improve memcg oom killer robustness

Back before smart OOM killing, when faulting tasks where killed directly on allocation failures, the arch-specific fault handlers needed special protection for the init process.

Now that all fault handlers call into the generic OOM killer (609838c "mm: invoke oom-killer from remaining unconverted page fault handlers"), which already provides init protection, the arch-specific leftovers can be removed.

Signed-off-by Johannes Weiner <[email protected]>

---
 arch/arc/mm/fault.c   | 5 -----
 arch/score/mm/fault.c | 6 ------
 arch/tile/mm/fault.c  | 6 ------
 3 files changed, 17 deletions(-)

arch/arc/mm/fault.c +0 -5

index 0fd1f0d..6b0bb41 100644
@@ -122,7 +122,6 @@ good_area:
 			goto bad_area;
 	}
 
-survive:
 	/*
 	 * If for any reason at all we couldn't handle the fault,
 	 * make sure we exit gracefully rather than endlessly redo

@@ -201,10 +200,6 @@ no_context:
 	die("Oops", regs, address);
 
 out_of_memory:
-	if (is_global_init(tsk)) {
-		yield();
-		goto survive;
-	}
 	up_read(&mm->mmap_sem);
 
 	if (user_mode(regs)) {

arch/score/mm/fault.c +0 -6

index 6b18fb0..4b71a62 100644
@@ -100,7 +100,6 @@ good_area:
 			goto bad_area;
 	}
 
-survive:
 	/*
 	* If for any reason at all we couldn't handle the fault,
 	* make sure we exit gracefully rather than endlessly redo

@@ -167,11 +166,6 @@ no_context:
 	*/
 out_of_memory:
 	up_read(&mm->mmap_sem);
-	if (is_global_init(tsk)) {
-		yield();
-		down_read(&mm->mmap_sem);
-		goto survive;
-	}
 	if (!user_mode(regs))
 		goto no_context;
 	pagefault_out_of_memory();

arch/tile/mm/fault.c +0 -6

index f7f99f9..ac553ee 100644
@@ -430,7 +430,6 @@ good_area:
 			goto bad_area;
 	}
 
- survive:
 	/*
 	 * If for any reason at all we couldn't handle the fault,
 	 * make sure we exit gracefully rather than endlessly redo

@@ -568,11 +567,6 @@ no_context:
  */
 out_of_memory:
 	up_read(&mm->mmap_sem);
-	if (is_global_init(tsk)) {
-		yield();
-		down_read(&mm->mmap_sem);
-		goto survive;
-	}
 	if (is_kernel_mode)
 		goto no_context;
 	pagefault_out_of_memory();

Signature · 1 line

1.8.3.2

Back before smart OOM killing, when faulting tasks where killed directly on allocation failures, the arch-specific fault handlers needed special protection for the init process.

Now that all fault handlers call into the generic OOM killer (609838c "mm: invoke oom-killer from remaining unconverted page fault handlers"), which already provides init protection, the arch-specific leftovers can be removed.

Signed-off-by Johannes Weiner <[email protected]>

Kernel faults are expected to handle OOM conditions gracefully (gup, uaccess etc.), so they should never invoke the OOM killer. Reserve this for faults triggered in user context when it is the only option.

Most architectures already do this, fix up the remaining few.

Signed-off-by Johannes Weiner <[email protected]>

---
 arch/arm/mm/fault.c       | 14 +++++++-------
 arch/arm64/mm/fault.c     | 14 +++++++-------
 arch/avr32/mm/fault.c     |  2 +-
 arch/mips/mm/fault.c      |  2 ++
 arch/um/kernel/trap.c     |  2 ++
 arch/unicore32/mm/fault.c | 14 +++++++-------
 6 files changed, 26 insertions(+), 22 deletions(-)

arch/arm/mm/fault.c +7 -7

index c97f794..217bcbf 100644
@@ -349,6 +349,13 @@ retry:
 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
 		return 0;
 
+	/*
+	 * If we are in kernel mode at this point, we
+	 * have no context to handle this fault with.
+	 */
+	if (!user_mode(regs))
+		goto no_context;
+
 	if (fault & VM_FAULT_OOM) {
 		/*
 		 * We ran out of memory, call the OOM killer, and return to

@@ -359,13 +366,6 @@ retry:
 		return 0;
 	}
 
-	/*
-	 * If we are in kernel mode at this point, we
-	 * have no context to handle this fault with.
-	 */
-	if (!user_mode(regs))
-		goto no_context;
-
 	if (fault & VM_FAULT_SIGBUS) {
 		/*
 		 * We had some memory, but were unable to

arch/arm64/mm/fault.c +7 -7

index 0ecac89..dab1cfd 100644
@@ -294,6 +294,13 @@ retry:
 			      VM_FAULT_BADACCESS))))
 		return 0;
 
+	/*
+	 * If we are in kernel mode at this point, we have no context to
+	 * handle this fault with.
+	 */
+	if (!user_mode(regs))
+		goto no_context;
+
 	if (fault & VM_FAULT_OOM) {
 		/*
 		 * We ran out of memory, call the OOM killer, and return to

@@ -304,13 +311,6 @@ retry:
 		return 0;
 	}
 
-	/*
-	 * If we are in kernel mode at this point, we have no context to
-	 * handle this fault with.
-	 */
-	if (!user_mode(regs))
-		goto no_context;
-
 	if (fault & VM_FAULT_SIGBUS) {
 		/*
 		 * We had some memory, but were unable to successfully fix up

arch/avr32/mm/fault.c +1 -1

index b2f2d2d..2ca27b0 100644
@@ -228,9 +228,9 @@ no_context:
 	 */
 out_of_memory:
 	up_read(&mm->mmap_sem);
-	pagefault_out_of_memory();
 	if (!user_mode(regs))
 		goto no_context;
+	pagefault_out_of_memory();
 	return;
 
 do_sigbus:

arch/mips/mm/fault.c +2 -0

index 85df1cd..94d3a31 100644
@@ -241,6 +241,8 @@ out_of_memory:
 	 * (which will retry the fault, or kill us if we got oom-killed).
 	 */
 	up_read(&mm->mmap_sem);
+	if (!user_mode(regs))
+		goto no_context;
 	pagefault_out_of_memory();
 	return;
 

arch/um/kernel/trap.c +2 -0

index 089f398..b2f5adf 100644
@@ -124,6 +124,8 @@ out_of_memory:
 	 * (which will retry the fault, or kill us if we got oom-killed).
 	 */
 	up_read(&mm->mmap_sem);
+	if (!is_user)
+		goto out_nosemaphore;
 	pagefault_out_of_memory();
 	return 0;
 }

arch/unicore32/mm/fault.c +7 -7

index f9b5c10..8ed3c45 100644
@@ -278,6 +278,13 @@ retry:
 	       (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
 		return 0;
 
+	/*
+	 * If we are in kernel mode at this point, we
+	 * have no context to handle this fault with.
+	 */
+	if (!user_mode(regs))
+		goto no_context;
+
 	if (fault & VM_FAULT_OOM) {
 		/*
 		 * We ran out of memory, call the OOM killer, and return to

@@ -288,13 +295,6 @@ retry:
 		return 0;
 	}
 
-	/*
-	 * If we are in kernel mode at this point, we
-	 * have no context to handle this fault with.
-	 */
-	if (!user_mode(regs))
-		goto no_context;
-
 	if (fault & VM_FAULT_SIGBUS) {
 		/*
 		 * We had some memory, but were unable to

Signature · 1 line

1.8.3.2

Kernel faults are expected to handle OOM conditions gracefully (gup, uaccess etc.), so they should never invoke the OOM killer. Reserve this for faults triggered in user context when it is the only option.

Most architectures already do this, fix up the remaining few.

Signed-off-by Johannes Weiner <[email protected]>

Quoted 4 lines · Johannes Weiner

Kernel faults are expected to handle OOM conditions gracefully (gup, uaccess etc.), so they should never invoke the OOM killer. Reserve this for faults triggered in user context when it is the only option.

Quoted 3 lines

Most architectures already do this, fix up the remaining few.

Signed-off-by Johannes Weiner <[email protected]>

OK. but now almost all arch have the same page fault handler. So, I think we can implement arch generic page fault handler in future. Ah, ok, never mind if you are not interest.

Acked-by KOSAKI Motohiro <[email protected]>

Unlike global OOM handling, memory cgroup code will invoke the OOM killer in any OOM situation because it has no way of telling faults occuring in kernel context - which could be handled more gracefully - from user-triggered faults.

Pass a flag that identifies faults originating in user space from the architecture-specific fault handlers to generic code so that memcg OOM handling can be improved.

