Re: [PATCH v4 01/13] clk: microchip: mpfs: fix clk_cfg array bounds violation

From: Conor.Dooley
Date: Fri Sep 09 2022 - 07:02:58 EST


On 08/09/2022 07:48, Conor Dooley wrote:
> On 08/09/2022 07:44, Claudiu Beznea - M18063 wrote:
>> On 30.08.2022 15:52, Conor Dooley wrote:
>>> Unnoticed in current code, there is an array bounds violation present
>>> during clock registration. This seems to fail gracefully in v6.0-rc1,
>>> and life carrys on. While converting the driver to use standard clock
>>> structs/ops, kernel panics were seen during boot when built with clang:
>>>
>>> [    0.581754] Unable to handle kernel NULL pointer dereference at virtual address 00000000000000b1
>>> [    0.591520] Oops [#1]
>>> [    0.594045] Modules linked in:
>>> [    0.597435] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 6.0.0-rc1-00011-g8e1459cf4eca #1
>>> [    0.606188] Hardware name: Microchip PolarFire-SoC Icicle Kit (DT)
>>> [    0.613012] epc : __clk_register+0x4a6/0x85c
>>> [    0.617759]  ra : __clk_register+0x49e/0x85c
>>> [    0.622489] epc : ffffffff803faf7c ra : ffffffff803faf74 sp : ffffffc80400b720
>>> [    0.630466]  gp : ffffffff810e93f8 tp : ffffffe77fe60000 t0 : ffffffe77ffb3800
>>> [    0.638443]  t1 : 000000000000000a t2 : ffffffffffffffff s0 : ffffffc80400b7c0
>>> [    0.646420]  s1 : 0000000000000001 a0 : 0000000000000001 a1 : 0000000000000000
>>> [    0.654396]  a2 : 0000000000000001 a3 : 0000000000000000 a4 : 0000000000000000
>>> [    0.662373]  a5 : ffffffff803a5810 a6 : 0000000200000022 a7 : 0000000000000006
>>> [    0.670350]  s2 : ffffffff81099d48 s3 : ffffffff80d6e28e s4 : 0000000000000028
>>> [    0.678327]  s5 : ffffffff810ed3c8 s6 : ffffffff810ed3d0 s7 : ffffffe77ffbc100
>>> [    0.686304]  s8 : ffffffe77ffb1540 s9 : ffffffe77ffb1540 s10: 0000000000000008
>>> [    0.694281]  s11: 0000000000000000 t3 : 00000000000000c6 t4 : 0000000000000007
>>> [    0.702258]  t5 : ffffffff810c78c0 t6 : ffffffe77ff88cd0
>>> [    0.708125] status: 0000000200000120 badaddr: 00000000000000b1 cause: 000000000000000d
>>> [    0.716869] [<ffffffff803fb892>] devm_clk_hw_register+0x62/0xaa
>>> [    0.723420] [<ffffffff80403412>] mpfs_clk_probe+0x1e0/0x244
>>>
>>> It fails on "clk_periph_timer" - which uses a different parent, that it
>>> tries to find using the macro:
>>> \#define PARENT_CLK(PARENT) (&mpfs_cfg_clks[CLK_##PARENT].cfg.hw)
>>>
>>> If parent is RTCREF, so the macro becomes: &mpfs_cfg_clks[33].cfg.hw
>>> which is well beyond the end of the array. Amazingly, builds with GCC
>>> 11.1 see no problem here, booting correctly and hooking the parent up
>>> etc. Builds with clang-15 do not, with the above panic.
>>>
>>> Drop the macro for the RTCREF and use the array directly to avoid the
>>> panic, using a newly added define that brings the index into the valid
>>> range.
>>>
>>> Fixes: 1c6a7ea32b8c ("clk: microchip: mpfs: add RTCREF clock control")
>>> CC: Nathan Chancellor <nathan@xxxxxxxxxx>
>>> Signed-off-by: Conor Dooley <conor.dooley@xxxxxxxxxxxxx>
>>> ---
>>>   drivers/clk/microchip/clk-mpfs.c | 5 ++++-
>>>   1 file changed, 4 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/clk/microchip/clk-mpfs.c b/drivers/clk/microchip/clk-mpfs.c
>>> index 070c3b896559..9e41f07b3fa6 100644
>>> --- a/drivers/clk/microchip/clk-mpfs.c
>>> +++ b/drivers/clk/microchip/clk-mpfs.c
>>> @@ -27,6 +27,8 @@
>>>   #define MSSPLL_POSTDIV_WIDTH    0x07u
>>>   #define MSSPLL_FIXED_DIV    4u
>>> +#define RTCREF_OFFSET        (CLK_RTCREF - CLK_ENVM)
>>> +
>>>   struct mpfs_clock_data {
>>>       void __iomem *base;
>>>       void __iomem *msspll_base;
>>> @@ -381,7 +383,8 @@ static struct mpfs_periph_hw_clock mpfs_periph_clks[] = {
>>>       CLK_PERIPH(CLK_MAC0, "clk_periph_mac0", PARENT_CLK(AHB), 1, 0),
>>>       CLK_PERIPH(CLK_MAC1, "clk_periph_mac1", PARENT_CLK(AHB), 2, 0),
>>>       CLK_PERIPH(CLK_MMC, "clk_periph_mmc", PARENT_CLK(AHB), 3, 0),
>>> -    CLK_PERIPH(CLK_TIMER, "clk_periph_timer", PARENT_CLK(RTCREF), 4, 0),
>>> +    CLK_PERIPH(CLK_TIMER, "clk_periph_timer",
>>> +           &mpfs_cfg_clks[CLK_RTCREF - RTCREF_OFFSET].hw, 4, 0),
>>
>> A personal taste: as clk IDs and clk indexes in mpfs_cfg_clks[] array are
>> different thing (ID for clk_periph_timer is already different) and the
>> CLK_RTCREF - RTCREF_OFFSET here is in the end CLK_ENVM = 3 maybe easier to
>> follow the code would be to add new macros like:
>>
>> #define CLK_CPU_OFF 0
>> #define CLK_AXI_OFF 1
>> #define CLK_AHB_OFF 2
>> #define CLK_RTCREF_OFF 3
>>
>> and change the CLK_PARENT() macro something as follows:
>>
>> #define PARENT_CLK(PARENT) (&mpfs_cfg_clks[CLK_##PARENT##_OFF].hw)
>
> Sure, but that is out-of-scope for this fix which needs backporting.

I think I mis-interpreted your suggestion. Gonna give it a go & unless
clang screams and yells at me I think I like it more than relying on
the relationships between the binding numbers.

Thanks,
Conor.