Merge branch 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus

Pull MIPS updates from Ralf Baechle:
 "These are the highlists of the main MIPS pull request for 4.4:

   - Add latencytop support
   - Support appended DTBs
   - VDSO support and initially use it for gettimeofday.
   - Drop the .MIPS.abiflags and ELF NOTE sections from vmlinux
   - Support for the 5KE, an internal test core.
   - Switch all MIPS platfroms to libata drivers.
   - Improved support, cleanups for ralink and Lantiq platforms.
   - Support for the new xilfpga platform.
   - A number of DTB improvments for BMIPS.
   - Improved support for CM and CPS.
   - Minor JZ4740 and BCM47xx enhancements"

* 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus: (120 commits)
  MIPS: idle: add case for CPU_5KE
  MIPS: Octeon: Support APPENDED_DTB
  MIPS: vmlinux: create a section for appended DTB
  MIPS: Clean up compat_siginfo_t
  MIPS: Fix PAGE_MASK definition
  MIPS: BMIPS: Enable GZIP ramdisk and timed printks
  MIPS: Add xilfpga defconfig
  MIPS: xilfpga: Add mipsfpga platform code
  MIPS: xilfpga: Add xilfpga device tree files.
  dt-bindings: MIPS: Document xilfpga bindings and boot style
  MIPS: Make MIPS_CMDLINE_DTB default
  MIPS: Make the kernel arguments from dtb available
  MIPS: Use USE_OF as the guard for appended dtb
  MIPS: BCM63XX: Use pr_* instead of printk
  MIPS: Loongson: Cleanup CONFIG_LOONGSON_SUSPEND.
  MIPS: lantiq: Disable xbar fpi burst mode
  MIPS: lantiq: Force the crossbar to big endian
  MIPS: lantiq: Initialize the USB core on boot
  MIPS: lantiq: Return correct value for fpi clock on ar9
  MIPS: ralink: Add missing clock on rt305x
  ...

4 files changed, 206 insertions, 38 deletions

diff --git a/arch/mips/mm/Makefile b/arch/mips/mm/Makefile
index 67ede4ef9b8d..b4c64bd3f723 100644
--- a/arch/mips/mm/Makefile
+++ b/arch/mips/mm/Makefile
@@ -28,3 +28,4 @@ obj-$(CONFIG_IP22_CPU_SCACHE)	+= sc-ip22.o
 obj-$(CONFIG_R5000_CPU_SCACHE)	+= sc-r5k.o
 obj-$(CONFIG_RM7000_CPU_SCACHE) += sc-rm7k.o
 obj-$(CONFIG_MIPS_CPU_SCACHE)	+= sc-mips.o
+obj-$(CONFIG_SCACHE_DEBUGFS)	+= sc-debugfs.o
diff --git a/arch/mips/mm/sc-debugfs.c b/arch/mips/mm/sc-debugfs.c
new file mode 100644
index 000000000000..5eefe3281b24
--- /dev/null
+++ b/arch/mips/mm/sc-debugfs.c
@@ -0,0 +1,81 @@
+/*
+ * Copyright (C) 2015 Imagination Technologies
+ * Author: Paul Burton <paul.burton@imgtec.com>
+ *
+ * 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 the
+ * Free Software Foundation; either version 2 of the License, or (at your
+ * option) any later version.
+ */
+
+#include <asm/bcache.h>
+#include <asm/debug.h>
+#include <asm/uaccess.h>
+#include <linux/debugfs.h>
+#include <linux/init.h>
+
+static ssize_t sc_prefetch_read(struct file *file, char __user *user_buf,
+				size_t count, loff_t *ppos)
+{
+	bool enabled = bc_prefetch_is_enabled();
+	char buf[3];
+
+	buf[0] = enabled ? 'Y' : 'N';
+	buf[1] = '\n';
+	buf[2] = 0;
+
+	return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
+}
+
+static ssize_t sc_prefetch_write(struct file *file,
+				 const char __user *user_buf,
+				 size_t count, loff_t *ppos)
+{
+	char buf[32];
+	ssize_t buf_size;
+	bool enabled;
+	int err;
+
+	buf_size = min(count, sizeof(buf) - 1);
+	if (copy_from_user(buf, user_buf, buf_size))
+		return -EFAULT;
+
+	buf[buf_size] = '\0';
+	err = strtobool(buf, &enabled);
+	if (err)
+		return err;
+
+	if (enabled)
+		bc_prefetch_enable();
+	else
+		bc_prefetch_disable();
+
+	return count;
+}
+
+static const struct file_operations sc_prefetch_fops = {
+	.open = simple_open,
+	.llseek = default_llseek,
+	.read = sc_prefetch_read,
+	.write = sc_prefetch_write,
+};
+
+static int __init sc_debugfs_init(void)
+{
+	struct dentry *dir, *file;
+
+	if (!mips_debugfs_dir)
+		return -ENODEV;
+
+	dir = debugfs_create_dir("l2cache", mips_debugfs_dir);
+	if (IS_ERR(dir))
+		return PTR_ERR(dir);
+
+	file = debugfs_create_file("prefetch", S_IRUGO | S_IWUSR, dir,
+				   NULL, &sc_prefetch_fops);
+	if (IS_ERR(file))
+		return PTR_ERR(file);
+
+	return 0;
+}
+late_initcall(sc_debugfs_init);
diff --git a/arch/mips/mm/sc-mips.c b/arch/mips/mm/sc-mips.c
index 53ea8391f9bb..3bd0597d9c3d 100644
--- a/arch/mips/mm/sc-mips.c
+++ b/arch/mips/mm/sc-mips.c
@@ -51,11 +51,69 @@ static void mips_sc_disable(void)
 	/* L2 cache is permanently enabled */
 }
 
