On 8/9/21 12:25 AM, kernel test robot wrote:
[...]
[ 9.433105][ T1] BUG: sleeping function called from invalid context at mm/page_alloc.c:5170 [ 9.434513][ T1] in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 1, name: swapper/0 [ 9.436116][ T1] no locks held by swapper/0/1. [ 9.436923][ T1] CPU: 1 PID: 1 Comm: swapper/0 Not tainted 5.14.0-rc4-00161-g08cb589cb3d2 #1 [ 9.438398][ T1] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.12.0-1 04/01/2014 [ 9.439880][ T1] Call Trace: [ 9.440404][ T1] ? dump_stack_lvl (lib/dump_stack.c:106) [ 9.441154][ T1] ? dump_stack (lib/dump_stack.c:113) [ 9.441827][ T1] ? ___might_sleep.cold (kernel/sched/core.c:9155) [ 9.442645][ T1] ? __might_sleep (kernel/sched/core.c:9109 (discriminator 14)) [ 9.444795][ T1] ? __alloc_pages (include/linux/kernel.h:104 mm/page_alloc.c:5170 mm/page_alloc.c:5380) [ 9.445583][ T1] ? mem_encrypt_init+0x1/0x1 [ 9.446389][ T1] ? destroy_args (mm/debug_vm_pgtable.c:1219) [ 9.447193][ T1] ? debug_vm_pgtable (mm/debug_vm_pgtable.c:1208 mm/debug_vm_pgtable.c:1237) [ 9.448076][ T1] ? destroy_args (mm/debug_vm_pgtable.c:1219) [ 9.448863][ T1] ? do_one_initcall (init/main.c:1282) [ 9.449676][ T1] ? parse_args (kernel/params.c:190) [ 9.450470][ T1] ? kernel_init_freeable (init/main.c:1354 init/main.c:1371 init/main.c:1390 init/main.c:1592) [ 9.451358][ T1] ? rest_init (init/main.c:1476) [ 9.452063][ T1] ? kernel_init (init/main.c:1486) [ 9.452759][ T1] ? ret_from_fork (arch/x86/entry/entry_32.S:775) [ 9.453815][ T1] ------------[ cut here ]------------
[...] Thanks for reporting the issue. The same issue was reported against v4 series before, but this time, it's raised on v5 series. Fortunately, this report provides more useful information so that I can reproduce the issue locally and I'm going to fix the issue in v6 since I need to address Anshuman's last comment on v5. As to the root cause, pte_alloc_map() is used in init_args(). The PTE entry is mapped in atomic mode when CONFIG_HIGHPTE is enabled on i386. It means GFP_KERNEL isn't right option to be used by alloc_pages() in init_args(). The fix is use pte_alloc() to replace pte_alloc_map() and args->pte will be mapped when it's used in debug_vm_pgtable() for PTE modifying tests. We need avoid atomic context lasting for too long. Thanks, Gavin
