Merge tag 'nios2-v5.1-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/lftan/nios2

Pull nios2 updates from Ley Foon Tan:
 "Most of updates are MMU related"

* tag 'nios2-v5.1-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/lftan/nios2:
  nios2: Fix update_mmu_cache preload the TLB with the new PTE
  nios2: update_mmu_cache preload the TLB with the new PTE
  nios2: User address TLB flush break after finding the matching entry
  nios2: flush_tlb_all use TLBMISC way auto-increment feature
  nios2: improve readability of tlb functions
  nios2: flush_tlb_mm flush only the pid
  nios2: flush_tlb_pid can just restore TLBMISC once
  nios2: TLBMISC writes do not require PID bits to be set
  nios2: Use an invalid TLB entry address helper function
  nios2: pte_clear does not need to flush TLB
  nios2: flush_tlb_page use PID based flush
  nios2: update_mmu_cache clear the old entry from the TLB
  nios2: remove redundant 'default n' from Kconfig-s
  nios2: ksyms: Add missing symbol exports

arch/nios2/Kconfig

@@ -123,7 +123,6 @@ config NIOS2_CMDLINE_IGNORE_DTB
 
 config NIOS2_PASS_CMDLINE
 	bool "Passed kernel command line from u-boot"
-	default n
 	help
 	  Use bootargs env variable from u-boot for kernel command line.
 	  will override "Default kernel command string".

arch/nios2/include/asm/pgtable.h

@@ -232,7 +232,6 @@ static inline void pte_clear(struct mm_struct *mm,
 	pte_val(null) = (addr >> PAGE_SHIFT) & 0xf;
 
 	set_pte_at(mm, addr, ptep, null);
-	flush_tlb_one(addr);
 }
 
 /*

arch/nios2/include/asm/tlbflush.h

@@ -26,21 +26,32 @@ struct mm_struct;
  *
  *  - flush_tlb_all() flushes all processes TLB entries
  *  - flush_tlb_mm(mm) flushes the specified mm context TLB entries
- *  - flush_tlb_page(vma, vmaddr) flushes one page
  *  - flush_tlb_range(vma, start, end) flushes a range of pages
+ *  - flush_tlb_page(vma, address) flushes a page
  *  - flush_tlb_kernel_range(start, end) flushes a range of kernel pages
+ *  - flush_tlb_kernel_page(address) flushes a kernel page
+ *
+ *  - reload_tlb_page(vma, address, pte) flushes the TLB for address like
+ *    flush_tlb_page, then replaces it with a TLB for pte.
  */
 extern void flush_tlb_all(void);
 extern void flush_tlb_mm(struct mm_struct *mm);
 extern void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
 			    unsigned long end);
 extern void flush_tlb_kernel_range(unsigned long start, unsigned long end);
-extern void flush_tlb_one(unsigned long vaddr);
 
 static inline void flush_tlb_page(struct vm_area_struct *vma,
-				unsigned long addr)
+				  unsigned long address)
 {
-	flush_tlb_one(addr);
+	flush_tlb_range(vma, address, address + PAGE_SIZE);
 }
 
+static inline void flush_tlb_kernel_page(unsigned long address)
+{
+	flush_tlb_kernel_range(address, address + PAGE_SIZE);
+}
+
+extern void reload_tlb_page(struct vm_area_struct *vma, unsigned long addr,
+			    pte_t pte);
+
 #endif /* _ASM_NIOS2_TLBFLUSH_H */

arch/nios2/kernel/nios2_ksyms.c

@@ -9,12 +9,20 @@
 #include <linux/export.h>
 #include <linux/string.h>
 
+#include <asm/cacheflush.h>
+#include <asm/pgtable.h>
+
 /* string functions */
 
 EXPORT_SYMBOL(memcpy);
 EXPORT_SYMBOL(memset);
 EXPORT_SYMBOL(memmove);
 
