Skip to contentSkip to conversation

LoongArch: mm: remove redundant pte occupancy check in __set_fixmap()

PATCH Superseded george 4 participants 8 messages May 8, 2025 – Aug 28, 2025 LoongArch

lore
Share & tools
lore

george

author
#0 lore raw

PATCH 1/1 LoongArch: mm: remove redundant pte occupancy check in __set_fixmap()

From: George Guo <[email protected]>

This check could falsely report errors when PTEs are remapped:
arch/loongarch/mm/init.c:204: bad pte 0000000107b601c3.

The check was unnecessary because:
- Fixmap slots are designed to be reusable
- Other architectures (x86/arm64) allow PTE overwrites
- Callers are responsible for proper mapping management

---
 arch/loongarch/mm/init.c | 4 ----
 1 file changed, 4 deletions(-)
arch/loongarch/mm/init.c +0 -4
index 06f11d9e4ec1..7b7e3b1aa00d 100644
@@ -200,10 +200,6 @@ void __init __set_fixmap(enum fixed_addresses idx,
 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
 
 	ptep = populate_kernel_pte(addr);
-	if (!pte_none(ptep_get(ptep))) {
-		pte_ERROR(*ptep);
-		return;
-	}
 
 	if (pgprot_val(flags))
 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
Signature · 1 line
2.34.1
#1 lore raw
Quoted 1 line · george

From: George Guo <[email protected]>

This is unnecessary. Please remove it.

Thanks,
Yanteng

Quoted 29 lines

This check could falsely report errors when PTEs are remapped:
arch/loongarch/mm/init.c:204: bad pte 0000000107b601c3.

The check was unnecessary because:
- Fixmap slots are designed to be reusable
- Other architectures (x86/arm64) allow PTE overwrites
- Callers are responsible for proper mapping management

---
  arch/loongarch/mm/init.c | 4 ----
  1 file changed, 4 deletions(-)
arch/loongarch/mm/init.c +0 -4
index 06f11d9e4ec1..7b7e3b1aa00d 100644
@@ -200,10 +200,6 @@ void __init __set_fixmap(enum fixed_addresses idx,
  	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
 
  	ptep = populate_kernel_pte(addr);
-	if (!pte_none(ptep_get(ptep))) {
-		pte_ERROR(*ptep);
-		return;
-	}
 
  	if (pgprot_val(flags))
  		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
#2 lore raw
Quoted 3 lines · On Thu, 8 May 2025 13:23:38 +0800 Yanteng Si <[email protected]> wrote:
Quoted 1 line · george

From: George Guo <[email protected]>

This is unnecessary. Please remove it.

Why?

I believe it is necessary. Without this line, the patch will be committed as

From: george <[email protected]>

- different display name
- different email address
- email address doesn't match signed-off-by

#3 lore raw
Quoted 8 lines · Andrew Morton
Quoted 3 lines · On Thu, 8 May 2025 13:23:38 +0800 Yanteng Si <[email protected]> wrote:
Quoted 1 line · george

From: George Guo <[email protected]>

This is unnecessary. Please remove it.

Why?

Quoted 9 lines

I believe it is necessary. Without this line, the patch will be committed as

From: george <[email protected]>

- different display name
- different email address
- email address doesn't match signed-off-by

OK, I see. It is necessary.

Thanks,
Yanteng

maintainer
#4 lore raw
In reply to george

Hi, George,

Quoted 21 lines · george

From: George Guo <[email protected]>

This check could falsely report errors when PTEs are remapped:
arch/loongarch/mm/init.c:204: bad pte 0000000107b601c3.

The check was unnecessary because:
- Fixmap slots are designed to be reusable
- Other architectures (x86/arm64) allow PTE overwrites
- Callers are responsible for proper mapping management

---
 arch/loongarch/mm/init.c | 4 ----
 1 file changed, 4 deletions(-)
arch/loongarch/mm/init.c +0 -0
index 06f11d9e4ec1..7b7e3b1aa00d 100644

arch/loongarch/mm/init.c void __init __set_fixmap(enum fixed_addresses idx, · line 206

Quoted 7 lines
arch/loongarch/mm/init.c +0 -4
@@ -200,10 +200,6 @@ void __init __set_fixmap(enum fixed_addresses idx,
        BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
 
        ptep = populate_kernel_pte(addr);
-       if (!pte_none(ptep_get(ptep))) {
-               pte_ERROR(*ptep);
-               return;
-       }

Resolve what problems? If other patches need this change, please submit them as a series.