Signed-off-by Johannes Weiner <[email protected]>

---
 arch/alpha/mm/fault.c      |  7 ++++---
 arch/arc/mm/fault.c        |  6 ++++--
 arch/arm/mm/fault.c        |  9 ++++++---
 arch/arm64/mm/fault.c      |  9 ++++++---
 arch/avr32/mm/fault.c      |  2 ++
 arch/cris/mm/fault.c       |  6 ++++--
 arch/frv/mm/fault.c        | 10 ++++++----
 arch/hexagon/mm/vm_fault.c |  6 ++++--
 arch/ia64/mm/fault.c       |  6 ++++--
 arch/m32r/mm/fault.c       | 10 ++++++----
 arch/m68k/mm/fault.c       |  2 ++
 arch/metag/mm/fault.c      |  6 ++++--
 arch/microblaze/mm/fault.c |  7 +++++--
 arch/mips/mm/fault.c       |  6 ++++--
 arch/mn10300/mm/fault.c    |  2 ++
 arch/openrisc/mm/fault.c   |  1 +
 arch/parisc/mm/fault.c     |  7 +++++--
 arch/powerpc/mm/fault.c    |  7 ++++---
 arch/s390/mm/fault.c       |  2 ++
 arch/score/mm/fault.c      |  7 ++++++-
 arch/sh/mm/fault.c         |  9 ++++++---
 arch/sparc/mm/fault_32.c   | 12 +++++++++---
 arch/sparc/mm/fault_64.c   |  8 +++++---
 arch/tile/mm/fault.c       |  7 +++++--
 arch/um/kernel/trap.c      | 20 ++++++++++++--------
 arch/unicore32/mm/fault.c  |  8 ++++++--
 arch/x86/mm/fault.c        |  8 +++++---
 arch/xtensa/mm/fault.c     |  2 ++
 include/linux/mm.h         |  1 +
 29 files changed, 132 insertions(+), 61 deletions(-)

arch/alpha/mm/fault.c +4 -3

index 0c4132d..98838a0 100644
@@ -89,8 +89,7 @@ do_page_fault(unsigned long address, unsigned long mmcsr,
 	const struct exception_table_entry *fixup;
 	int fault, si_code = SEGV_MAPERR;
 	siginfo_t info;
-	unsigned int flags = (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-			      (cause > 0 ? FAULT_FLAG_WRITE : 0));
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 	/* As of EV6, a load into $31/$f31 is a prefetch, and never faults
 	   (or is suppressed by the PALcode).  Support that for older CPUs

@@ -115,7 +114,8 @@ do_page_fault(unsigned long address, unsigned long mmcsr,
 	if (address >= TASK_SIZE)
 		goto vmalloc_fault;
 #endif
-
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
 retry:
 	down_read(&mm->mmap_sem);
 	vma = find_vma(mm, address);

@@ -142,6 +142,7 @@ retry:
 	} else {
 		if (!(vma->vm_flags & VM_WRITE))
 			goto bad_area;
+		flags |= FAULT_FLAG_WRITE;
 	}
 
 	/* If for any reason at all we couldn't handle the fault,

arch/arc/mm/fault.c +4 -2

index 6b0bb41..d63f3de 100644
@@ -60,8 +60,7 @@ void do_page_fault(struct pt_regs *regs, unsigned long address)
 	siginfo_t info;
 	int fault, ret;
 	int write = regs->ecr_cause & ECR_C_PROTV_STORE;  /* ST/EX */
-	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-				(write ? FAULT_FLAG_WRITE : 0);
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 	/*
 	 * We fault-in kernel-space virtual memory on-demand. The

@@ -89,6 +88,8 @@ void do_page_fault(struct pt_regs *regs, unsigned long address)
 	if (in_atomic() || !mm)
 		goto no_context;
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
 retry:
 	down_read(&mm->mmap_sem);
 	vma = find_vma(mm, address);

@@ -117,6 +118,7 @@ good_area:
 	if (write) {
 		if (!(vma->vm_flags & VM_WRITE))
 			goto bad_area;
+		flags |= FAULT_FLAG_WRITE;
 	} else {
 		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 			goto bad_area;

arch/arm/mm/fault.c +6 -3

index 217bcbf..eb8830a 100644
@@ -261,9 +261,7 @@ do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 	struct task_struct *tsk;
 	struct mm_struct *mm;
 	int fault, sig, code;
-	int write = fsr & FSR_WRITE;
-	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-				(write ? FAULT_FLAG_WRITE : 0);
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 	if (notify_page_fault(regs, fsr))
 		return 0;

@@ -282,6 +280,11 @@ do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 	if (in_atomic() || !mm)
 		goto no_context;
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
+	if (fsr & FSR_WRITE)
+		flags |= FAULT_FLAG_WRITE;
+
 	/*
 	 * As per x86, we may deadlock here.  However, since the kernel only
 	 * validly references user space from well defined areas of the code,

arch/arm64/mm/fault.c +6 -3

index dab1cfd..12205b4 100644
@@ -208,9 +208,7 @@ static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
 	struct task_struct *tsk;
 	struct mm_struct *mm;
 	int fault, sig, code;
-	bool write = (esr & ESR_WRITE) && !(esr & ESR_CM);
-	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-		(write ? FAULT_FLAG_WRITE : 0);
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 	tsk = current;
 	mm  = tsk->mm;

@@ -226,6 +224,11 @@ static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
 	if (in_atomic() || !mm)
 		goto no_context;
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
+	if ((esr & ESR_WRITE) && !(esr & ESR_CM))
+		flags |= FAULT_FLAG_WRITE;
+
 	/*
 	 * As per x86, we may deadlock here. However, since the kernel only
 	 * validly references user space from well defined areas of the code,

arch/avr32/mm/fault.c +2 -0

index 2ca27b0..0eca933 100644
@@ -86,6 +86,8 @@ asmlinkage void do_page_fault(unsigned long ecr, struct pt_regs *regs)
 
 	local_irq_enable();
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
 retry:
 	down_read(&mm->mmap_sem);
 

arch/cris/mm/fault.c +4 -2

index 73312ab..1790f22 100644
@@ -58,8 +58,7 @@ do_page_fault(unsigned long address, struct pt_regs *regs,
 	struct vm_area_struct * vma;
 	siginfo_t info;
 	int fault;
-	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-				((writeaccess & 1) ? FAULT_FLAG_WRITE : 0);
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 	D(printk(KERN_DEBUG
 		 "Page fault for %lX on %X at %lX, prot %d write %d\n",

@@ -117,6 +116,8 @@ do_page_fault(unsigned long address, struct pt_regs *regs,
 	if (in_atomic() || !mm)
 		goto no_context;
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
 retry:
 	down_read(&mm->mmap_sem);
 	vma = find_vma(mm, address);

@@ -155,6 +156,7 @@ retry:
 	} else if (writeaccess == 1) {
 		if (!(vma->vm_flags & VM_WRITE))
 			goto bad_area;
+		flags |= FAULT_FLAG_WRITE;
 	} else {
 		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 			goto bad_area;

arch/frv/mm/fault.c +6 -4

index 331c1e2..9a66372 100644
@@ -34,11 +34,11 @@ asmlinkage void do_page_fault(int datammu, unsigned long esr0, unsigned long ear
 	struct vm_area_struct *vma;
 	struct mm_struct *mm;
 	unsigned long _pme, lrai, lrad, fixup;
+	unsigned long flags = 0;
 	siginfo_t info;
 	pgd_t *pge;
 	pud_t *pue;
 	pte_t *pte;
-	int write;
 	int fault;
 
 #if 0

@@ -81,6 +81,9 @@ asmlinkage void do_page_fault(int datammu, unsigned long esr0, unsigned long ear
 	if (in_atomic() || !mm)
 		goto no_context;
 
+	if (user_mode(__frame))
+		flags |= FAULT_FLAG_USER;
+
 	down_read(&mm->mmap_sem);
 
 	vma = find_vma(mm, ear0);

@@ -129,7 +132,6 @@ asmlinkage void do_page_fault(int datammu, unsigned long esr0, unsigned long ear
  */
  good_area:
 	info.si_code = SEGV_ACCERR;
-	write = 0;
 	switch (esr0 & ESR0_ATXC) {
 	default:
 		/* handle write to write protected page */

@@ -140,7 +142,7 @@ asmlinkage void do_page_fault(int datammu, unsigned long esr0, unsigned long ear
 #endif
 		if (!(vma->vm_flags & VM_WRITE))
 			goto bad_area;
-		write = 1;
+		flags |= FAULT_FLAG_WRITE;
 		break;
 
 		 /* handle read from protected page */

@@ -162,7 +164,7 @@ asmlinkage void do_page_fault(int datammu, unsigned long esr0, unsigned long ear
 	 * make sure we exit gracefully rather than endlessly redo
 	 * the fault.
 	 */
-	fault = handle_mm_fault(mm, vma, ear0, write ? FAULT_FLAG_WRITE : 0);
+	fault = handle_mm_fault(mm, vma, ear0, flags);
 	if (unlikely(fault & VM_FAULT_ERROR)) {
 		if (fault & VM_FAULT_OOM)
 			goto out_of_memory;

arch/hexagon/mm/vm_fault.c +4 -2

index 1bd276d..8704c93 100644
@@ -53,8 +53,7 @@ void do_page_fault(unsigned long address, long cause, struct pt_regs *regs)
 	int si_code = SEGV_MAPERR;
 	int fault;
 	const struct exception_table_entry *fixup;
-	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-				 (cause > 0 ? FAULT_FLAG_WRITE : 0);
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 	/*
 	 * If we're in an interrupt or have no user context,

