Merge branch 'parisc-for-3.11' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux

Pull parisc updates from Helge Deller:
 "The PA-RISC updates for v3.11 include a gcc miscompilation fix,
  gzip-compressed vmlinuz support, a fix in the PCI code for ATI FireGL
  support on c8000 machines, a fix to prevent that %sr1 is being
  clobbered and a few smaller optimizations and documentation updates"

* 'parisc-for-3.11' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Fix gcc miscompilation in pa_memcpy()
  parisc: Ensure volatile space register %sr1 is not clobbered
  parisc: optimize mtsp(0,sr) inline assembly
  parisc: switch to gzip-compressed vmlinuz kernel
  parisc: document the shadow registers
  parisc: more capabilities info in /proc/cpuinfo
  parisc: fix LMMIO mismatch between PAT length and MASK register

Documentation/parisc/registers

@@ -77,6 +77,14 @@ PSW default E value		0
 Shadow Registers		used by interruption handler code
 TOC enable bit			1
 
+=========================================================================
+
+The PA-RISC architecture defines 7 registers as "shadow registers".
+Those are used in RETURN FROM INTERRUPTION AND RESTORE instruction to reduce
+the state save and restore time by eliminating the need for general register
+(GR) saves and restores in interruption handlers.
+Shadow registers are the GRs 1, 8, 9, 16, 17, 24, and 25.
+
 =========================================================================
 Register usage notes, originally from John Marvin, with some additional
 notes from Randolph Chung.

arch/parisc/Makefile

@@ -17,6 +17,8 @@
 # Mike Shaver, Helge Deller and Martin K. Petersen
 #
 
+KBUILD_IMAGE := vmlinuz
+
 KBUILD_DEFCONFIG := default_defconfig
 
 NM		= sh $(srctree)/arch/parisc/nm
@@ -92,7 +94,7 @@ PALOCONF := $(shell if [ -f $(src)/palo.conf ]; then echo $(src)/palo.conf; \
 	else echo $(obj)/palo.conf; \
 	fi)
 
-palo: vmlinux
+palo: vmlinuz
 	@if test ! -x "$(PALO)"; then \
 		echo 'ERROR: Please install palo first (apt-get install palo)';\
 		echo 'or build it from source and install it somewhere in your $$PATH';\
@@ -107,10 +109,14 @@ palo: vmlinux
 	fi
 	$(PALO) -f $(PALOCONF)
 
-# Shorthands for known targets not supported by parisc, use vmlinux as default
-Image zImage bzImage: vmlinux
+# Shorthands for known targets not supported by parisc, use vmlinux/vmlinuz as default
+Image: vmlinux
+zImage bzImage: vmlinuz
+
+vmlinuz: vmlinux
+	@gzip -cf -9 $< > $@
 
-install: vmlinux
+install: vmlinuz
 	sh $(src)/arch/parisc/install.sh \
 			$(KERNELRELEASE) $< System.map "$(INSTALL_PATH)"
 
@@ -119,6 +125,7 @@ MRPROPER_FILES	+= palo.conf
 
 define archhelp
 	@echo  '* vmlinux	- Uncompressed kernel image (./vmlinux)'
+	@echo  '  vmlinuz	- Compressed kernel image (./vmlinuz)'
 	@echo  '  palo		- Bootable image (./lifimage)'
 	@echo  '  install	- Install kernel using'
 	@echo  '		  (your) ~/bin/$(INSTALLKERNEL) or'

arch/parisc/defpalo.conf

@@ -4,7 +4,7 @@
 # Most people using 'make palo' want a bootable file, usable for
 # network or tape booting for example.
 --init-tape=lifimage
---recoverykernel=vmlinux
+--recoverykernel=vmlinuz
 
 ########## Pick your ROOT here! ##########
 # You need at least one 'root='!
@@ -12,10 +12,10 @@
 # If you want a root ramdisk, use the next 2 lines
 #   (Edit the ramdisk image name!!!!)
 --ramdisk=ram-disk-image-file
---commandline=0/vmlinux HOME=/ root=/dev/ram initrd=0/ramdisk
+--commandline=0/vmlinuz HOME=/ root=/dev/ram initrd=0/ramdisk panic_timeout=60 panic=-1
 
 # If you want NFS root, use the following command line (Edit the HOSTNAME!!!)
-#--commandline=0/vmlinux HOME=/ root=/dev/nfs nfsroot=HOSTNAME ip=bootp
+#--commandline=0/vmlinuz HOME=/ root=/dev/nfs nfsroot=HOSTNAME ip=bootp
 
