Message-ID: <cb57eb20-4812-674c-5488-277449cc06f8@oracle.com>
Date:   Mon, 30 Aug 2021 16:51:34 -0700
From:   Mike Kravetz <mike.kravetz@...cle.com>
To:     yaozhenguo <yaozhenguo1@...il.com>
Cc:     corbet@....net, akpm@...ux-foundation.org, yaozhenguo@...com,
        willy@...radead.org, linux-kernel@...r.kernel.org,
        linux-doc@...r.kernel.org, linux-mm@...ck.org
Subject: Re: [v2] hugetlbfs: Extend the definition of hugepages parameter to
 support node allocation

Thank you for continuing to work on this.  Some comments below.

On 8/26/21 6:58 PM, yaozhenguo wrote:
> We can specify the number of hugepages to allocate at boot. But the
> hugepages is balanced in all nodes at present. In some scenarios,
> we only need hugepags in one node. For example: DPDK needs hugepages
> which is in the same node as NIC. if DPDK needs four hugepags of 1G
> size in node1 and system has 16 numa nodes. We must reserve 64 hugepags
> in kernel cmdline. But, only four hugepages is used. The others should
> be free after boot.If the system memory is low(for example: 64G), it will
> be an impossible task. So, extend hugepages kernel parameter to specify
> node number of hugepages to allocate at boot.
> For example add following parameter:
> 
> hugepagesz=1G hugepages=0:1,1:3
> 
> It will allocate 1 hugepags in node0 and 3 hugepages in node1.
                     hugepages
> 
> Signed-off-by: yaozhenguo <yaozhenguo1@...il.com>
> ---
> v2: 1. add checking for max node to avoid array out of bounds.
>     2. fix wrong max_huge_pages after failed allocation.
>     3. fix wrong behavior when parsing parameter: hugepages=0:1,2,3:4.
>     4. return 0 when parsing invalid parameter in hugepages_setup
> ---
>  .../admin-guide/kernel-parameters.txt         |   8 +-
>  include/linux/hugetlb.h                       |   1 +
>  mm/hugetlb.c                                  | 133 ++++++++++++++++--
>  3 files changed, 126 insertions(+), 16 deletions(-)
> 
> diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> index bdb22006f..64a128924 100644
> --- a/Documentation/admin-guide/kernel-parameters.txt
> +++ b/Documentation/admin-guide/kernel-parameters.txt
> @@ -1588,9 +1588,11 @@
>  			the number of pages of hugepagesz to be allocated.
>  			If this is the first HugeTLB parameter on the command
>  			line, it specifies the number of pages to allocate for
> -			the default huge page size.  See also
> -			Documentation/admin-guide/mm/hugetlbpage.rst.
> -			Format: <integer>
> +			the default huge page size. If using node format, It
> +			specifies numbers of hugepage in a specific node.
> +			See also Documentation/admin-guide/mm/hugetlbpage.rst.

Documentation/admin-guide/mm/hugetlbpage.rst also describes the
hugepages parameter.  It should also be updated.

> +			Format: <integer> or (node format)
> +				<node>:<numbers>[,<node>:<numbers>]
>  
>  	hugepagesz=
>  			[HW] The size of the HugeTLB pages.  This is used in
> diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
> index f7ca1a387..5939ecd4f 100644
> --- a/include/linux/hugetlb.h
> +++ b/include/linux/hugetlb.h
> @@ -605,6 +605,7 @@ struct hstate {
>  	unsigned long nr_overcommit_huge_pages;
>  	struct list_head hugepage_activelist;
>  	struct list_head hugepage_freelists[MAX_NUMNODES];
> +	unsigned int max_huge_pages_node[MAX_NUMNODES];
>  	unsigned int nr_huge_pages_node[MAX_NUMNODES];
>  	unsigned int free_huge_pages_node[MAX_NUMNODES];
>  	unsigned int surplus_huge_pages_node[MAX_NUMNODES];
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index dfc940d52..9e9d94b2a 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -66,6 +66,7 @@ static struct hstate * __initdata parsed_hstate;
>  static unsigned long __initdata default_hstate_max_huge_pages;
>  static bool __initdata parsed_valid_hugepagesz = true;
>  static bool __initdata parsed_default_hugepagesz;
> +static unsigned int default_hugepages_in_node[MAX_NUMNODES] __initdata;
>  
>  /*
>   * Protects updates to hugepage_freelists, hugepage_activelist, nr_huge_pages,
> @@ -2842,10 +2843,65 @@ static void __init gather_bootmem_prealloc(void)
>  	}
>  }
>  
> +static void __init hugetlb_hstate_alloc_pages_onenode(struct hstate *h, int nid)
> +{
> +	unsigned long i;
> +	char buf[32];
> +
> +	for (i = 0; i < h->max_huge_pages_node[nid]; ++i) {
> +		if (hstate_is_gigantic(h)) {
> +			struct huge_bootmem_page *m;
> +			void *addr;
> +
> +			addr = memblock_alloc_try_nid_raw(
> +					huge_page_size(h), huge_page_size(h),
> +					0, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
> +			if (!addr)
> +				break;
> +			m = addr;
> +			BUG_ON(!IS_ALIGNED(virt_to_phys(m), huge_page_size(h)));
> +			/* Put them into a private list first because mem_map is not up yet */
> +			INIT_LIST_HEAD(&m->list);
> +			list_add(&m->list, &huge_boot_pages);
> +			m->hstate = h;
> +		} else {
> +			struct page *page;
> +
> +			gfp_t gfp_mask = htlb_alloc_mask(h) | __GFP_THISNODE;
> +
> +			page = alloc_fresh_huge_page(h, gfp_mask, nid,
> +					&node_states[N_MEMORY], NULL);
> +			if (!page)
> +				break;
> +			put_page(page); /* free it into the hugepage allocator */
> +		}
> +	}
> +	if (i == h->max_huge_pages_node[nid])
> +		return;
> +
> +	string_get_size(huge_page_size(h), 1, STRING_UNITS_2, buf, 32);
> +	pr_warn("HugeTLB: allocating %u of page size %s failed node%d.  Only allocated %lu hugepages.\n",
> +		h->max_huge_pages_node[nid], buf, nid, i);
> +	h->max_huge_pages_node[nid] = i;
> +	h->max_huge_pages -= (h->max_huge_pages_node[nid] - i);
> +}
> +
>  static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
>  {
>  	unsigned long i;
>  	nodemask_t *node_alloc_noretry;
> +	bool hugetlb_node_set = false;
> +
> +	/* do node alloc */
> +	for (i = 0; i < nodes_weight(node_states[N_MEMORY]); i++) {
> +		if (h->max_huge_pages_node[i] > 0) {
> +			hugetlb_hstate_alloc_pages_onenode(h, i);
> +			hugetlb_node_set = true;
> +		}
> +	}

