On 11/01/2017 12:56 PM, Thomas Gleixner wrote:
> On Wed, 25 Oct 2017, Prarit Bhargava wrote:
>> diff --git a/arch/x86/include/asm/processor.h
>> b/arch/x86/include/asm/processor.h
>> index b390ff76e58f..f4ab1edf4e24 100644
>> --- a/arch/x86/include/asm/processor.h
>> +++ b/arch/x86/include/asm/processor.h
>> @@ -124,8 +124,10 @@ struct cpuinfo_x86 {
>> u16 booted_cores;
>> /* Physical processor id: */
>> u16 phys_proc_id;
>> - /* Logical processor id: */
>> + /* Logical processor (package) id: */
>> u16 logical_proc_id;
>> + /* Physical package ID */
>> + u16 phys_pkg_id;
>
> How is this new field used aside of being written to and how is it
> different from phys_proc_id? AFAICT, it's the same as all callers to
> topology_update_package_map() are handing in cpu_data->phys_proc_id.
>
I've removed this in v5.
>> +/**
>> + * topology_phys_to_logical_pkg - Map a physical package id to a logical
>> + *
>> + * Returns logical package id or -1 if not found
>> + */
>> +int topology_phys_to_logical_pkg(unsigned int phys_pkg)
>> +{
>> + int log_pkg;
>> +
>> + for (log_pkg = 0; log_pkg < logical_packages; log_pkg++)
>> + if (logical_to_physical_pkg_map[log_pkg] == phys_pkg)
>> + return log_pkg;
>> +
>> + return -1;
>> +}
>> +EXPORT_SYMBOL(topology_phys_to_logical_pkg);
>
> ....
>
>> +
>> + /* Allocate and copy a new array */
>> + ltp_pkg_map_new = kmalloc(logical_packages * sizeof(u16), GFP_KERNEL);
>> + BUG_ON(!ltp_pkg_map_new);
>> + if (logical_to_physical_pkg_map) {
>> + memcpy(ltp_pkg_map_new, logical_to_physical_pkg_map,
>> + logical_packages * sizeof(u16));
>> + kfree(logical_to_physical_pkg_map);
>> }
>> - physical_to_logical_pkg[pkg] = new;
>> + logical_to_physical_pkg_map = ltp_pkg_map_new;
>
> This lacks serialization and is therefore broken against a concurrent
> topology_phys_to_logical_pkg() call for obvious reasons. The current user
> is probably safe, but this really needs to be fixed now.
I'm using a spin_lock_irq in v5.
I am finishing testing on 4S & 8S systems now.
P.
>
> Thanks,
>
> tglx
>
