[PATCH v5 07/22] powerpc/mm: Use hardware assistance in TLB handlers on the 8xx

From: Christophe Leroy
Date: Tue Sep 25 2018 - 12:50:57 EST

Next message: Christophe Leroy: "[PATCH v5 08/22] powerpc/mm: Enable 512k hugepage support with HW assistance on the 8xx"
Previous message: Christophe Leroy: "[PATCH v5 06/22] powerpc/8xx: Temporarily disable 16k pages and 512k hugepages"
In reply to: Christophe Leroy: "[PATCH v5 06/22] powerpc/8xx: Temporarily disable 16k pages and 512k hugepages"
Next in thread: Christophe Leroy: "[PATCH v5 08/22] powerpc/mm: Enable 512k hugepage support with HW assistance on the 8xx"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Today, on the 8xx the TLB handlers do SW tablewalk by doing all
the calculation in ASM, in order to match with the Linux page
table structure.

The 8xx offers hardware assistance which allows significant size
reduction of the TLB handlers, hence also reduces the time spent
in the handlers.

However, using this HW assistance implies some constraints on the
page table structure:
- Regardless of the main page size used (4k or 16k), the
level 1 table (PGD) contains 1024 entries and each PGD entry covers
a 4Mbytes area which is managed by a level 2 table (PTE) containing
also 1024 entries each describing a 4k page.
- 16k pages require 4 identifical entries in the L2 table
- 512k pages PTE have to be spread every 128 bytes in the L2 table
- 8M pages PTE are at the address pointed by the L1 entry and each
8M page require 2 identical entries in the PGD.

This patch modifies the TLB handlers to use HW assistance for 4K PAGES.

Before that patch, the mean time spent in TLB miss handlers is:
- ITLB miss: 80 ticks
- DTLB miss: 62 ticks
After that patch, the mean time spent in TLB miss handlers is:
- ITLB miss: 72 ticks
- DTLB miss: 54 ticks
So the improvement is 10% for ITLB and 13% for DTLB misses

Signed-off-by: Christophe Leroy <christophe.leroy@xxxxxx>
---
arch/powerpc/kernel/head_8xx.S | 97 +++++++++++++-----------------------------
arch/powerpc/mm/8xx_mmu.c | 4 +-
2 files changed, 32 insertions(+), 69 deletions(-)

diff --git a/arch/powerpc/kernel/head_8xx.S b/arch/powerpc/kernel/head_8xx.S
index 9b31721b522c..50e97027b507 100644
--- a/arch/powerpc/kernel/head_8xx.S
+++ b/arch/powerpc/kernel/head_8xx.S
@@ -292,7 +292,7 @@ SystemCall:
. = 0x1100
/*
* For the MPC8xx, this is a software tablewalk to load the instruction
- * TLB. The task switch loads the M_TW register with the pointer to the first
+ * TLB. The task switch loads the M_TWB register with the pointer to the first
* level table.
* If we discover there is no second level table (value is zero) or if there
* is an invalid pte, we load that into the TLB, which causes another fault
@@ -314,7 +314,7 @@ SystemCall:
InstructionTLBMiss:
mtspr SPRN_SPRG_SCRATCH0, r10
mtspr SPRN_SPRG_SCRATCH1, r11
-#if defined(ITLB_MISS_KERNEL) || defined(CONFIG_HUGETLB_PAGE)
+#ifdef ITLB_MISS_KERNEL
mtspr SPRN_SPRG_SCRATCH2, r12
#endif