+static void mips_sc_prefetch_enable(void)
+{
+	unsigned long pftctl;
+
+	if (mips_cm_revision() < CM_REV_CM2_5)
+		return;
+
+	/*
+	 * If there is one or more L2 prefetch unit present then enable
+	 * prefetching for both code & data, for all ports.
+	 */
+	pftctl = read_gcr_l2_pft_control();
+	if (pftctl & CM_GCR_L2_PFT_CONTROL_NPFT_MSK) {
+		pftctl &= ~CM_GCR_L2_PFT_CONTROL_PAGEMASK_MSK;
+		pftctl |= PAGE_MASK & CM_GCR_L2_PFT_CONTROL_PAGEMASK_MSK;
+		pftctl |= CM_GCR_L2_PFT_CONTROL_PFTEN_MSK;
+		write_gcr_l2_pft_control(pftctl);
+
+		pftctl = read_gcr_l2_pft_control_b();
+		pftctl |= CM_GCR_L2_PFT_CONTROL_B_PORTID_MSK;
+		pftctl |= CM_GCR_L2_PFT_CONTROL_B_CEN_MSK;
+		write_gcr_l2_pft_control_b(pftctl);
+	}
+}
+
+static void mips_sc_prefetch_disable(void)
+{
+	unsigned long pftctl;
+
+	if (mips_cm_revision() < CM_REV_CM2_5)
+		return;
+
+	pftctl = read_gcr_l2_pft_control();
+	pftctl &= ~CM_GCR_L2_PFT_CONTROL_PFTEN_MSK;
+	write_gcr_l2_pft_control(pftctl);
+
+	pftctl = read_gcr_l2_pft_control_b();
+	pftctl &= ~CM_GCR_L2_PFT_CONTROL_B_PORTID_MSK;
+	pftctl &= ~CM_GCR_L2_PFT_CONTROL_B_CEN_MSK;
+	write_gcr_l2_pft_control_b(pftctl);
+}
+
+static bool mips_sc_prefetch_is_enabled(void)
+{
+	unsigned long pftctl;
+
+	if (mips_cm_revision() < CM_REV_CM2_5)
+		return false;
+
+	pftctl = read_gcr_l2_pft_control();
+	if (!(pftctl & CM_GCR_L2_PFT_CONTROL_NPFT_MSK))
+		return false;
+	return !!(pftctl & CM_GCR_L2_PFT_CONTROL_PFTEN_MSK);
+}
+
 static struct bcache_ops mips_sc_ops = {
 	.bc_enable = mips_sc_enable,
 	.bc_disable = mips_sc_disable,
 	.bc_wback_inv = mips_sc_wback_inv,
-	.bc_inv = mips_sc_inv
+	.bc_inv = mips_sc_inv,
+	.bc_prefetch_enable = mips_sc_prefetch_enable,
+	.bc_prefetch_disable = mips_sc_prefetch_disable,
+	.bc_prefetch_is_enabled = mips_sc_prefetch_is_enabled,
 };
 
