powerpc/set_memory: Avoid spinlock recursion in change_page_attr()
authorChristophe Leroy <christophe.leroy@csgroup.eu>
Fri, 24 Dec 2021 11:07:33 +0000 (11:07 +0000)
committerMichael Ellerman <mpe@ellerman.id.au>
Sat, 12 Feb 2022 11:47:42 +0000 (22:47 +1100)
commita4c182ecf33584b9b2d1aa9dad073014a504c01f
treee80b8d18ea36178a39711c62dac75fe6c1558544
parent4ee83a2cfbc46c13f2a08fe6d48dbcede53cdbf8
powerpc/set_memory: Avoid spinlock recursion in change_page_attr()

Commit 1f9ad21c3b38 ("powerpc/mm: Implement set_memory() routines")
included a spin_lock() to change_page_attr() in order to
safely perform the three step operations. But then
commit 9f7853d7609d ("powerpc/mm: Fix set_memory_*() against
concurrent accesses") modify it to use pte_update() and do
the operation safely against concurrent access.

In the meantime, Maxime reported some spinlock recursion.

[   15.351649] BUG: spinlock recursion on CPU#0, kworker/0:2/217
[   15.357540]  lock: init_mm+0x3c/0x420, .magic: dead4ead, .owner: kworker/0:2/217, .owner_cpu: 0
[   15.366563] CPU: 0 PID: 217 Comm: kworker/0:2 Not tainted 5.15.0+ #523
[   15.373350] Workqueue: events do_free_init
[   15.377615] Call Trace:
[   15.380232] [e4105ac0] [800946a4] do_raw_spin_lock+0xf8/0x120 (unreliable)
[   15.387340] [e4105ae0] [8001f4ec] change_page_attr+0x40/0x1d4
[   15.393413] [e4105b10] [801424e0] __apply_to_page_range+0x164/0x310
[   15.400009] [e4105b60] [80169620] free_pcp_prepare+0x1e4/0x4a0
[   15.406045] [e4105ba0] [8016c5a0] free_unref_page+0x40/0x2b8
[   15.411979] [e4105be0] [8018724c] kasan_depopulate_vmalloc_pte+0x6c/0x94
[   15.418989] [e4105c00] [801424e0] __apply_to_page_range+0x164/0x310
[   15.425451] [e4105c50] [80187834] kasan_release_vmalloc+0xbc/0x134
[   15.431898] [e4105c70] [8015f7a8] __purge_vmap_area_lazy+0x4e4/0xdd8
[   15.438560] [e4105d30] [80160d10] _vm_unmap_aliases.part.0+0x17c/0x24c
[   15.445283] [e4105d60] [801642d0] __vunmap+0x2f0/0x5c8
[   15.450684] [e4105db0] [800e32d0] do_free_init+0x68/0x94
[   15.456181] [e4105dd0] [8005d094] process_one_work+0x4bc/0x7b8
[   15.462283] [e4105e90] [8005d614] worker_thread+0x284/0x6e8
[   15.468227] [e4105f00] [8006aaec] kthread+0x1f0/0x210
[   15.473489] [e4105f40] [80017148] ret_from_kernel_thread+0x14/0x1c

Remove the read / modify / write sequence to make the operation atomic
and remove the spin_lock() in change_page_attr().

To do the operation atomically, we can't use pte modification helpers
anymore. Because all platforms have different combination of bits, it
is not easy to use those bits directly. But all have the
_PAGE_KERNEL_{RO/ROX/RW/RWX} set of flags. All we need it to compare
two sets to know which bits are set or cleared.

For instance, by comparing _PAGE_KERNEL_ROX and _PAGE_KERNEL_RO you
know which bit gets cleared and which bit get set when changing exec
permission.

Reported-by: Maxime Bizon <mbizon@freebox.fr>
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/all/20211212112152.GA27070@sakura/
Link: https://lore.kernel.org/r/43c3c76a1175ae6dc1a3d3b5c3f7ecb48f683eea.1640344012.git.christophe.leroy@csgroup.eu
arch/powerpc/mm/pageattr.c