1. Patchew
  2. linux
  3. View series

[PATCH v3] kho: init alloc tags when restoring pages from reserved memory

ranxiaokai627@163.com posted 1 patch 2 weeks, 2 days ago
There is a newer version of this series
kernel/liveupdate/kexec_handover.c | 9 +++++++--
1 file changed, 7 insertions(+), 2 deletions(-)
[PATCH v3] kho: init alloc tags when restoring pages from reserved memory
Posted by ranxiaokai627@163.com 2 weeks, 2 days ago 
From: Ran Xiaokai <ran.xiaokai@zte.com.cn>

Memblock pages (including reserved memory) should have their allocation
tags initialized to CODETAG_EMPTY via clear_page_tag_ref() before being
released to the page allocator. When kho restores pages through
kho_restore_page(), missing this call causes mismatched
allocation/deallocation tracking and below warning message:

alloc_tag was not set
WARNING: include/linux/alloc_tag.h:164 at ___free_pages+0xb8/0x260, CPU#1: swapper/0/1
RIP: 0010:___free_pages+0xb8/0x260
 kho_restore_vmalloc+0x187/0x2e0
 kho_test_init+0x3c4/0xa30
 do_one_initcall+0x62/0x2b0
 kernel_init_freeable+0x25b/0x480
 kernel_init+0x1a/0x1c0
 ret_from_fork+0x2d1/0x360

Add missing clear_page_tag_ref() annotation in kho_restore_page() to
fix this.

Signed-off-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
---
It is based on linux-next 20260120. I dont konw whether this base is ok ?

 kernel/liveupdate/kexec_handover.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/kernel/liveupdate/kexec_handover.c b/kernel/liveupdate/kexec_handover.c
index 43d9d2661163..8f7e09d67e1b 100644
--- a/kernel/liveupdate/kexec_handover.c
+++ b/kernel/liveupdate/kexec_handover.c
@@ -219,8 +219,11 @@ static int __kho_preserve_order(struct kho_mem_track *track, unsigned long pfn,
 /* For physically contiguous 0-order pages. */
 static void kho_init_pages(struct page *page, unsigned long nr_pages)
 {
-	for (unsigned long i = 0; i < nr_pages; i++)
+	for (unsigned long i = 0; i < nr_pages; i++) {
 		set_page_count(page + i, 1);
+		/* Mark the codetag as empty to avoid accounting mismatch */
+		clear_page_tag_ref(page + i);
+	}
 }
 
 static void kho_init_folio(struct page *page, unsigned int order)
@@ -236,6 +239,9 @@ static void kho_init_folio(struct page *page, unsigned int order)
 
 	if (order > 0)
 		prep_compound_page(page, order);
+
+	/* Mark the codetag as empty to avoid accounting mismatch */
+	clear_page_tag_ref(page);
 }
 
 static struct page *kho_restore_page(phys_addr_t phys, bool is_folio)
@@ -265,7 +271,6 @@ static struct page *kho_restore_page(phys_addr_t phys, bool is_folio)
 	else
 		kho_init_pages(page, nr_pages);
 
-	clear_page_tag_ref(page);
 	adjust_managed_page_count(page, nr_pages);
 	return page;
 }
-- 
2.25.1
Re: [PATCH v3] kho: init alloc tags when restoring pages from reserved memory
Posted by Andrew Morton 2 weeks, 2 days ago 
On Wed, 21 Jan 2026 14:34:30 +0000 ranxiaokai627@163.com wrote:

> From: Ran Xiaokai <ran.xiaokai@zte.com.cn>
> 
> Memblock pages (including reserved memory) should have their allocation
> tags initialized to CODETAG_EMPTY via clear_page_tag_ref() before being
> released to the page allocator. When kho restores pages through
> kho_restore_page(), missing this call causes mismatched
> allocation/deallocation tracking and below warning message:
> 
> alloc_tag was not set
> WARNING: include/linux/alloc_tag.h:164 at ___free_pages+0xb8/0x260, CPU#1: swapper/0/1
> RIP: 0010:___free_pages+0xb8/0x260
>  kho_restore_vmalloc+0x187/0x2e0
>  kho_test_init+0x3c4/0xa30
>  do_one_initcall+0x62/0x2b0
>  kernel_init_freeable+0x25b/0x480
>  kernel_init+0x1a/0x1c0
>  ret_from_fork+0x2d1/0x360
> 
> Add missing clear_page_tag_ref() annotation in kho_restore_page() to
> fix this.
> 
> Signed-off-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
> ---
> It is based on linux-next 20260120. I dont konw whether this base is ok ?

It's awkward.

Your v2 patch was based on Linus mainline.  This is appropriate, as the
patch should be sent to Linus soon and it has cc:stable, so -stable
maintainers will try to backport it into earlier kernels.

However your v3 patch is dependent upon other material ("kho: simplify
page initialization in kho_restore_page()") which is scheduled for
6.20(?)-rc1.

For a prompt, backportable merge it's best to base the fix on latest
Linus mainline, please.

You didn't actually describe why v3 is different from v2.  If the
v2->v3 changes are just nice-to-have then let's redo those and base
them on linux-next in the usual fashion.

