* [PATCH 1/7] Add various hugetlb arm high level hooks
2012-02-13 9:44 [RFC-PATCH V2] ARM hugetlb support Bill Carson
@ 2012-02-13 9:44 ` Bill Carson
2012-02-28 17:30 ` Catalin Marinas
` (2 more replies)
2012-02-13 9:44 ` [PATCH 2/7] Add various hugetlb page table fix Bill Carson
` (5 subsequent siblings)
6 siblings, 3 replies; 29+ messages in thread
From: Bill Carson @ 2012-02-13 9:44 UTC (permalink / raw)
To: linux-arm-kernel
Signed-off-by: Bill Carson <bill4carson@gmail.com>
---
arch/arm/include/asm/hugetlb.h | 178 ++++++++++++++++++++++++++++++++++++++++
arch/arm/include/asm/page.h | 15 ++++
arch/arm/mm/hugetlb.c | 132 +++++++++++++++++++++++++++++
3 files changed, 325 insertions(+), 0 deletions(-)
create mode 100644 arch/arm/include/asm/hugetlb.h
create mode 100644 arch/arm/mm/hugetlb.c
diff --git a/arch/arm/include/asm/hugetlb.h b/arch/arm/include/asm/hugetlb.h
new file mode 100644
index 0000000..3c34528
--- /dev/null
+++ b/arch/arm/include/asm/hugetlb.h
@@ -0,0 +1,178 @@
+/*
+ * hugetlb.h, ARM Huge Tlb Page support.
+ *
+ * Copyright (c) Bill Carson
+ *
+ * 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.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ *
+ */
+
+#ifndef __ASM_HUGETLB_H
+#define __ASM_HUGETLB_H
+
+#include <asm/page.h>
+#include <asm/pgtable-2level.h>
+#include <asm/tlb.h>
+
+
+/* 2M and 16M hugepage linux ptes are stored in mmu_context_t->huge_linux_pte
+ *
+ * 2M hugepage
+ * ===========
+ * one huge linux pte caters to two HW ptes,
+ *
+ * 16M hugepage
+ * ============
+ * one huge linux pte caters for sixteen HW ptes,
+ *
+ * The number of huge linux ptes depends on PAGE_OFFSET configuration
+ * which is defined as following:
+ */
+#define HUGE_LINUX_PTE_COUNT ( PAGE_OFFSET >> HPAGE_SHIFT)
+#define HUGE_LINUX_PTE_SIZE (HUGE_LINUX_PTE_COUNT * sizeof(pte_t *))
+#define HUGE_LINUX_PTE_INDEX(addr) (addr >> HPAGE_SHIFT)
+
+static inline int is_hugepage_only_range(struct mm_struct *mm,
+ unsigned long addr,
+ unsigned long len)
+{
+ return 0;
+}
+
+static inline int prepare_hugepage_range(struct file *file,
+ unsigned long addr,
+ unsigned long len)
+{
+ struct hstate *h = hstate_file(file);
+ /* addr/len should be aligned with huge page size */
+ if (len & ~huge_page_mask(h))
+ return -EINVAL;
+ if (addr & ~huge_page_mask(h))
+ return -EINVAL;
+
+ return 0;
+}
+
+static inline void hugetlb_prefault_arch_hook(struct mm_struct *mm)
+{
+}
+
+static inline void hugetlb_free_pgd_range(struct mmu_gather *tlb,
+ unsigned long addr, unsigned long end,
+ unsigned long floor, unsigned long ceiling)
+{
+}
+
+static inline void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
+ pte_t *ptep, pte_t pte)
+{
+ pgd_t *pgd;
+ pud_t *pud;
+ pmd_t *pmd;
+ pte_t *linuxpte = mm->context.huge_linux_pte;
+
+ BUG_ON(linuxpte == NULL);
+ BUG_ON(HUGE_LINUX_PTE_INDEX(addr) >= HUGE_LINUX_PTE_COUNT);
+ BUG_ON(ptep != &linuxpte[HUGE_LINUX_PTE_INDEX(addr)]);
+
+ /* set huge linux pte first */