@@ -65,6 +64,8 @@ void do_page_fault(unsigned long address, long cause, struct pt_regs *regs)
 
 	local_irq_enable();
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
 retry:
 	down_read(&mm->mmap_sem);
 	vma = find_vma(mm, address);

@@ -96,6 +97,7 @@ good_area:
 	case FLT_STORE:
 		if (!(vma->vm_flags & VM_WRITE))
 			goto bad_area;
+		flags |= FAULT_FLAG_WRITE;
 		break;
 	}
 

arch/ia64/mm/fault.c +4 -2

index 6cf0341..7225dad 100644
@@ -90,8 +90,6 @@ ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *re
 	mask = ((((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT)
 		| (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT));
 
-	flags |= ((mask & VM_WRITE) ? FAULT_FLAG_WRITE : 0);
-
 	/* mmap_sem is performance critical.... */
 	prefetchw(&mm->mmap_sem);
 

@@ -119,6 +117,10 @@ ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *re
 	if (notify_page_fault(regs, TRAP_BRKPT))
 		return;
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
+	if (mask & VM_WRITE)
+		flags |= FAULT_FLAG_WRITE;
 retry:
 	down_read(&mm->mmap_sem);
 

arch/m32r/mm/fault.c +6 -4

index 3cdfa9c..e9c6a80 100644
@@ -78,7 +78,7 @@ asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long error_code,
 	struct mm_struct *mm;
 	struct vm_area_struct * vma;
 	unsigned long page, addr;
-	int write;
+	unsigned long flags = 0;
 	int fault;
 	siginfo_t info;
 

@@ -117,6 +117,9 @@ asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long error_code,
 	if (in_atomic() || !mm)
 		goto bad_area_nosemaphore;
 
+	if (error_code & ACE_USERMODE)
+		flags |= FAULT_FLAG_USER;
+
 	/* When running in the kernel we expect faults to occur only to
 	 * addresses in user space.  All other faults represent errors in the
 	 * kernel and should generate an OOPS.  Unfortunately, in the case of an

@@ -166,14 +169,13 @@ asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long error_code,
  */
 good_area:
 	info.si_code = SEGV_ACCERR;
-	write = 0;
 	switch (error_code & (ACE_WRITE|ACE_PROTECTION)) {
 		default:	/* 3: write, present */
 			/* fall through */
 		case ACE_WRITE:	/* write, not present */
 			if (!(vma->vm_flags & VM_WRITE))
 				goto bad_area;
-			write++;
+			flags |= FAULT_FLAG_WRITE;
 			break;
 		case ACE_PROTECTION:	/* read, present */
 		case 0:		/* read, not present */

@@ -194,7 +196,7 @@ good_area:
 	 */
 	addr = (address & PAGE_MASK);
 	set_thread_fault_code(error_code);
-	fault = handle_mm_fault(mm, vma, addr, write ? FAULT_FLAG_WRITE : 0);
+	fault = handle_mm_fault(mm, vma, addr, flags);
 	if (unlikely(fault & VM_FAULT_ERROR)) {
 		if (fault & VM_FAULT_OOM)
 			goto out_of_memory;

arch/m68k/mm/fault.c +2 -0

index a563727..eb1d61f 100644
@@ -88,6 +88,8 @@ int do_page_fault(struct pt_regs *regs, unsigned long address,
 	if (in_atomic() || !mm)
 		goto no_context;
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
 retry:
 	down_read(&mm->mmap_sem);
 

arch/metag/mm/fault.c +4 -2

index 8fddf46..332680e 100644
@@ -53,8 +53,7 @@ int do_page_fault(struct pt_regs *regs, unsigned long address,
 	struct vm_area_struct *vma, *prev_vma;
 	siginfo_t info;
 	int fault;
-	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-				(write_access ? FAULT_FLAG_WRITE : 0);
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 	tsk = current;
 

@@ -109,6 +108,8 @@ int do_page_fault(struct pt_regs *regs, unsigned long address,
 	if (in_atomic() || !mm)
 		goto no_context;
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
 retry:
 	down_read(&mm->mmap_sem);
 

@@ -121,6 +122,7 @@ good_area:
 	if (write_access) {
 		if (!(vma->vm_flags & VM_WRITE))
 			goto bad_area;
+		flags |= FAULT_FLAG_WRITE;
 	} else {
 		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
 			goto bad_area;

arch/microblaze/mm/fault.c +5 -2

index 731f739..fa4cf52 100644
@@ -92,8 +92,7 @@ void do_page_fault(struct pt_regs *regs, unsigned long address,
 	int code = SEGV_MAPERR;
 	int is_write = error_code & ESR_S;
 	int fault;
-	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-					 (is_write ? FAULT_FLAG_WRITE : 0);
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 	regs->ear = address;
 	regs->esr = error_code;

@@ -121,6 +120,9 @@ void do_page_fault(struct pt_regs *regs, unsigned long address,
 		die("Weird page fault", regs, SIGSEGV);
 	}
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
+
 	/* When running in the kernel we expect faults to occur only to
 	 * addresses in user space.  All other faults represent errors in the
 	 * kernel and should generate an OOPS.  Unfortunately, in the case of an

@@ -199,6 +201,7 @@ good_area:
 	if (unlikely(is_write)) {
 		if (unlikely(!(vma->vm_flags & VM_WRITE)))
 			goto bad_area;
+		flags |= FAULT_FLAG_WRITE;
 	/* a read */
 	} else {
 		/* protection fault */

arch/mips/mm/fault.c +4 -2

index 94d3a31..becc42b 100644
@@ -42,8 +42,7 @@ static void __kprobes __do_page_fault(struct pt_regs *regs, unsigned long write,
 	const int field = sizeof(unsigned long) * 2;
 	siginfo_t info;
 	int fault;
-	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-						 (write ? FAULT_FLAG_WRITE : 0);
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 #if 0
 	printk("Cpu%d[%s:%d:%0*lx:%ld:%0*lx]\n", raw_smp_processor_id(),

@@ -93,6 +92,8 @@ static void __kprobes __do_page_fault(struct pt_regs *regs, unsigned long write,
 	if (in_atomic() || !mm)
 		goto bad_area_nosemaphore;
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
 retry:
 	down_read(&mm->mmap_sem);
 	vma = find_vma(mm, address);

@@ -114,6 +115,7 @@ good_area:
 	if (write) {
 		if (!(vma->vm_flags & VM_WRITE))
 			goto bad_area;
+		flags |= FAULT_FLAG_WRITE;
 	} else {
 		if (cpu_has_rixi) {
 			if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {

arch/mn10300/mm/fault.c +2 -0

index 8a2e6de..3516cbd 100644
@@ -171,6 +171,8 @@ asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long fault_code,
 	if (in_atomic() || !mm)
 		goto no_context;
 
+	if ((fault_code & MMUFCR_xFC_ACCESS) == MMUFCR_xFC_ACCESS_USR)
+		flags |= FAULT_FLAG_USER;
 retry:
 	down_read(&mm->mmap_sem);
 

arch/openrisc/mm/fault.c +1 -0

index 4a41f84..0703acf 100644
@@ -86,6 +86,7 @@ asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long address,
 	if (user_mode(regs)) {
 		/* Exception was in userspace: reenable interrupts */
 		local_irq_enable();
+		flags |= FAULT_FLAG_USER;
 	} else {
 		/* If exception was in a syscall, then IRQ's may have
 		 * been enabled or disabled.  If they were enabled,

arch/parisc/mm/fault.c +5 -2

index f247a34..d10d27a 100644
@@ -180,6 +180,10 @@ void do_page_fault(struct pt_regs *regs, unsigned long code,
 	if (in_atomic() || !mm)
 		goto no_context;
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
+	if (acc_type & VM_WRITE)
+		flags |= FAULT_FLAG_WRITE;
 retry:
 	down_read(&mm->mmap_sem);
 	vma = find_vma_prev(mm, address, &prev_vma);

@@ -203,8 +207,7 @@ good_area:
 	 * fault.
 	 */
 
-	fault = handle_mm_fault(mm, vma, address,
-			flags | ((acc_type & VM_WRITE) ? FAULT_FLAG_WRITE : 0));
+	fault = handle_mm_fault(mm, vma, address, flags);
 
 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 		return;

arch/powerpc/mm/fault.c +4 -3

index 8726779..d9196c9 100644
@@ -223,9 +223,6 @@ int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
 	is_write = error_code & ESR_DST;
 #endif /* CONFIG_4xx || CONFIG_BOOKE */
 
-	if (is_write)
-		flags |= FAULT_FLAG_WRITE;
-
 #ifdef CONFIG_PPC_ICSWX
 	/*
 	 * we need to do this early because this "data storage

@@ -280,6 +277,9 @@ int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
 
 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
+
 	/* When running in the kernel we expect faults to occur only to
 	 * addresses in user space.  All other faults represent errors in the
 	 * kernel and should generate an OOPS.  Unfortunately, in the case of an

