Patchwork [1/1,v5] arm: fault.c: Port OOM changes into do_page_fault

login
register
mail settings
Submitter Kautuk Consul
Date Nov. 27, 2011, 5:58 a.m.
Message ID <CAFPAmTTWVwq8H6HVzdvTcN-jcSoNQ2-TfLWT33ks8_kpGND+SA@mail.gmail.com>
Download mbox | patch
Permalink /patch/127863/
State New
Headers show

Comments

Kautuk Consul - Nov. 27, 2011, 5:58 a.m.
>
> Thanks.  I think this is fine.  Please put it in the patch system so it's
> not forgotten, and I'll get to it sometime during the next week.  Thanks.
>

Thanks. I don't know what you mean by "put it in the patch system" as I already
did a "git send-email" for this patch to this mailing list.

So, please find the v5 patch file attached to this email for any
further reference.
Will Deacon - Nov. 27, 2011, 12:56 p.m.
On Sun, Nov 27, 2011 at 05:58:33AM +0000, Kautuk Consul wrote:
> >
> > Thanks.  I think this is fine.  Please put it in the patch system so it's
> > not forgotten, and I'll get to it sometime during the next week.  Thanks.
> >
> 
> Thanks. I don't know what you mean by "put it in the patch system" as I already
> did a "git send-email" for this patch to this mailing list.
> 
> So, please find the v5 patch file attached to this email for any
> further reference.

You should read the documentation here:

http://www.arm.linux.org.uk/developer/patches/info.php

and then you can use git send-email --suppress-cc=all to email your patch to
the patch system, making sure you have the correct KernelVersion tag.

Will
Kautuk Consul - Dec. 1, 2011, 7:40 a.m.
Thanks for the info, Will.

Just FYI, I have entered the patch into the patch system and it can be
accessed via :
http://www.arm.linux.org.uk/developer/patches/viewpatch.php?id=7178/1

Please inform me if I need to make any more changes to this with
respect to the status or
any other field(s).
Currently, the patch status is "Incoming".

Patch

From 9b7a0e6a271843246cfcc6a4dbb867d27f1868d1 Mon Sep 17 00:00:00 2001
From: Kautuk Consul <consul.kautuk@gmail.com>
Date: Tue, 22 Nov 2011 13:30:35 -0500
Subject: [PATCH 1/1 v5] arm: fault.c: Port OOM changes into do_page_fault

Commit d065bd810b6deb67d4897a14bfe21f8eb526ba99
(mm: retry page fault when blocking on disk transfer) and
commit 37b23e0525d393d48a7d59f870b3bc061a30ccdb
(x86,mm: make pagefault killable)

The above commits introduced changes into the x86 pagefault handler
for making the page fault handler retryable as well as killable.

These changes reduce the mmap_sem hold time, which is crucial
during OOM killer invocation.

Port these changes to ARM.

Without these changes, my ARM board encounters many hang and livelock
scenarios.
After applying this patch, OOM feature performance improves according to
my testing.

Signed-off-by: Kautuk Consul <consul.kautuk@gmail.com>
---
 arch/arm/mm/fault.c |   58 ++++++++++++++++++++++++++++++++++----------------
 1 files changed, 39 insertions(+), 19 deletions(-)

diff --git a/arch/arm/mm/fault.c b/arch/arm/mm/fault.c
index aa33949..4aabeae 100644
--- a/arch/arm/mm/fault.c
+++ b/arch/arm/mm/fault.c
@@ -231,7 +231,7 @@  static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
 
 static int __kprobes
 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
-		struct task_struct *tsk)
+		unsigned int flags, struct task_struct *tsk)
 {
 	struct vm_area_struct *vma;
 	int fault;
@@ -253,18 +253,7 @@  good_area:
 		goto out;
 	}
 
-	/*
-	 * If for any reason at all we couldn't handle the fault, make
-	 * sure we exit gracefully rather than endlessly redo the fault.
-	 */
-	fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, (fsr & FSR_WRITE) ? FAULT_FLAG_WRITE : 0);
-	if (unlikely(fault & VM_FAULT_ERROR))
-		return fault;
-	if (fault & VM_FAULT_MAJOR)
-		tsk->maj_flt++;
-	else
-		tsk->min_flt++;
-	return fault;
+	return handle_mm_fault(mm, vma, addr & PAGE_MASK, flags);
 
 check_stack:
 	if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
@@ -279,6 +268,9 @@  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);
 
 	if (notify_page_fault(regs, fsr))
 		return 0;
@@ -305,6 +297,7 @@  do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 	if (!down_read_trylock(&mm->mmap_sem)) {
 		if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
 			goto no_context;
+retry:
 		down_read(&mm->mmap_sem);
 	} else {
 		/*
@@ -320,14 +313,41 @@  do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 #endif
 	}
 
-	fault = __do_page_fault(mm, addr, fsr, tsk);
-	up_read(&mm->mmap_sem);
+	fault = __do_page_fault(mm, addr, fsr, flags, tsk);
+
+	/* 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 ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
+		return 0;
+
+	/*
+	 * Major/minor page fault accounting is only done on the
+	 * initial attempt. If we go through a retry, it is extremely
+	 * likely that the page will be found in page cache at that point.
+	 */
 
 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
-	if (fault & VM_FAULT_MAJOR)
-		perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, addr);
-	else if (fault & VM_FAULT_MINOR)
-		perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, addr);
+	if (flags & FAULT_FLAG_ALLOW_RETRY) {
+		if (fault & VM_FAULT_MAJOR) {
+			tsk->maj_flt++;
+			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
+					regs, addr);
+		} else {
+			tsk->min_flt++;
+			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
+					regs, addr);
+		}
+		if (fault & VM_FAULT_RETRY) {
+			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
+			* of starvation. */
+			flags &= ~FAULT_FLAG_ALLOW_RETRY;
+			goto retry;
+		}
+	}
+
+	up_read(&mm->mmap_sem);
 
 	/*
 	 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
-- 
1.7.5.4