+ *ptep = pte;
+
+ /* then set hardware pte */
+ addr &= HPAGE_MASK;
+ pgd = pgd_offset(mm, addr);
+ pud = pud_offset(pgd, addr);
+ pmd = pmd_offset(pud, addr);
+ set_hugepte_at(mm, addr, pmd, pte);
+}
+
+static inline pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
+ unsigned long addr, pte_t *ptep)
+{
+ pte_t pte = *ptep;
+ pte_t fake = L_PTE_YOUNG;
+ pgd_t *pgd;
+ pud_t *pud;
+ pmd_t *pmd;
+
+ /* clear linux pte */
+ *ptep = 0;
+
+ /* let set_hugepte_at clear HW entry */
+ addr &= HPAGE_MASK;
+ pgd = pgd_offset(mm, addr);
+ pud = pud_offset(pgd, addr);
+ pmd = pmd_offset(pud, addr);
+ set_hugepte_at(mm, addr, pmd, fake);
+ return pte;
+}
+
+static inline void huge_ptep_clear_flush(struct vm_area_struct *vma,
+ unsigned long addr, pte_t *ptep)
+{
+ if (HPAGE_SHIFT == SUPERSECTION_SHIFT)
+ flush_tlb_page(vma, addr & SUPERSECTION_MASK);
+ else {
+ flush_tlb_page(vma, addr & SECTION_MASK);
+ flush_tlb_page(vma, (addr & SECTION_MASK)^0x100000);
+ }
+}
+
+static inline int huge_pte_none(pte_t pte)
+{
+ return pte_none(pte);
+}
+
+static inline pte_t huge_pte_wrprotect(pte_t pte)
+{
+ return pte_wrprotect(pte);
+}
+
+static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
+ unsigned long addr, pte_t *ptep)
+{
+ pte_t old_pte = *ptep;
+ set_huge_pte_at(mm, addr, ptep, pte_wrprotect(old_pte));
+}
+
+static inline pte_t huge_ptep_get(pte_t *ptep)
+{
+ return *ptep;
+}
+
+static inline int huge_ptep_set_access_flags(struct vm_area_struct *vma,
+ unsigned long addr,
+ pte_t *ptep, pte_t pte,
+ int dirty)
+{
+ int changed = !pte_same(huge_ptep_get(ptep), pte);
+ if (changed) {
+ set_huge_pte_at(vma->vm_mm, addr, ptep, pte);
+ huge_ptep_clear_flush(vma, addr, &pte);
+ }
+
+ return changed;
+}
+
+static inline int arch_prepare_hugepage(struct page *page)
+{
+ return 0;
+}
+
+static inline void arch_release_hugepage(struct page *page)
+{
+}
+
+#endif /* __ASM_HUGETLB_H */
+
diff --git a/arch/arm/include/asm/page.h b/arch/arm/include/asm/page.h
index 97b440c..3e6769a 100644
--- a/arch/arm/include/asm/page.h
+++ b/arch/arm/include/asm/page.h
@@ -15,6 +15,21 @@
#define PAGE_SIZE (_AC(1,UL) << PAGE_SHIFT)
#define PAGE_MASK (~(PAGE_SIZE-1))
+#ifdef CONFIG_HUGEPAGE_SIZE_2MB
+/* we have 2MB hugepage for two 1MB section mapping */
+#define HPAGE_SHIFT (SECTION_SHIFT + 1)
+#define HPAGE_SIZE (_AC(1, UL) << HPAGE_SHIFT)
+#define HPAGE_MASK (~(HPAGE_SIZE - 1))
+#define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT)
+#endif
+
+#ifdef CONFIG_HUGEPAGE_SIZE_16MB
+#define HPAGE_SHIFT SUPERSECTION_SHIFT
+#define HPAGE_SIZE (_AC(1, UL) << HPAGE_SHIFT)
+#define HPAGE_MASK (~(HPAGE_SIZE - 1))
+#define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT)
+#endif
+
#ifndef __ASSEMBLY__
#ifndef CONFIG_MMU
diff --git a/arch/arm/mm/hugetlb.c b/arch/arm/mm/hugetlb.c
new file mode 100644
index 0000000..165bd8f
--- /dev/null
+++ b/arch/arm/mm/hugetlb.c
@@ -0,0 +1,132 @@
+/*
+ * hugetlb.c, ARM Huge Tlb Page support.
