Thread (45 messages) 45 messages, 5 authors, 2016-02-18

Re: [PATCH V6 20/35] powerpc/mm: Don't track subpage valid bit in pte_t

From: Aneesh Kumar K.V <hidden>
Date: 2015-12-03 18:58:12

Anshuman Khandual [off-list ref] writes:
On 12/01/2015 09:06 AM, Aneesh Kumar K.V wrote:
quoted
This free up 11 bits in pte_t. In the later patch we also change
the pte_t format so that we can start supporting migration pte
at pmd level. We now track 4k subpage valid bit as below

If we have _PAGE_COMBO set, we override the _PAGE_F_GIX_SHIFT
and _PAGE_F_SECOND. Together we have 4 bits, each of them
used to indicate whether any of the 4 4k subpage in that group
is valid. ie,

[ group 1 bit ]   [ group 2 bit ]  ..... [ group 4 ]
[ subpage 1 - 4]  [ subpage 5- 8]  ..... [ subpage 13 - 16]

We still track each 4k subpage slot number and secondary hash
information in the second half of pgtable_t. Removing the subpage
tracking have some significant overhead on aim9 and ebizzy benchmark and
to support THP with 4K subpage, we do need a pgtable_t of 4096 bytes.

Signed-off-by: Aneesh Kumar K.V <redacted>
We removed 16 bits which used to track individual sub pages and added
back 4 bits to track newly created sub page groups. So we save 12 bits
there. How it is 11 bits ? Have we added back one more bit into pte_t
some where for sub page purpose which I am missing ?

We added back 5 bits, (I guess you are missing _PAGE_HASHPTE )

quoted
---
 arch/powerpc/include/asm/book3s/64/hash-4k.h  | 10 +-------
 arch/powerpc/include/asm/book3s/64/hash-64k.h | 35 ++++++---------------------
 arch/powerpc/include/asm/book3s/64/hash.h     | 10 ++++----
 arch/powerpc/mm/hash64_64k.c                  | 34 ++++++++++++++++++++++++--
.....
......
quoted hunk ↗ jump to hunk
diff --git a/arch/powerpc/mm/hash64_64k.c b/arch/powerpc/mm/hash64_64k.c
quoted
index 9ffeae2cbb57..f1b86ba63430 100644
--- a/arch/powerpc/mm/hash64_64k.c
+++ b/arch/powerpc/mm/hash64_64k.c
@@ -15,6 +15,35 @@
 #include <linux/mm.h>
 #include <asm/machdep.h>
 #include <asm/mmu.h>
+/*
+ * index from 0 - 15
+ */
+bool __rpte_sub_valid(real_pte_t rpte, unsigned long index)
+{
+	unsigned long g_idx;
+	unsigned long ptev = pte_val(rpte.pte);
+
+	g_idx = (ptev & _PAGE_COMBO_VALID) >> _PAGE_F_GIX_SHIFT;
+	index = index >> 2;
+	if (g_idx & (0x1 << index))
+		return true;
+	else
+		return false;
+}
This function checks for validity of the sub page group not individual
sub page index. Because we dont track individual sub page index validity
any more, wondering if that is even possible.
That is correct. We are not tracking individual subpage validity after
this patch. That is explained in the commit message.

quoted
+/*
+ * index from 0 - 15
+ */
+static unsigned long mark_subptegroup_valid(unsigned long ptev, unsigned long index)
+{
+	unsigned long g_idx;
+
+	if (!(ptev & _PAGE_COMBO))
+		return ptev;
+	index = index >> 2;
+	g_idx = 0x1 << index;
+
+	return ptev | (g_idx << _PAGE_F_GIX_SHIFT);
+}
 
 int __hash_page_4K(unsigned long ea, unsigned long access, unsigned long vsid,
 		   pte_t *ptep, unsigned long trap, unsigned long flags,
@@ -102,7 +131,7 @@ int __hash_page_4K(unsigned long ea, unsigned long access, unsigned long vsid,
 	 */
 	if (!(old_pte & _PAGE_COMBO)) {
 		flush_hash_page(vpn, rpte, MMU_PAGE_64K, ssize, flags);
-		old_pte &= ~_PAGE_HPTE_SUB;
+		old_pte &= ~_PAGE_HASHPTE | _PAGE_F_GIX | _PAGE_F_SECOND;
Or  use _PAGE_COMBO_VALID directly instead ?