 # If you have root on a disk partition, use this (Edit the partition name!!!)
-#--commandline=0/vmlinux HOME=/ root=/dev/sda1
+#--commandline=0/vmlinuz HOME=/ root=/dev/sda1

arch/parisc/include/asm/special_insns.h

@@ -32,9 +32,12 @@ static inline void set_eiem(unsigned long val)
 	cr;				\
 })
 
-#define mtsp(gr, cr) \
+#define mtsp(val, cr) \
+	{ if (__builtin_constant_p(val) && ((val) == 0)) \
+	 __asm__ __volatile__("mtsp %%r0,%0" : : "i" (cr) : "memory"); \
+	else \
 	 __asm__ __volatile__("mtsp %0,%1" \
 		: /* no outputs */ \
-		: "r" (gr), "i" (cr) : "memory")
+		: "r" (val), "i" (cr) : "memory"); }
 
 #endif /* __PARISC_SPECIAL_INSNS_H */

arch/parisc/include/asm/tlbflush.h

@@ -63,13 +63,14 @@ static inline void flush_tlb_mm(struct mm_struct *mm)
 static inline void flush_tlb_page(struct vm_area_struct *vma,
 	unsigned long addr)
 {
-	unsigned long flags;
+	unsigned long flags, sid;
 
 	/* For one page, it's not worth testing the split_tlb variable */
 
 	mb();
-	mtsp(vma->vm_mm->context,1);
+	sid = vma->vm_mm->context;
 	purge_tlb_start(flags);
+	mtsp(sid, 1);
 	pdtlb(addr);
 	pitlb(addr);
 	purge_tlb_end(flags);

arch/parisc/install.sh

@@ -26,13 +26,13 @@ if [ -x /sbin/${INSTALLKERNEL} ]; then exec /sbin/${INSTALLKERNEL} "$@"; fi
 
 # Default install
 
-if [ -f $4/vmlinux ]; then
-	mv $4/vmlinux $4/vmlinux.old
+if [ -f $4/vmlinuz ]; then
+	mv $4/vmlinuz $4/vmlinuz.old
 fi
 
 if [ -f $4/System.map ]; then
 	mv $4/System.map $4/System.old
 fi
 
-cat $2 > $4/vmlinux
+cat $2 > $4/vmlinuz
 cp $3 $4/System.map

arch/parisc/kernel/cache.c

@@ -440,8 +440,8 @@ void __flush_tlb_range(unsigned long sid, unsigned long start,
 	else {
 		unsigned long flags;
 
-		mtsp(sid, 1);
 		purge_tlb_start(flags);
+		mtsp(sid, 1);
 		if (split_tlb) {
 			while (npages--) {
 				pdtlb(start);

arch/parisc/kernel/processor.c

@@ -371,10 +371,23 @@ show_cpuinfo (struct seq_file *m, void *v)
 
 		seq_printf(m, "capabilities\t:");
 		if (boot_cpu_data.pdc.capabilities & PDC_MODEL_OS32)
-			seq_printf(m, " os32");
+			seq_puts(m, " os32");
 		if (boot_cpu_data.pdc.capabilities & PDC_MODEL_OS64)
-			seq_printf(m, " os64");
-		seq_printf(m, "\n");
+			seq_puts(m, " os64");
+		if (boot_cpu_data.pdc.capabilities & PDC_MODEL_IOPDIR_FDC)
+			seq_puts(m, " iopdir_fdc");
+		switch (boot_cpu_data.pdc.capabilities & PDC_MODEL_NVA_MASK) {
+		case PDC_MODEL_NVA_SUPPORTED:
+			seq_puts(m, " nva_supported");
+			break;
+		case PDC_MODEL_NVA_SLOW:
+			seq_puts(m, " nva_slow");
+			break;
+		case PDC_MODEL_NVA_UNSUPPORTED:
+			seq_puts(m, " needs_equivalent_aliasing");
+			break;
+		}
+		seq_printf(m, " (0x%02lx)\n", boot_cpu_data.pdc.capabilities);
 
 		seq_printf(m, "model\t\t: %s\n"
 				"model name\t: %s\n",

arch/parisc/lib/memcpy.c

@@ -2,6 +2,7 @@
  *    Optimized memory copy routines.
  *
  *    Copyright (C) 2004 Randolph Chung <tausq@debian.org>
+ *    Copyright (C) 2013 Helge Deller <deller@gmx.de>
  *
  *    This program is free software; you can redistribute it and/or modify
  *    it under the terms of the GNU General Public License as published by
@@ -153,17 +154,21 @@ static inline void prefetch_dst(const void *addr)
 #define prefetch_dst(addr) do { } while(0)
 #endif
 
