From 853075183f2f0ffd5723e88aa3505aa1ea43a8eb Mon Sep 17 00:00:00 2001
From: Laura Abbott <lauraa@codeaurora.org>
Date: Mon, 14 Apr 2014 19:42:04 -0700
Subject: [PATCH] arm: Allow remapping lowmem as 4K pages

Lowmem is currently mapped with sections (1MB/2MB) whenever
possible due to TLB performance boosts. Sections cannot easily
be changed at runtime however which makes implementing certain
features annoying. Add an option to map lowmem with 4K patches.
This is intended to be used as a debugging feature and should
NOT be used for performance testing.

Change-Id: I9612a99b8e05a022f5ba7e568f21307cf66b5667
Signed-off-by: Laura Abbott <lauraa@codeaurora.org>
---
 arch/arm/Kconfig.debug |  11 +++++
 arch/arm/mm/mmu.c      | 105 +++++++++++++++++++++++++++++++++++++++--
 2 files changed, 112 insertions(+), 4 deletions(-)

diff --git a/arch/arm/Kconfig.debug b/arch/arm/Kconfig.debug
index 45270cc7697c..25e9e03dcd1e 100644
--- a/arch/arm/Kconfig.debug
+++ b/arch/arm/Kconfig.debug
@@ -75,6 +75,17 @@ config DEBUG_USER
 	      8 - SIGSEGV faults
 	     16 - SIGBUS faults
 
+config FORCE_PAGES
+	bool "Force lowmem to be mapped with 4K pages"
+        help
+          There are some advanced debug features that can only be done when
+          memory is mapped with pages instead of sections. Enable this option
+          to always map lowmem pages with pages. This may have a performance
+          cost due to increased TLB pressure.
+
+          If unsure say N.
+
+
 # These options are only for real kernel hackers who want to get their hands dirty.
 config DEBUG_LL
 	bool "Kernel low-level debugging functions (read help!)"
diff --git a/arch/arm/mm/mmu.c b/arch/arm/mm/mmu.c
index 0fe05401b69c..53b669203c4c 100644
--- a/arch/arm/mm/mmu.c
+++ b/arch/arm/mm/mmu.c
@@ -393,11 +393,13 @@ int set_memory_##_name(unsigned long addr, int numpages) \
 	unsigned long size = PAGE_SIZE*numpages; \
 	unsigned end = start + size; \
 \
-	if (start < MODULES_VADDR || start >= MODULES_END) \
-		return -EINVAL;\
+	if (!IS_ENABLED(CONFIG_FORCE_PAGES)) { \
+		if (start < MODULES_VADDR || start >= MODULES_END) \
+			return -EINVAL;\
 \
-	if (end < MODULES_VADDR || end >= MODULES_END) \
-		return -EINVAL; \
+		if (end < MODULES_VADDR || end >= MODULES_END) \
+			return -EINVAL; \
+	} \
 \
 	apply_to_page_range(&init_mm, start, size, callback, NULL); \
 	flush_tlb_kernel_range(start, end); \
@@ -1567,6 +1569,100 @@ static void __init map_lowmem(void)
 	}
 }
 
+#ifdef CONFIG_FORCE_PAGES
+/*
+ * remap a PMD into pages
+ * We split a single pmd here none of this two pmd nonsense
+ */
+static noinline void split_pmd(pmd_t *pmd, unsigned long addr,
+				unsigned long end, unsigned long pfn,
+				const struct mem_type *type)
+{
+	pte_t *pte, *start_pte;
+
+	start_pte = early_alloc(PTE_HWTABLE_OFF + PTE_HWTABLE_SIZE);
+
+	pte = start_pte;
+
+	do {
+		set_pte_ext(pte, pfn_pte(pfn, type->prot_pte), 0);
+		pfn++;
+	} while (pte++, addr += PAGE_SIZE, addr != end);
+
+	*pmd = __pmd((__pa(start_pte) + PTE_HWTABLE_OFF) | type->prot_l1);
+	mb();
+	flush_pmd_entry(pmd);
+	flush_tlb_all();
+}
+
+/*
+ * It's significantly easier to remap as pages later after all memory is
+ * mapped. Everything is sections so all we have to do is split
+ */
+static void __init remap_pages(void)
+{
+	struct memblock_region *reg;
+
+	for_each_memblock(memory, reg) {
+		phys_addr_t phys_start = reg->base;
+		phys_addr_t phys_end = reg->base + reg->size;
+		unsigned long addr = (unsigned long)__va(phys_start);
+		unsigned long end = (unsigned long)__va(phys_end);
+		pmd_t *pmd = NULL;
+		unsigned long next;
+		unsigned long pfn = __phys_to_pfn(phys_start);
+		bool fixup = false;
+		unsigned long saved_start = addr;
+
+		if (phys_end > arm_lowmem_limit)
+			end = (unsigned long)__va(arm_lowmem_limit);
+		if (phys_start >= phys_end)
+			break;
+
+		pmd = pmd_offset(
+			pud_offset(pgd_offset(&init_mm, addr), addr), addr);
+
+#ifndef	CONFIG_ARM_LPAE
+		if (addr & SECTION_SIZE) {
+			fixup = true;
+			pmd_empty_section_gap((addr - SECTION_SIZE) & PMD_MASK);
+			pmd++;
+		}
+
+		if (end & SECTION_SIZE)
+			pmd_empty_section_gap(end);
+#endif
+
+		do {
+			next = addr + SECTION_SIZE;
+
+			if (pmd_none(*pmd) || pmd_bad(*pmd))
+				split_pmd(pmd, addr, next, pfn,
+						&mem_types[MT_MEMORY]);
+			pmd++;
+			pfn += SECTION_SIZE >> PAGE_SHIFT;
+
+		} while (addr = next, addr < end);
+
+		if (fixup) {
+			/*
+			 * Put a faulting page table here to avoid detecting no
+			 * pmd when accessing an odd section boundary. This
+			 * needs to be faulting to help catch errors and avoid
+			 * speculation
+			 */
+			pmd = pmd_off_k(saved_start);
+			pmd[0] = pmd[1] & ~1;
+		}
+	}
+}
+#else
+static void __init remap_pages(void)
+{
+
+}
+#endif
+
 /*
  * paging_init() sets up the page tables, initialises the zone memory
  * maps, and sets up the zero page, bad page and bad page tables.
@@ -1581,6 +1677,7 @@ void __init paging_init(struct machine_desc *mdesc)
 	prepare_page_table();
 	map_lowmem();
 	dma_contiguous_remap();
+	remap_pages();
 	devicemaps_init(mdesc);
 	kmap_init();
 	tcm_init();