+/* memory management */
+
+EXPORT_SYMBOL(empty_zero_page);
+EXPORT_SYMBOL(flush_icache_range);
+
 /*
  * libgcc functions - functions that are used internally by the
  * compiler...  (prototypes are not correct though, but that
@@ -31,3 +39,7 @@ DECLARE_EXPORT(__udivsi3);
 DECLARE_EXPORT(__umoddi3);
 DECLARE_EXPORT(__umodsi3);
 DECLARE_EXPORT(__muldi3);
+DECLARE_EXPORT(__ucmpdi2);
+DECLARE_EXPORT(__lshrdi3);
+DECLARE_EXPORT(__ashldi3);
+DECLARE_EXPORT(__ashrdi3);

arch/nios2/mm/cacheflush.c

@@ -198,12 +198,15 @@ void flush_dcache_page(struct page *page)
 EXPORT_SYMBOL(flush_dcache_page);
 
 void update_mmu_cache(struct vm_area_struct *vma,
-		      unsigned long address, pte_t *pte)
+		      unsigned long address, pte_t *ptep)
 {
-	unsigned long pfn = pte_pfn(*pte);
+	pte_t pte = *ptep;
+	unsigned long pfn = pte_pfn(pte);
 	struct page *page;
 	struct address_space *mapping;
 
+	reload_tlb_page(vma, address, pte);
+
 	if (!pfn_valid(pfn))
 		return;
 

arch/nios2/mm/fault.c

@@ -270,7 +270,7 @@ asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long cause,
 		if (!pte_present(*pte_k))
 			goto no_context;
 
-		flush_tlb_one(address);
+		flush_tlb_kernel_page(address);
 		return;
 	}
 }

arch/nios2/mm/tlb.c

@@ -23,10 +23,6 @@
 	((((1UL << (cpuinfo.tlb_ptr_sz - cpuinfo.tlb_num_ways_log2))) - 1) \
 		<< PAGE_SHIFT)
 
-/* Used as illegal PHYS_ADDR for TLB mappings
- */
-#define MAX_PHYS_ADDR 0
-
 static void get_misc_and_pid(unsigned long *misc, unsigned long *pid)
 {
 	*misc  = RDCTL(CTL_TLBMISC);
@@ -35,28 +31,23 @@ static void get_misc_and_pid(unsigned long *misc, unsigned long *pid)
 }
 
 /*
- * All entries common to a mm share an asid.  To effectively flush these
- * entries, we just bump the asid.
+ * This provides a PTEADDR value for addr that will cause a TLB miss
+ * (fast TLB miss). TLB invalidation replaces entries with this value.
  */
-void flush_tlb_mm(struct mm_struct *mm)
+static unsigned long pteaddr_invalid(unsigned long addr)
 {
-	if (current->mm == mm)
-		flush_tlb_all();
-	else
-		memset(&mm->context, 0, sizeof(mm_context_t));
+	return ((addr | 0xC0000000UL) >> PAGE_SHIFT) << 2;
 }
 
 /*
  * This one is only used for pages with the global bit set so we don't care
  * much about the ASID.
  */
-void flush_tlb_one_pid(unsigned long addr, unsigned long mmu_pid)
+static void replace_tlb_one_pid(unsigned long addr, unsigned long mmu_pid, unsigned long tlbacc)
 {
 	unsigned int way;
 	unsigned long org_misc, pid_misc;
 
-	pr_debug("Flush tlb-entry for vaddr=%#lx\n", addr);
-
 	/* remember pid/way until we return. */
 	get_misc_and_pid(&org_misc, &pid_misc);
 
@@ -67,30 +58,48 @@ void flush_tlb_one_pid(unsigned long addr, unsigned long mmu_pid)
 		unsigned long tlbmisc;
 		unsigned long pid;
 
-		tlbmisc = pid_misc | TLBMISC_RD | (way << TLBMISC_WAY_SHIFT);
+		tlbmisc = TLBMISC_RD | (way << TLBMISC_WAY_SHIFT);
 		WRCTL(CTL_TLBMISC, tlbmisc);
+
 		pteaddr = RDCTL(CTL_PTEADDR);
+		if (((pteaddr >> 2) & 0xfffff) != (addr >> PAGE_SHIFT))
+			continue;
+
 		tlbmisc = RDCTL(CTL_TLBMISC);
 		pid = (tlbmisc >> TLBMISC_PID_SHIFT) & TLBMISC_PID_MASK;
-		if (((((pteaddr >> 2) & 0xfffff)) == (addr >> PAGE_SHIFT)) &&
-				pid == mmu_pid) {
-			unsigned long vaddr = CONFIG_NIOS2_IO_REGION_BASE +
-				((PAGE_SIZE * cpuinfo.tlb_num_lines) * way) +
-				(addr & TLB_INDEX_MASK);
-			pr_debug("Flush entry by writing %#lx way=%dl pid=%ld\n",
-				vaddr, way, (pid_misc >> TLBMISC_PID_SHIFT));
-
-			WRCTL(CTL_PTEADDR, (vaddr >> 12) << 2);
-			tlbmisc = pid_misc | TLBMISC_WE |
+		if (pid != mmu_pid)
+			continue;
+
+		tlbmisc = (mmu_pid << TLBMISC_PID_SHIFT) | TLBMISC_WE |
 			  (way << TLBMISC_WAY_SHIFT);
 		WRCTL(CTL_TLBMISC, tlbmisc);
-			WRCTL(CTL_TLBACC, (MAX_PHYS_ADDR >> PAGE_SHIFT));
-		}
+		if (tlbacc == 0)
+			WRCTL(CTL_PTEADDR, pteaddr_invalid(addr));
+		WRCTL(CTL_TLBACC, tlbacc);
+		/*
+		 * There should be only a single entry that maps a
+		 * particular {address,pid} so break after a match.
+		 */
+		break;
 	}
 
