Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Vineet Gupta <Vineet.Gupta1@synopsys.com>
To:     <linux-snps-arc@lists.infradead.org>
CC:     <paltsev@snyopsys.com>, <linux-kernel@vger.kernel.org>,
        Vineet Gupta <Vineet.Gupta1@synopsys.com>,
        Peter Zijlstra <peterz@infradead.org>
Subject: [PATCH 9/9] ARC: mm: do_page_fault refactor #8: release mmap_sem sooner
Date:   Tue, 14 May 2019 17:29:36 -0700
Message-ID: <1557880176-24964-10-git-send-email-vgupta@synopsys.com>
In-Reply-To: <1557880176-24964-1-git-send-email-vgupta@synopsys.com>
References: <1557880176-24964-1-git-send-email-vgupta@synopsys.com>
MIME-Version: 1.0
Content-Type: text/plain
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

In case of successful page fault handling, this patch releases mmap_sem
before updating the perf stat event for major/minor faults. So even
though the contention reduction is NOT supe rhigh, it is still an
improvement.

There's an additional code size improvement as we only have 2 up_read()
calls now.

Note to myself:
--------------

1. Given the way it is done, we are forced to move @bad_area label earlier
   causing the various "goto bad_area" cases to hit perf stat code.

 - PERF_COUNT_SW_PAGE_FAULTS is NOW updated for access errors which is what
   arm/arm64 seem to be doing as well (with slightly different code)
 - PERF_COUNT_SW_PAGE_FAULTS_{MAJ,MIN} must NOT be updated for the
   error case which is guarded by now setting @fault initial value
   to VM_FAULT_ERROR which serves both cases when handle_mm_fault()
   returns error or is not called at all.

2. arm/arm64 use two homebrew fault flags VM_FAULT_BAD{MAP,MAPACCESS}
   which I was inclined to add too but seems not needed for ARC

 - given that we have everything is 1 function we cabn stil use goto
 - we setup si_code at the right place (arm* do that in the end)
 - we init fault already to error value which guards entry into perf
   stats event update

Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
---
 arch/arc/mm/fault.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/arch/arc/mm/fault.c b/arch/arc/mm/fault.c
index 20402217d9da..e93ea06c214c 100644
--- a/arch/arc/mm/fault.c
+++ b/arch/arc/mm/fault.c
@@ -68,7 +68,7 @@ void do_page_fault(unsigned long address, struct pt_regs *regs)
 	struct mm_struct *mm = tsk->mm;
 	int sig, si_code = SEGV_MAPERR;
 	unsigned int write = 0, exec = 0, mask;
-	vm_fault_t fault;			/* handle_mm_fault() output */
+	vm_fault_t fault = VM_FAULT_ERROR;	/* handle_mm_fault() output */
 	unsigned int flags;			/* handle_mm_fault() input */
 
 	/*
@@ -155,6 +155,9 @@ void do_page_fault(unsigned long address, struct pt_regs *regs)
 		}
 	}
 
+bad_area:
+	up_read(&mm->mmap_sem);
+
 	/*
 	 * Major/minor page fault accounting
 	 * (in case of retry we only land here once)
@@ -173,13 +176,9 @@ void do_page_fault(unsigned long address, struct pt_regs *regs)
 		}
 
 		/* Normal return path: fault Handled Gracefully */
-		up_read(&mm->mmap_sem);
 		return;
 	}
 
-bad_area:
-	up_read(&mm->mmap_sem);
-
 	if (!user_mode(regs))
 		goto no_context;
 
-- 
2.7.4