+ *
+ * Copyright (c) Bill Carson
+ *
+ * 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.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ * See the GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ *
+ */
+
+#include <linux/hugetlb.h>
+
+pte_t *huge_pte_alloc(struct mm_struct *mm, unsigned long addr,
+ unsigned long sz)
+{
+ pte_t *linuxpte = mm->context.huge_linux_pte;
+ int index;
+
+ if (linuxpte == NULL) {
+ linuxpte = kzalloc(HUGE_LINUX_PTE_SIZE, GFP_ATOMIC);
+ if (linuxpte == NULL) {
+ printk(KERN_ERR "Cannot allocate memory for huge linux pte\n");
+ return NULL;
+ }
+ mm->context.huge_linux_pte = linuxpte;
+ }
+ /* huge page mapping only cover user space address */
+ BUG_ON(HUGE_LINUX_PTE_INDEX(addr) >= HUGE_LINUX_PTE_COUNT);
+ index = HUGE_LINUX_PTE_INDEX(addr);
+ return &linuxpte[HUGE_LINUX_PTE_INDEX(addr)];
+}
+
+pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr)
+{
+ pgd_t *pgd;
+ pud_t *pud;
+ pmd_t *pmd = NULL;
+ pte_t *linuxpte = mm->context.huge_linux_pte;
+
+ /* check this mapping exist at pmd level */
+ pgd = pgd_offset(mm, addr);
+ if (pgd_present(*pgd)) {
+ pud = pud_offset(pgd, addr);
+ pmd = pmd_offset(pud, addr);
+ if (!pmd_present(*pmd))
+ return NULL;
+ }
+
+ BUG_ON(HUGE_LINUX_PTE_INDEX(addr) >= HUGE_LINUX_PTE_COUNT);
+ BUG_ON((*pmd & PMD_TYPE_MASK) != PMD_TYPE_SECT);
+ return &linuxpte[HUGE_LINUX_PTE_INDEX(addr)];
+}
+
+int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep)
+{
+ return 0;
+}
+
+struct page *follow_huge_addr(struct mm_struct *mm, unsigned long address,
+ int write)
+{
+ return ERR_PTR(-EINVAL);
+}
+
+int pmd_huge(pmd_t pmd)
+{
+ return (pmd_val(pmd) & PMD_TYPE_MASK) == PMD_TYPE_SECT;
+}
+
+int pud_huge(pud_t pud)
+{
+ return 0; } struct page * follow_huge_pmd(struct mm_struct *mm, unsigned long address, pmd_t *pmd, int write)
+{
+ struct page *page = NULL;
+ unsigned long pfn;
+
+ BUG_ON((pmd_val(*pmd) & PMD_TYPE_MASK) != PMD_TYPE_SECT);
+ pfn = ((pmd_val(*pmd) & HPAGE_MASK) >> PAGE_SHIFT);
+ page = pfn_to_page(pfn);
+ return page;
+}
+
+static int __init add_huge_page_size(unsigned long long size)
+{
+ int shift = __ffs(size);
+ u32 mmfr3 = 0;
+
+ /* Check that it is a page size supported by the hardware and
+ * that it fits within pagetable and slice limits. */
+ if (!is_power_of_2(size) || (shift != HPAGE_SHIFT))