@@ -323,12 +323,11 @@ InstructionTLBMiss:
*/
mfspr r10, SPRN_SRR0 /* Get effective address of fault */
INVALIDATE_ADJACENT_PAGES_CPU15(r11, r10)
+ mtspr SPRN_MD_EPN, r10
/* Only modules will cause ITLB Misses as we always
* pin the first 8MB of kernel memory */
-#if defined(ITLB_MISS_KERNEL) || defined(CONFIG_HUGETLB_PAGE)
- mfcr r12
-#endif
#ifdef ITLB_MISS_KERNEL
+ mfcr r12
#if defined(SIMPLE_KERNEL_ADDRESS) && defined(CONFIG_PIN_TLB_TEXT)
andis. r11, r10, 0x8000 /* Address >= 0x80000000 */
#else
@@ -341,7 +340,7 @@ InstructionTLBMiss:
#endif
#endif
#endif
- mfspr r11, SPRN_M_TW /* Get level 1 table */
+ mfspr r11, SPRN_M_TWB /* Get level 1 table */
#ifdef ITLB_MISS_KERNEL
#if defined(SIMPLE_KERNEL_ADDRESS) && defined(CONFIG_PIN_TLB_TEXT)
beq+ 3f
@@ -351,23 +350,17 @@ InstructionTLBMiss:
#ifndef CONFIG_PIN_TLB_TEXT
blt cr7, ITLBMissLinear
#endif
- lis r11, (swapper_pg_dir-PAGE_OFFSET)@ha
+ rlwinm r11, r11, 0, 20, 31
+ oris r11, r11, (swapper_pg_dir - PAGE_OFFSET)@ha
3:
#endif
- /* Insert level 1 index */
- rlwimi r11, r10, 32 - ((PAGE_SHIFT - 2) << 1), (PAGE_SHIFT - 2) << 1, 29
lwz r11, (swapper_pg_dir-PAGE_OFFSET)@l(r11) /* Get the level 1 entry */

- /* Extract level 2 index */
- rlwinm r10, r10, 32 - (PAGE_SHIFT - 2), 32 - PAGE_SHIFT, 29
-#ifdef CONFIG_HUGETLB_PAGE
- mtcr r11
- bt- 28, 10f /* bit 28 = Large page (8M) */
-#endif
- rlwimi r10, r11, 0, 0, 32 - PAGE_SHIFT - 1 /* Add level 2 base */
+ mtspr SPRN_MD_TWC, r11
+ mfspr r10, SPRN_MD_TWC
lwz r10, 0(r10) /* Get the pte */
-4:
-#if defined(ITLB_MISS_KERNEL) || defined(CONFIG_HUGETLB_PAGE)
+
+#ifdef ITLB_MISS_KERNEL
mtcr r12
#endif
/* Load the MI_TWC with the attributes for this "segment." */
@@ -392,7 +385,7 @@ InstructionTLBMiss:
/* Restore registers */
0: mfspr r10, SPRN_SPRG_SCRATCH0
mfspr r11, SPRN_SPRG_SCRATCH1
-#if defined(ITLB_MISS_KERNEL) || defined(CONFIG_HUGETLB_PAGE)
+#ifdef ITLB_MISS_KERNEL
mfspr r12, SPRN_SPRG_SCRATCH2
#endif
rfi
@@ -405,20 +398,12 @@ InstructionTLBMiss:
stw r10, (itlb_miss_counter - PAGE_OFFSET)@l(0)
mfspr r10, SPRN_SPRG_SCRATCH0
mfspr r11, SPRN_SPRG_SCRATCH1
-#if defined(ITLB_MISS_KERNEL) || defined(CONFIG_HUGETLB_PAGE)
+#ifdef ITLB_MISS_KERNEL
mfspr r12, SPRN_SPRG_SCRATCH2
#endif
rfi
#endif

