linux-mm.kvack.org archive mirror
 help / color / mirror / Atom feed
From: Wei Yang <richard.weiyang@gmail.com>
To: Zi Yan <ziy@nvidia.com>
Cc: Wei Yang <richard.weiyang@gmail.com>,
	wang lian <lianux.mm@gmail.com>,
	Baolin Wang <baolin.wang@linux.alibaba.com>,
	David Hildenbrand <david@redhat.com>,
	linux-mm@kvack.org, Andrew Morton <akpm@linux-foundation.org>,
	Lorenzo Stoakes <lorenzo.stoakes@oracle.com>,
	"Liam R. Howlett" <Liam.Howlett@oracle.com>,
	Nico Pache <npache@redhat.com>,
	Ryan Roberts <ryan.roberts@arm.com>, Dev Jain <dev.jain@arm.com>,
	Barry Song <baohua@kernel.org>, Vlastimil Babka <vbabka@suse.cz>,
	Mike Rapoport <rppt@kernel.org>,
	Suren Baghdasaryan <surenb@google.com>,
	Michal Hocko <mhocko@suse.com>, Shuah Khan <shuah@kernel.org>,
	linux-kernel@vger.kernel.org, linux-kselftest@vger.kernel.org
Subject: Re: [PATCH v2 2/3] selftests/mm: add check_folio_orders() helper.
Date: Sat, 9 Aug 2025 20:18:36 +0000	[thread overview]
Message-ID: <20250809201836.jegaanplfcjak44f@master> (raw)
In-Reply-To: <20250808190144.797076-3-ziy@nvidia.com>

On Fri, Aug 08, 2025 at 03:01:43PM -0400, Zi Yan wrote:
>The helper gathers an folio order statistics of folios within a virtual
>address range and checks it against a given order list. It aims to provide
>a more precise folio order check instead of just checking the existence of
>PMD folios.
>
>Signed-off-by: Zi Yan <ziy@nvidia.com>
>---
> .../selftests/mm/split_huge_page_test.c       |   4 +-
> tools/testing/selftests/mm/vm_util.c          | 133 ++++++++++++++++++
> tools/testing/selftests/mm/vm_util.h          |   7 +
> 3 files changed, 141 insertions(+), 3 deletions(-)
>
>diff --git a/tools/testing/selftests/mm/split_huge_page_test.c b/tools/testing/selftests/mm/split_huge_page_test.c
>index cb364c5670c6..5ab488fab1cd 100644
>--- a/tools/testing/selftests/mm/split_huge_page_test.c
>+++ b/tools/testing/selftests/mm/split_huge_page_test.c
>@@ -34,8 +34,6 @@ uint64_t pmd_pagesize;
> #define PID_FMT_OFFSET "%d,0x%lx,0x%lx,%d,%d"
> #define PATH_FMT "%s,0x%lx,0x%lx,%d"
> 
>-#define PFN_MASK     ((1UL<<55)-1)
>-#define KPF_THP      (1UL<<22)
> #define GET_ORDER(nr_pages)    (31 - __builtin_clz(nr_pages))
> 
> int is_backed_by_thp(char *vaddr, int pagemap_file, int kpageflags_file)
>@@ -49,7 +47,7 @@ int is_backed_by_thp(char *vaddr, int pagemap_file, int kpageflags_file)
> 
> 		if (kpageflags_file) {
> 			pread(kpageflags_file, &page_flags, sizeof(page_flags),
>-				(paddr & PFN_MASK) * sizeof(page_flags));
>+				PAGEMAP_PFN(paddr) * sizeof(page_flags));
> 

is_backed_by_thp() shares similar logic as get_page_flags(), I am thinking we can
leverage get_page_flags() here.

> 			return !!(page_flags & KPF_THP);
> 		}
>diff --git a/tools/testing/selftests/mm/vm_util.c b/tools/testing/selftests/mm/vm_util.c
>index 6a239aa413e2..41d50b74b2f6 100644
>--- a/tools/testing/selftests/mm/vm_util.c
>+++ b/tools/testing/selftests/mm/vm_util.c
>@@ -338,6 +338,139 @@ int detect_hugetlb_page_sizes(size_t sizes[], int max)
> 	return count;
> }
> 
>+static int get_page_flags(char *vaddr, int pagemap_file, int kpageflags_file,
>+			  uint64_t *flags)
>+{

Nit.

In vm_util.c, we usually name the file descriptor as xxx_fd.

