35dcae535a
Compiling clock driver with CONFIG_UBSAN enabled shows the following trace:
UBSAN: shift-out-of-bounds in drivers/clk/ralink/clk-mt7621.c:121:15
shift exponent 131072 is too large for 32-bit type 'long unsigned int'
CPU: 1 PID: 1 Comm: swapper/0 Not tainted 5.15.86 #0
Stack : ...
Call Trace:
[<80009a58>] show_stack+0x38/0x118
[<8045ce04>] dump_stack_lvl+0x60/0x80
[<80458868>] ubsan_epilogue+0x10/0x54
[<804590e0>] __ubsan_handle_shift_out_of_bounds+0x118/0x190
[<804c9a10>] mt7621_gate_is_enabled+0x98/0xa0
[<804bb774>] clk_core_is_enabled+0x34/0x90
[<80aad73c>] clk_disable_unused_subtree+0x98/0x1e4
[<80aad6d4>] clk_disable_unused_subtree+0x30/0x1e4
[<80aad6d4>] clk_disable_unused_subtree+0x30/0x1e4
[<80aad900>] clk_disable_unused+0x78/0x120
[<80002030>] do_one_initcall+0x54/0x1f0
[<80a922a4>] kernel_init_freeable+0x280/0x31c
[<808047c4>] kernel_init+0x20/0x118
[<80003e58>] ret_from_kernel_thread+0x14/0x1c
Shifting a value (131032) larger than the type (32 bit unsigned integer)
is undefined behaviour in C.
The problem is in 'mt7621_gate_is_enabled()' function which is using the
'BIT()' kernel macro with the bit index for the clock gate to check if the
bit is set. When the clock gates structure is created driver is already
setting 'bit_idx' using 'BIT()' macro, so we are wrongly applying an extra
'BIT()' mask here. Removing it solve the problem and makes this function
correct. However when clock gating is correctly working, the kernel starts
disabling those clocks that are not requested. Some drivers for this SoC
are older than this clock driver itself. So to avoid the kernel to disable
clocks that have been enabled until now, we must apply 'CLK_IS_CRITICAL'
flag on gates initialization code.
Fixes:
|
||
|---|---|---|
| .. | ||
| clk-mt7621.c | ||
| Kconfig | ||
| Makefile | ||