Nope we are not clearning a combo related detail there. We are
invalidating an old 64K mapping there. (yes, the values are matching but
we should not use the _PAGE_COMBO_VALID there).

quoted
 		goto htab_insert_hpte;
 	}
 	/*
@@ -192,7 +221,8 @@ repeat:
 	/* __real_pte use pte_val() any idea why ? FIXME!! */
 	rpte.hidx &= ~(0xfUL << (subpg_index << 2));
 	*hidxp = rpte.hidx  | (slot << (subpg_index << 2));
-	new_pte |= (_PAGE_HPTE_SUB0 >> subpg_index);
+	new_pte = mark_subptegroup_valid(new_pte, subpg_index);
+	new_pte |=  _PAGE_HASHPTE;
 	/*
 	 * check __real_pte for details on matching smp_rmb()
 	 */
diff --git a/arch/powerpc/mm/hash_low_64.S b/arch/powerpc/mm/hash_low_64.S
index 6b4d4c1d0628..359839a57f26 100644
--- a/arch/powerpc/mm/hash_low_64.S
+++ b/arch/powerpc/mm/hash_low_64.S
@@ -285,7 +285,7 @@ htab_modify_pte:
 
 	/* Secondary group ? if yes, get a inverted hash value */
 	mr	r5,r28
-	andi.	r0,r31,_PAGE_SECONDARY
+	andi.	r0,r31,_PAGE_F_SECOND
 	beq	1f
 	not	r5,r5
 1:
@@ -473,11 +473,7 @@ ht64_insert_pte:
 	lis	r0,_PAGE_HPTEFLAGS@h
 	ori	r0,r0,_PAGE_HPTEFLAGS@l
 	andc	r30,r30,r0
-#ifdef CONFIG_PPC_64K_PAGES
-	oris	r30,r30,_PAGE_HPTE_SUB0@h
-#else
 	ori	r30,r30,_PAGE_HASHPTE
-#endif
 	/* Phyical address in r5 */
 	rldicl	r5,r31,64-PTE_RPN_SHIFT,PTE_RPN_SHIFT
 	sldi	r5,r5,PAGE_SHIFT
diff --git a/arch/powerpc/mm/hugetlbpage-hash64.c b/arch/powerpc/mm/hugetlbpage-hash64.c
index d94b1af53a93..7584e8445512 100644
--- a/arch/powerpc/mm/hugetlbpage-hash64.c
+++ b/arch/powerpc/mm/hugetlbpage-hash64.c
@@ -91,11 +91,8 @@ int __hash_page_huge(unsigned long ea, unsigned long access, unsigned long vsid,
 		pa = pte_pfn(__pte(old_pte)) << PAGE_SHIFT;
 
 		/* clear HPTE slot informations in new PTE */
-#ifdef CONFIG_PPC_64K_PAGES
-		new_pte = (new_pte & ~_PAGE_HPTEFLAGS) | _PAGE_HPTE_SUB0;
-#else
 		new_pte = (new_pte & ~_PAGE_HPTEFLAGS) | _PAGE_HASHPTE;
-#endif
+
 		/* Add in WIMG bits */
 		rflags |= (new_pte & (_PAGE_WRITETHRU | _PAGE_NO_CACHE |
 				      _PAGE_COHERENT | _PAGE_GUARDED));
diff --git a/arch/powerpc/mm/pgtable_64.c b/arch/powerpc/mm/pgtable_64.c
index d692ae31cfc7..3967e3cce03e 100644
--- a/arch/powerpc/mm/pgtable_64.c
+++ b/arch/powerpc/mm/pgtable_64.c
@@ -625,7 +625,7 @@ void pmdp_splitting_flush(struct vm_area_struct *vma,
 	"1:	ldarx	%0,0,%3\n\
 		andi.	%1,%0,%6\n\
 		bne-	1b \n\
-		ori	%1,%0,%4 \n\
+		oris	%1,%0,%4@h \n\
Why is this change required ?

Because _PAGE_SPLITTING is beyond 16 bits now ?

-aneesh
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help