 /*
@@ -162,13 +220,13 @@ static inline int __init mips_sc_probe(void)
 		return 0;
 
 	tmp = (config2 >> 8) & 0x0f;
-	if (0 <= tmp && tmp <= 7)
+	if (tmp <= 7)
 		c->scache.sets = 64 << tmp;
 	else
 		return 0;
 
 	tmp = (config2 >> 0) & 0x0f;
-	if (0 <= tmp && tmp <= 7)
+	if (tmp <= 7)
 		c->scache.ways = tmp + 1;
 	else
 		return 0;
@@ -186,6 +244,7 @@ int mips_sc_init(void)
 	int found = mips_sc_probe();
 	if (found) {
 		mips_sc_enable();
+		mips_sc_prefetch_enable();
 		bcops = &mips_sc_ops;
 	}
 	return found;
diff --git a/arch/mips/mm/tlbex.c b/arch/mips/mm/tlbex.c
index 323d1d302f2b..32e0be27673f 100644
--- a/arch/mips/mm/tlbex.c
+++ b/arch/mips/mm/tlbex.c
@@ -311,6 +311,7 @@ static struct uasm_label labels[128];
 static struct uasm_reloc relocs[128];
 
 static int check_for_high_segbits;
+static bool fill_includes_sw_bits;
 
 static unsigned int kscratch_used_mask;
 
@@ -630,8 +631,14 @@ static void build_tlb_write_entry(u32 **p, struct uasm_label **l,
 static __maybe_unused void build_convert_pte_to_entrylo(u32 **p,
 							unsigned int reg)
 {
-	if (cpu_has_rixi) {
-		UASM_i_ROTR(p, reg, reg, ilog2(_PAGE_GLOBAL));
+	if (cpu_has_rixi && _PAGE_NO_EXEC) {
+		if (fill_includes_sw_bits) {
+			UASM_i_ROTR(p, reg, reg, ilog2(_PAGE_GLOBAL));
+		} else {
+			UASM_i_SRL(p, reg, reg, ilog2(_PAGE_NO_EXEC));
+			UASM_i_ROTR(p, reg, reg,
+				    ilog2(_PAGE_GLOBAL) - ilog2(_PAGE_NO_EXEC));
+		}
 	} else {
 #ifdef CONFIG_PHYS_ADDR_T_64BIT
 		uasm_i_dsrl_safe(p, reg, reg, ilog2(_PAGE_GLOBAL));
@@ -1005,21 +1012,7 @@ static void build_update_entries(u32 **p, unsigned int tmp, unsigned int ptep)
 	 * 64bit address support (36bit on a 32bit CPU) in a 32bit
 	 * Kernel is a special case. Only a few CPUs use it.
 	 */
-#ifdef CONFIG_PHYS_ADDR_T_64BIT
-	if (cpu_has_64bits) {
-		uasm_i_ld(p, tmp, 0, ptep); /* get even pte */
-		uasm_i_ld(p, ptep, sizeof(pte_t), ptep); /* get odd pte */
-		if (cpu_has_rixi) {
-			UASM_i_ROTR(p, tmp, tmp, ilog2(_PAGE_GLOBAL));
-			UASM_i_MTC0(p, tmp, C0_ENTRYLO0); /* load it */
-			UASM_i_ROTR(p, ptep, ptep, ilog2(_PAGE_GLOBAL));
-		} else {
-			uasm_i_dsrl_safe(p, tmp, tmp, ilog2(_PAGE_GLOBAL)); /* convert to entrylo0 */
-			UASM_i_MTC0(p, tmp, C0_ENTRYLO0); /* load it */
-			uasm_i_dsrl_safe(p, ptep, ptep, ilog2(_PAGE_GLOBAL)); /* convert to entrylo1 */
-		}
-		UASM_i_MTC0(p, ptep, C0_ENTRYLO1); /* load it */
-	} else {
+	if (config_enabled(CONFIG_PHYS_ADDR_T_64BIT) && !cpu_has_64bits) {
 		int pte_off_even = sizeof(pte_t) / 2;
 		int pte_off_odd = pte_off_even + sizeof(pte_t);
 #ifdef CONFIG_XPA
@@ -1043,31 +1036,23 @@ static void build_update_entries(u32 **p, unsigned int tmp, unsigned int ptep)
 		uasm_i_mthc0(p, tmp, C0_ENTRYLO0);
 		uasm_i_mthc0(p, ptep, C0_ENTRYLO1);
 #endif
+		return;
 	}
-#else
+
 	UASM_i_LW(p, tmp, 0, ptep); /* get even pte */
 	UASM_i_LW(p, ptep, sizeof(pte_t), ptep); /* get odd pte */
 	if (r45k_bvahwbug())
 		build_tlb_probe_entry(p);
-	if (cpu_has_rixi) {
-		UASM_i_ROTR(p, tmp, tmp, ilog2(_PAGE_GLOBAL));
-		if (r4k_250MHZhwbug())
-			UASM_i_MTC0(p, 0, C0_ENTRYLO0);
-		UASM_i_MTC0(p, tmp, C0_ENTRYLO0); /* load it */
-		UASM_i_ROTR(p, ptep, ptep, ilog2(_PAGE_GLOBAL));
-	} else {
-		UASM_i_SRL(p, tmp, tmp, ilog2(_PAGE_GLOBAL)); /* convert to entrylo0 */
-		if (r4k_250MHZhwbug())
-			UASM_i_MTC0(p, 0, C0_ENTRYLO0);
-		UASM_i_MTC0(p, tmp, C0_ENTRYLO0); /* load it */
-		UASM_i_SRL(p, ptep, ptep, ilog2(_PAGE_GLOBAL)); /* convert to entrylo1 */
-		if (r45k_bvahwbug())
-			uasm_i_mfc0(p, tmp, C0_INDEX);
-	}
+	build_convert_pte_to_entrylo(p, tmp);
+	if (r4k_250MHZhwbug())
+		UASM_i_MTC0(p, 0, C0_ENTRYLO0);
+	UASM_i_MTC0(p, tmp, C0_ENTRYLO0); /* load it */
+	build_convert_pte_to_entrylo(p, ptep);
+	if (r45k_bvahwbug())
+		uasm_i_mfc0(p, tmp, C0_INDEX);
 	if (r4k_250MHZhwbug())
 		UASM_i_MTC0(p, 0, C0_ENTRYLO1);
 	UASM_i_MTC0(p, ptep, C0_ENTRYLO1); /* load it */
-#endif
 }
 