 	WRCTL(CTL_TLBMISC, org_misc);
 }
 
+static void flush_tlb_one_pid(unsigned long addr, unsigned long mmu_pid)
+{
+	pr_debug("Flush tlb-entry for vaddr=%#lx\n", addr);
+
+	replace_tlb_one_pid(addr, mmu_pid, 0);
+}
+
+static void reload_tlb_one_pid(unsigned long addr, unsigned long mmu_pid, pte_t pte)
+{
+	pr_debug("Reload tlb-entry for vaddr=%#lx\n", addr);
+
+	replace_tlb_one_pid(addr, mmu_pid, pte_val(pte));
+}
+
 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
 			unsigned long end)
 {
@@ -102,19 +111,18 @@ void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
 	}
 }
 
-void flush_tlb_kernel_range(unsigned long start, unsigned long end)
+void reload_tlb_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
 {
-	while (start < end) {
-		flush_tlb_one(start);
-		start += PAGE_SIZE;
-	}
+	unsigned long mmu_pid = get_pid_from_context(&vma->vm_mm->context);
+
+	reload_tlb_one_pid(addr, mmu_pid, pte);
 }
 
 /*
  * This one is only used for pages with the global bit set so we don't care
  * much about the ASID.
  */
-void flush_tlb_one(unsigned long addr)
+static void flush_tlb_one(unsigned long addr)
 {
 	unsigned int way;
 	unsigned long org_misc, pid_misc;
@@ -130,30 +138,33 @@ void flush_tlb_one(unsigned long addr)
 		unsigned long pteaddr;
 		unsigned long tlbmisc;
 
-		tlbmisc = pid_misc | TLBMISC_RD | (way << TLBMISC_WAY_SHIFT);
+		tlbmisc = TLBMISC_RD | (way << TLBMISC_WAY_SHIFT);
 		WRCTL(CTL_TLBMISC, tlbmisc);
-		pteaddr = RDCTL(CTL_PTEADDR);
-		tlbmisc = RDCTL(CTL_TLBMISC);
 
-		if ((((pteaddr >> 2) & 0xfffff)) == (addr >> PAGE_SHIFT)) {
-			unsigned long vaddr = CONFIG_NIOS2_IO_REGION_BASE +
-				((PAGE_SIZE * cpuinfo.tlb_num_lines) * way) +
-				(addr & TLB_INDEX_MASK);
+		pteaddr = RDCTL(CTL_PTEADDR);
+		if (((pteaddr >> 2) & 0xfffff) != (addr >> PAGE_SHIFT))
+			continue;
 
-			pr_debug("Flush entry by writing %#lx way=%dl pid=%ld\n",
-				vaddr, way, (pid_misc >> TLBMISC_PID_SHIFT));
+		pr_debug("Flush entry by writing way=%dl pid=%ld\n",
+			 way, (pid_misc >> TLBMISC_PID_SHIFT));
 
-			tlbmisc = pid_misc | TLBMISC_WE |
-				(way << TLBMISC_WAY_SHIFT);
-			WRCTL(CTL_PTEADDR, (vaddr >> 12) << 2);
+		tlbmisc = TLBMISC_WE | (way << TLBMISC_WAY_SHIFT);
 		WRCTL(CTL_TLBMISC, tlbmisc);
-			WRCTL(CTL_TLBACC, (MAX_PHYS_ADDR >> PAGE_SHIFT));
-		}
+		WRCTL(CTL_PTEADDR, pteaddr_invalid(addr));
+		WRCTL(CTL_TLBACC, 0);
 	}
 