-#ifdef CONFIG_HUGETLB_PAGE
-10: /* 8M pages */
- /* Level 2 base */
- rlwinm r10, r11, 0, ~HUGEPD_SHIFT_MASK
- lwz r10, 0(r10) /* Get the pte */
- b 4b
-#endif
-
. = 0x1200
DataStoreTLBMiss:
mtspr SPRN_SPRG_SCRATCH0, r10
@@ -432,7 +417,7 @@ DataStoreTLBMiss:
mfspr r10, SPRN_MD_EPN
rlwinm r11, r10, 16, 0xfff8
cmpli cr0, r11, PAGE_OFFSET@h
- mfspr r11, SPRN_M_TW /* Get level 1 table */
+ mfspr r11, SPRN_M_TWB /* Get level 1 table */
blt+ 3f
rlwinm r11, r10, 16, 0xfff8
#ifndef CONFIG_PIN_TLB_IMMR
@@ -445,24 +430,16 @@ DataStoreTLBMiss:
patch_site 0b, patch__dtlbmiss_immr_jmp
#endif
blt cr7, DTLBMissLinear
- lis r11, (swapper_pg_dir-PAGE_OFFSET)@ha
+ mfspr r11, SPRN_M_TWB /* Get level 1 table */
+ rlwinm r11, r11, 0, 20, 31
+ oris r11, r11, (swapper_pg_dir - PAGE_OFFSET)@ha
3:
-
- /* Insert level 1 index */
- rlwimi r11, r10, 32 - ((PAGE_SHIFT - 2) << 1), (PAGE_SHIFT - 2) << 1, 29
lwz r11, (swapper_pg_dir-PAGE_OFFSET)@l(r11) /* Get the level 1 entry */

- /* We have a pte table, so load fetch the pte from the table.
- */
- /* Extract level 2 index */
- rlwinm r10, r10, 32 - (PAGE_SHIFT - 2), 32 - PAGE_SHIFT, 29
-#ifdef CONFIG_HUGETLB_PAGE
- mtcr r11
- bt- 28, 10f /* bit 28 = Large page (8M) */
-#endif
- rlwimi r10, r11, 0, 0, 32 - PAGE_SHIFT - 1 /* Add level 2 base */
+ mtspr SPRN_MD_TWC, r11
+ mfspr r10, SPRN_MD_TWC
lwz r10, 0(r10) /* Get the pte */
-4:
+
mtcr r12

/* Insert the Guarded flag into the TWC from the Linux PTE.
@@ -517,15 +494,6 @@ DataStoreTLBMiss:
rfi
#endif

-#ifdef CONFIG_HUGETLB_PAGE
-10: /* 8M pages */
- /* Extract level 2 index */
- /* Level 2 base */
- rlwinm r10, r11, 0, ~HUGEPD_SHIFT_MASK
- lwz r10, 0(r10) /* Get the pte */
- b 4b
-#endif
-
/* This is an instruction TLB error on the MPC8xx. This could be due
* to many reasons, such as executing guarded memory or illegal instruction
* addresses. There is nothing to do but handle a big time error fault.
@@ -696,9 +664,10 @@ FixupDAR:/* Entry point for dcbx workaround. */
mtspr SPRN_SPRG_SCRATCH2, r10
/* fetch instruction from memory. */
mfspr r10, SPRN_SRR0
+ mtspr SPRN_MD_EPN, r10
rlwinm r11, r10, 16, 0xfff8
cmpli cr0, r11, PAGE_OFFSET@h
- mfspr r11, SPRN_M_TW /* Get level 1 table */
+ mfspr r11, SPRN_M_TWB /* Get level 1 table */
blt+ 3f
rlwinm r11, r10, 16, 0xfff8

