Skip to contentSkip to conversation

mm: simplify the VM_BUG_ON condition in pmdp_huge_clear_flush()

PATCH Applied Miaohe Lin 2 participants 3 messages Feb 1, 2021 – Feb 2, 2021

Subsystems (2)
lore
Share & tools
lore

Miaohe Lin

author
#0 lore raw

PATCH 1/1 mm: simplify the VM_BUG_ON condition in pmdp_huge_clear_flush()

The condition (A && !C && !D) || !A is equivalent to !A || (A && !C && !D) and can be further simplified to !A || (!C && !D).

---
 mm/pgtable-generic.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
mm/pgtable-generic.c +2 -2
index 9578db83e312..fa1375f3e3b2 100644
@@ -135,8 +135,8 @@ pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma, unsigned long address,
 {
 	pmd_t pmd;
 	VM_BUG_ON(address & ~HPAGE_PMD_MASK);
-	VM_BUG_ON((pmd_present(*pmdp) && !pmd_trans_huge(*pmdp) &&
-			   !pmd_devmap(*pmdp)) || !pmd_present(*pmdp));
+	VM_BUG_ON(!pmd_present(*pmdp) || (!pmd_trans_huge(*pmdp) &&
+					  !pmd_devmap(*pmdp)));
 	pmd = pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
 	flush_pmd_tlb_range(vma, address, address + HPAGE_PMD_SIZE);
 	return pmd;
Signature · 1 line
2.19.1
maintainer
#1 lore raw
Quoted 8 lines · On Mon, 1 Feb 2021 06:43:19 -0500 Miaohe Lin <[email protected]> wrote:

The condition (A && !C && !D) || !A is equivalent to !A || (A && !C && !D) and can be further simplified to !A || (!C && !D).

..

mm/pgtable-generic.c pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma, unsigned long address, · line 142

Quoted 10 lines
mm/pgtable-generic.c +2 -2
@@ -135,8 +135,8 @@ pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma, unsigned long address,
 {
 	pmd_t pmd;
 	VM_BUG_ON(address & ~HPAGE_PMD_MASK);
-	VM_BUG_ON((pmd_present(*pmdp) && !pmd_trans_huge(*pmdp) &&
-			   !pmd_devmap(*pmdp)) || !pmd_present(*pmdp));
+	VM_BUG_ON(!pmd_present(*pmdp) || (!pmd_trans_huge(*pmdp) &&
+					  !pmd_devmap(*pmdp)));
 	pmd = pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
 	flush_pmd_tlb_range(vma, address, address + HPAGE_PMD_SIZE);
 	return pmd;

True, and the resulting code is still readable enough.

But a problem with such a complex expression is that the developer will have trouble figuring out why the BUG actually triggered.

If we had a VM_BUG_ON_PMD() then we could print the pmd's value and permit diagnosis from that. But we don't have such a thing.

So I suggest that it would be better to have

	VM_BUG_ON((pmd_present(*pmdp) && !pmd_trans_huge(*pmdp) &&
			   !pmd_devmap(*pmdp)));
	VM_BUG_ON(!pmd_present(*pmdp));

This way, the BUG()'s file-n-line output will tell us more about why the kernel went splat.

I suppose maybe this could be optimized the same way, as

	VM_BUG_ON(!pmd_present(*pmdp));
	/* Below assumes pmd_present() is true */
	VM_BUG_ON(!pmd_trans_huge(*pmdp) && !pmd_devmap(*pmdp));

Which works because VM_BUG_ON is, depending up Kconfig, either a no-op or a noreturn-if-it-triggered. I'm not sure if I like this trick much though.

author
#2 lore raw

Hi:

Quoted 23 lines · Andrew Morton
Quoted 8 lines · On Mon, 1 Feb 2021 06:43:19 -0500 Miaohe Lin <[email protected]> wrote:

The condition (A && !C && !D) || !A is equivalent to !A || (A && !C && !D) and can be further simplified to !A || (!C && !D).

..

mm/pgtable-generic.c pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma, unsigned long address, · line 142

Quoted 10 lines
mm/pgtable-generic.c +2 -2
@@ -135,8 +135,8 @@ pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma, unsigned long address,
 {
 	pmd_t pmd;
 	VM_BUG_ON(address & ~HPAGE_PMD_MASK);
-	VM_BUG_ON((pmd_present(*pmdp) && !pmd_trans_huge(*pmdp) &&
-			   !pmd_devmap(*pmdp)) || !pmd_present(*pmdp));
+	VM_BUG_ON(!pmd_present(*pmdp) || (!pmd_trans_huge(*pmdp) &&
+					  !pmd_devmap(*pmdp)));
 	pmd = pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
 	flush_pmd_tlb_range(vma, address, address + HPAGE_PMD_SIZE);
 	return pmd;

True, and the resulting code is still readable enough.

Quoted 3 lines

But a problem with such a complex expression is that the developer will have trouble figuring out why the BUG actually triggered.

Agree! We can determine which condition is failing through the line number __but__ we can't figure out exactly which one triggered BUG for a complex expression.

Quoted 13 lines

If we had a VM_BUG_ON_PMD() then we could print the pmd's value and permit diagnosis from that. But we don't have such a thing.

So I suggest that it would be better to have

	VM_BUG_ON((pmd_present(*pmdp) && !pmd_trans_huge(*pmdp) &&
			   !pmd_devmap(*pmdp)));
	VM_BUG_ON(!pmd_present(*pmdp));

This way, the BUG()'s file-n-line output will tell us more about why the kernel went splat.

Quoted 5 lines

I suppose maybe this could be optimized the same way, as

	VM_BUG_ON(!pmd_present(*pmdp));
	/* Below assumes pmd_present() is true */
	VM_BUG_ON(!pmd_trans_huge(*pmdp) && !pmd_devmap(*pmdp));

This one looks good and provide more information than before. I can send another patch to do this (and feel free to merge into this one), should I ?

Many thanks.

Quoted 6 lines

Which works because VM_BUG_ON is, depending up Kconfig, either a no-op or a noreturn-if-it-triggered. I'm not sure if I like this trick much though.

.