[PATCH] fwctl: Fix class init ordering to avoid NULL pointer dereference on device removal

[Richard Cheng <icheng@nvidia.com>]

CXL is linked before fwctl in drivers/Makefile. Both use `module_init,
so `cxl_pci_driver_init()` runs first. When `cxl_pci_probe()` calls
`fwctl_register()` and then `device_add()`, fwctl_class is not yet
registered because fwctl_init() hasn't run, causing `class_to_subsys()`
to return NULL and skip knode_class initialization.
On device removal, `class_to_subsys()` returns non-NULL, and
`device_del()` calls `klist_del()` on the uninitialized knode,
triggering a NULL pointer dereference [1].

Fixes: 858ce2f56b52 ("cxl: Add FWCTL support to CXL")
Signed-off-by: Richard Cheng <icheng@nvidia.com>
Reviewed-by: Kai-Heng Feng <kaihengf@nvidia.com>
---
[1]:
The error is triggered on with 7.0.0-rc6 kernel with CXL
device.
The PCI topology is as below
```
$ sudo lspci -tv
-[0001:00]---00.0-[01]--+-00.0  Mellanox Technologies CX8 Family [ConnectX-8]
                        +-00.1  Mellanox Technologies CX8 Family [ConnectX-8]
                        +-00.2  Mellanox Technologies CX8 Family [ConnectX-8]
                        \-00.3  Mellanox Technologies CX8 Family [ConnectX-8]
-[0002:00]---00.0-[01]--
-+-[0003:00]---00.0-[01]----00.0  Montage Technology Co., Ltd. Device c002
 \-[0003:80]---00.0-[81]----00.0  Montage Technology Co., Ltd. Device c002
-[0004:00]---00.0-[01]--
-+-[0005:00]---00.0-[01]----00.0  Samsung Electronics Co Ltd Device a810
 +-[0005:40]---00.0-[41]----00.0  Samsung Electronics Co Ltd Device a810
 +-[0005:c0]---00.0-[c1]----00.0  Intel Corporation I210 Gigabit Network Connection
 \-[0005:e0]---00.0-[e1-e2]----00.0-[e2]--+-00.0  ASPEED Technology, Inc. ASPEED Graphics Family
                                          \-02.0  ASPEED Technology, Inc. Device 2603
-+-[0006:00]---00.0-[01]--
 \-[0006:80]---00.0-[81]--
```
The CXL device is on 0003:01:00.0 CXL [0502]: Montage Technology Co.,
Ltd. Device [1b00:c002] (rev 03) and another one is 0003:81:00.0 CXL
[0502]: Montage Technology Co., Ltd. Device [1b00:c002] (rev 03). The
one we are targeting is 0003:01:00.0.

The kernel should be built with CONFIG_FWCTL=y and
CONFIG_CXL_FEATURES=y, otherwise the bug won't be triggered.