How does this interact with the hugetlb_cma parameter?  Currently, boot
time allocation of gigantic pages is skipped if hugetlb_cma is
specified.  Looks like the above code will allocate them.

This also brings up the quesiton,  Should hugetlb_cma be changed to be
node specific as well?

For now, I would keep the current behavior and skip all boot time
allocation of gigantic pages if hugetlb_cma.

> +
> +	if (hugetlb_node_set)
> +		return;
>  
>  	if (!hstate_is_gigantic(h)) {
>  		/*
> @@ -3580,6 +3636,9 @@ static int __init hugetlb_init(void)
>  				default_hstate_max_huge_pages;
>  		}
>  	}
> +	for (i = 0; i < nodes_weight(node_states[N_MEMORY]); i++)
> +		if (default_hugepages_in_node[i] > 0)
> +			default_hstate.max_huge_pages_node[i] = default_hugepages_in_node[i];
>  
>  	hugetlb_cma_check();
>  	hugetlb_init_hstates();
> @@ -3649,6 +3708,11 @@ static int __init hugepages_setup(char *s)
>  {
>  	unsigned long *mhp;
>  	static unsigned long *last_mhp;
> +	unsigned int node = NUMA_NO_NODE;
> +	int ret;
> +	int count;
> +	unsigned long tmp;
> +	char *p = s;
>  
>  	if (!parsed_valid_hugepagesz) {
>  		pr_warn("HugeTLB: hugepages=%s does not follow a valid hugepagesz, ignoring\n", s);
> @@ -3656,25 +3720,68 @@ static int __init hugepages_setup(char *s)
>  		return 0;
>  	}
>  
> -	/*
> -	 * !hugetlb_max_hstate means we haven't parsed a hugepagesz= parameter
> -	 * yet, so this hugepages= parameter goes to the "default hstate".
> -	 * Otherwise, it goes with the previously parsed hugepagesz or
> -	 * default_hugepagesz.
> -	 */
> -	else if (!hugetlb_max_hstate)
> -		mhp = &default_hstate_max_huge_pages;
> -	else
> -		mhp = &parsed_hstate->max_huge_pages;
> +	while (*p) {
> +		count = 0;
> +		ret = sscanf(p, "%lu%n", &tmp, &count);
> +		if (ret != 1) {
> +			pr_warn("HugeTLB: Invalid hugepages parameter %s\n", p);
> +			return 0;
> +		}
> +		/* Parameter is node format */
> +		if (p[count] == ':') {
> +			node = tmp;
> +			p += count + 1;
> +			if (node < 0 || node >= nodes_weight(node_states[N_MEMORY])) {
> +				pr_warn("HugeTLB: Invalid hugepages parameter node:%d\n", node);
> +				return 0;
> +			}
> +			if (!hugetlb_max_hstate)
> +				mhp = (unsigned long *)
> +					&(default_hugepages_in_node[node]);
> +			else
> +				mhp = (unsigned long *)
> +					&(parsed_hstate->max_huge_pages_node[node]);
> +			/* Parse hugepages */
> +			ret = sscanf(p, "%lu%n", mhp, &count);
> +			if (ret != 1) {
> +				pr_warn("HugeTLB: Invalid hugepages parameter %s\n", p);
> +				return 0;
> +			}
> +			if (!hugetlb_max_hstate)
> +				default_hstate_max_huge_pages += *mhp;
> +			else
> +				parsed_hstate->max_huge_pages += *mhp;
> +			/* Go to parse next node*/
> +			if (p[count] == ',')
> +				p += count + 1;
> +			else
> +				break;
> +		} else if (p == s) {
> +
> +			/*
> +			 * !hugetlb_max_hstate means we haven't parsed a hugepagesz= parameter
> +			 * yet, so this hugepages= parameter goes to the "default hstate".
> +			 * Otherwise, it goes with the previously parsed hugepagesz or
> +			 * default_hugepagesz.
> +			 */
> +			if (!hugetlb_max_hstate) {
> +				default_hstate_max_huge_pages = tmp;
> +				mhp = &default_hstate_max_huge_pages;
> +			} else {
> +				parsed_hstate->max_huge_pages = tmp;
> +				mhp = &parsed_hstate->max_huge_pages;
> +			}
> +			break;
> +		}
> +		pr_warn("HugeTLB: Invalid hugepages parameter %s\n", p);
> +		return 0;
> +	}
>  
>  	if (mhp == last_mhp) {
>  		pr_warn("HugeTLB: hugepages= specified twice without interleaving hugepagesz=, ignoring hugepages=%s\n", s);
>  		return 0;
>  	}