>+	unsigned long pfn;
>+	size_t count;
>+
>+	pfn = pagemap_get_pfn(pagemap_file, vaddr);
>+	/*
>+	 * Treat non-present page as a page without any flag, so that
>+	 * gather_folio_orders() just record the current folio order.
>+	 */
>+	if (pfn == -1UL) {
>+		*flags = 0;
>+		return 0;
>+	}
>+
>+	count = pread(kpageflags_file, flags, sizeof(*flags),
>+		      pfn * sizeof(*flags));
>+
>+	if (count != sizeof(*flags))
>+		return -1;
>+
>+	return 0;
>+}
>+

Maybe a simple document here would be helpful.

>+static int gather_folio_orders(char *vaddr_start, size_t len,
>+			       int pagemap_file, int kpageflags_file,
>+			       int orders[], int nr_orders)
>+{
>+	uint64_t page_flags = 0;
>+	int cur_order = -1;
>+	char *vaddr;
>+
>+	if (!pagemap_file || !kpageflags_file)
>+		return -1;
>+	if (nr_orders <= 0)
>+		return -1;
>+
>+	for (vaddr = vaddr_start; vaddr < vaddr_start + len; ) {
>+		char *next_folio_vaddr;
>+		int status;
>+
>+		if (get_page_flags(vaddr, pagemap_file, kpageflags_file, &page_flags))
>+			return -1;
>+
>+		/* all order-0 pages with possible false postive (non folio) */
>+		if (!(page_flags & (KPF_COMPOUND_HEAD | KPF_COMPOUND_TAIL))) {
>+			orders[0]++;
>+			vaddr += psize();
>+			continue;
>+		}
>+
>+		/* skip non thp compound pages */
>+		if (!(page_flags & KPF_THP)) {
>+			vaddr += psize();
>+			continue;
>+		}
>+
>+		/* vpn points to part of a THP at this point */
>+		if (page_flags & KPF_COMPOUND_HEAD)
>+			cur_order = 1;
>+		else {
>+			/* not a head nor a tail in a THP? */
>+			if (!(page_flags & KPF_COMPOUND_TAIL))
>+				return -1;
>+			continue;
>+		}
>+
>+		next_folio_vaddr = vaddr + (1UL << (cur_order + pshift()));
>+
>+		if (next_folio_vaddr >= vaddr_start + len)
>+			break;

Would we skip order 1 folio at the last position?

For example, vaddr_start is 0x2000, len is 0x2000 and the folio at vaddr_start
is an order 1 folio, whose size is exactly 0x2000.

Then we will get next_folio_vaddr == vaddr_start + len.

Could that happen?

>+
>+		while (!(status = get_page_flags(next_folio_vaddr, pagemap_file,
>+						 kpageflags_file,
>+						 &page_flags))) {
>+			/* next compound head page or order-0 page */
>+			if ((page_flags & KPF_COMPOUND_HEAD) ||
>+			    !(page_flags & (KPF_COMPOUND_HEAD |
>+			      KPF_COMPOUND_TAIL))) {

Maybe we can put them into one line.

>+				if (cur_order < nr_orders) {
>+					orders[cur_order]++;
>+					cur_order = -1;
>+					vaddr = next_folio_vaddr;
>+				}
>+				break;
>+			}
>+
>+			/* not a head nor a tail in a THP? */
>+			if (!(page_flags & KPF_COMPOUND_TAIL))
>+				return -1;
>+
>+			cur_order++;
>+			next_folio_vaddr = vaddr + (1UL << (cur_order + pshift()));
>+		}

The while loop share similar logic as the outer for loop. Is it possible
reduce some duplication?

>+
>+		if (status)
>+			return status;
>+	}
>+	if (cur_order > 0 && cur_order < nr_orders)
>+		orders[cur_order]++;
>+	return 0;
>+}
>+
>+int check_folio_orders(char *vaddr_start, size_t len, int pagemap_file,
>+			int kpageflags_file, int orders[], int nr_orders)
>+{
>+	int *vaddr_orders;
>+	int status;
>+	int i;
>+
>+	vaddr_orders = (int *)malloc(sizeof(int) * nr_orders);
>+

I took a look into thp_setting.h, where defines an array with NR_ORDERS
element which is 20. Maybe we can leverage it here, since we don't expect the
order to be larger.