 	WRCTL(CTL_TLBMISC, org_misc);
 }
 
+void flush_tlb_kernel_range(unsigned long start, unsigned long end)
+{
+	while (start < end) {
+		flush_tlb_one(start);
+		start += PAGE_SIZE;
+	}
+}
+
 void dump_tlb_line(unsigned long line)
 {
 	unsigned int way;
@@ -177,7 +188,7 @@ void dump_tlb_line(unsigned long line)
 		tlbmisc = RDCTL(CTL_TLBMISC);
 		tlbacc = RDCTL(CTL_TLBACC);
 
-		if ((tlbacc << PAGE_SHIFT) != (MAX_PHYS_ADDR & PAGE_MASK)) {
+		if ((tlbacc << PAGE_SHIFT) != 0) {
 			pr_debug("-- way:%02x vpn:0x%08lx phys:0x%08lx pid:0x%02lx flags:%c%c%c%c%c\n",
 				way,
 				(pteaddr << (PAGE_SHIFT-2)),
@@ -203,8 +214,9 @@ void dump_tlb(void)
 		dump_tlb_line(i);
 }
 
-void flush_tlb_pid(unsigned long pid)
+void flush_tlb_pid(unsigned long mmu_pid)
 {
+	unsigned long addr = 0;
 	unsigned int line;
 	unsigned int way;
 	unsigned long org_misc, pid_misc;
@@ -213,55 +225,65 @@ void flush_tlb_pid(unsigned long pid)
 	get_misc_and_pid(&org_misc, &pid_misc);
 
 	for (line = 0; line < cpuinfo.tlb_num_lines; line++) {
-		WRCTL(CTL_PTEADDR, line << 2);
+		WRCTL(CTL_PTEADDR, pteaddr_invalid(addr));
 
 		for (way = 0; way < cpuinfo.tlb_num_ways; way++) {
-			unsigned long pteaddr;
 			unsigned long tlbmisc;
-			unsigned long tlbacc;
+			unsigned long pid;
 
-			tlbmisc = pid_misc | TLBMISC_RD |
-				(way << TLBMISC_WAY_SHIFT);
+			tlbmisc = TLBMISC_RD | (way << TLBMISC_WAY_SHIFT);
 			WRCTL(CTL_TLBMISC, tlbmisc);
-			pteaddr = RDCTL(CTL_PTEADDR);
 			tlbmisc = RDCTL(CTL_TLBMISC);
-			tlbacc = RDCTL(CTL_TLBACC);
+			pid = (tlbmisc >> TLBMISC_PID_SHIFT) & TLBMISC_PID_MASK;
+			if (pid != mmu_pid)
+				continue;
 
-			if (((tlbmisc>>TLBMISC_PID_SHIFT) & TLBMISC_PID_MASK)
-				== pid) {
-				tlbmisc = pid_misc | TLBMISC_WE |
-					(way << TLBMISC_WAY_SHIFT);
+			tlbmisc = TLBMISC_WE | (way << TLBMISC_WAY_SHIFT);
 			WRCTL(CTL_TLBMISC, tlbmisc);
-				WRCTL(CTL_TLBACC,
-					(MAX_PHYS_ADDR >> PAGE_SHIFT));
+			WRCTL(CTL_TLBACC, 0);
 		}
+
+		addr += PAGE_SIZE;
 	}
 
 	WRCTL(CTL_TLBMISC, org_misc);
+}
+
+/*
+ * All entries common to a mm share an asid.  To effectively flush these
+ * entries, we just bump the asid.
+ */
+void flush_tlb_mm(struct mm_struct *mm)
+{
+	if (current->mm == mm) {
+		unsigned long mmu_pid = get_pid_from_context(&mm->context);
+		flush_tlb_pid(mmu_pid);
+	} else {
+		memset(&mm->context, 0, sizeof(mm_context_t));
 	}
 }
 
 void flush_tlb_all(void)
 {
-	int i;
-	unsigned long vaddr = CONFIG_NIOS2_IO_REGION_BASE;
+	unsigned long addr = 0;
+	unsigned int line;
 	unsigned int way;
-	unsigned long org_misc, pid_misc, tlbmisc;
+	unsigned long org_misc, pid_misc;
 
 	/* remember pid/way until we return */
 	get_misc_and_pid(&org_misc, &pid_misc);
-	pid_misc |= TLBMISC_WE;
+
+	/* Start at way 0, way is auto-incremented after each TLBACC write */
+	WRCTL(CTL_TLBMISC, TLBMISC_WE);
 