@@ -408,6 +408,7 @@ good_area:
 	} else if (is_write) {
 		if (!(vma->vm_flags & VM_WRITE))
 			goto bad_area;
+		flags |= FAULT_FLAG_WRITE;
 	/* a read */
 	} else {
 		/* protection fault */

arch/s390/mm/fault.c +2 -0

index f00aefb..6fa7b05 100644
@@ -302,6 +302,8 @@ static inline int do_exception(struct pt_regs *regs, int access)
 	address = trans_exc_code & __FAIL_ADDR_MASK;
 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 	flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
+	if (regs->psw.mask & PSW_MASK_PSTATE)
+		flags |= FAULT_FLAG_USER;
 	if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
 		flags |= FAULT_FLAG_WRITE;
 	down_read(&mm->mmap_sem);

arch/score/mm/fault.c +6 -1

index 4b71a62..52238983 100644
@@ -47,6 +47,7 @@ asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long write,
 	struct task_struct *tsk = current;
 	struct mm_struct *mm = tsk->mm;
 	const int field = sizeof(unsigned long) * 2;
+	unsigned long flags = 0;
 	siginfo_t info;
 	int fault;
 

@@ -75,6 +76,9 @@ asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long write,
 	if (in_atomic() || !mm)
 		goto bad_area_nosemaphore;
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
+
 	down_read(&mm->mmap_sem);
 	vma = find_vma(mm, address);
 	if (!vma)

@@ -95,6 +99,7 @@ good_area:
 	if (write) {
 		if (!(vma->vm_flags & VM_WRITE))
 			goto bad_area;
+		flags |= FAULT_FLAG_WRITE;
 	} else {
 		if (!(vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)))
 			goto bad_area;

@@ -105,7 +110,7 @@ good_area:
 	* make sure we exit gracefully rather than endlessly redo
 	* the fault.
 	*/
-	fault = handle_mm_fault(mm, vma, address, write);
+	fault = handle_mm_fault(mm, vma, address, flags);
 	if (unlikely(fault & VM_FAULT_ERROR)) {
 		if (fault & VM_FAULT_OOM)
 			goto out_of_memory;

arch/sh/mm/fault.c +6 -3

index 1f49c28..541dc61 100644
@@ -400,9 +400,7 @@ asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
 	struct mm_struct *mm;
 	struct vm_area_struct * vma;
 	int fault;
-	int write = error_code & FAULT_CODE_WRITE;
-	unsigned int flags = (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-			      (write ? FAULT_FLAG_WRITE : 0));
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 	tsk = current;
 	mm = tsk->mm;

@@ -476,6 +474,11 @@ good_area:
 
 	set_thread_fault_code(error_code);
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
+	if (error_code & FAULT_CODE_WRITE)
+		flags |= FAULT_FLAG_WRITE;
+
 	/*
 	 * If for any reason at all we couldn't handle the fault,
 	 * make sure we exit gracefully rather than endlessly redo

arch/sparc/mm/fault_32.c +9 -3

index e98bfda..59dbd46 100644
@@ -177,8 +177,7 @@ asmlinkage void do_sparc_fault(struct pt_regs *regs, int text_fault, int write,
 	unsigned long g2;
 	int from_user = !(regs->psr & PSR_PS);
 	int fault, code;
-	unsigned int flags = (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-			      (write ? FAULT_FLAG_WRITE : 0));
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 	if (text_fault)
 		address = regs->pc;

@@ -235,6 +234,11 @@ good_area:
 			goto bad_area;
 	}
 
+	if (from_user)
+		flags |= FAULT_FLAG_USER;
+	if (write)
+		flags |= FAULT_FLAG_WRITE;
+
 	/*
 	 * If for any reason at all we couldn't handle the fault,
 	 * make sure we exit gracefully rather than endlessly redo

@@ -383,6 +387,7 @@ static void force_user_fault(unsigned long address, int write)
 	struct vm_area_struct *vma;
 	struct task_struct *tsk = current;
 	struct mm_struct *mm = tsk->mm;
+	unsigned int flags = FAULT_FLAG_USER;
 	int code;
 
 	code = SEGV_MAPERR;

@@ -402,11 +407,12 @@ good_area:
 	if (write) {
 		if (!(vma->vm_flags & VM_WRITE))
 			goto bad_area;
+		flags |= FAULT_FLAG_WRITE;
 	} else {
 		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 			goto bad_area;
 	}
-	switch (handle_mm_fault(mm, vma, address, write ? FAULT_FLAG_WRITE : 0)) {
+	switch (handle_mm_fault(mm, vma, address, flags)) {
 	case VM_FAULT_SIGBUS:
 	case VM_FAULT_OOM:
 		goto do_sigbus;

arch/sparc/mm/fault_64.c +5 -3

index 5062ff3..c08b9bb 100644
@@ -314,8 +314,9 @@ asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
 		} else {
 			bad_kernel_pc(regs, address);
 			return;
-		}
-	}
+		}		
+	} else
+		flags |= FAULT_FLAG_USER;
 
 	/*
 	 * If we're in an interrupt or have no user

@@ -418,13 +419,14 @@ good_area:
 		    vma->vm_file != NULL)
 			set_thread_fault_code(fault_code |
 					      FAULT_CODE_BLKCOMMIT);
+
+		flags |= FAULT_FLAG_WRITE;
 	} else {
 		/* Allow reads even for write-only mappings */
 		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 			goto bad_area;
 	}
 
-	flags |= ((fault_code & FAULT_CODE_WRITE) ? FAULT_FLAG_WRITE : 0);
 	fault = handle_mm_fault(mm, vma, address, flags);
 
 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))

arch/tile/mm/fault.c +5 -2

index ac553ee..3ff289f 100644
@@ -280,8 +280,7 @@ static int handle_page_fault(struct pt_regs *regs,
 	if (!is_page_fault)
 		write = 1;
 
-	flags = (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-		 (write ? FAULT_FLAG_WRITE : 0));
+	flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 	is_kernel_mode = (EX1_PL(regs->ex1) != USER_PL);
 

@@ -365,6 +364,9 @@ static int handle_page_fault(struct pt_regs *regs,
 		goto bad_area_nosemaphore;
 	}
 
+	if (!is_kernel_mode)
+		flags |= FAULT_FLAG_USER;
+
 	/*
 	 * When running in the kernel we expect faults to occur only to
 	 * addresses in user space.  All other faults represent errors in the

@@ -425,6 +427,7 @@ good_area:
 #endif
 		if (!(vma->vm_flags & VM_WRITE))
 			goto bad_area;
+		flags |= FAULT_FLAG_WRITE;
 	} else {
 		if (!is_page_fault || !(vma->vm_flags & VM_READ))
 			goto bad_area;

arch/um/kernel/trap.c +12 -8

index b2f5adf..5c3aef7 100644
@@ -30,8 +30,7 @@ int handle_page_fault(unsigned long address, unsigned long ip,
 	pmd_t *pmd;
 	pte_t *pte;
 	int err = -EFAULT;
-	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-				 (is_write ? FAULT_FLAG_WRITE : 0);
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 	*code_out = SEGV_MAPERR;
 

@@ -42,6 +41,8 @@ int handle_page_fault(unsigned long address, unsigned long ip,
 	if (in_atomic())
 		goto out_nosemaphore;
 
+	if (is_user)
+		flags |= FAULT_FLAG_USER;
 retry:
 	down_read(&mm->mmap_sem);
 	vma = find_vma(mm, address);

@@ -58,12 +59,15 @@ retry:
 
 good_area:
 	*code_out = SEGV_ACCERR;
-	if (is_write && !(vma->vm_flags & VM_WRITE))
-		goto out;
-
-	/* Don't require VM_READ|VM_EXEC for write faults! */
-	if (!is_write && !(vma->vm_flags & (VM_READ | VM_EXEC)))
-		goto out;
+	if (is_write) {
+		if (!(vma->vm_flags & VM_WRITE))
+			goto out;
+		flags |= FAULT_FLAG_WRITE;
+	} else {
+		/* Don't require VM_READ|VM_EXEC for write faults! */
+		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
+			goto out;
+	}
 
 	do {
 		int fault;

arch/unicore32/mm/fault.c +6 -2

index 8ed3c45..0dc922d 100644
@@ -209,8 +209,7 @@ static int do_pf(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 	struct task_struct *tsk;
 	struct mm_struct *mm;
 	int fault, sig, code;
-	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-				 ((!(fsr ^ 0x12)) ? FAULT_FLAG_WRITE : 0);
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 	tsk = current;
 	mm = tsk->mm;