With `sudo setpci -v -s 0003:00:00.0 CAP_EXP+0x10.b=0x10:0x10` to bring
its root port link down and the error log in dmesg is like the following
```
[  890.137377] pcieport 0003:00:00.0: pciehp: Slot(0-2): Link Down
[  890.145201] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000020
[  890.145203] Mem abort info:
[  890.145205]   ESR = 0x0000000096000006
[  890.145207]   EC = 0x25: DABT (current EL), IL = 32 bits
[  890.145208]   SET = 0, FnV = 0
[  890.145209]   EA = 0, S1PTW = 0
[  890.145211]   FSC = 0x06: level 2 translation fault
[  890.145212] Data abort info:
[  890.145213]   ISV = 0, ISS = 0x00000006, ISS2 = 0x00000000
[  890.145214]   CM = 0, WnR = 0, TnD = 0, TagAccess = 0
[  890.145215]   GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
[  890.145216] user pgtable: 4k pages, 48-bit VAs, pgdp=00000001d1937000
[  890.145218] [0000000000000020] pgd=08000001d193e403, p4d=08000001d193e403, pud=08000001d193d403, pmd=0000000000000000
[  890.145223] Internal error: Oops: 0000000096000006 [#1]  SMP
[  890.214749] Modules linked in: nft_masq nft_ct nft_reject_ipv4 nf_reject_ipv4 nft_reject act_csum cls_u32 sch_htb nft_chain_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 nf_tables bridge stp llc qrtr cfg80211 binfmt_misc nls_iso8859_1 ast cxl_pmu dax_hmem nvidia_cspmu acpi_power_meter coresight_trbe sbsa_gwdt ipmi_ssif acpi_ipmi cxl_acpi arm_smmuv3_pmu coresight arm_cspmu_module arm_spe_pmu ipmi_devintf ipmi_msghandler cxl_pmem cppc_cpufreq sch_fq_codel dm_multipath nvme_fabrics efi_pstore nfnetlink dmi_sysfs ip_tables x_tables autofs4 btrfs libblake2b raid10 raid456 async_raid6_recov async_memcpy async_pq async_xor async_tx xor xor_neon raid6_pq raid1 raid0 linear mlx5_ib macsec ib_uverbs ib_core mlx5_dpll ghash_ce sm4_ce_gcm nvme sm4_ce_ccm mlx5_core nvme_core sm4_ce mlxfw nvme_keyring sm4_ce_cipher tls sm4 igb nvme_auth sm3_ce arm_smccc_trng i2c_algo_bit hkdf psample aes_neon_bs aes_neon_blk aes_ce_blk
[  890.294573] CPU: 28 UID: 0 PID: 1350 Comm: irq/156-pciehp Not tainted 7.0.0-rc6-richard+ #3 PREEMPT(full)
[  890.304108] Hardware name:  , BIOS buildbrain-gcid-sbios-44706962 Mon Mar 30 03:04:08 PM UTC 2026
[  890.312959] pstate: 61400009 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
[  890.319660] pc : klist_put+0x2c/0x140
[  890.323441] lr : klist_del+0x18/0x38
[  890.327050] sp : ffff80008cfd39e0
[  890.330229] x29: ffff80008cfd39e0 x28: 0000000000000018 x27: ffffd9f0a1a9deb0
[  890.337275] x26: ffffd9f0a1a9a2b8 x25: ffff00009c3ebe78 x24: ffff00009c3ebe00
[  890.344321] x23: ffff00008d1cc800 x22: 0000000000000001 x21: ffff00009c3ebe68
[  890.351367] x20: 0000000000000000 x19: ffff00009afc7188 x18: ffff80008cfb50a8
[  890.358413] x17: 0000000000000000 x16: 0000000000000000 x15: 633d4d4554535953
[  890.365459] x14: 42555300302e306d x13: 5300302e306d656d x12: 5f756d702f302e30
[  890.372504] x11: 0035333033343d4d x10: 0000000000000000 x9 : ffffd9f0a31b9a5c
[  890.379551] x8 : 0101010101010101 x7 : 7fff7f7f7f7f7f7f x6 : 339eff3033746f62
[  890.386597] x5 : 0000000000000000 x4 : 0000000000000000 x3 : 0000000000000000
[  890.393642] x2 : 0000000000000000 x1 : 0000000000000001 x0 : 0000000000000000
[  890.400689] Call trace:
[  890.402923]  klist_put+0x2c/0x140 (P)
[  890.406703]  klist_del+0x18/0x38
[  890.409968]  device_del+0x120/0x3c8
[  890.413148]  cdev_device_del+0x2c/0xa8
[  890.416928]  fwctl_unregister+0x11c/0x128
[  890.420967]  free_memdev_fwctl+0x24/0x50
[  890.424919]  devm_action_release+0x20/0x48
[  890.428786]  release_nodes+0x68/0xc8
[  890.432481]  devres_release_all+0x9c/0x130
[  890.436348]  device_unbind_cleanup+0x24/0xb0
[  890.440730]  device_release_driver_internal+0x234/0x2e0
[  890.445627]  device_release_driver+0x24/0x50
[  890.450096]  pci_stop_bus_device+0x88/0x100
[  890.453962]  pci_stop_and_remove_bus_device+0x24/0x58
[  890.459118]  pciehp_unconfigure_device+0xb4/0x1e0
[  890.463758]  pciehp_disable_slot+0x7c/0x190
[  890.467710]  pciehp_handle_presence_or_link_change+0x94/0x518
[  890.473554]  pciehp_ist+0x1c8/0x310
[  890.476819]  irq_thread_fn+0x38/0xd0
[  890.480599]  irq_thread+0x1ac/0x450
[  890.483779]  kthread+0x13c/0x150
[  890.487215]  ret_from_fork+0x10/0x20
[  890.490739] Code: 12001c36 f9400014 927ffa94 aa1403e0 (f9401295)
[  890.496668] ---[ end trace 0000000000000000 ]---
[  890.552081] genirq: exiting task "irq/156-pciehp" (1350) is an active IRQ thread (irq 156)
```


Best regards,
Richard Cheng.
---
 drivers/fwctl/main.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/fwctl/main.c b/drivers/fwctl/main.c
index bc6378506296..098c3824ad75 100644
--- a/drivers/fwctl/main.c
+++ b/drivers/fwctl/main.c
@@ -415,7 +415,7 @@ static void __exit fwctl_exit(void)
 	unregister_chrdev_region(fwctl_dev, FWCTL_MAX_DEVICES);
 }
 
-module_init(fwctl_init);
+subsys_initcall(fwctl_init);
 module_exit(fwctl_exit);
 MODULE_DESCRIPTION("fwctl device firmware access framework");
 MODULE_LICENSE("GPL");
-- 
2.43.0