 	/* Map each TLB entry to physcal address 0 with no-access and a
 	   bad ptbase */
-	for (way = 0; way < cpuinfo.tlb_num_ways; way++) {
-		tlbmisc = pid_misc | (way << TLBMISC_WAY_SHIFT);
-		for (i = 0; i < cpuinfo.tlb_num_lines; i++) {
-			WRCTL(CTL_PTEADDR, ((vaddr) >> PAGE_SHIFT) << 2);
-			WRCTL(CTL_TLBMISC, tlbmisc);
-			WRCTL(CTL_TLBACC, (MAX_PHYS_ADDR >> PAGE_SHIFT));
-			vaddr += 1UL << 12;
-		}
+	for (line = 0; line < cpuinfo.tlb_num_lines; line++) {
+		WRCTL(CTL_PTEADDR, pteaddr_invalid(addr));
+		for (way = 0; way < cpuinfo.tlb_num_ways; way++)
+			WRCTL(CTL_TLBACC, 0);
+
+		addr += PAGE_SIZE;
 	}
 
 	/* restore pid/way */
@@ -270,6 +292,10 @@ void flush_tlb_all(void)
 
 void set_mmu_pid(unsigned long pid)
 {
-	WRCTL(CTL_TLBMISC, (RDCTL(CTL_TLBMISC) & TLBMISC_WAY) |
-		((pid & TLBMISC_PID_MASK) << TLBMISC_PID_SHIFT));
+	unsigned long tlbmisc;
+
+	tlbmisc = RDCTL(CTL_TLBMISC);
+	tlbmisc = (tlbmisc & TLBMISC_WAY);
+	tlbmisc |= (pid & TLBMISC_PID_MASK) << TLBMISC_PID_SHIFT;
+	WRCTL(CTL_TLBMISC, tlbmisc);
 }

arch/nios2/platform/Kconfig.platform

@@ -17,7 +17,6 @@ comment "Device tree"
 
 config NIOS2_DTB_AT_PHYS_ADDR
 	bool "DTB at physical address"
-	default n
 	help
 	  When enabled you can select a physical address to load the dtb from.
 	  Normally this address is passed by a bootloader such as u-boot but
@@ -37,7 +36,6 @@ config NIOS2_DTB_PHYS_ADDR
 
 config NIOS2_DTB_SOURCE_BOOL
 	bool "Compile and link device tree into kernel image"
-	default n
 	help
 	  This allows you to specify a dts (device tree source) file
 	  which will be compiled and linked into the kernel image.
@@ -62,21 +60,18 @@ config NIOS2_ARCH_REVISION
 
 config NIOS2_HW_MUL_SUPPORT
 	bool "Enable MUL instruction"
-	default n
 	help
 	  Set to true if you configured the Nios II to include the MUL
 	  instruction.  This will enable the -mhw-mul compiler flag.
 
 config NIOS2_HW_MULX_SUPPORT
 	bool "Enable MULX instruction"
-	default n
 	help
 	  Set to true if you configured the Nios II to include the MULX
 	  instruction.  Enables the -mhw-mulx compiler flag.
 
 config NIOS2_HW_DIV_SUPPORT
 	bool "Enable DIV instruction"
-	default n
 	help
 	  Set to true if you configured the Nios II to include the DIV
 	  instruction.  Enables the -mhw-div compiler flag.
@@ -84,7 +79,6 @@ config NIOS2_HW_DIV_SUPPORT
 config NIOS2_BMX_SUPPORT
 	bool "Enable BMX instructions"
 	depends on NIOS2_ARCH_REVISION = 2
-	default n
 	help
 	  Set to true if you configured the Nios II R2 to include
 	  the BMX Bit Manipulation Extension instructions. Enables
@@ -93,7 +87,6 @@ config NIOS2_BMX_SUPPORT
 config NIOS2_CDX_SUPPORT
 	bool "Enable CDX instructions"
 	depends on NIOS2_ARCH_REVISION = 2
-	default n
 	help
 	  Set to true if you configured the Nios II R2 to include
 	  the CDX Bit Manipulation Extension instructions. Enables
@@ -101,13 +94,11 @@ config NIOS2_CDX_SUPPORT
 
 config NIOS2_FPU_SUPPORT
 	bool "Custom floating point instr support"
-	default n
 	help
 	  Enables the -mcustom-fpu-cfg=60-1 compiler flag.
 
 config NIOS2_CI_SWAB_SUPPORT
 	bool "Byteswap custom instruction"
-	default n
 	help
 	  Use the byteswap (endian converter) Nios II custom instruction provided
 	  by Altera and which can be enabled in QSYS builder. This accelerates