@@ -708,17 +677,17 @@ FixupDAR:/* Entry point for dcbx workaround. */
/* create physical page address from effective address */
tophys(r11, r10)
blt- cr7, 201f
- lis r11, (swapper_pg_dir-PAGE_OFFSET)@ha
- /* Insert level 1 index */
-3: rlwimi r11, r10, 32 - ((PAGE_SHIFT - 2) << 1), (PAGE_SHIFT - 2) << 1, 29
+ mfspr r11, SPRN_M_TWB /* Get level 1 table */
+ rlwinm r11, r11, 0, 20, 31
+ oris r11, r11, (swapper_pg_dir - PAGE_OFFSET)@ha
+3:
lwz r11, (swapper_pg_dir-PAGE_OFFSET)@l(r11) /* Get the level 1 entry */
+ mtspr SPRN_MD_TWC, r11
mtcr r11
+ mfspr r11, SPRN_MD_TWC
+ lwz r11, 0(r11) /* Get the pte */
bt 28,200f /* bit 28 = Large page (8M) */
bt 29,202f /* bit 29 = Large page (8M or 512K) */
- rlwinm r11, r11,0,0,19 /* Extract page descriptor page address */
- /* Insert level 2 index */
- rlwimi r11, r10, 32 - (PAGE_SHIFT - 2), 32 - PAGE_SHIFT, 29
- lwz r11, 0(r11) /* Get the pte */
/* concat physical page address(r11) and page offset(r10) */
rlwimi r11, r10, 0, 32 - PAGE_SHIFT, 31
201: lwz r11,0(r11)
@@ -740,18 +709,12 @@ FixupDAR:/* Entry point for dcbx workaround. */
141: mfspr r10,SPRN_SPRG_SCRATCH2
b DARFixed /* Nope, go back to normal TLB processing */

- /* concat physical page address(r11) and page offset(r10) */
200:
- rlwinm r11, r10, 0, ~HUGEPD_SHIFT_MASK
- lwz r11, 0(r11) /* Get the pte */
/* concat physical page address(r11) and page offset(r10) */
rlwimi r11, r10, 0, 32 - PAGE_SHIFT_8M, 31
b 201b

202:
- rlwinm r11, r11, 0, 0, 32 + PAGE_SHIFT_512K - (PAGE_SHIFT << 1) - 1
- rlwimi r11, r10, 32 - (PAGE_SHIFT_512K - 2), 32 + PAGE_SHIFT_512K - (PAGE_SHIFT << 1), 29
- lwz r11, 0(r11) /* Get the pte */
/* concat physical page address(r11) and page offset(r10) */
rlwimi r11, r10, 0, 32 - PAGE_SHIFT_512K, 31
b 201b
@@ -867,7 +830,7 @@ start_here:

lis r6, swapper_pg_dir@ha
tophys(r6,r6)
- mtspr SPRN_M_TW, r6
+ mtspr SPRN_M_TWB, r6

bl early_init /* We have to do this with MMU on */

diff --git a/arch/powerpc/mm/8xx_mmu.c b/arch/powerpc/mm/8xx_mmu.c
index d39f3af03221..896e710e5697 100644
--- a/arch/powerpc/mm/8xx_mmu.c
+++ b/arch/powerpc/mm/8xx_mmu.c
@@ -174,12 +174,12 @@ void set_context(unsigned long id, pgd_t *pgd)
*(ptr + 1) = pgd;
#endif

- /* Register M_TW will contain base address of level 1 table minus the
+ /* Register M_TWB will contain base address of level 1 table minus the
* lower part of the kernel PGDIR base address, so that all accesses to
* level 1 table are done relative to lower part of kernel PGDIR base
* address.
*/
- mtspr(SPRN_M_TW, __pa(pgd) - offset);
+ mtspr(SPRN_M_TWB, __pa(pgd) - offset);

/* Update context */
mtspr(SPRN_M_CASID, id - 1);
--
2.13.3

Next message: Christophe Leroy: "[PATCH v5 08/22] powerpc/mm: Enable 512k hugepage support with HW assistance on the 8xx"
Previous message: Christophe Leroy: "[PATCH v5 06/22] powerpc/8xx: Temporarily disable 16k pages and 512k hugepages"
In reply to: Christophe Leroy: "[PATCH v5 06/22] powerpc/8xx: Temporarily disable 16k pages and 512k hugepages"
Next in thread: Christophe Leroy: "[PATCH v5 08/22] powerpc/mm: Enable 512k hugepage support with HW assistance on the 8xx"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]