Re: [PATCH v3] mm/alloc_tag: Fix the kmemleak false positive issue in the allocation of the percpu variable tag->counters
From: Catalin Marinas
Date: Fri Jun 20 2025 - 06:44:27 EST
On Fri, Jun 20, 2025 at 05:31:02PM +0800, Hao Ge wrote:
> From: Hao Ge <gehao@xxxxxxxxxx>
>
> When loading a module, as long as the module has memory
> allocation operations, kmemleak produces a false positive
> report that resembles the following:
>
> unreferenced object (percpu) 0x7dfd232a1650 (size 16):
> comm "modprobe", pid 1301, jiffies 4294940249
> hex dump (first 16 bytes on cpu 2):
> 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
> backtrace (crc 0):
> kmemleak_alloc_percpu+0xb4/0xd0
> pcpu_alloc_noprof+0x700/0x1098
> load_module+0xd4/0x348
> codetag_module_init+0x20c/0x450
> codetag_load_module+0x70/0xb8
> load_module+0xef8/0x1608
> init_module_from_file+0xec/0x158
> idempotent_init_module+0x354/0x608
> __arm64_sys_finit_module+0xbc/0x150
> invoke_syscall+0xd4/0x258
> el0_svc_common.constprop.0+0xb4/0x240
> do_el0_svc+0x48/0x68
> el0_svc+0x40/0xf8
> el0t_64_sync_handler+0x10c/0x138
> el0t_64_sync+0x1ac/0x1b0
>
> This is because the module can only indirectly reference alloc_tag_counters
> through the alloc_tag section, which misleads kmemleak.
>
> However, we don't have a kmemleak ignore interface for percpu
> allocations yet. So let's create one and invoke it for tag->counters.
>
> Fixes: 12ca42c23775 ("alloc_tag: allocate percpu counters for module tags dynamically")
> Signed-off-by: Hao Ge <gehao@xxxxxxxxxx>
Reviewed-by: Catalin Marinas <catalin.marinas@xxxxxxx>