[PATCH] riscv: Report error when repeatedly recording CPU hardware ID

[Conor Dooley]

On Sat, Aug 24, 2024 at 10:41:28AM +0800, qiaozhe wrote:
> 
> 
> 
> On 2024/8/23 20:57, Conor Dooley wrote:
> 
> > On Fri, Aug 23, 2024 at 05:11:00PM +0800, Zhe Qiao wrote:
> >> In the of_parse_and_init_cpus() function, when the __cpuid_to_hartid_map[]
> >> array records the CPU hardware ID, if the same CPU hardware attribute has 
> >> been recorded, an error report is issued, thereby ensuring the uniqueness
> >> of the CPU hardware ID recorded in the __cpuid_to_hartid_map[] array.
> > Why is this actually required? On what system did you encounter this?
> 
> This is not actually a patch submitted for problems encountered in actual 
> development environments, but rather a comparison of ARM architecture when 
> I was learning Linux kernel and found similar judgments on ARM architecture.

Okay, it's good that you didn't find such a bad dtb "in the wild" :)

> In addition, if the same attribute exists on the CPU hardware ID and is 
> recorded in __cpuid_to_hartid_map[], the kernel may need to make a judgment 
> on this error.
> 
> 
> >> Signed-off-by: Zhe Qiao <qiaozhe at iscas.ac.cn>
> >> ---
> >>  arch/riscv/kernel/smpboot.c | 16 ++++++++++++++++
> >>  1 file changed, 16 insertions(+)
> >>
> >> diff --git a/arch/riscv/kernel/smpboot.c b/arch/riscv/kernel/smpboot.c
> >> index 0f8f1c95ac38..698f9fe791f7 100644
> >> --- a/arch/riscv/kernel/smpboot.c
> >> +++ b/arch/riscv/kernel/smpboot.c
> >> @@ -118,6 +118,16 @@ static void __init acpi_parse_and_init_cpus(void)
> >>  #define acpi_parse_and_init_cpus(...)	do { } while (0)
> >>  #endif
> >>  
> >> +static bool __init is_mpidr_duplicate(unsigned int cpuid, u64 hart)
> >> +{
> >> +	unsigned int i;
> >> +
> >> +	for (i = 1; (i < cpuid) && (i < NR_CPUS); i++)
> >> +		if (cpuid_to_hartid_map(i) == hart)
> >> +			return true;
> >> +	return false;
> >> +}
> >> +
> >>  static void __init of_parse_and_init_cpus(void)
> >>  {
> >>  	struct device_node *dn;
> >> @@ -131,6 +141,12 @@ static void __init of_parse_and_init_cpus(void)
> >>  		if (rc < 0)
> >>  			continue;
> >>  
> >> +		if (is_mpidr_duplicate(cpuid, hart)) {
> >> +			pr_err("%pOF: duplicate cpu reg properties in the DT\n",
> >> +				dn);
> >> +			continue;
> > 
> > Why would we continue in this case? If the devicetree is this broken,
> > why shouldn't we just BUG() and abort immediately?
> 
> This is because I did not find any judgment on this issue in the previous code 
> during the analysis process, so I did not take more aggressive measures in this 
> regard, but only issued an error alarm.

What do you think though? Should we continue to boot in this case?
If you read the function a bit further, you'll see that we abort boot
if there are two instances of the boot CPU. Do you think the same should
be done for all CPUs?


Thanks,
Conor.

> 
> >> +		}
> >> +
> >>  		if (hart == cpuid_to_hartid_map(0)) {
> >>  			BUG_ON(found_boot_cpu);
> >>  			found_boot_cpu = 1;
> >> -- 
> >> 2.43.0
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-riscv/attachments/20240826/c2dcf22e/attachment.sig>