_up+0x7c/0x80
> >   dsa_port_phylink_mac_link_up from phylink_resolve+0x304/0x3d0
> >   phylink_resolve from process_one_work+0x214/0x31c
> >   process_one_work from worker_thread+0x254/0x2d4
> >   worker_thread from kthread+0xfc/0x108
> >   kthread from ret_from_fork+0x14/0x2c
> > ...
> >  ksz8863-switch gpio-0:00 lan2 (uninitialized): PHY [dsa-0.0:01] driver [Micrel KSZ8851 Ethernet MAC or KSZ886X Switch] (irq=POLL)
> >  ksz8863-switch gpio-0:00: nonfatal error -34 setting MTU to 1500 on port 1
> >  device eth0 entered promiscuous mode
> >  DSA: tree 0 setup
> >  ---[ end trace 0000000000000000 ]---
> 
> Hi Oleksij,
> Is this Bug related to fix in 
> https://lore.kernel.org/lkml/20220829105810.577903823@xxxxxxxxxxxxxxxxxxx/
> . 
> It is observed in ksz8794 switch. I think after applying this bug fix
> patch it should work. I don't have ksz8 series to test. I ran the
> regression only for ksz9 series switches. 

I find it unlikely that the cited patch will fix a NULL pointer
dereference in ksz_get_gbit(). But rather, some pointer to a structure
is NULL, and we then dereference a member located at its offset 0x5, no?

My eyes are on this:

	const u8 *bitval = dev->info->xmii_ctrl1;

		data8 |= FIELD_PREP(P_GMII_1GBIT_M, bitval[P_GMII_NOT_1GBIT]);
							   ~~~~~~~~~~~~~~~~
							   this is coincidentally
							   also 5

See, looking at the struct ksz_chip_data[] array element for KSZ8873
that Oleksij mentions as broken, I do not see xmii_ctrl1 and xmii_ctrl2
as being pointers to anything.

	[KSZ8830] = {
		.chip_id = KSZ8830_CHIP_ID,
		.dev_name = "KSZ8863/KSZ8873",
		.num_vlans = 16,
		.num_alus = 0,
		.num_statics = 8,
		.cpu_ports = 0x4,	/* can be configured as cpu port */
		.port_cnt = 3,
		.ops = &ksz8_dev_ops,
		.mib_names = ksz88xx_mib_names,
		.mib_cnt = ARRAY_SIZE(ksz88xx_mib_names),
		.reg_mib_cnt = MIB_COUNTER_NUM,
		.regs = ksz8863_regs,
		.masks = ksz8863_masks,
		.shifts = ksz8863_shifts,
		.supports_mii = {false, false, true},
		.supports_rmii = {false, false, true},
		.internal_phy = {true, true, false},
	},

Should we point them to ksz8795_xmii_ctrl0 and ksz8795_xmii_ctrl1? I don't know.
Could you find out what these should be set to?