+ return -EINVAL;
+
+ /* If user wants super-section support, then check if our cpu
+ * has this feature supported in ID_MMFR3 */
+ if (shift == SUPERSECTION_SHIFT) {
+ __asm__("mrc p15, 0, %0, c0, c1, 7\n" : "=r" (mmfr3));
+ if (mmfr3 & 0xF0000000) {
+ printk("Super-Section is NOT supported by this CPU, mmfr3:0x%x\n", mmfr3);
+ return -EINVAL;
+ }
+ }
+
+ /* Return if huge page size has already been setup */
+ if (size_to_hstate(size))
+ return 0;
+
+ hugetlb_add_hstate(shift - PAGE_SHIFT);
+ return 0;
+}
+
+static int __init hugepage_setup_sz(char *str)
+{
+ unsigned long long size;
+
+ size = memparse(str, &str);
+ if (add_huge_page_size(size) != 0)
+ printk(KERN_WARNING "Invalid huge page size specified(%llu)\n",
+ size);
+
+ return 1;
+}
+__setup("hugepagesz=", hugepage_setup_sz);
--
1.7.1
^ permalink raw reply related [flat|nested] 29+ messages in thread* [PATCH 1/7] Add various hugetlb arm high level hooks
2012-02-13 9:44 ` [PATCH 1/7] Add various hugetlb arm high level hooks Bill Carson
@ 2012-02-28 17:30 ` Catalin Marinas
2012-02-29 2:34 ` bill4carson
2012-02-29 10:32 ` Catalin Marinas
2012-02-29 12:31 ` Arnd Bergmann
2 siblings, 1 reply; 29+ messages in thread
From: Catalin Marinas @ 2012-02-28 17:30 UTC (permalink / raw)
To: linux-arm-kernel
On Mon, Feb 13, 2012 at 09:44:22AM +0000, Bill Carson wrote:
> diff --git a/arch/arm/include/asm/page.h b/arch/arm/include/asm/page.h
> index 97b440c..3e6769a 100644
> --- a/arch/arm/include/asm/page.h
> +++ b/arch/arm/include/asm/page.h
> @@ -15,6 +15,21 @@
> #define PAGE_SIZE (_AC(1,UL) << PAGE_SHIFT)
> #define PAGE_MASK (~(PAGE_SIZE-1))
>
> +#ifdef CONFIG_HUGEPAGE_SIZE_2MB
> +/* we have 2MB hugepage for two 1MB section mapping */
> +#define HPAGE_SHIFT (SECTION_SHIFT + 1)
> +#define HPAGE_SIZE (_AC(1, UL) << HPAGE_SHIFT)
> +#define HPAGE_MASK (~(HPAGE_SIZE - 1))
> +#define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT)
> +#endif
> +
> +#ifdef CONFIG_HUGEPAGE_SIZE_16MB
> +#define HPAGE_SHIFT SUPERSECTION_SHIFT
> +#define HPAGE_SIZE (_AC(1, UL) << HPAGE_SHIFT)
> +#define HPAGE_MASK (~(HPAGE_SIZE - 1))
> +#define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT)
> +#endif
Ah, you still have these just config time options. Can you not make an
hpage_shift variable like PowerPC or IA-64?
