Re: [PATCH 15/26] x86-64, NUMA: Unify the rest of memblk registration

[Yinghai Lu <yinghai@kernel.org>]

On 02/12/2011 09:10 AM, Tejun Heo wrote:
> Move the remaining memblk registration logic from acpi_scan_nodes() to
> numa_register_memblks() and initmem_init().
> 
> This applies nodes_cover_memory() sanity check, memory node sorting
> and node_online() checking, which were only applied to acpi, to all
> init methods.
> 
> As all memblk registration is moved to common code, active range
> clearing is moved to initmem_init() too and removed from bad_srat().
> 
> Signed-off-by: Tejun Heo <tj@kernel.org>
> Cc: Yinghai Lu <yinghai@kernel.org>
> Cc: Brian Gerst <brgerst@gmail.com>
> Cc: Cyrill Gorcunov <gorcunov@gmail.com>
> Cc: Shaohui Zheng <shaohui.zheng@intel.com>
> Cc: David Rientjes <rientjes@google.com>
> Cc: Ingo Molnar <mingo@elte.hu>
> Cc: H. Peter Anvin <hpa@linux.intel.com>
> ---
>  arch/x86/mm/amdtopology_64.c |    6 ---
>  arch/x86/mm/numa_64.c        |   71 +++++++++++++++++++++++++++++++++++++++---
>  arch/x86/mm/srat_64.c        |   59 ----------------------------------
>  3 files changed, 66 insertions(+), 70 deletions(-)
> 
> diff --git a/arch/x86/mm/amdtopology_64.c b/arch/x86/mm/amdtopology_64.c
> index 48ec374..9c9f46a 100644
> --- a/arch/x86/mm/amdtopology_64.c
> +++ b/arch/x86/mm/amdtopology_64.c
> @@ -262,11 +262,5 @@ void __init amd_fake_nodes(const struct bootnode *nodes, int nr_nodes)
>  
>  int __init amd_scan_nodes(void)
>  {
> -	int i;
> -
> -	for_each_node_mask(i, node_possible_map)
> -		setup_node_bootmem(i, numa_nodes[i].start, numa_nodes[i].end);
> -
> -	numa_init_array();
>  	return 0;
>  }
> diff --git a/arch/x86/mm/numa_64.c b/arch/x86/mm/numa_64.c
> index 2e2ca94..062649d 100644
> --- a/arch/x86/mm/numa_64.c
> +++ b/arch/x86/mm/numa_64.c
> @@ -287,6 +287,37 @@ setup_node_bootmem(int nodeid, unsigned long start, unsigned long end)
>  	node_set_online(nodeid);
>  }
>  
> +/*
> + * Sanity check to catch more bad NUMA configurations (they are amazingly
> + * common).  Make sure the nodes cover all memory.
> + */
> +static int __init nodes_cover_memory(const struct bootnode *nodes)
> +{
> +	unsigned long numaram, e820ram;
> +	int i;
> +
> +	numaram = 0;
> +	for_each_node_mask(i, mem_nodes_parsed) {
> +		unsigned long s = nodes[i].start >> PAGE_SHIFT;
> +		unsigned long e = nodes[i].end >> PAGE_SHIFT;
> +		numaram += e - s;
> +		numaram -= __absent_pages_in_range(i, s, e);
> +		if ((long)numaram < 0)
> +			numaram = 0;
> +	}
> +
> +	e820ram = max_pfn -
> +		(memblock_x86_hole_size(0, max_pfn<<PAGE_SHIFT) >> PAGE_SHIFT);
> +	/* We seem to lose 3 pages somewhere. Allow 1M of slack. */
> +	if ((long)(e820ram - numaram) >= (1<<(20 - PAGE_SHIFT))) {
> +		printk(KERN_ERR "NUMA: nodes only cover %luMB of your %luMB e820 RAM. Not used.\n",
> +			(numaram << PAGE_SHIFT) >> 20,
> +			(e820ram << PAGE_SHIFT) >> 20);
> +		return 0;
> +	}
> +	return 1;
> +}
> +
>  static int __init numa_register_memblks(void)
>  {
>  	int i;
> @@ -349,6 +380,25 @@ static int __init numa_register_memblks(void)
>  		memblock_x86_register_active_regions(memblk_nodeid[i],
>  				node_memblk_range[i].start >> PAGE_SHIFT,
>  				node_memblk_range[i].end >> PAGE_SHIFT);
> +
> +	/* for out of order entries */
> +	sort_node_map();
> +	if (!nodes_cover_memory(numa_nodes))
> +		return -EINVAL;
> +
> +	/* Finally register nodes. */
> +	for_each_node_mask(i, node_possible_map)
> +		setup_node_bootmem(i, numa_nodes[i].start, numa_nodes[i].end);
> +
> +	/*
> +	 * Try again in case setup_node_bootmem missed one due to missing
> +	 * bootmem.
> +	 */
> +	for_each_node_mask(i, node_possible_map)
> +		if (!node_online(i))
> +			setup_node_bootmem(i, numa_nodes[i].start,
> +					   numa_nodes[i].end);
> +
>  	return 0;
>  }

please don't put setup_node_bootmem calling into numa_register_memblks()
that is not related.

put the calling in initmem_init() directly is more reasonable.