This check for 'mhp == last_mhp' after setting parameters will result in
a change of behavior.  For example, consider the following command line
options:

	hugepagesz=2M hugepages=2 hugepages=5

Before this change,

# dmesg | grep -i huge
[    0.000000] Command line: BOOT_IMAGE=/vmlinuz-5.14.0+.old root=/dev/mapper/fedora_new--host-root ro rd.lvm.lv=fedora_new-host/root rd.lvm.lv=fedora_new-host/swap console=tty0 console=ttyS0,115200 audit=0 transparent_hugepage=always hugetlb_free_vmemmap=on hugepagesz=2M hugepages=2 hugepages=5
[    0.054206] Kernel command line: BOOT_IMAGE=/vmlinuz-5.14.0+.old root=/dev/mapper/fedora_new--host-root ro rd.lvm.lv=fedora_new-host/root rd.lvm.lv=fedora_new-host/swap console=tty0 console=ttyS0,115200 audit=0 transparent_hugepage=always hugetlb_free_vmemmap=on hugepagesz=2M hugepages=2 hugepages=5
[    0.054271] HugeTLB: can free 6 vmemmap pages for hugepages-2048kB
[    0.054282] HugeTLB: hugepages= specified twice without interleaving hugepagesz=, ignoring hugepages=5
[    0.054286] Unknown command line parameters: BOOT_IMAGE=/vmlinuz-5.14.0+.old hugepages=5
[    0.412456] HugeTLB: can free 4094 vmemmap pages for hugepages-1048576kB
[    0.424527] HugeTLB registered 2.00 MiB page size, pre-allocated 2 pages
[    0.425459] HugeTLB registered 1.00 GiB page size, pre-allocated 0 pages
[    5.168890]     hugepages=5

After this change,

# dmesg | grep -i huge
[    0.000000] Command line: BOOT_IMAGE=/vmlinuz-5.14.0+ root=/dev/mapper/fedora_new--host-root ro rd.lvm.lv=fedora_new-host/root rd.lvm.lv=fedora_new-host/swap console=tty0 console=ttyS0,115200 audit=0 transparent_hugepage=always hugetlb_free_vmemmap=on hugepagesz=2M hugepages=2 hugepages=5
[    0.054569] Kernel command line: BOOT_IMAGE=/vmlinuz-5.14.0+ root=/dev/mapper/fedora_new--host-root ro rd.lvm.lv=fedora_new-host/root rd.lvm.lv=fedora_new-host/swap console=tty0 console=ttyS0,115200 audit=0 transparent_hugepage=always hugetlb_free_vmemmap=on hugepagesz=2M hugepages=2 hugepages=5
[    0.054636] HugeTLB: can free 6 vmemmap pages for hugepages-2048kB
[    0.054647] HugeTLB: hugepages= specified twice without interleaving hugepagesz=, ignoring hugepages=5
[    0.054651] Unknown command line parameters: BOOT_IMAGE=/vmlinuz-5.14.0+ hugepages=5
[    0.413714] HugeTLB: can free 4094 vmemmap pages for hugepages-1048576kB
[    0.427774] HugeTLB registered 2.00 MiB page size, pre-allocated 5 pages
[    0.428628] HugeTLB registered 1.

-- 
Mike Kravetz

>  
> -	if (sscanf(s, "%lu", mhp) <= 0)
> -		*mhp = 0;
> -
>  	/*
>  	 * Global state is always initialized later in hugetlb_init.
>  	 * But we need to allocate gigantic hstates here early to still
> 

Powered by blists - more mailing lists