(I haven't yet reviewed the rest of the code)
--
Catalin
^ permalink raw reply [flat|nested] 29+ messages in thread
* [PATCH 1/7] Add various hugetlb arm high level hooks
2012-02-28 17:30 ` Catalin Marinas
@ 2012-02-29 2:34 ` bill4carson
2012-02-29 9:39 ` Catalin Marinas
0 siblings, 1 reply; 29+ messages in thread
From: bill4carson @ 2012-02-29 2:34 UTC (permalink / raw)
To: linux-arm-kernel
On 2012?02?29? 01:30, Catalin Marinas wrote:
> On Mon, Feb 13, 2012 at 09:44:22AM +0000, Bill Carson wrote:
>> diff --git a/arch/arm/include/asm/page.h b/arch/arm/include/asm/page.h
>> index 97b440c..3e6769a 100644
>> --- a/arch/arm/include/asm/page.h
>> +++ b/arch/arm/include/asm/page.h
>> @@ -15,6 +15,21 @@
>> #define PAGE_SIZE (_AC(1,UL)<< PAGE_SHIFT)
>> #define PAGE_MASK (~(PAGE_SIZE-1))
>>
>> +#ifdef CONFIG_HUGEPAGE_SIZE_2MB
>> +/* we have 2MB hugepage for two 1MB section mapping */
>> +#define HPAGE_SHIFT (SECTION_SHIFT + 1)
>> +#define HPAGE_SIZE (_AC(1, UL)<< HPAGE_SHIFT)
>> +#define HPAGE_MASK (~(HPAGE_SIZE - 1))
>> +#define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT)
>> +#endif
>> +
>> +#ifdef CONFIG_HUGEPAGE_SIZE_16MB
>> +#define HPAGE_SHIFT SUPERSECTION_SHIFT
>> +#define HPAGE_SIZE (_AC(1, UL)<< HPAGE_SHIFT)
>> +#define HPAGE_MASK (~(HPAGE_SIZE - 1))
>> +#define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT)
>> +#endif
>
> Ah, you still have these just config time options. Can you not make an
> hpage_shift variable like PowerPC or IA-64?
>
Hi, Catalin
Thank you for reviewing this patch.
unsigned int hpage_shift = SECTION_SHIFT + 1; /* default to 2MB page */
#define HPAGE_SHIFT hpage_shift
then user could configure hpage_shift through "hugepagesz" parameter.
I think this is what I am supposed to do,
If so, hugepage_setup_sz/cpu_v7_set_hugepte_ext needs modification
> (I haven't yet reviewed the rest of the code)
>
So, should I send another version now, or wait until you walk through
the reset of patch set?
And again, thanks for the patch reviewing:)
--
I am a slow learner
but I will keep trying to fight for my dreams!
--bill
^ permalink raw reply [flat|nested] 29+ messages in thread
* [PATCH 1/7] Add various hugetlb arm high level hooks
2012-02-29 2:34 ` bill4carson
@ 2012-02-29 9:39 ` Catalin Marinas
2012-02-29 10:21 ` bill4carson
0 siblings, 1 reply; 29+ messages in thread
From: Catalin Marinas @ 2012-02-29 9:39 UTC (permalink / raw)
To: linux-arm-kernel
On Wed, Feb 29, 2012 at 02:34:34AM +0000, bill4carson wrote:
> On 2012?02?29? 01:30, Catalin Marinas wrote:
> > On Mon, Feb 13, 2012 at 09:44:22AM +0000, Bill Carson wrote:
> >> diff --git a/arch/arm/include/asm/page.h b/arch/arm/include/asm/page.h
> >> index 97b440c..3e6769a 100644
> >> --- a/arch/arm/include/asm/page.h
> >> +++ b/arch/arm/include/asm/page.h
> >> @@ -15,6 +15,21 @@
> >> #define PAGE_SIZE (_AC(1,UL)<< PAGE_SHIFT)
> >> #define PAGE_MASK (~(PAGE_SIZE-1))
> >>
> >> +#ifdef CONFIG_HUGEPAGE_SIZE_2MB
> >> +/* we have 2MB hugepage for two 1MB section mapping */
> >> +#define HPAGE_SHIFT (SECTION_SHIFT + 1)
> >> +#define HPAGE_SIZE (_AC(1, UL)<< HPAGE_SHIFT)
> >> +#define HPAGE_MASK (~(HPAGE_SIZE - 1))
> >> +#define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT)
> >> +#endif
> >> +
> >> +#ifdef CONFIG_HUGEPAGE_SIZE_16MB
> >> +#define HPAGE_SHIFT SUPERSECTION_SHIFT
> >> +#define HPAGE_SIZE (_AC(1, UL)<< HPAGE_SHIFT)
> >> +#define HPAGE_MASK (~(HPAGE_SIZE - 1))
> >> +#define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT)
> >> +#endif
> >
> > Ah, you still have these just config time options. Can you not make an
> > hpage_shift variable like PowerPC or IA-64?
