Shrink the size of tlb handler and fix vmalloc()

By combining swapper_pg_dir and module_pg_dir, several if conditions can be eliminated from the tlb exception handler. The reason they can be combined is that, the effective virtual address of vmalloc returned is at the bottom, and of module_alloc returned is at the top. It also fixes the bug in vmalloc(), which happens when its return address is not covered by the first pgd.

Signed-off-by Wu Fei <[email protected]>

---
 arch/mips/include/asm/pgtable-64.h |   11 ++------
 arch/mips/mm/init.c                |    3 --
 arch/mips/mm/pgtable-64.c          |    3 --
 arch/mips/mm/tlbex.c               |   49 ------------------------------------
 4 files changed, 3 insertions(+), 63 deletions(-)

arch/mips/include/asm/pgtable-64.h +3 -8

index 4ed9d1b..9cd5089 100644
@@ -109,13 +109,13 @@
 
 #define VMALLOC_START		MAP_BASE
 #define VMALLOC_END	\
-	(VMALLOC_START + PTRS_PER_PGD * PTRS_PER_PMD * PTRS_PER_PTE * PAGE_SIZE)
+	(VMALLOC_START + \
+	 PTRS_PER_PGD * PTRS_PER_PMD * PTRS_PER_PTE * PAGE_SIZE - (1UL << 32))
 #if defined(CONFIG_MODULES) && defined(KBUILD_64BIT_SYM32) && \
 	VMALLOC_START != CKSSEG
 /* Load modules into 32bit-compatible segment. */
 #define MODULE_START	CKSSEG
 #define MODULE_END	(FIXADDR_START-2*PAGE_SIZE)
-extern pgd_t module_pg_dir[PTRS_PER_PGD];
 #endif
 
 #define pte_ERROR(e) \

@@ -188,12 +188,7 @@ static inline void pud_clear(pud_t *pudp)
 #define __pmd_offset(address)	pmd_index(address)
 
 /* to find an entry in a kernel page-table-directory */
-#ifdef MODULE_START
-#define pgd_offset_k(address) \
-	((address) >= MODULE_START ? module_pg_dir : pgd_offset(&init_mm, 0UL))
-#else
-#define pgd_offset_k(address) pgd_offset(&init_mm, 0UL)
-#endif
+#define pgd_offset_k(address) pgd_offset(&init_mm, address)
 
 #define pgd_index(address)	(((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
 #define pmd_index(address)	(((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))

arch/mips/mm/init.c +0 -3

index 0e82050..38c79c5 100644
@@ -475,9 +475,6 @@ unsigned long pgd_current[NR_CPUS];
  */
 pgd_t swapper_pg_dir[_PTRS_PER_PGD] __page_aligned(_PGD_ORDER);
 #ifdef CONFIG_64BIT
-#ifdef MODULE_START
-pgd_t module_pg_dir[PTRS_PER_PGD] __page_aligned(PGD_ORDER);
-#endif
 pmd_t invalid_pmd_table[PTRS_PER_PMD] __page_aligned(PMD_ORDER);
 #endif
 pte_t invalid_pte_table[PTRS_PER_PTE] __page_aligned(PTE_ORDER);

arch/mips/mm/pgtable-64.c +0 -3

index e4b565a..1121019 100644
@@ -59,9 +59,6 @@ void __init pagetable_init(void)
 
 	/* Initialize the entire pgd.  */
 	pgd_init((unsigned long)swapper_pg_dir);
-#ifdef MODULE_START
-	pgd_init((unsigned long)module_pg_dir);
-#endif
 	pmd_init((unsigned long)invalid_pmd_table, (unsigned long)invalid_pte_table);
 
 	pgd_base = swapper_pg_dir;

arch/mips/mm/tlbex.c +0 -49

index 9a17bf8..bc66f57 100644
@@ -499,11 +499,7 @@ build_get_pmde64(u32 **p, struct uasm_label **l, struct uasm_reloc **r,
 	 * The vmalloc handling is not in the hotpath.
 	 */
 	uasm_i_dmfc0(p, tmp, C0_BADVADDR);
-#ifdef MODULE_START
-	uasm_il_bltz(p, r, tmp, label_module_alloc);
-#else
 	uasm_il_bltz(p, r, tmp, label_vmalloc);
-#endif
 	/* No uasm_i_nop needed here, since the next insn doesn't touch TMP. */
 