+#define PA_MEMCPY_OK		0
+#define PA_MEMCPY_LOAD_ERROR	1
+#define PA_MEMCPY_STORE_ERROR	2
+
 /* Copy from a not-aligned src to an aligned dst, using shifts. Handles 4 words
  * per loop.  This code is derived from glibc. 
  */
-static inline unsigned long copy_dstaligned(unsigned long dst, unsigned long src, unsigned long len, unsigned long o_dst, unsigned long o_src, unsigned long o_len)
+static inline unsigned long copy_dstaligned(unsigned long dst,
+					unsigned long src, unsigned long len)
 {
 	/* gcc complains that a2 and a3 may be uninitialized, but actually
 	 * they cannot be.  Initialize a2/a3 to shut gcc up.
 	 */
 	register unsigned int a0, a1, a2 = 0, a3 = 0;
 	int sh_1, sh_2;
-	struct exception_data *d;
 
 	/* prefetch_src((const void *)src); */
 
@@ -197,7 +202,7 @@ static inline unsigned long copy_dstaligned(unsigned long dst, unsigned long src
 			goto do2;
 		case 0:
 			if (len == 0)
-				return 0;
+				return PA_MEMCPY_OK;
 			/* a3 = ((unsigned int *) src)[0];
 			   a0 = ((unsigned int *) src)[1]; */
 			ldw(s_space, 0, src, a3, cda_ldw_exc);
@@ -256,42 +261,35 @@ static inline unsigned long copy_dstaligned(unsigned long dst, unsigned long src
 	preserve_branch(handle_load_error);
 	preserve_branch(handle_store_error);
 
-	return 0;
+	return PA_MEMCPY_OK;
 
 handle_load_error:
 	__asm__ __volatile__ ("cda_ldw_exc:\n");
-	d = &__get_cpu_var(exception_data);
-	DPRINTF("cda_ldw_exc: o_len=%lu fault_addr=%lu o_src=%lu ret=%lu\n",
-		o_len, d->fault_addr, o_src, o_len - d->fault_addr + o_src);
-	return o_len * 4 - d->fault_addr + o_src;
+	return PA_MEMCPY_LOAD_ERROR;
 
 handle_store_error:
 	__asm__ __volatile__ ("cda_stw_exc:\n");
-	d = &__get_cpu_var(exception_data);
-	DPRINTF("cda_stw_exc: o_len=%lu fault_addr=%lu o_dst=%lu ret=%lu\n",
-		o_len, d->fault_addr, o_dst, o_len - d->fault_addr + o_dst);
-	return o_len * 4 - d->fault_addr + o_dst;
+	return PA_MEMCPY_STORE_ERROR;
 }
 
 
-/* Returns 0 for success, otherwise, returns number of bytes not transferred. */
-static unsigned long pa_memcpy(void *dstp, const void *srcp, unsigned long len)
+/* Returns PA_MEMCPY_OK, PA_MEMCPY_LOAD_ERROR or PA_MEMCPY_STORE_ERROR.
+ * In case of an access fault the faulty address can be read from the per_cpu
+ * exception data struct. */
+static unsigned long pa_memcpy_internal(void *dstp, const void *srcp,
+					unsigned long len)
 {
 	register unsigned long src, dst, t1, t2, t3;
 	register unsigned char *pcs, *pcd;
 	register unsigned int *pws, *pwd;
 	register double *pds, *pdd;
-	unsigned long ret = 0;
-	unsigned long o_dst, o_src, o_len;
-	struct exception_data *d;
+	unsigned long ret;
 
 	src = (unsigned long)srcp;
 	dst = (unsigned long)dstp;
 	pcs = (unsigned char *)srcp;
 	pcd = (unsigned char *)dstp;
 
-	o_dst = dst; o_src = src; o_len = len;
-
 	/* prefetch_src((const void *)srcp); */
 
 	if (len < THRESHOLD)
@@ -401,7 +399,7 @@ static unsigned long pa_memcpy(void *dstp, const void *srcp, unsigned long len)
 		len--;
 	}
 
-	return 0;
+	return PA_MEMCPY_OK;
 
 unaligned_copy:
 	/* possibly we are aligned on a word, but not on a double... */
@@ -438,8 +436,7 @@ static unsigned long pa_memcpy(void *dstp, const void *srcp, unsigned long len)
 		src = (unsigned long)pcs;
 	}
 
-	ret = copy_dstaligned(dst, src, len / sizeof(unsigned int), 
-		o_dst, o_src, o_len);
+	ret = copy_dstaligned(dst, src, len / sizeof(unsigned int));
 	if (ret)
 		return ret;
 
