Re: [PATCH] powerpc/pseries/iommu: Add TCEs for 16GB pages when RAM is pre-mapped

[Harsh Prateek Bora <harshpb@linux.ibm.com>]

On 07/05/26 12:57 am, Gaurav Batra wrote:
> 
> On 5/5/26 11:04 PM, Harsh Prateek Bora wrote:
>>
>>
>> On 06/05/26 2:31 am, Gaurav Batra wrote:
>>>
>>> On 5/5/26 1:49 PM, Harsh Prateek Bora wrote:
>>>>
>>>>
>>>> On 05/05/26 2:24 am, Gaurav Batra wrote:
>>>>> @@ -2431,7 +2437,7 @@ static int iommu_mem_notifier(struct 
>>>>> notifier_block *nb, unsigned long action,
>>>>>           spin_lock(&dma_win_list_lock);
>>>>>           list_for_each_entry(window, &dma_win_list, list) {
>>>>>               if (window->direct && (arg->start_pfn << PAGE_SHIFT) <
>>>>> -                ddw_memory_hotplug_max()) {
>>>>> +                pseries_ddw_max_ram) {
>>>>>                   ret |= tce_setrange_multi_pSeriesLP(arg->start_pfn,
>>>>>                           arg->nr_pages, window->prop);
>>>>
>>>> I think not only start_pfn, but end_pfn also needs to be within allowed
>>>> range, which may require clamping arg->nr_pages if crossing the limits.
>>> The reason to only check for start_pfn is because the range given 
>>> will either be in the RAM or pmemory. It can never cross the boundary
>>
>> Usually as an API, we do not trust the caller, hence the check for
>> start_pfn. However, if at all we need to go ahead with that assumption,
>> may be document that as a code comment and/or in the commit log also?
> 
> I will document it. I am wondering, if instead of capping it to 
> start+nrpages,
> 
> maybe, return an error. If in the future, we get an error, this needs to be
> 
> investigated.
> 

I am not sure erroring out for crossing RAM/pmem boundary is a right
thing to do here if it can hinder persistent memory onlining (since
there is an existing code comment which says it can get called when
onlining persistent memory as well). In that case, a WARN may be a
better choice.


>>
>>>>
>>>>>               }
>>>>> @@ -2444,7 +2450,7 @@ static int iommu_mem_notifier(struct 
>>>>> notifier_block *nb, unsigned long action,
>>>>>           spin_lock(&dma_win_list_lock);
>>>>>           list_for_each_entry(window, &dma_win_list, list) {
>>>>>               if (window->direct && (arg->start_pfn << PAGE_SHIFT) <
>>>>> -                ddw_memory_hotplug_max()) {
>>>>> +                pseries_ddw_max_ram) {
>>>>>                   ret |= tce_clearrange_multi_pSeriesLP(arg- 
>>>>> >start_pfn,
>>>>>                           arg->nr_pages, window->prop);
>>>>>               }
>>>>
>>