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
  ...

2 files changed, 85 insertions, 0 deletions

diff --git a/arch/s390/include/asm/fpu/api.h b/arch/s390/include/asm/fpu/api.h
index 8ae236b0f80b..6aba6fc406ad 100644
--- a/arch/s390/include/asm/fpu/api.h
+++ b/arch/s390/include/asm/fpu/api.h
@@ -1,6 +1,41 @@
 /*
  * In-kernel FPU support functions
  *
+ *
+ * Consider these guidelines before using in-kernel FPU functions:
+ *
+ *  1. Use kernel_fpu_begin() and kernel_fpu_end() to enclose all in-kernel
+ *     use of floating-point or vector registers and instructions.
+ *
+ *  2. For kernel_fpu_begin(), specify the vector register range you want to
+ *     use with the KERNEL_VXR_* constants. Consider these usage guidelines:
+ *
+ *     a) If your function typically runs in process-context, use the lower
+ *	  half of the vector registers, for example, specify KERNEL_VXR_LOW.
+ *     b) If your function typically runs in soft-irq or hard-irq context,
+ *	  prefer using the upper half of the vector registers, for example,
+ *	  specify KERNEL_VXR_HIGH.
+ *
+ *     If you adhere to these guidelines, an interrupted process context
+ *     does not require to save and restore vector registers because of
+ *     disjoint register ranges.
+ *
+ *     Also note that the __kernel_fpu_begin()/__kernel_fpu_end() functions
+ *     includes logic to save and restore up to 16 vector registers at once.
+ *
+ *  3. You can nest kernel_fpu_begin()/kernel_fpu_end() by using different
+ *     struct kernel_fpu states.  Vector registers that are in use by outer
+ *     levels are saved and restored.  You can minimize the save and restore
+ *     effort by choosing disjoint vector register ranges.
+ *
+ *  5. To use vector floating-point instructions, specify the KERNEL_FPC
+ *     flag to save and restore floating-point controls in addition to any
+ *     vector register range.
+ *
+ *  6. To use floating-point registers and instructions only, specify the
+ *     KERNEL_FPR flag.  This flag triggers a save and restore of vector
+ *     registers V0 to V15 and floating-point controls.
+ *
  * Copyright IBM Corp. 2015
  * Author(s): Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
  */
@@ -8,6 +43,8 @@
 #ifndef _ASM_S390_FPU_API_H
 #define _ASM_S390_FPU_API_H
 
+#include <linux/preempt.h>
+
 void save_fpu_regs(void);
 
 static inline int test_fp_ctl(u32 fpc)
@@ -27,4 +64,42 @@ static inline int test_fp_ctl(u32 fpc)
 	return rc;
 }
 
+#define KERNEL_VXR_V0V7		1
+#define KERNEL_VXR_V8V15	2
+#define KERNEL_VXR_V16V23	4
+#define KERNEL_VXR_V24V31	8
+#define KERNEL_FPR		16
+#define KERNEL_FPC		256
+
+#define KERNEL_VXR_LOW		(KERNEL_VXR_V0V7|KERNEL_VXR_V8V15)
+#define KERNEL_VXR_MID		(KERNEL_VXR_V8V15|KERNEL_VXR_V16V23)
+#define KERNEL_VXR_HIGH		(KERNEL_VXR_V16V23|KERNEL_VXR_V24V31)
+
+#define KERNEL_FPU_MASK		(KERNEL_VXR_LOW|KERNEL_VXR_HIGH|KERNEL_FPR)
+
+struct kernel_fpu;
+
+/*
+ * Note the functions below must be called with preemption disabled.
+ * Do not enable preemption before calling __kernel_fpu_end() to prevent
+ * an corruption of an existing kernel FPU state.
+ *
+ * Prefer using the kernel_fpu_begin()/kernel_fpu_end() pair of functions.
+ */
+void __kernel_fpu_begin(struct kernel_fpu *state, u32 flags);
+void __kernel_fpu_end(struct kernel_fpu *state);
+
+
+static inline void kernel_fpu_begin(struct kernel_fpu *state, u32 flags)
+{
+	preempt_disable();
+	__kernel_fpu_begin(state, flags);
+}
+
+static inline void kernel_fpu_end(struct kernel_fpu *state)
+{
+	__kernel_fpu_end(state);
+	preempt_enable();
+}
+
 #endif /* _ASM_S390_FPU_API_H */
diff --git a/arch/s390/include/asm/fpu/types.h b/arch/s390/include/asm/fpu/types.h
index fe937c9b6471..bce255ead72b 100644
--- a/arch/s390/include/asm/fpu/types.h
+++ b/arch/s390/include/asm/fpu/types.h
@@ -24,4 +24,14 @@ struct fpu {
 /* VX array structure for address operand constraints in inline assemblies */
 struct vx_array { __vector128 _[__NUM_VXRS]; };
 
+/* In-kernel FPU state structure */
+struct kernel_fpu {
+	u32	    mask;
+	u32	    fpc;
+	union {
+		freg_t fprs[__NUM_FPRS];
+		__vector128 vxrs[__NUM_VXRS];
+	};
+};
+
 #endif /* _ASM_S390_FPU_TYPES_H */