 struct mips_huge_tlb_info {
@@ -2299,6 +2284,10 @@ static void config_htw_params(void)
 	/* re-initialize the PTI field including the even/odd bit */
 	pwfield &= ~MIPS_PWFIELD_PTI_MASK;
 	pwfield |= PAGE_SHIFT << MIPS_PWFIELD_PTI_SHIFT;
+	if (CONFIG_PGTABLE_LEVELS >= 3) {
+		pwfield &= ~MIPS_PWFIELD_MDI_MASK;
+		pwfield |= PMD_SHIFT << MIPS_PWFIELD_MDI_SHIFT;
+	}
 	/* Set the PTEI right shift */
 	ptei = _PAGE_GLOBAL_SHIFT << MIPS_PWFIELD_PTEI_SHIFT;
 	pwfield |= ptei;
@@ -2320,9 +2309,11 @@ static void config_htw_params(void)
 
 	pwsize = ilog2(PTRS_PER_PGD) << MIPS_PWSIZE_GDW_SHIFT;
 	pwsize |= ilog2(PTRS_PER_PTE) << MIPS_PWSIZE_PTW_SHIFT;
+	if (CONFIG_PGTABLE_LEVELS >= 3)
+		pwsize |= ilog2(PTRS_PER_PMD) << MIPS_PWSIZE_MDW_SHIFT;
 
 	/* If XPA has been enabled, PTEs are 64-bit in size. */
-	if (read_c0_pagegrain() & PG_ELPA)
+	if (config_enabled(CONFIG_64BITS) || (read_c0_pagegrain() & PG_ELPA))
 		pwsize |= 1;
 
 	write_c0_pwsize(pwsize);
@@ -2360,6 +2351,41 @@ static void config_xpa_params(void)
 #endif
 }
 
+static void check_pabits(void)
+{
+	unsigned long entry;
+	unsigned pabits, fillbits;
+
+	if (!cpu_has_rixi || !_PAGE_NO_EXEC) {
+		/*
+		 * We'll only be making use of the fact that we can rotate bits
+		 * into the fill if the CPU supports RIXI, so don't bother
+		 * probing this for CPUs which don't.
+		 */
+		return;
+	}
+
+	write_c0_entrylo0(~0ul);
+	back_to_back_c0_hazard();
+	entry = read_c0_entrylo0();
+
+	/* clear all non-PFN bits */
+	entry &= ~((1 << MIPS_ENTRYLO_PFN_SHIFT) - 1);
+	entry &= ~(MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI);
+
+	/* find a lower bound on PABITS, and upper bound on fill bits */
+	pabits = fls_long(entry) + 6;
+	fillbits = max_t(int, (int)BITS_PER_LONG - pabits, 0);
+
+	/* minus the RI & XI bits */
+	fillbits -= min_t(unsigned, fillbits, 2);
+
+	if (fillbits >= ilog2(_PAGE_NO_EXEC))
+		fill_includes_sw_bits = true;
+
+	pr_debug("Entry* registers contain %u fill bits\n", fillbits);
+}
+
 void build_tlb_refill_handler(void)
 {
 	/*
@@ -2370,6 +2396,7 @@ void build_tlb_refill_handler(void)
 	static int run_once = 0;
 
 	output_pgtable_bits_defines();
+	check_pabits();
 
 #ifdef CONFIG_64BIT
 	check_for_high_segbits = current_cpu_data.vmbits > (PGDIR_SHIFT + PGD_ORDER + PAGE_SHIFT - 3);