Unless I'm missing something, your well-reviewed, decently-tested v2
patch remains suitable for upstreaming during 6.18-rcX?
Re: [PATCH v3] kho: init alloc tags when restoring pages from reserved memory
Posted by ranxiaokai627@163.com 2 weeks, 2 days ago 
>> From: Ran Xiaokai <ran.xiaokai@zte.com.cn>
>> 
>> Memblock pages (including reserved memory) should have their allocation
>> tags initialized to CODETAG_EMPTY via clear_page_tag_ref() before being
>> released to the page allocator. When kho restores pages through
>> kho_restore_page(), missing this call causes mismatched
>> allocation/deallocation tracking and below warning message:
>> 
>> alloc_tag was not set
>> WARNING: include/linux/alloc_tag.h:164 at ___free_pages+0xb8/0x260, CPU#1: swapper/0/1
>> RIP: 0010:___free_pages+0xb8/0x260
>>  kho_restore_vmalloc+0x187/0x2e0
>>  kho_test_init+0x3c4/0xa30
>>  do_one_initcall+0x62/0x2b0
>>  kernel_init_freeable+0x25b/0x480
>>  kernel_init+0x1a/0x1c0
>>  ret_from_fork+0x2d1/0x360
>> 
>> Add missing clear_page_tag_ref() annotation in kho_restore_page() to
>> fix this.
>> 
>> Signed-off-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
>> ---
>> It is based on linux-next 20260120. I dont konw whether this base is ok ?
>
>It's awkward.
>
>Your v2 patch was based on Linus mainline.  This is appropriate, as the
>patch should be sent to Linus soon and it has cc:stable, so -stable
>maintainers will try to backport it into earlier kernels.
>
>However your v3 patch is dependent upon other material ("kho: simplify
>page initialization in kho_restore_page()") which is scheduled for
>6.20(?)-rc1.

I think i misunderstood Pratyush's last reply:
"I suggested a re-roll of this patch based on top of my cleanup patches
[1], since I think with those the end result is a bit nicer."

>For a prompt, backportable merge it's best to base the fix on latest
>Linus mainline, please.
>
>You didn't actually describe why v3 is different from v2.  If the
>v2->v3 changes are just nice-to-have then let's redo those and base
>them on linux-next in the usual fashion.

>Unless I'm missing something, your well-reviewed, decently-tested v2
>patch remains suitable for upstreaming during 6.18-rcX

v2 version just fixed the folio case(compound page), but didn't fix the
contiguous order 0 pages case. So i think it is better to send a v3 version
base on lastest Linus tree and drop the v2 version.
Re: [PATCH v3] kho: init alloc tags when restoring pages from reserved memory
Posted by Pratyush Yadav 2 weeks, 1 day ago 
Hi Ran,

On Thu, Jan 22 2026, ranxiaokai627@163.com wrote:

>>> From: Ran Xiaokai <ran.xiaokai@zte.com.cn>
>>> 
>>> Memblock pages (including reserved memory) should have their allocation
>>> tags initialized to CODETAG_EMPTY via clear_page_tag_ref() before being
>>> released to the page allocator. When kho restores pages through
>>> kho_restore_page(), missing this call causes mismatched
>>> allocation/deallocation tracking and below warning message:
>>> 
>>> alloc_tag was not set
>>> WARNING: include/linux/alloc_tag.h:164 at ___free_pages+0xb8/0x260, CPU#1: swapper/0/1
>>> RIP: 0010:___free_pages+0xb8/0x260
>>>  kho_restore_vmalloc+0x187/0x2e0
>>>  kho_test_init+0x3c4/0xa30
>>>  do_one_initcall+0x62/0x2b0
>>>  kernel_init_freeable+0x25b/0x480
>>>  kernel_init+0x1a/0x1c0
>>>  ret_from_fork+0x2d1/0x360
>>> 
>>> Add missing clear_page_tag_ref() annotation in kho_restore_page() to
>>> fix this.
>>> 
>>> Signed-off-by: Ran Xiaokai <ran.xiaokai@zte.com.cn>
>>> ---
>>> It is based on linux-next 20260120. I dont konw whether this base is ok ?
>>
>>It's awkward.
>>
>>Your v2 patch was based on Linus mainline.  This is appropriate, as the
>>patch should be sent to Linus soon and it has cc:stable, so -stable
>>maintainers will try to backport it into earlier kernels.
>>
>>However your v3 patch is dependent upon other material ("kho: simplify
>>page initialization in kho_restore_page()") which is scheduled for
>>6.20(?)-rc1.
>
> I think i misunderstood Pratyush's last reply:
> "I suggested a re-roll of this patch based on top of my cleanup patches
> [1], since I think with those the end result is a bit nicer."

I was giving context to Andrew about the whole thing.

I thought it was a good idea when I suggested it to you, but at the time
I didn't think that this will go in the hotfixes branch. If it goes in
hotfixes, it doesn't make sense to base it on a series for the next
kernel.

Sorry for the confusion.

>
>>For a prompt, backportable merge it's best to base the fix on latest
>>Linus mainline, please.
>>
>>You didn't actually describe why v3 is different from v2.  If the
>>v2->v3 changes are just nice-to-have then let's redo those and base
>>them on linux-next in the usual fashion.
>
>>Unless I'm missing something, your well-reviewed, decently-tested v2
>>patch remains suitable for upstreaming during 6.18-rcX
>
> v2 version just fixed the folio case(compound page), but didn't fix the
> contiguous order 0 pages case. So i think it is better to send a v3 version
> base on lastest Linus tree and drop the v2 version.

Yep, that would be the idea. Resend the changes fixing both compound and
non-compound cases on top of Linus' tree and ignore my "simplify page
initialization" series.

And then I can later resend my series on top of your patch.

-- 
Regards,
Pratyush Yadav

Patchew 2.1-devel-b67e4c2

© 2016 - 2026 Red Hat, Inc.