>+	if (!vaddr_orders)
>+		ksft_exit_fail_msg("Cannot allocate memory for vaddr_orders");
>+
>+	memset(vaddr_orders, 0, sizeof(int) * nr_orders);
>+	status = gather_folio_orders(vaddr_start, len, pagemap_file,
>+				     kpageflags_file, vaddr_orders, nr_orders);
>+	if (status)
>+		return status;
>+
>+	status = 0;
>+	for (i = 0; i < nr_orders; i++)
>+		if (vaddr_orders[i] != orders[i]) {
>+			ksft_print_msg("order %d: expected: %d got %d\n", i,
>+				       orders[i], vaddr_orders[i]);
>+			status = -1;
>+		}
>+
>+	return status;
>+}
>+
> /* If `ioctls' non-NULL, the allowed ioctls will be returned into the var */
> int uffd_register_with_ioctls(int uffd, void *addr, uint64_t len,
> 			      bool miss, bool wp, bool minor, uint64_t *ioctls)
>diff --git a/tools/testing/selftests/mm/vm_util.h b/tools/testing/selftests/mm/vm_util.h
>index 1843ad48d32b..02e3f1e7065b 100644
>--- a/tools/testing/selftests/mm/vm_util.h
>+++ b/tools/testing/selftests/mm/vm_util.h
>@@ -18,6 +18,11 @@
> #define PM_SWAP                       BIT_ULL(62)
> #define PM_PRESENT                    BIT_ULL(63)
> 
>+#define KPF_COMPOUND_HEAD             BIT_ULL(15)
>+#define KPF_COMPOUND_TAIL             BIT_ULL(16)
>+#define KPF_THP                       BIT_ULL(22)
>+
>+
> /*
>  * Ignore the checkpatch warning, we must read from x but don't want to do
>  * anything with it in order to trigger a read page fault. We therefore must use
>@@ -85,6 +90,8 @@ bool check_huge_shmem(void *addr, int nr_hpages, uint64_t hpage_size);
> int64_t allocate_transhuge(void *ptr, int pagemap_fd);
> unsigned long default_huge_page_size(void);
> int detect_hugetlb_page_sizes(size_t sizes[], int max);
>+int check_folio_orders(char *vaddr_start, size_t len, int pagemap_file,
>+			int kpageflags_file, int orders[], int nr_orders);
> 
> int uffd_register(int uffd, void *addr, uint64_t len,
> 		  bool miss, bool wp, bool minor);
>-- 
>2.47.2

-- 
Wei Yang
Help you, Help me


  reply	other threads:[~2025-08-09 20:18 UTC|newest]

Thread overview: 19+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-08-08 19:01 [PATCH v2 0/3] Better split_huge_page_test result check Zi Yan
2025-08-08 19:01 ` [PATCH v2 1/3] mm/huge_memory: add new_order and offset to split_huge_pages*() pr_debug Zi Yan
2025-08-09 18:45   ` Wei Yang
2025-08-10 16:55   ` Donet Tom
2025-08-11  1:32   ` wang lian
2025-08-11  6:45   ` Baolin Wang
2025-08-11  6:58   ` Barry Song
2025-08-11  7:55   ` David Hildenbrand
2025-08-08 19:01 ` [PATCH v2 2/3] selftests/mm: add check_folio_orders() helper Zi Yan
2025-08-09 20:18   ` Wei Yang [this message]
2025-08-11 18:39     ` Zi Yan
2025-08-11 21:28       ` Wei Yang
2025-08-10 16:49   ` Donet Tom
2025-08-11 18:40     ` Zi Yan
2025-08-11  7:52   ` Baolin Wang
2025-08-11 18:41     ` Zi Yan
2025-08-08 19:01 ` [PATCH v2 3/3] selftests/mm: check after-split folio orders in split_huge_page_test Zi Yan
2025-08-10 16:53   ` Donet Tom
2025-08-11 18:20     ` Zi Yan

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20250809201836.jegaanplfcjak44f@master \
    --to=richard.weiyang@gmail.com \
    --cc=Liam.Howlett@oracle.com \
    --cc=akpm@linux-foundation.org \
    --cc=baohua@kernel.org \
    --cc=baolin.wang@linux.alibaba.com \
    --cc=david@redhat.com \
    --cc=dev.jain@arm.com \
    --cc=lianux.mm@gmail.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-kselftest@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=lorenzo.stoakes@oracle.com \
    --cc=mhocko@suse.com \
    --cc=npache@redhat.com \
    --cc=rppt@kernel.org \
    --cc=ryan.roberts@arm.com \
    --cc=shuah@kernel.org \
    --cc=surenb@google.com \
    --cc=vbabka@suse.cz \
    --cc=ziy@nvidia.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).