ARC: mm: HIGHMEM: kmap API implementation

Implement kmap* API for ARC.

This enables
 - permanent kernel maps (pkmaps): :kmap() API
 - fixmap : kmap_atomic()

We use a very simple/uniform approach for both (unlike some of the other
arches). So fixmap doesn't use the customary compile time address stuff.
The important semantic is sleep'ability (pkmap) vs. not (fixmap) which
the API guarantees.

Note that this patch only enables highmem for subsequent PAE40 support
as there is no real highmem for ARC in pure 32-bit paradigm as explained
below.

ARC has 2:2 address split of the 32-bit address space with lower half
being translated (virtual) while upper half unstranslated
(0x8000_0000 to 0xFFFF_FFFF). kernel itself is linked at base of
unstranslated space (i.e. 0x8000_0000 onwards), which is mapped to say
DDR 0x0 by external Bus Glue logic (outside the core). So kernel can
potentially access 1.75G worth of memory directly w/o need for highmem.
(the top 256M is taken by uncached peripheral space from 0xF000_0000 to
0xFFFF_FFFF)

In PAE40, hardware can address memory beyond 4G (0x1_0000_0000) while
the logical/virtual addresses remain 32-bits. Thus highmem is required
for kernel proper to be able to access these pages for it's own purposes
(user space is agnostic to this anyways).

Signed-off-by: Alexey Brodkin <abrodkin@synopsys.com>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>

1 files changed, 140 insertions, 0 deletions

diff --git a/arch/arc/mm/highmem.c b/arch/arc/mm/highmem.c
new file mode 100644
index 000000000000..065ee6bfa82a
--- /dev/null
+++ b/arch/arc/mm/highmem.c
@@ -0,0 +1,140 @@
+/*
+ * Copyright (C) 2015 Synopsys, Inc. (www.synopsys.com)
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#include <linux/bootmem.h>
+#include <linux/export.h>
+#include <linux/highmem.h>
+#include <asm/processor.h>
+#include <asm/pgtable.h>
+#include <asm/pgalloc.h>
+#include <asm/tlbflush.h>
+
+/*
+ * HIGHMEM API:
+ *
+ * kmap() API provides sleep semantics hence refered to as "permanent maps"
+ * It allows mapping LAST_PKMAP pages, using @last_pkmap_nr as the cursor
+ * for book-keeping
+ *
+ * kmap_atomic() can't sleep (calls pagefault_disable()), thus it provides
+ * shortlived ala "temporary mappings" which historically were implemented as
+ * fixmaps (compile time addr etc). Their book-keeping is done per cpu.
+ *
+ *	Both these facts combined (preemption disabled and per-cpu allocation)
+ *	means the total number of concurrent fixmaps will be limited to max
+ *	such allocations in a single control path. Thus KM_TYPE_NR (another
+ *	historic relic) is a small'ish number which caps max percpu fixmaps
+ *
+ * ARC HIGHMEM Details
+ *
+ * - the kernel vaddr space from 0x7z to 0x8z (currently used by vmalloc/module)
+ *   is now shared between vmalloc and kmap (non overlapping though)
+ *
+ * - Both fixmap/pkmap use a dedicated page table each, hooked up to swapper PGD
+ *   This means each only has 1 PGDIR_SIZE worth of kvaddr mappings, which means
+ *   2M of kvaddr space for typical config (8K page and 11:8:13 traversal split)
+ *
+ * - fixmap anyhow needs a limited number of mappings. So 2M kvaddr == 256 PTE
+ *   slots across NR_CPUS would be more than sufficient (generic code defines
+ *   KM_TYPE_NR as 20).
+ *
+ * - pkmap being preemptible, in theory could do with more than 256 concurrent
+ *   mappings. However, generic pkmap code: map_new_virtual(), doesn't traverse
+ *   the PGD and only works with a single page table @pkmap_page_table, hence
+ *   sets the limit
+ */
+
+extern pte_t * pkmap_page_table;
+static pte_t * fixmap_page_table;
+
+void *kmap(struct page *page)
+{
+	BUG_ON(in_interrupt());
+	if (!PageHighMem(page))
+		return page_address(page);
+
+	return kmap_high(page);
+}
+
+void *kmap_atomic(struct page *page)
+{
+	int idx, cpu_idx;
+	unsigned long vaddr;
+
+	preempt_disable();
+	pagefault_disable();
+	if (!PageHighMem(page))
+		return page_address(page);
+
+	cpu_idx = kmap_atomic_idx_push();
+	idx = cpu_idx + KM_TYPE_NR * smp_processor_id();
+	vaddr = FIXMAP_ADDR(idx);
+
+	set_pte_at(&init_mm, vaddr, fixmap_page_table + idx,
+		   mk_pte(page, kmap_prot));
+
+	return (void *)vaddr;
+}
+EXPORT_SYMBOL(kmap_atomic);
+
+void __kunmap_atomic(void *kv)
+{
+	unsigned long kvaddr = (unsigned long)kv;
+
+	if (kvaddr >= FIXMAP_BASE && kvaddr < (FIXMAP_BASE + FIXMAP_SIZE)) {
+
+		/*
+		 * Because preemption is disabled, this vaddr can be associated
+		 * with the current allocated index.
+		 * But in case of multiple live kmap_atomic(), it still relies on
+		 * callers to unmap in right order.
+		 */
+		int cpu_idx = kmap_atomic_idx();
+		int idx = cpu_idx + KM_TYPE_NR * smp_processor_id();
+
+		WARN_ON(kvaddr != FIXMAP_ADDR(idx));
+
+		pte_clear(&init_mm, kvaddr, fixmap_page_table + idx);
+		local_flush_tlb_kernel_range(kvaddr, kvaddr + PAGE_SIZE);
+
+		kmap_atomic_idx_pop();
+	}
+
+	pagefault_enable();
+	preempt_enable();
+}
+EXPORT_SYMBOL(__kunmap_atomic);
+
+noinline pte_t *alloc_kmap_pgtable(unsigned long kvaddr)
+{
+	pgd_t *pgd_k;
+	pud_t *pud_k;
+	pmd_t *pmd_k;
+	pte_t *pte_k;
+
+	pgd_k = pgd_offset_k(kvaddr);
+	pud_k = pud_offset(pgd_k, kvaddr);
+	pmd_k = pmd_offset(pud_k, kvaddr);
+
+	pte_k = (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
+	pmd_populate_kernel(&init_mm, pmd_k, pte_k);
+	return pte_k;
+}
+
+void kmap_init(void)
+{
+	/* Due to recursive include hell, we can't do this in processor.h */
+	BUILD_BUG_ON(PAGE_OFFSET < (VMALLOC_END + FIXMAP_SIZE + PKMAP_SIZE));
+
+	BUILD_BUG_ON(KM_TYPE_NR > PTRS_PER_PTE);
+	pkmap_page_table = alloc_kmap_pgtable(PKMAP_BASE);
+
+	BUILD_BUG_ON(LAST_PKMAP > PTRS_PER_PTE);
+	fixmap_page_table = alloc_kmap_pgtable(FIXMAP_BASE);
+}