[PATCH] powerpc: Clean up tm_abort duplication in hash_utils_64.c

The same logic appears twice and should probably be pulled out into a function.

Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Rui Teng <redacted>
---
 arch/powerpc/mm/hash_utils_64.c | 45 +++++++++++++++++------------------------
 1 file changed, 19 insertions(+), 26 deletions(-)

diff --git a/arch/powerpc/mm/hash_utils_64.c b/arch/powerpc/mm/hash_utils_64.c
index 0821556..69ef702 100644
--- a/arch/powerpc/mm/hash_utils_64.c
+++ b/arch/powerpc/mm/hash_utils_64.c

@@ -1460,6 +1460,23 @@ out_exit:
 	local_irq_restore(flags);
 }
 
+/*
+ * Transactions are not aborted by tlbiel, only tlbie.
+ * Without, syncing a page back to a block device w/ PIO could pick up
+ * transactional data (bad!) so we force an abort here.  Before the
+ * sync the page will be made read-only, which will flush_hash_page.
+ * BIG ISSUE here: if the kernel uses a page from userspace without
+ * unmapping it first, it may see the speculated version.
+ */
+void local_tm_abort(int local)
+{
+	if (local && cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
+		MSR_TM_ACTIVE(current->thread.regs->msr)) {
+		tm_enable();
+		tm_abort(TM_CAUSE_TLBI);
+	}
+}
+
 /* WARNING: This is called from hash_low_64.S, if you change this prototype,
  *          do not forget to update the assembly call site !
  */

@@ -1487,19 +1504,7 @@ void flush_hash_page(unsigned long vpn, real_pte_t pte, int psize, int ssize,
 	} pte_iterate_hashed_end();
 
 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
-	/* Transactions are not aborted by tlbiel, only tlbie.
-	 * Without, syncing a page back to a block device w/ PIO could pick up
-	 * transactional data (bad!) so we force an abort here.  Before the
-	 * sync the page will be made read-only, which will flush_hash_page.
-	 * BIG ISSUE here: if the kernel uses a page from userspace without
-	 * unmapping it first, it may see the speculated version.
-	 */
-	if (local && cpu_has_feature(CPU_FTR_TM) &&
-	    current->thread.regs &&
-	    MSR_TM_ACTIVE(current->thread.regs->msr)) {
-		tm_enable();
-		tm_abort(TM_CAUSE_TLBI);
-	}
+	local_tm_abort(local);
 #endif
 }
 

@@ -1558,19 +1563,7 @@ void flush_hash_hugepage(unsigned long vsid, unsigned long addr,
 	}
 tm_abort:
 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
-	/* Transactions are not aborted by tlbiel, only tlbie.
-	 * Without, syncing a page back to a block device w/ PIO could pick up
-	 * transactional data (bad!) so we force an abort here.  Before the
-	 * sync the page will be made read-only, which will flush_hash_page.
-	 * BIG ISSUE here: if the kernel uses a page from userspace without
-	 * unmapping it first, it may see the speculated version.
-	 */
-	if (local && cpu_has_feature(CPU_FTR_TM) &&
-	    current->thread.regs &&
-	    MSR_TM_ACTIVE(current->thread.regs->msr)) {
-		tm_enable();
-		tm_abort(TM_CAUSE_TLBI);
-	}
+	local_tm_abort(local);
 #endif
 	return;
 }

-- 
2.7.4

Am Freitag, 26 August 2016, 11:50:10 schrieb Rui Teng:

Since local_tm_abort is only used in this file, it should be static.

Also, since both places calling it are guarded by 
CONFIG_PPC_TRANSACTIONAL_MEM, wouldn't it be cleaner if the #ifdef was here 
instead and the #else block defined an empty static inline function? Then 
the call sites wouldn't need to be guarded.

-- 
[]'s
Thiago Jung Bauermann
IBM Linux Technology Center

On 9/1/16 11:46 PM, Thiago Jung Bauermann wrote:

Am Freitag, 26 August 2016, 11:50:10 schrieb Rui Teng:

quoted

The same logic appears twice and should probably be pulled out into a
function.

Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Rui Teng <redacted>
---
 arch/powerpc/mm/hash_utils_64.c | 45
+++++++++++++++++------------------------ 1 file changed, 19
insertions(+), 26 deletions(-)

diff --git a/arch/powerpc/mm/hash_utils_64.c

b/arch/powerpc/mm/hash_utils_64.c index 0821556..69ef702 100644

--- a/arch/powerpc/mm/hash_utils_64.c
+++ b/arch/powerpc/mm/hash_utils_64.c

@@ -1460,6 +1460,23 @@ out_exit:
 	local_irq_restore(flags);
 }

+/*
+ * Transactions are not aborted by tlbiel, only tlbie.
+ * Without, syncing a page back to a block device w/ PIO could pick up
+ * transactional data (bad!) so we force an abort here.  Before the
+ * sync the page will be made read-only, which will flush_hash_page.
+ * BIG ISSUE here: if the kernel uses a page from userspace without
+ * unmapping it first, it may see the speculated version.
+ */
+void local_tm_abort(int local)
+{
+	if (local && cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
+		MSR_TM_ACTIVE(current->thread.regs->msr)) {
+		tm_enable();
+		tm_abort(TM_CAUSE_TLBI);
+	}
+}
+

Since local_tm_abort is only used in this file, it should be static.

OK

Also, since both places calling it are guarded by
CONFIG_PPC_TRANSACTIONAL_MEM, wouldn't it be cleaner if the #ifdef was here
instead and the #else block defined an empty static inline function? Then
the call sites wouldn't need to be guarded.

I have considered this style before, but I am worried about the call
stacks increased by empty function and forgot the inline function.
Will send v2 with your comments.

Thanks!