kmod: igb_main: Use slow_virt_to_phys() instead of virt_to_phys()

[MarcinMiklas]

In kernel >=v6.8.0 virt_to_phys() returns address which is not usable from userspace. It was working fine in kernels up to v6.7.x.

I failed to find which exactly commit (or maybe kernel config option) causes that change of behaviour. What I understand, neither virt_to_phys() nor slow_virt_to_phys() should be used in drivers for dma transfers, we are triggering dma transfers from userspace which is even more evil.

[MarcinMiklas]

I checked the issue on debian 12 machine (previously I was working on ubuntu 22.04) and problem occurs also on kernel 6.1, so it is rather kernel configuration which causes it. I didn't share the details of the crash before so I am doing it now.

Here is dmesg output:

[ 612.276866] igb_avb 0000:02:00.0 enp2s0: igb: enp2s0 NIC Link is Down
[ 613.985478] igb_avb 0000:02:00.0 enp2s0: igb: enp2s0 NIC Link is Up 100 Mbps Full Duplex, Flow Control: None
[ 1150.528882] bind to iface 0000:02:00.0
[ 1150.528894] checking against adapter name 0000:02:00.0
[ 1150.528966] openavb_harness: Corrupted page table at address 7f3e515bd000
[ 1150.528971] PGD 8000000031e02067 P4D 8000000031e02067 PUD 31ef2067 PMD 32640067 PTE 80001ad28088d237
[ 1150.528980] Bad pagetable: 000f [#2] PREEMPT SMP PTI
[ 1150.528984] CPU: 1 PID: 5580 Comm: openavb_harness Tainted: G D W OE 6.1.0-28-amd64 #1 Debian 6.1.119-1
[ 1150.528990] Hardware name: FUJITSU D3544-Sx/D3544-Sx, BIOS V5.0.0.13 R1.2.0 for D3544-Sxx 03/29/2018
[ 1150.528993] RIP: 0033:0x7f3e51aeb6fa
[ 1150.528999] Code: 83 ea c0 48 39 fa 72 e8 0f 11 07 0f 11 47 10 0f 11 47 20 0f 11 47 30 c3 0f 1f 84 00 00 00 00 00 40 0f b6 c6 48 89 d1 48 89 fa aa 48 89 d0 c3 66 48 0f 7e c6 83 fa 08 7d 16 83 fa 04 7d 1a 83
[ 1150.529003] RSP: 002b:00007ffc100a8888 EFLAGS: 00010206
[ 1150.529007] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000001000
[ 1150.529010] RDX: 00007f3e515bd000 RSI: 0000000000000000 RDI: 00007f3e515bd000
[ 1150.529012] RBP: 0000564b75e4eb30 R08: 0000564b75e4ebc0 R09: 00001ad28088d000
[ 1150.529015] R10: 00007f3e51a5f000 R11: 00007f3e51aeb640 R12: 0000000000000000
[ 1150.529017] R13: 0000000000000000 R14: 0000000000000004 R15: 00007ffc100a88b0
[ 1150.529020] FS: 00007f3e51640800 GS: 0000000000000000

The issue is triggered from user space in function igb_allocate_queues() which can be found in lib/igb.c file.

Exactly this memset is causing it.
memset((void *)adapter->tx_rings[i].tx_base, 0, ubuf.mmap_size);

[gisburn]

@MarcinMiklas Did you ever find out which kernel config caused this issue ?