Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux

Pull s390 updates from Martin Schwidefsky:
 "There are a couple of new things for s390 with this merge request:

   - a new scheduling domain "drawer" is added to reflect the unusual
     topology found on z13 machines.  Performance tests showed up to 8
     percent gain with the additional domain.

   - the new crc-32 checksum crypto module uses the vector-galois-field
     multiply and sum SIMD instruction to speed up crc-32 and crc-32c.

   - proper __ro_after_init support, this requires RO_AFTER_INIT_DATA in
     the generic vmlinux.lds linker script definitions.

   - kcov instrumentation support.  A prerequisite for that is the
     inline assembly basic block cleanup, which is the reason for the
     net/iucv/iucv.c change.

   - support for 2GB pages is added to the hugetlbfs backend.

  Then there are two removals:

   - the oprofile hardware sampling support is dead code and is removed.
     The oprofile user space uses the perf interface nowadays.

   - the ETR clock synchronization is removed, this has been superseeded
     be the STP clock synchronization.  And it always has been
     "interesting" code..

  And the usual bug fixes and cleanups"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux: (82 commits)
  s390/pci: Delete an unnecessary check before the function call "pci_dev_put"
  s390/smp: clean up a condition
  s390/cio/chp : Remove deprecated create_singlethread_workqueue
  s390/chsc: improve channel path descriptor determination
  s390/chsc: sanitize fmt check for chp_desc determination
  s390/cio: make fmt1 channel path descriptor optional
  s390/chsc: fix ioctl CHSC_INFO_CU command
  s390/cio/device_ops: fix kernel doc
  s390/cio: allow to reset channel measurement block
  s390/console: Make preferred console handling more consistent
  s390/mm: fix gmap tlb flush issues
  s390/mm: add support for 2GB hugepages
  s390: have unique symbol for __switch_to address
  s390/cpuinfo: show maximum thread id
  s390/ptrace: clarify bits in the per_struct
  s390: stack address vs thread_info
  s390: remove pointless load within __switch_to
  s390: enable kcov support
  s390/cpumf: use basic block for ecctr inline assembly
  s390/hypfs: use basic block for diag inline assembly
  ...

1 files changed, 11 insertions, 20 deletions

diff --git a/arch/s390/include/asm/tlbflush.h b/arch/s390/include/asm/tlbflush.h
index a2e6ef32e054..1a691ef740cf 100644
--- a/arch/s390/include/asm/tlbflush.h
+++ b/arch/s390/include/asm/tlbflush.h
@@ -5,6 +5,7 @@
 #include <linux/sched.h>
 #include <asm/processor.h>
 #include <asm/pgalloc.h>
+#include <asm/pgtable.h>
 
 /*
  * Flush all TLB entries on the local CPU.
@@ -44,17 +45,9 @@ void smp_ptlb_all(void);
  */
 static inline void __tlb_flush_global(void)
 {
-	register unsigned long reg2 asm("2");
-	register unsigned long reg3 asm("3");
-	register unsigned long reg4 asm("4");
-	long dummy;
-
-	dummy = 0;
-	reg2 = reg3 = 0;
-	reg4 = ((unsigned long) &dummy) + 1;
-	asm volatile(
-		"	csp	%0,%2"
-		: : "d" (reg2), "d" (reg3), "d" (reg4), "m" (dummy) : "cc" );
+	unsigned int dummy = 0;
+
+	csp(&dummy, 0, 0);
 }
 
 /*
@@ -64,7 +57,7 @@ static inline void __tlb_flush_global(void)
 static inline void __tlb_flush_full(struct mm_struct *mm)
 {
 	preempt_disable();
-	atomic_add(0x10000, &mm->context.attach_count);
+	atomic_inc(&mm->context.flush_count);
 	if (cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id()))) {
 		/* Local TLB flush */
 		__tlb_flush_local();
@@ -76,21 +69,19 @@ static inline void __tlb_flush_full(struct mm_struct *mm)
 			cpumask_copy(mm_cpumask(mm),
 				     &mm->context.cpu_attach_mask);
 	}
-	atomic_sub(0x10000, &mm->context.attach_count);
+	atomic_dec(&mm->context.flush_count);
 	preempt_enable();
 }
 
 /*
- * Flush TLB entries for a specific ASCE on all CPUs.
+ * Flush TLB entries for a specific ASCE on all CPUs. Should never be used
+ * when more than one asce (e.g. gmap) ran on this mm.
  */
 static inline void __tlb_flush_asce(struct mm_struct *mm, unsigned long asce)
 {
-	int active, count;
-
 	preempt_disable();
-	active = (mm == current->active_mm) ? 1 : 0;
-	count = atomic_add_return(0x10000, &mm->context.attach_count);
-	if (MACHINE_HAS_TLB_LC && (count & 0xffff) <= active &&
+	atomic_inc(&mm->context.flush_count);
+	if (MACHINE_HAS_TLB_LC &&
 	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id()))) {
 		__tlb_flush_idte_local(asce);
 	} else {
@@ -103,7 +94,7 @@ static inline void __tlb_flush_asce(struct mm_struct *mm, unsigned long asce)
 			cpumask_copy(mm_cpumask(mm),
 				     &mm->context.cpu_attach_mask);
 	}
-	atomic_sub(0x10000, &mm->context.attach_count);
+	atomic_dec(&mm->context.flush_count);
 	preempt_enable();
 }