@@ -222,6 +221,11 @@ static int do_pf(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 	if (in_atomic() || !mm)
 		goto no_context;
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
+	if (!(fsr ^ 0x12))
+		flags |= FAULT_FLAG_WRITE;
+
 	/*
 	 * As per x86, we may deadlock here.  However, since the kernel only
 	 * validly references user space from well defined areas of the code,

arch/x86/mm/fault.c +5 -3

index 654be4a..6d77c38 100644
@@ -1011,9 +1011,7 @@ __do_page_fault(struct pt_regs *regs, unsigned long error_code)
 	unsigned long address;
 	struct mm_struct *mm;
 	int fault;
-	int write = error_code & PF_WRITE;
-	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
-					(write ? FAULT_FLAG_WRITE : 0);
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
 	tsk = current;
 	mm = tsk->mm;

@@ -1083,6 +1081,7 @@ __do_page_fault(struct pt_regs *regs, unsigned long error_code)
 	if (user_mode_vm(regs)) {
 		local_irq_enable();
 		error_code |= PF_USER;
+		flags |= FAULT_FLAG_USER;
 	} else {
 		if (regs->flags & X86_EFLAGS_IF)
 			local_irq_enable();

@@ -1109,6 +1108,9 @@ __do_page_fault(struct pt_regs *regs, unsigned long error_code)
 		return;
 	}
 
+	if (error_code & PF_WRITE)
+		flags |= FAULT_FLAG_WRITE;
+
 	/*
 	 * When running in the kernel we expect faults to occur only to
 	 * addresses in user space.  All other faults represent errors in

arch/xtensa/mm/fault.c +2 -0

index 4b7bc8d..70fa7bc 100644
@@ -72,6 +72,8 @@ void do_page_fault(struct pt_regs *regs)
 	       address, exccause, regs->pc, is_write? "w":"", is_exec? "x":"");
 #endif
 
+	if (user_mode(regs))
+		flags |= FAULT_FLAG_USER;
 retry:
 	down_read(&mm->mmap_sem);
 	vma = find_vma(mm, address);

include/linux/mm.h +1 -0

index d5c82dc..c51fc32 100644
@@ -170,6 +170,7 @@ extern pgprot_t protection_map[16];
 #define FAULT_FLAG_RETRY_NOWAIT	0x10	/* Don't drop mmap_sem and wait when retrying */
 #define FAULT_FLAG_KILLABLE	0x20	/* The fault task is in SIGKILL killable region */
 #define FAULT_FLAG_TRIED	0x40	/* second try */
+#define FAULT_FLAG_USER		0x80	/* The fault originated in userspace */
 
 /*
  * vm_fault is filled by the the pagefault handler and passed to the vma's

Signature · 1 line

1.8.3.2

Quoted 5 lines · Johannes Weiner

Unlike global OOM handling, memory cgroup code will invoke the OOM killer in any OOM situation because it has no way of telling faults occuring in kernel context - which could be handled more gracefully - from user-triggered faults.

Quoted 5 lines

Pass a flag that identifies faults originating in user space from the architecture-specific fault handlers to generic code so that memcg OOM handling can be improved.

Signed-off-by Johannes Weiner <[email protected]>

Looks good to me but I guess maintainers of the affected archs should be CCed

Reviewed-by Michal Hocko <[email protected]>

The x86 fault handler bails in the middle of error handling when the task has a fatal signal pending. For a subsequent patch this is a problem in OOM situations because it relies on pagefault_out_of_memory() being called even when the task has been killed, to perform proper per-task OOM state unwinding.

Shortcutting the fault like this is a rather minor optimization that saves a few instructions in rare cases. Just remove it for user-triggered faults.

Use the opportunity to split the fault retry handling from actual fault errors and add locking documentation that reads suprisingly similar to ARM's.

Signed-off-by Johannes Weiner <[email protected]>

---
 arch/x86/mm/fault.c | 35 +++++++++++++++++------------------
 1 file changed, 17 insertions(+), 18 deletions(-)

arch/x86/mm/fault.c +17 -18

index 6d77c38..3aaeffc 100644
@@ -842,23 +842,15 @@ do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
 	force_sig_info_fault(SIGBUS, code, address, tsk, fault);
 }
 
-static noinline int
+static noinline void
 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
 	       unsigned long address, unsigned int fault)
 {
-	/*
-	 * Pagefault was interrupted by SIGKILL. We have no reason to
-	 * continue pagefault.
-	 */
-	if (fatal_signal_pending(current)) {
-		if (!(fault & VM_FAULT_RETRY))
-			up_read(&current->mm->mmap_sem);
-		if (!(error_code & PF_USER))
-			no_context(regs, error_code, address, 0, 0);
-		return 1;
+	if (fatal_signal_pending(current) && !(error_code & PF_USER)) {
+		up_read(&current->mm->mmap_sem);
+		no_context(regs, error_code, address, 0, 0);
+		return;
 	}
-	if (!(fault & VM_FAULT_ERROR))
-		return 0;
 
 	if (fault & VM_FAULT_OOM) {
 		/* Kernel mode? Handle exceptions or die: */

@@ -866,7 +858,7 @@ mm_fault_error(struct pt_regs *regs, unsigned long error_code,
 			up_read(&current->mm->mmap_sem);
 			no_context(regs, error_code, address,
 				   SIGSEGV, SEGV_MAPERR);
-			return 1;
+			return;
 		}
 
 		up_read(&current->mm->mmap_sem);

@@ -884,7 +876,6 @@ mm_fault_error(struct pt_regs *regs, unsigned long error_code,
 		else
 			BUG();
 	}
-	return 1;
 }
 
 static int spurious_fault_check(unsigned long error_code, pte_t *pte)

@@ -1189,9 +1180,17 @@ good_area:
 	 */
 	fault = handle_mm_fault(mm, vma, address, flags);
 
-	if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) {
-		if (mm_fault_error(regs, error_code, address, fault))
-			return;
+	/*
+	 * If we need to retry but a fatal signal is pending, handle the
+	 * signal first. We do not need to release the mmap_sem because it
+	 * would already be released in __lock_page_or_retry in mm/filemap.c.
+	 */
+	if (unlikely((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)))
+		return;
+
+	if (unlikely(fault & VM_FAULT_ERROR)) {
+		mm_fault_error(regs, error_code, address, fault);
+		return;
 	}
 
 	/*

Signature · 1 line

1.8.3.2

Quoted 6 lines · Johannes Weiner

The x86 fault handler bails in the middle of error handling when the task has a fatal signal pending. For a subsequent patch this is a problem in OOM situations because it relies on pagefault_out_of_memory() being called even when the task has been killed, to perform proper per-task OOM state unwinding.

Quoted 3 lines

Shortcutting the fault like this is a rather minor optimization that saves a few instructions in rare cases. Just remove it for user-triggered faults.

OK, I thought that this optimization tries to prevent calling OOM because the current might release some memory but that wasn't the intention of b80ef10e8 (x86: Move do_page_fault()'s error path under unlikely()).

Quoted 5 lines

Use the opportunity to split the fault retry handling from actual fault errors and add locking documentation that reads suprisingly similar to ARM's.

Signed-off-by Johannes Weiner <[email protected]>

Reviewed-by Michal Hocko <[email protected]>

Quoted 11 lines · Michal Hocko

Quoted 9 lines · Johannes Weiner

The x86 fault handler bails in the middle of error handling when the task has a fatal signal pending. For a subsequent patch this is a problem in OOM situations because it relies on pagefault_out_of_memory() being called even when the task has been killed, to perform proper per-task OOM state unwinding.

Shortcutting the fault like this is a rather minor optimization that saves a few instructions in rare cases. Just remove it for user-triggered faults.

Quoted 4 lines

OK, I thought that this optimization tries to prevent calling OOM because the current might release some memory but that wasn't the intention of b80ef10e8 (x86: Move do_page_fault()'s error path under unlikely()).

out_of_memory() also checks the caller for pending signals, so it would not actually invoke the OOM killer if the caller is already dying.

Quoted 7 lines

Quoted 5 lines

Use the opportunity to split the fault retry handling from actual fault errors and add locking documentation that reads suprisingly similar to ARM's.

Signed-off-by Johannes Weiner <[email protected]>

Reviewed-by Michal Hocko <[email protected]>

Thanks!

The x86 fault handler bails in the middle of error handling when the task has a fatal signal pending. For a subsequent patch this is a problem in OOM situations because it relies on pagefault_out_of_memory() being called even when the task has been killed, to perform proper per-task OOM state unwinding.

Shortcutting the fault like this is a rather minor optimization that saves a few instructions in rare cases. Just remove it for user-triggered faults.

Use the opportunity to split the fault retry handling from actual fault errors and add locking documentation that reads suprisingly similar to ARM's.

Signed-off-by Johannes Weiner <[email protected]>

System calls and kernel faults (uaccess, gup) can handle an out of memory situation gracefully and just return -ENOMEM.

Enable the memcg OOM killer only for user faults, where it's really the only option available.