@@ -454,17 +451,41 @@ static unsigned long pa_memcpy(void *dstp, const void *srcp, unsigned long len)
 
 handle_load_error:
 	__asm__ __volatile__ ("pmc_load_exc:\n");
-	d = &__get_cpu_var(exception_data);
-	DPRINTF("pmc_load_exc: o_len=%lu fault_addr=%lu o_src=%lu ret=%lu\n",
-		o_len, d->fault_addr, o_src, o_len - d->fault_addr + o_src);
-	return o_len - d->fault_addr + o_src;
+	return PA_MEMCPY_LOAD_ERROR;
 
 handle_store_error:
 	__asm__ __volatile__ ("pmc_store_exc:\n");
+	return PA_MEMCPY_STORE_ERROR;
+}
+
+
+/* Returns 0 for success, otherwise, returns number of bytes not transferred. */
+static unsigned long pa_memcpy(void *dstp, const void *srcp, unsigned long len)
+{
+	unsigned long ret, fault_addr, reference;
+	struct exception_data *d;
+
+	ret = pa_memcpy_internal(dstp, srcp, len);
+	if (likely(ret == PA_MEMCPY_OK))
+		return 0;
+
+	/* if a load or store fault occured we can get the faulty addr */
 	d = &__get_cpu_var(exception_data);
-	DPRINTF("pmc_store_exc: o_len=%lu fault_addr=%lu o_dst=%lu ret=%lu\n",
-		o_len, d->fault_addr, o_dst, o_len - d->fault_addr + o_dst);
-	return o_len - d->fault_addr + o_dst;
+	fault_addr = d->fault_addr;
+
+	/* error in load or store? */
+	if (ret == PA_MEMCPY_LOAD_ERROR)
+		reference = (unsigned long) srcp;
+	else
+		reference = (unsigned long) dstp;
+
+	DPRINTF("pa_memcpy: fault type = %lu, len=%lu fault_addr=%lu ref=%lu\n",
+		ret, len, fault_addr, reference);
+
+	if (fault_addr >= reference)
+		return len - (fault_addr - reference);
+	else
+		return len;
 }
 
 #ifdef __KERNEL__

drivers/parisc/lba_pci.c

@@ -613,6 +613,54 @@ truncate_pat_collision(struct resource *root, struct resource *new)
 	return 0;	/* truncation successful */
 }
 
+/*
+ * extend_lmmio_len: extend lmmio range to maximum length
+ *
+ * This is needed at least on C8000 systems to get the ATI FireGL card
+ * working. On other systems we will currently not extend the lmmio space.
+ */
+static unsigned long
+extend_lmmio_len(unsigned long start, unsigned long end, unsigned long lba_len)
+{
+	struct resource *tmp;
+
+	pr_debug("LMMIO mismatch: PAT length = 0x%lx, MASK register = 0x%lx\n",
+		end - start, lba_len);
+
+	lba_len = min(lba_len+1, 256UL*1024*1024); /* limit to 256 MB */
+
+	pr_debug("LBA: lmmio_space [0x%lx-0x%lx] - original\n", start, end);
+
+	if (boot_cpu_data.cpu_type < mako) {
+		pr_info("LBA: Not a C8000 system - not extending LMMIO range.\n");
+		return end;
+	}
+
+	end += lba_len;
+	if (end < start) /* fix overflow */
+		end = -1ULL;
+
+	pr_debug("LBA: lmmio_space [0x%lx-0x%lx] - current\n", start, end);
+
+	/* first overlap */
+	for (tmp = iomem_resource.child; tmp; tmp = tmp->sibling) {
+		pr_debug("LBA: testing %pR\n", tmp);
+		if (tmp->start == start)
+			continue; /* ignore ourself */
+		if (tmp->end < start)
+			continue;
+		if (tmp->start > end)
+			continue;
+		if (end >= tmp->start)
+			end = tmp->start - 1;
+	}
+
+	pr_info("LBA: lmmio_space [0x%lx-0x%lx] - new\n", start, end);
+
+	/* return new end */
+	return end;
+}
+
 #else
 #define truncate_pat_collision(r,n)  (0)
 #endif
@@ -994,6 +1042,14 @@ lba_pat_resources(struct parisc_device *pa_dev, struct lba_device *lba_dev)
 		case PAT_LMMIO:
 			/* used to fix up pre-initialized MEM BARs */
 			if (!lba_dev->hba.lmmio_space.flags) {
+				unsigned long lba_len;
+
+				lba_len = ~READ_REG32(lba_dev->hba.base_addr
+						+ LBA_LMMIO_MASK);
+				if ((p->end - p->start) != lba_len)
+					p->end = extend_lmmio_len(p->start,
+						p->end, lba_len);
+
 				sprintf(lba_dev->hba.lmmio_name,
 						"PCI%02x LMMIO",
 						(int)lba_dev->hba.bus_num.start);