>  
> @@ -713,15 +763,14 @@ static int dummy_numa_init(void)
>  	node_set(0, cpu_nodes_parsed);
>  	node_set(0, mem_nodes_parsed);
>  	numa_add_memblk(0, 0, (u64)max_pfn << PAGE_SHIFT);
> +	numa_nodes[0].start = 0;
> +	numa_nodes[0].end = (u64)max_pfn << PAGE_SHIFT;
>  
>  	return 0;
>  }
>  
>  static int dummy_scan_nodes(void)
>  {
> -	setup_node_bootmem(0, 0, max_pfn << PAGE_SHIFT);
> -	numa_init_array();
> -
>  	return 0;
>  }
>  
> @@ -757,6 +806,7 @@ void __init initmem_init(void)
>  		memset(node_memblk_range, 0, sizeof(node_memblk_range));
>  		memset(memblk_nodeid, 0, sizeof(memblk_nodeid));
>  		memset(numa_nodes, 0, sizeof(numa_nodes));
> +		remove_all_active_ranges();
>  
>  		if (numa_init[i]() < 0)
>  			continue;
> @@ -781,8 +831,19 @@ void __init initmem_init(void)
>  		if (numa_register_memblks() < 0)
>  			continue;
>  
> -		if (!scan_nodes[i]())
> -			return;
> +		if (scan_nodes[i]() < 0)
> +			continue;
> +
> +		for (j = 0; j < nr_cpu_ids; j++) {
> +			int nid = early_cpu_to_node(j);
> +
> +			if (nid == NUMA_NO_NODE)
> +				continue;
> +			if (!node_online(nid))
> +				numa_clear_node(j);
> +		}
> +		numa_init_array();
> +		return;
>  	}
>  	BUG();
>  }
> diff --git a/arch/x86/mm/srat_64.c b/arch/x86/mm/srat_64.c
> index 755d157..4a2c33b 100644
> --- a/arch/x86/mm/srat_64.c
> +++ b/arch/x86/mm/srat_64.c
> @@ -44,7 +44,6 @@ static __init void bad_srat(void)
>  		numa_nodes[i].start = numa_nodes[i].end = 0;
>  		nodes_add[i].start = nodes_add[i].end = 0;
>  	}
> -	remove_all_active_ranges();
>  }
>  
>  static __init inline int srat_disabled(void)
> @@ -259,35 +258,6 @@ acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *ma)
>  		update_nodes_add(node, start, end);
>  }
>  
> -/* Sanity check to catch more bad SRATs (they are amazingly common).
> -   Make sure the PXMs cover all memory. */
> -static int __init nodes_cover_memory(const struct bootnode *nodes)
> -{
> -	int i;
> -	unsigned long pxmram, e820ram;
> -
> -	pxmram = 0;
> -	for_each_node_mask(i, mem_nodes_parsed) {
> -		unsigned long s = nodes[i].start >> PAGE_SHIFT;
> -		unsigned long e = nodes[i].end >> PAGE_SHIFT;
> -		pxmram += e - s;
> -		pxmram -= __absent_pages_in_range(i, s, e);
> -		if ((long)pxmram < 0)
> -			pxmram = 0;
> -	}
> -
> -	e820ram = max_pfn - (memblock_x86_hole_size(0, max_pfn<<PAGE_SHIFT)>>PAGE_SHIFT);
> -	/* We seem to lose 3 pages somewhere. Allow 1M of slack. */
> -	if ((long)(e820ram - pxmram) >= (1<<(20 - PAGE_SHIFT))) {
> -		printk(KERN_ERR
> -	"SRAT: PXMs only cover %luMB of your %luMB e820 RAM. Not used.\n",
> -			(pxmram << PAGE_SHIFT) >> 20,
> -			(e820ram << PAGE_SHIFT) >> 20);
> -		return 0;
> -	}
> -	return 1;
> -}
> -
>  void __init acpi_numa_arch_fixup(void) {}
>  
>  int __init x86_acpi_numa_init(void)
> @@ -303,37 +273,8 @@ int __init x86_acpi_numa_init(void)
>  /* Use the information discovered above to actually set up the nodes. */
>  int __init acpi_scan_nodes(void)
>  {
> -	int i;
> -
>  	if (acpi_numa <= 0)
>  		return -1;
> -
> -	/* for out of order entries in SRAT */
> -	sort_node_map();
> -	if (!nodes_cover_memory(numa_nodes)) {
> -		bad_srat();
> -		return -1;
> -	}
> -
> -	/* Finally register nodes */
> -	for_each_node_mask(i, node_possible_map)
> -		setup_node_bootmem(i, numa_nodes[i].start, numa_nodes[i].end);
> -	/* Try again in case setup_node_bootmem missed one due
> -	   to missing bootmem */
> -	for_each_node_mask(i, node_possible_map)
> -		if (!node_online(i))
> -			setup_node_bootmem(i, numa_nodes[i].start,
> -					   numa_nodes[i].end);
> -
> -	for (i = 0; i < nr_cpu_ids; i++) {
> -		int node = early_cpu_to_node(i);
> -
> -		if (node == NUMA_NO_NODE)
> -			continue;
> -		if (!node_online(node))
> -			numa_clear_node(i);
> -	}
> -	numa_init_array();
>  	return 0;
>  }
>