Signed-off-by Johannes Weiner <[email protected]>

---
 include/linux/memcontrol.h | 23 +++++++++++++++++++++++
 include/linux/sched.h      |  3 +++
 mm/filemap.c               | 11 ++++++++++-
 mm/memcontrol.c            |  2 +-
 mm/memory.c                | 40 ++++++++++++++++++++++++++++++----------
 5 files changed, 67 insertions(+), 12 deletions(-)

include/linux/memcontrol.h +0 -0

index 7b4d9d7..9bb5eeb 100644

include/linux/memcontrol.h +15 -0

+ * @new: true to enable, false to disable
+ *
+ * Toggle whether a failed memcg charge should invoke the OOM killer
+ * or just return -ENOMEM.  Returns the previous toggle state.
+ */
+static inline bool mem_cgroup_xchg_may_oom(struct task_struct *p, bool new)
+{
+	bool old;
+
+	old = p->memcg_oom.may_oom;
+	p->memcg_oom.may_oom = new;
+
+	return old;
+}
+
 #ifdef CONFIG_MEMCG_SWAP
 extern int do_swap_account;
 #endif

mm/filemap.c +0 -0

index a6981fe..2932810 100644

mm/filemap.c +3 -0

 		do_async_mmap_readahead(vma, ra, file, page, offset);
+		mem_cgroup_xchg_may_oom(current, may_oom);
 	} else if (!page) {
 		/* No page in the page cache at all */
+		may_oom = mem_cgroup_xchg_may_oom(current, 0);
 		do_sync_mmap_readahead(vma, ra, file, offset);
+		mem_cgroup_xchg_may_oom(current, may_oom);
 		count_vm_event(PGMAJFAULT);
 		mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
 		ret = VM_FAULT_MAJOR;

mm/memory.c +0 -0

index f2ab2a8..5ea7b47 100644

mm/memory.c +14 -0

@@ -3851,6 +3843,34 @@ retry:
 	return handle_pte_fault(mm, vma, address, pte, pmd, flags);
 }
 
+int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
+		    unsigned long address, unsigned int flags)
+{
+	int ret;
+
+	__set_current_state(TASK_RUNNING);
+
+	count_vm_event(PGFAULT);
+	mem_cgroup_count_vm_event(mm, PGFAULT);
+
+	/* do counter updates before entering really critical section. */
+	check_sync_rss_stat(current);
+
+	/*

mm/memory.c +4 -0

+
+	return ret;
+}
+
 #ifndef __PAGETABLE_PUD_FOLDED
 /*
  * Allocate page upper directory.

Signature · 1 line

1.8.3.2

Quoted 7 lines · Johannes Weiner

System calls and kernel faults (uaccess, gup) can handle an out of memory situation gracefully and just return -ENOMEM.

Enable the memcg OOM killer only for user faults, where it's really the only option available.

Signed-off-by Johannes Weiner <[email protected]>

It looks OK to me, but I have few comments bellow. Nothing really huge but I do not like mem_cgroup_xchg_may_oom for !MEMCG.

Quoted 16 lines

---
 include/linux/memcontrol.h | 23 +++++++++++++++++++++++
 include/linux/sched.h      |  3 +++
 mm/filemap.c               | 11 ++++++++++-
 mm/memcontrol.c            |  2 +-
 mm/memory.c                | 40 ++++++++++++++++++++++++++++++----------
 5 files changed, 67 insertions(+), 12 deletions(-)

include/linux/memcontrol.h +0 -0

index 7b4d9d7..9bb5eeb 100644
@@ -125,6 +125,24 @@ extern void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
 extern void mem_cgroup_replace_page_cache(struct page *oldpage,
 					struct page *newpage);
 

include/linux/memcontrol.h extern void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, · line 130

Quoted 3 lines

include/linux/memcontrol.h +3 -0

+/**
+ * mem_cgroup_xchg_may_oom - toggle the memcg OOM killer for a task
+ * @p: task

Is this ever safe to call on !current? If not then I wouldn't allow to give p as a parameter.

include/linux/memcontrol.h int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) · line 1642

Quoted 5 lines

include/linux/memcontrol.h +1 -0

@@ -1634,10 +1639,14 @@ int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 		 * We found the page, so try async readahead before
 		 * waiting for the lock.
 		 */
+		may_oom = mem_cgroup_xchg_may_oom(current, 0);

s/0/false/

below ditto

Quoted 10 lines

include/linux/memcontrol.h +3 -0

 		do_async_mmap_readahead(vma, ra, file, page, offset);
+		mem_cgroup_xchg_may_oom(current, may_oom);
 	} else if (!page) {
 		/* No page in the page cache at all */
+		may_oom = mem_cgroup_xchg_may_oom(current, 0);
 		do_sync_mmap_readahead(vma, ra, file, offset);
+		mem_cgroup_xchg_may_oom(current, may_oom);
 		count_vm_event(PGMAJFAULT);
 		mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
 		ret = VM_FAULT_MAJOR;

[...]

Quoted 4 lines

mm/memory.c +0 -0

index f2ab2a8..5ea7b47 100644

[...]

Quoted 24 lines

mm/memory.c +20 -0

@@ -3851,6 +3843,34 @@ retry:
 	return handle_pte_fault(mm, vma, address, pte, pmd, flags);
 }
 
+int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
+		    unsigned long address, unsigned int flags)
+{
+	int ret;
+
+	__set_current_state(TASK_RUNNING);
+
+	count_vm_event(PGFAULT);
+	mem_cgroup_count_vm_event(mm, PGFAULT);
+
+	/* do counter updates before entering really critical section. */
+	check_sync_rss_stat(current);
+
+	/*
+	 * Enable the memcg OOM handling for faults triggered in user
+	 * space.  Kernel faults are handled more gracefully.
+	 */
+	if (flags & FAULT_FLAG_USER)
+		WARN_ON(mem_cgroup_xchg_may_oom(current, true) == true);
+

mm/memory.c retry: · line 3869

Quoted 4 lines

mm/memory.c +4 -0

+	ret = __handle_mm_fault(mm, vma, address, flags);
+
+	if (flags & FAULT_FLAG_USER)
+		WARN_ON(mem_cgroup_xchg_may_oom(current, false) == false);

Ohh, I see why you used !new in mem_cgroup_xchg_may_oom for !MEMCG case above. This could be fixed easily if you add mem_cgroup_{enable,disable}_oom which would be empty for !MEMCG.

System calls and kernel faults (uaccess, gup) can handle an out of memory situation gracefully and just return -ENOMEM.

Enable the memcg OOM killer only for user faults, where it's really the only option available.

Signed-off-by Johannes Weiner <[email protected]>

---
  include/linux/memcontrol.h | 23 +++++++++++++++++++++++
  include/linux/sched.h      |  3 +++
  mm/filemap.c               | 11 ++++++++++-
  mm/memcontrol.c            |  2 +-
  mm/memory.c                | 40 ++++++++++++++++++++++++++++++----------
  5 files changed, 67 insertions(+), 12 deletions(-)

include/linux/memcontrol.h +7 -0

index 7b4d9d7..9bb5eeb 100644
@@ -125,6 +125,24 @@ extern void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
  extern void mem_cgroup_replace_page_cache(struct page *oldpage,
  					struct page *newpage);
 
+/**
+ * mem_cgroup_xchg_may_oom - toggle the memcg OOM killer for a task
+ * @p: task
+ * @new: true to enable, false to disable
+ *
+ * Toggle whether a failed memcg charge should invoke the OOM killer
+ * or just return -ENOMEM.  Returns the previous toggle state.

mm/filemap.c +9 -1

@@ -1626,7 +1627,11 @@ int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
  		return VM_FAULT_SIGBUS;
 
  	/*
-	 * Do we have something in the page cache already?
+	 * Do we have something in the page cache already?  Either
+	 * way, try readahead, but disable the memcg OOM killer for it
+	 * as readahead is optional and no errors are propagated up
+	 * the fault stack.  The OOM killer is enabled while trying to
+	 * instantiate the faulting page individually below.
  	 */
  	page = find_get_page(mapping, offset);
  	if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {

@@ -1634,10 +1639,14 @@ int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
  		 * We found the page, so try async readahead before
  		 * waiting for the lock.
  		 */
+		may_oom = mem_cgroup_xchg_may_oom(current, 0);
  		do_async_mmap_readahead(vma, ra, file, page, offset);
+		mem_cgroup_xchg_may_oom(current, may_oom);
  	} else if (!page) {
  		/* No page in the page cache at all */
+		may_oom = mem_cgroup_xchg_may_oom(current, 0);
  		do_sync_mmap_readahead(vma, ra, file, offset);
+		mem_cgroup_xchg_may_oom(current, may_oom);
  		count_vm_event(PGMAJFAULT);
  		mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
  		ret = VM_FAULT_MAJOR;

mm/memcontrol.c +1 -1

index 00a7a66..30ae46a 100644
@@ -2614,7 +2614,7 @@ static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
  		return CHARGE_RETRY;
 
  	/* If we don't need to call oom-killer at el, return immediately */
-	if (!oom_check)
+	if (!oom_check || !current->memcg_oom.may_oom)
  		return CHARGE_NOMEM;
  	/* check OOM */
  	if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize)))

mm/memory.c +11 -10

index f2ab2a8..5ea7b47 100644
@@ -3752,22 +3752,14 @@ unlock:
  /*
   * By the time we get here, we already hold the mm semaphore
   */
-int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
-		unsigned long address, unsigned int flags)
+static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
+			     unsigned long address, unsigned int flags)
  {
  	pgd_t *pgd;
  	pud_t *pud;
  	pmd_t *pmd;
  	pte_t *pte;
 
-	__set_current_state(TASK_RUNNING);
-
-	count_vm_event(PGFAULT);
-	mem_cgroup_count_vm_event(mm, PGFAULT);
-
-	/* do counter updates before entering really critical section. */
-	check_sync_rss_stat(current);
-
  	if (unlikely(is_vm_hugetlb_page(vma)))
  		return hugetlb_fault(mm, vma, address, flags);
 

@@ -3851,6 +3843,34 @@ retry:
  	return handle_pte_fault(mm, vma, address, pte, pmd, flags);
  }
 
+int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
+		    unsigned long address, unsigned int flags)
+{
+	int ret;
+
+	__set_current_state(TASK_RUNNING);
+
+	count_vm_event(PGFAULT);
+	mem_cgroup_count_vm_event(mm, PGFAULT);

Quoted 36 lines · Johannes Weiner

System calls and kernel faults (uaccess, gup) can handle an out of memory situation gracefully and just return -ENOMEM.

