Re: [sched_ext/for-6.11]: Issue with BPF Scheduler during CPU Hotplug

[Hari Bathini]

On 24/09/24 10:27 pm, Song Liu wrote:
> On Tue, Sep 24, 2024 at 5:00 AM Aboorva Devarajan
> <aboorvad@xxxxxxxxxxxxx> wrote:
> > On Tue, 2024-09-24 at 10:03 +0200, Alexei Starovoitov wrote:
> > > On Mon, Sep 23, 2024 at 8:21 PM Tejun Heo <tj@xxxxxxxxxx> wrote:
> > > > Hello,
> > > >
> > > > (cc'ing Alexei and Andrii for the BPF part)
> > > >
> > > > On Mon, Sep 23, 2024 at 08:26:32PM +0530, Aboorva Devarajan wrote:
> > > > > Sharing the crash logs observed in PowerPC here for general reference, FYI:
> > > > >
> > > > > [ 8638.891964] Kernel attempted to read user page (a8) - exploit attempt? (uid: 0)
> > > > > [ 8638.892002] BUG: Kernel NULL pointer dereference on read at 0x000000a8
> > > > > [ 8638.892019] Faulting instruction address: 0xc0000000004e7cc0
> > > > > [ 8638.892038] Oops: Kernel access of bad area, sig: 11 [#1]
> > > > > [ 8638.892060] LE PAGE_SIZE=64K MMU=Radix SMP NR_CPUS=2048 NUMA PowerNV
> > > > > [ 8638.892080] Modules linked in: nf_conntrack_netlink nfnetlink xfrm_user xfrm_algo xt_addrtype
> > > > > br_netfilter xt_CHECKSUM xt_MASQUERADE xt_conntrack ipt_REJECT nf_reject_ipv4 xt_tcpudp
> > > > >   ip6table_mangle ip6table_nat iptable_mangle iptable_nat nf_nat nf_conntrack nf_defrag_ipv6
> > > > > nf_defrag_ipv4 ebtable_filter ebtables vhost_vsock vmw_vsock_virtio_transport_common ip6tabl
> > > > > e_filter ip6_tables vhost vhost_iotlb iptable_filter vsock bridge stp llc kvm_hv kvm joydev
> > > > > input_leds mac_hid at24 ofpart cmdlinepart uio_pdrv_genirq ibmpowernv opal_prd ipmi_powernv
> > > > > powernv_flash uio binfmt_misc sch_fq_codel nfsd mtd ipmi_devintf ipmi_msghandler auth_rpcgss
> > > > > jc42 ramoops reed_solomon ip_tables x_tables autofs4 raid10 raid456 async_raid6_recov async
> > > > > _memcpy async_pq async_xor async_tx raid1 raid0 dm_mirror dm_region_hash dm_log mlx5_ib ib_uverbs
> > > > > ib_core mlx5_core hid_generic usbhid hid ast i2c_algo_bit drm_shmem_helper drm_kms_hel
> > > > > per vmx_crypto drm mlxfw crct10dif_vpmsum crc32c_vpmsum psample tls tg3 ahci libahci
> > > > > drm_panel_orientation_quirks
> > > > > [ 8638.892621] CPU: 62 UID: 0 PID: 5591 Comm: kworker/62:2 Not tainted 6.11.0-rc4+ #2
> > > > > [ 8638.892663] Hardware name: 8335-GTW POWER9 0x4e1203 opal:skiboot-v6.5.3-35-g1851b2a06 PowerNV
> > > > > [ 8638.892693] Workqueue: events bpf_prog_free_deferred
> > > > > [ 8638.892735] NIP:  c0000000004e7cc0 LR: c0000000004e7bbc CTR: c0000000003a9b30
> > > > > [ 8638.892798] REGS: c000000ea4cbf7f0 TRAP: 0300   Not tainted  (6.11.0-rc4+)
> > > > > [ 8638.892862] MSR:  9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE>  CR: 42a00284  XER: 00000000
> > > > > [ 8638.892915] CFAR: c0000000004e7bb8 DAR: 00000000000000a8 DSISR: 40000000 IRQMASK: 1
> > > > > [ 8638.892915] GPR00: c0000000004e7bbc c000000ea4cbfa90 c000000002837f00 0000000000000005
> > > > > [ 8638.892915] GPR04: 0000000000000015 0000000000000009 0000000000000009 c000000004840b00
> > > > > [ 8638.892915] GPR08: ffffffffffffffff 00000000ffffe000 ffffffffffffffff 000001937b55db50
> > > > > [ 8638.892915] GPR12: 0000000000200000 c000007ffdfac300 c0000000031b1fc8 0000000000010000
> > > > > [ 8638.892915] GPR16: c00000000000018e 000000007fffffff 0000000000000000 000000000000e1c0
> > > > > [ 8638.892915] GPR20: 61c8864680b583eb 0000000000000000 0000000000000000 00000000000de1d5
> > > > > [ 8638.892915] GPR24: 0000000000000000 c000000003da4408 c000000003da4400 c000000003da43f8
> > > > > [ 8638.892915] GPR24: 0000000000000000 c000000003da4408 c000000003da4400 c000000003da43f8
> > > > > [ 8638.892915] GPR28: 0000000000000000 0000000000000000 0000000000000000 c000000ea4cbfa90
> > > > > [ 8638.893350] NIP [c0000000004e7cc0] walk_to_pmd+0x80/0x240
>
> With "BUG: Kernel NULL pointer dereference on read at 0x000000a8" (from above),
> it appears bpf_arch_text_invalidate() is racing with
> text_area_cpu_down_mm(), which
> sets cpu_patching_context.mm to NULL?
>
> Am I going in the right direction?

Hi Song,

Yeah. That is the most plausible explanation for it.
I have just about started looking at this.
Will update if I find anything different..

Thanks
Hari