Huacai

Quoted 7 lines
arch/loongarch/mm/init.c +0 -1
        if (pgprot_val(flags))
                set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
--

2.34.1

author
#5 lore raw
Quoted 26 lines · On Sat, 10 May 2025 15:34:01 +0800 Huacai Chen <[email protected]> wrote:

Hi, George,

Quoted 23 lines · george

From: George Guo <[email protected]>

This check could falsely report errors when PTEs are remapped:
arch/loongarch/mm/init.c:204: bad pte 0000000107b601c3.

The check was unnecessary because:
- Fixmap slots are designed to be reusable
- Other architectures (x86/arm64) allow PTE overwrites
- Callers are responsible for proper mapping management

---
 arch/loongarch/mm/init.c | 4 ----
 1 file changed, 4 deletions(-)
arch/loongarch/mm/init.c +0 -0
index 06f11d9e4ec1..7b7e3b1aa00d 100644
@@ -200,10 +200,6 @@ void __init __set_fixmap(enum fixed_addresses
 idx, BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
 
Quoted 7 lines

arch/loongarch/mm/init.c void __init __set_fixmap(enum fixed_addresses idx, · line 206

Quoted 5 lines
arch/loongarch/mm/init.c +0 -4
        ptep = populate_kernel_pte(addr);
-       if (!pte_none(ptep_get(ptep))) {
-               pte_ERROR(*ptep);
-               return;
-       }

Resolve what problems? If other patches need this change, please submit them as a series.

Here is a test case to show the bug:
#include <linux/module.h>
#include <linux/vmalloc.h>
#include <linux/mm.h>
#include <asm/io.h>
#include <asm-generic/fixmap.h>

#define ALLOC_SIZE 8192 // 8KB

static char *vmalloc_mem = NULL;

static int __init fixmap_test_init(void)

{
	struct page *page;
	phys_addr_t phys;
	void *addr;
	char c;
	void __iomem *base;
	char *charp;

	pr_info("%s module loaded!\n", THIS_MODULE->name);

	addr = vmalloc_mem = vmalloc(ALLOC_SIZE);
	if (!vmalloc_mem) {
		pr_err("Failed to allocate memory with vmalloc\n");
		return -ENOMEM;
	}

	vmalloc_mem[4099] = 'a';
	vmalloc_mem[0] = 'g';
	pr_info("Allocated vmalloc memory at virtual address: %px,
	vmalloc_mem[0]=%c\n", vmalloc_mem, vmalloc_mem[0]);

	//vrit addr -> phys addr
	page = vmalloc_to_page(addr);
	phys = page_to_phys(page) + offset_in_page(addr);
	pr_info("phys = 0x%lx\n", phys);


	__set_fixmap(FIX_TEXT_POKE0, phys & PAGE_MASK, FIXMAP_PAGE_IO);
	base = (void __iomem *)__fix_to_virt(FIX_TEXT_POKE0);
	base += phys & ~PAGE_MASK;
	charp = (char *)base;
	pr_info("fixmap virt addr=0x%lx,charp[0] = %c\n",base,
	charp[0]);
	__set_fixmap(FIX_TEXT_POKE0, 0, FIXMAP_PAGE_CLEAR);

	return 0;
}

static void __exit fixmap_test_exit(void)

{
	if (vmalloc_mem) {
		vfree(vmalloc_mem);
		pr_info("Freed vmalloc memory\n");
	}

	pr_info("%s module unloaded!\n", THIS_MODULE->name);
}

module_init(fixmap_test_init);
module_exit(fixmap_test_exit);

MODULE_LICENSE("GPL");
MODULE_AUTHOR("George Guo <[email protected]>"); MODULE_DESCRIPTION("fixmap test");

And below is the log for test:

arm64:

first insmod:

[10476.162720] fixmap_test module loaded!
[10476.163390] Allocated vmalloc memory at virtual address:
ffff80008008f000, vmalloc_mem[0]=g
[10476.164497] phys = 0x10fab2000
[10476.165106] fixmap virt addr=0xffffffffff5fc000,charp[0] = g

rmmod then insmod again:

[10514.031689] Freed vmalloc memory
[10514.032268] fixmap_test module unloaded!
[10516.391483] fixmap_test module loaded!
[10516.392135] Allocated vmalloc memory at virtual address:
ffff800080097000, vmalloc_mem[0]=g
[10516.393205] phys = 0x10dc2c000
[10516.393663] fixmap virt addr=0xffffffffff5fc000,charp[0] = g

loongarch:

first insmod:

[   43.983300] [   T1399] fixmap test module loaded
[   43.983570] [   T1399] Allocated vmalloc memory at virtual address:
ffff80001b45c000, vmalloc_mem[0]=g
[   43.983973] [   T1399] phys = 0x107b60000
[   43.984535] [   T1399] fixmap virt addr=0xfffffffffffd8000,charp[0]

= g

rmmod then insmod again:

[   52.266230] [   T1417] Freed vmalloc memory
[   55.344169] [   T1421] fixmap test module loaded
[   55.344441] [   T1421] Allocated vmalloc memory at virtual address:
ffff80001b344000, vmalloc_mem[0]=g
[   55.344915] [   T1421] phys = 0x10ae04000
[   55.345503] [   T1421] arch/loongarch/mm/init.c:204: bad pte
0000000107b601c3.
[   55.346087] [   T1421] fixmap virt addr=0xfffffffffffd8000,charp[0]

= \x00

Quoted 10 lines

Huacai

Quoted 7 lines
arch/loongarch/mm/init.c +0 -1
        if (pgprot_val(flags))
                set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
--

2.34.1

author
#6 lore raw

PATCH v2 1/1 LoongArch: mm: remove redundant pte occupancy check in __set_fixmap()

From: George Guo <[email protected]>

This check could falsely report errors when PTEs are remapped:
arch/loongarch/mm/init.c:204: bad pte 0000000107b601c3.

The check was unnecessary because:
- Fixmap slots are designed to be reusable
- Callers are responsible for proper mapping management

Changes in v2:
- modify commit messages

---
 arch/loongarch/mm/init.c | 4 ----
 1 file changed, 4 deletions(-)
arch/loongarch/mm/init.c +0 -4
index c3e4586a7975..626c99b803a3 100644
@@ -192,10 +192,6 @@ void __init __set_fixmap(enum fixed_addresses idx,
 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
 
 	ptep = populate_kernel_pte(addr);
-	if (!pte_none(ptep_get(ptep))) {
-		pte_ERROR(*ptep);
-		return;
-	}
 
 	if (pgprot_val(flags))
 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
Signature · 1 line
2.34.1
maintainer
#7 lore raw
Quoted 23 lines · george

From: George Guo <[email protected]>

This check could falsely report errors when PTEs are remapped:
arch/loongarch/mm/init.c:204: bad pte 0000000107b601c3.

The check was unnecessary because:
- Fixmap slots are designed to be reusable
- Callers are responsible for proper mapping management

Changes in v2:
- modify commit messages

---
 arch/loongarch/mm/init.c | 4 ----
 1 file changed, 4 deletions(-)
arch/loongarch/mm/init.c +0 -0
index c3e4586a7975..626c99b803a3 100644

arch/loongarch/mm/init.c void __init __set_fixmap(enum fixed_addresses idx, · line 198

Quoted 7 lines
arch/loongarch/mm/init.c +0 -4
@@ -192,10 +192,6 @@ void __init __set_fixmap(enum fixed_addresses idx,
        BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
 
        ptep = populate_kernel_pte(addr);
-       if (!pte_none(ptep_get(ptep))) {
-               pte_ERROR(*ptep);
-               return;
-       }

This only makes sense if we have FIX_TEXT_POKE, if without it, fixmap won't change at runtime.

Huacai

Quoted 6 lines
arch/loongarch/mm/init.c +0 -1
        if (pgprot_val(flags))
                set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
--

2.34.1