Enable the memcg OOM killer only for user faults, where it's really the only option available.

Signed-off-by Johannes Weiner <[email protected]>

---
  include/linux/memcontrol.h | 23 +++++++++++++++++++++++
  include/linux/sched.h      |  3 +++
  mm/filemap.c               | 11 ++++++++++-
  mm/memcontrol.c            |  2 +-
  mm/memory.c                | 40 ++++++++++++++++++++++++++++++----------
  5 files changed, 67 insertions(+), 12 deletions(-)

include/linux/memcontrol.h +13 -0

index 7b4d9d7..9bb5eeb 100644
@@ -125,6 +125,24 @@ extern void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
  extern void mem_cgroup_replace_page_cache(struct page *oldpage,
  					struct page *newpage);
 
+/**
+ * mem_cgroup_xchg_may_oom - toggle the memcg OOM killer for a task
+ * @p: task
+ * @new: true to enable, false to disable
+ *
+ * Toggle whether a failed memcg charge should invoke the OOM killer
+ * or just return -ENOMEM.  Returns the previous toggle state.
+ */
+static inline bool mem_cgroup_xchg_may_oom(struct task_struct *p, bool new)
+{
+	bool old;
+
+	old = p->memcg_oom.may_oom;

include/linux/memcontrol.h extern void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, · line 144

Quoted 4 lines

include/linux/memcontrol.h +4 -0

+	p->memcg_oom.may_oom = new;
+
+	return old;
+}

The name of xchg strongly suggest the function use compare-swap op. So, it seems misleading name. I suggest just use "set_*" or something else. In linux kernel, many setter functions already return old value. Don't mind.

Quoted 7 lines

include/linux/sched.h +0 -0

index fc09d21..4b3effc 100644
@@ -1398,6 +1398,9 @@ struct task_struct {
  		unsigned long memsw_nr_pages; /* uncharged mem+swap usage */
  	} memcg_batch;

include/linux/sched.h struct task_struct { · line 1403

Quoted 4 lines

include/linux/sched.h +3 -0

  	unsigned int memcg_kmem_skip_account;
+	struct memcg_oom_info {
+		unsigned int may_oom:1;
+	} memcg_oom;

This ":1" makes slower but doesn't diet any memory space, right? I suggest to use bool. If anybody need to diet in future, he may change it to bit field. That's ok, let's stop too early and questionable micro optimization.

mm/filemap.c int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) · line 1620

Quoted 9 lines

mm/filemap.c +1 -0

index a6981fe..2932810 100644
@@ -1617,6 +1617,7 @@ int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
  	struct file_ra_state *ra = &file->f_ra;
  	struct inode *inode = mapping->host;
  	pgoff_t offset = vmf->pgoff;
+	unsigned int may_oom;

Why don't you use bool? your mem_cgroup_xchg_may_oom() uses bool and it seems cleaner more.

Quoted 18 lines

mm/filemap.c +14 -0

@@ -3851,6 +3843,34 @@ retry:
  	return handle_pte_fault(mm, vma, address, pte, pmd, flags);
  }
 
+int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
+		    unsigned long address, unsigned int flags)
+{
+	int ret;
+
+	__set_current_state(TASK_RUNNING);
+
+	count_vm_event(PGFAULT);
+	mem_cgroup_count_vm_event(mm, PGFAULT);
+
+	/* do counter updates before entering really critical section. */
+	check_sync_rss_stat(current);
+
+	/*

mm/filemap.c retry: · line 3864

Quoted 5 lines

mm/filemap.c +5 -0

+	 * Enable the memcg OOM handling for faults triggered in user
+	 * space.  Kernel faults are handled more gracefully.
+	 */
+	if (flags & FAULT_FLAG_USER)
+		WARN_ON(mem_cgroup_xchg_may_oom(current, true) == true);

Please don't assume WARN_ON never erase any code. I'm not surprised if embedded guys replace WARN_ON with nop in future.

Quoted 37 lines · KOSAKI Motohiro

Quoted 36 lines · Johannes Weiner

System calls and kernel faults (uaccess, gup) can handle an out of memory situation gracefully and just return -ENOMEM.

Enable the memcg OOM killer only for user faults, where it's really the only option available.

Signed-off-by Johannes Weiner <[email protected]>

---
   include/linux/memcontrol.h | 23 +++++++++++++++++++++++
   include/linux/sched.h      |  3 +++
   mm/filemap.c               | 11 ++++++++++-
   mm/memcontrol.c            |  2 +-
   mm/memory.c                | 40 ++++++++++++++++++++++++++++++----------
   5 files changed, 67 insertions(+), 12 deletions(-)

include/linux/memcontrol.h +13 -0

index 7b4d9d7..9bb5eeb 100644
@@ -125,6 +125,24 @@ extern void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
   extern void mem_cgroup_replace_page_cache(struct page *oldpage,
   					struct page *newpage);
 
+/**
+ * mem_cgroup_xchg_may_oom - toggle the memcg OOM killer for a task
+ * @p: task
+ * @new: true to enable, false to disable
+ *
+ * Toggle whether a failed memcg charge should invoke the OOM killer
+ * or just return -ENOMEM.  Returns the previous toggle state.
+ */
+static inline bool mem_cgroup_xchg_may_oom(struct task_struct *p, bool new)
+{
+	bool old;
+
+	old = p->memcg_oom.may_oom;

Quoted 8 lines

include/linux/memcontrol.h extern void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, · line 144

Quoted 4 lines

include/linux/memcontrol.h +4 -0

+	p->memcg_oom.may_oom = new;
+
+	return old;
+}

The name of xchg strongly suggest the function use compare-swap op. So, it seems misleading name. I suggest just use "set_*" or something else. In linux kernel, many setter functions already return old value. Don't mind.

I renamed it to bool mem_cgroup_toggle_oom(bool onoff) when I incorporated Michal's feedback, would you be okay with that?

Quoted 16 lines

include/linux/sched.h struct task_struct { · line 1403

Quoted 12 lines

include/linux/sched.h +3 -0

index fc09d21..4b3effc 100644
@@ -1398,6 +1398,9 @@ struct task_struct {
   		unsigned long memsw_nr_pages; /* uncharged mem+swap usage */
   	} memcg_batch;
   	unsigned int memcg_kmem_skip_account;
+	struct memcg_oom_info {
+		unsigned int may_oom:1;
+	} memcg_oom;

This ":1" makes slower but doesn't diet any memory space, right? I suggest to use bool. If anybody need to diet in future, he may change it to bit field. That's ok, let's stop too early and questionable micro optimization.

It should sit in the same word as the memcg_kmem_skip_account, plus I'm adding another bit in the next patch (in_memcg_oom), so we save space. It's also the OOM path, so anything but performance critical.

Quoted 11 lines

mm/filemap.c int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) · line 1620

Quoted 9 lines

mm/filemap.c +1 -0

index a6981fe..2932810 100644
@@ -1617,6 +1617,7 @@ int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
   	struct file_ra_state *ra = &file->f_ra;
   	struct inode *inode = mapping->host;
   	pgoff_t offset = vmf->pgoff;
+	unsigned int may_oom;

Why don't you use bool? your mem_cgroup_xchg_may_oom() uses bool and it seems cleaner more.

Yup, forgot to convert it with the interface, I changed it to bool.

Quoted 9 lines

mm/filemap.c int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)

Quoted 6 lines

mm/filemap.c +6 -0

+	/*
+	 * Enable the memcg OOM handling for faults triggered in user
+	 * space.  Kernel faults are handled more gracefully.
+	 */
+	if (flags & FAULT_FLAG_USER)
+		WARN_ON(mem_cgroup_xchg_may_oom(current, true) == true);

Please don't assume WARN_ON never erase any code. I'm not surprised if embedded guys replace WARN_ON with nop in future.

That would be really messed up.

But at the same time, the WARN_ON() obfuscates what's going on a little bit, so putting it separately should make the code more readable. I'll change it.