 #ifdef CONFIG_SMP

@@ -556,52 +552,7 @@ build_get_pgd_vmalloc64(u32 **p, struct uasm_label **l, struct uasm_reloc **r,
 {
 	long swpd = (long)swapper_pg_dir;
 
-#ifdef MODULE_START
-	long modd = (long)module_pg_dir;
-
-	uasm_l_module_alloc(l, *p);
-	/*
-	 * Assumption:
-	 * VMALLOC_START >= 0xc000000000000000UL
-	 * MODULE_START >= 0xe000000000000000UL
-	 */
-	UASM_i_SLL(p, ptr, bvaddr, 2);
-	uasm_il_bgez(p, r, ptr, label_vmalloc);
-
-	if (uasm_in_compat_space_p(MODULE_START) &&
-	    !uasm_rel_lo(MODULE_START)) {
-		uasm_i_lui(p, ptr, uasm_rel_hi(MODULE_START)); /* delay slot */
-	} else {
-		/* unlikely configuration */
-		uasm_i_nop(p); /* delay slot */
-		UASM_i_LA(p, ptr, MODULE_START);
-	}
-	uasm_i_dsubu(p, bvaddr, bvaddr, ptr);
-
-	if (uasm_in_compat_space_p(modd) && !uasm_rel_lo(modd)) {
-		uasm_il_b(p, r, label_vmalloc_done);
-		uasm_i_lui(p, ptr, uasm_rel_hi(modd));
-	} else {
-		UASM_i_LA_mostly(p, ptr, modd);
-		uasm_il_b(p, r, label_vmalloc_done);
-		if (uasm_in_compat_space_p(modd))
-			uasm_i_addiu(p, ptr, ptr, uasm_rel_lo(modd));
-		else
-			uasm_i_daddiu(p, ptr, ptr, uasm_rel_lo(modd));
-	}
-
 	uasm_l_vmalloc(l, *p);
-	if (uasm_in_compat_space_p(MODULE_START) &&
-	    !uasm_rel_lo(MODULE_START) &&
-	    MODULE_START << 32 == VMALLOC_START)
-		uasm_i_dsll32(p, ptr, ptr, 0);	/* typical case */
-	else
-		UASM_i_LA(p, ptr, VMALLOC_START);
-#else
-	uasm_l_vmalloc(l, *p);
-	UASM_i_LA(p, ptr, VMALLOC_START);
-#endif
-	uasm_i_dsubu(p, bvaddr, bvaddr, ptr);
 
 	if (uasm_in_compat_space_p(swpd) && !uasm_rel_lo(swpd)) {
 		uasm_il_b(p, r, label_vmalloc_done);

This patch tries to shrink the size of the 64bit tlb handler and also fix an vmalloc bug at the same time.

By combining the swapper_pg_dir and module_pg_dir, several checks in tlb handler, particularly build_get_pgd_vmalloc64, are not necessary. The reason they can be combined is that, the effective virtual address of vmalloc returned is at the bottom, and of module_alloc returned is at the top.

In the normal case of 4KB page size:
VMALLOC_START, VMALLOC_END 0xc0000000 00000000 - 0xc0000100 00000000
MODULE_START, MODULE_END 0xffffffff c0000000 - 0xffffffff +xxxxxxx
Change it to:
VMALLOC_START, VMALLOC_END 0xc0000000 00000000 - 0xc00000ff 00000000
MODULE_START, MODULE_END 0xffffffff c0000000 - 0xffffffff +xxxxxxx
We use the least 40 bits to traverse the page table, the change makes it still one-to-one mapping without more checking. "+" is in the range of [c,d,e,f], so there even are big holes bewteen them.

With this patch, the tlb refill handler only contains about 28 instructions, instead of the original 38.

And this patch also fix a bug in vmalloc, which happens when its returned address is not covered by the first pgd. e.g. if we do two vmallocs, the first returned address is 0xc0000000 00000000, and the 2nd is 0xc0000000 40000000,

vmalloc -> __vmalloc_node -> __vmalloc_area_node -> __vmalloc_area_node -> map_vm_area -> pgd_offset_k

pgd_offset_k doesn't use the address to index the pgd, just return the first one:

  #define pgd_offset_k(address) \ 
        ((address) >= MODULE_START ? module_pg_dir : pgd_offset(&init_mm, 0UL))     

This is wrong, then the 2 addresses are mapped to the same pte. This bug doesn't happen because even in the 4KB page case, one pgd can cover 1GB size, and it looks like the system won't vmalloc so much area.