Thanks for your input!

The memcg OOM handling is incredibly fragile and can deadlock. When a task fails to charge memory, it invokes the OOM killer and loops right there in the charge code until it succeeds. Comparably, any other task that enters the charge path at this point will go to a waitqueue right then and there and sleep until the OOM situation is resolved. The problem is that these tasks may hold filesystem locks and the mmap_sem; locks that the selected OOM victim may need to exit.

For example, in one reported case, the task invoking the OOM killer was about to charge a page cache page during a write(), which holds the i_mutex. The OOM killer selected a task that was just entering truncate() and trying to acquire the i_mutex:

OOM invoking task:

[<ffffffff8110a9c1>] mem_cgroup_handle_oom+0x241/0x3b0
[<ffffffff8110b5ab>] T.1146+0x5ab/0x5c0
[<ffffffff8110c22e>] mem_cgroup_cache_charge+0xbe/0xe0
[<ffffffff810ca28c>] add_to_page_cache_locked+0x4c/0x140
[<ffffffff810ca3a2>] add_to_page_cache_lru+0x22/0x50
[<ffffffff810ca45b>] grab_cache_page_write_begin+0x8b/0xe0
[<ffffffff81193a18>] ext3_write_begin+0x88/0x270
[<ffffffff810c8fc6>] generic_file_buffered_write+0x116/0x290
[<ffffffff810cb3cc>] __generic_file_aio_write+0x27c/0x480
[<ffffffff810cb646>] generic_file_aio_write+0x76/0xf0           # takes ->i_mutex
[<ffffffff8111156a>] do_sync_write+0xea/0x130
[<ffffffff81112183>] vfs_write+0xf3/0x1f0
[<ffffffff81112381>] sys_write+0x51/0x90
[<ffffffff815b5926>] system_call_fastpath+0x18/0x1d

[<ffffffffffffffff>] 0xffffffffffffffff

OOM kill victim:

[<ffffffff811109b8>] do_truncate+0x58/0xa0              # takes i_mutex
[<ffffffff81121c90>] do_last+0x250/0xa30
[<ffffffff81122547>] path_openat+0xd7/0x440
[<ffffffff811229c9>] do_filp_open+0x49/0xa0
[<ffffffff8110f7d6>] do_sys_open+0x106/0x240
[<ffffffff8110f950>] sys_open+0x20/0x30
[<ffffffff815b5926>] system_call_fastpath+0x18/0x1d

[<ffffffffffffffff>] 0xffffffffffffffff

The OOM handling task will retry the charge indefinitely while the OOM killed task is not releasing any resources.

A similar scenario can happen when the kernel OOM killer for a memcg is disabled and a userspace task is in charge of resolving OOM situations. In this case, ALL tasks that enter the OOM path will be made to sleep on the OOM waitqueue and wait for userspace to free resources or increase the group's limit. But a userspace OOM handler is prone to deadlock itself on the locks held by the waiting tasks. For example one of the sleeping tasks may be stuck in a brk() call with the mmap_sem held for writing but the userspace handler, in order to pick an optimal victim, may need to read files from /proc/<pid>, which tries to acquire the same mmap_sem for reading and deadlocks.

This patch changes the way tasks behave after detecting a memcg OOM and makes sure nobody loops or sleeps with locks held:

1. When OOMing in a user fault, invoke the OOM killer and restart the fault instead of looping on the charge attempt. This way, the OOM victim can not get stuck on locks the looping task may hold.

2. When OOMing in a user fault but somebody else is handling it (either the kernel OOM killer or a userspace handler), don't go to sleep in the charge context. Instead, remember the OOMing memcg in the task struct and then fully unwind the page fault stack with -ENOMEM. pagefault_out_of_memory() will then call back into the memcg code to check if the -ENOMEM came from the memcg, and then either put the task to sleep on the memcg's OOM waitqueue or just restart the fault. The OOM victim can no longer get stuck on any lock a sleeping task may hold.

This relies on the memcg OOM killer only being enabled when an allocation failure will result in a call to pagefault_out_of_memory().

While reworking the OOM routine, also remove a needless OOM waitqueue wakeup when invoking the killer. In addition to the wakeup implied in the kill signal delivery, only uncharges and limit increases, things that actually change the memory situation, should poke the waitqueue.

Reported-by Reported-by: azurIt <[email protected]>

Debugged-by Michal Hocko <[email protected]>

Signed-off-by Johannes Weiner <[email protected]>

mm/memcontrol.c +0 -0

index 30ae46a..029a3a8 100644

mm/memcontrol.c +2 -4

@@ -2189,31 +2191,20 @@ static void memcg_oom_recover(struct mem_cgroup *memcg)
 }
 
 /*
- * try to call OOM killer. returns false if we should exit memory-reclaim loop.
+ * try to call OOM killer
  */
-static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask,
-				  int order)
+static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
 {
-	struct oom_wait_info owait;

mm/memcontrol.c +40 -14

+	if (!current->memcg_oom.may_oom)
+		return;
+
+	current->memcg_oom.in_memcg_oom = 1;
 
 	/* At first, try to OOM lock hierarchy under memcg.*/
 	spin_lock(&memcg_oom_lock);
 	locked = mem_cgroup_oom_lock(memcg);
-	/*
-	 * Even if signal_pending(), we can't quit charge() loop without
-	 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
-	 * under OOM is always welcomed, use TASK_KILLABLE here.
-	 */
-	prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
 	if (!locked || memcg->oom_kill_disable)
 		need_to_kill = false;
 	if (locked)

@@ -2221,24 +2212,100 @@ static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask,
 	spin_unlock(&memcg_oom_lock);
 
 	if (need_to_kill) {
-		finish_wait(&memcg_oom_waitq, &owait.wait);
 		mem_cgroup_out_of_memory(memcg, mask, order);
 	} else {
-		schedule();
-		finish_wait(&memcg_oom_waitq, &owait.wait);
+		/*
+		 * A system call can just return -ENOMEM, but if this
+		 * is a page fault and somebody else is handling the
+		 * OOM already, we need to sleep on the OOM waitqueue
+		 * for this memcg until the situation is resolved.
+		 * Which can take some time because it might be
+		 * handled by a userspace task.
+		 *
+		 * However, this is the charge context, which means
+		 * that we may sit on a large call stack and hold
+		 * various filesystem locks, the mmap_sem etc. and we
+		 * don't want the OOM handler to deadlock on them
+		 * while we sit here and wait.  Store the current OOM
+		 * context in the task_struct, then return -ENOMEM.
+		 * At the end of the page fault handler, with the
+		 * stack unwound, pagefault_out_of_memory() will check
+		 * back with us by calling
+		 * mem_cgroup_oom_synchronize(), possibly putting the
+		 * task to sleep.
+		 */
+		mem_cgroup_mark_under_oom(memcg);
+		current->memcg_oom.wakeups = atomic_read(&memcg->oom_wakeups);
+		css_get(&memcg->css);
+		current->memcg_oom.wait_on_memcg = memcg;
 	}
-	spin_lock(&memcg_oom_lock);
-	if (locked)
+
+	if (locked) {
+		spin_lock(&memcg_oom_lock);
 		mem_cgroup_oom_unlock(memcg);
-	memcg_wakeup_oom(memcg);
-	spin_unlock(&memcg_oom_lock);
+		/*
+		 * Sleeping tasks might have been killed, make sure
+		 * they get scheduled so they can exit.
+		 */
+		if (need_to_kill)
+			memcg_oom_recover(memcg);
+		spin_unlock(&memcg_oom_lock);
+	}
+}
 
-	mem_cgroup_unmark_under_oom(memcg);

Quoted 12 lines · Johannes Weiner

(quoted diff) +2 -4

@@ -2189,31 +2191,20 @@ static void memcg_oom_recover(struct mem_cgroup *memcg)
 }
 
 /*
- * try to call OOM killer. returns false if we should exit memory-reclaim loop.
+ * try to call OOM killer
  */
-static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask,
-				  int order)
+static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
 {
-	struct oom_wait_info owait;

Review static void memcg_oom_recover(struct mem_cgroup *memcg) · line 2206

Quoted 10 lines

(quoted diff) +1 -8

-	bool locked, need_to_kill;
+	bool locked, need_to_kill = true;
 
-	owait.memcg = memcg;
-	owait.wait.flags = 0;
-	owait.wait.func = memcg_oom_wake_function;
-	owait.wait.private = current;
-	INIT_LIST_HEAD(&owait.wait.task_list);
-	need_to_kill = true;
-	mem_cgroup_mark_under_oom(memcg);

You are marking memcg under_oom only for the sleepers. So if we have no sleepers then the memcg will never report it is under oom which is a behavior change. On the other hand who-ever relies on under_oom under such conditions (it would basically mean a busy loop reading memory.oom_control) would be racy anyway so it is questionable it matters at all. At least now when we do not have any active notification that under_oom has changed.

Anyway, this shouldn't be a part of this patch so if you want it because it saves a pointless hierarchy traversal then make it a separate patch with explanation why the new behavior is still OK.