>
> Thank you for reviewing this patch.
I haven't finished yet (another email to come :)).
> unsigned int hpage_shift = SECTION_SHIFT + 1; /* default to 2MB page */
> #define HPAGE_SHIFT hpage_shift
Something like that. You could even use PMD_SHIFT which is 21 already
and it makes it clearer that we use huge pages at the pmd level.
> So, should I send another version now, or wait until you walk through
> the reset of patch set?
Just wait, I'll have a look through all the patches.
BTW, there are several coding style issues, I won't go through them but
it would help if you have a look at Documentation/CodingStyle
(especially comment style, if/else brackets).
Thanks.
--
Catalin
^ permalink raw reply [flat|nested] 29+ messages in thread
* [PATCH 1/7] Add various hugetlb arm high level hooks
2012-02-29 9:39 ` Catalin Marinas
@ 2012-02-29 10:21 ` bill4carson
2012-02-29 10:23 ` Catalin Marinas
0 siblings, 1 reply; 29+ messages in thread
From: bill4carson @ 2012-02-29 10:21 UTC (permalink / raw)
To: linux-arm-kernel
On 2012?02?29? 17:39, Catalin Marinas wrote:
> On Wed, Feb 29, 2012 at 02:34:34AM +0000, bill4carson wrote:
>> On 2012?02?29? 01:30, Catalin Marinas wrote:
>>> On Mon, Feb 13, 2012 at 09:44:22AM +0000, Bill Carson wrote:
>>>> diff --git a/arch/arm/include/asm/page.h b/arch/arm/include/asm/page.h
>>>> index 97b440c..3e6769a 100644
>>>> --- a/arch/arm/include/asm/page.h
>>>> +++ b/arch/arm/include/asm/page.h
>>>> @@ -15,6 +15,21 @@
>>>> #define PAGE_SIZE (_AC(1,UL)<< PAGE_SHIFT)
>>>> #define PAGE_MASK (~(PAGE_SIZE-1))
>>>>
>>>> +#ifdef CONFIG_HUGEPAGE_SIZE_2MB
>>>> +/* we have 2MB hugepage for two 1MB section mapping */
>>>> +#define HPAGE_SHIFT (SECTION_SHIFT + 1)
>>>> +#define HPAGE_SIZE (_AC(1, UL)<< HPAGE_SHIFT)
>>>> +#define HPAGE_MASK (~(HPAGE_SIZE - 1))
>>>> +#define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT)
>>>> +#endif
>>>> +
>>>> +#ifdef CONFIG_HUGEPAGE_SIZE_16MB
>>>> +#define HPAGE_SHIFT SUPERSECTION_SHIFT
>>>> +#define HPAGE_SIZE (_AC(1, UL)<< HPAGE_SHIFT)
>>>> +#define HPAGE_MASK (~(HPAGE_SIZE - 1))
>>>> +#define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT)
>>>> +#endif
>>>
>>> Ah, you still have these just config time options. Can you not make an
>>> hpage_shift variable like PowerPC or IA-64?
>>
>> Thank you for reviewing this patch.
>
> I haven't finished yet (another email to come :)).
>
I'm looking forward any comments/suggestions ^_^
>> unsigned int hpage_shift = SECTION_SHIFT + 1; /* default to 2MB page */
>> #define HPAGE_SHIFT hpage_shift
>
> Something like that. You could even use PMD_SHIFT which is 21 already
> and it makes it clearer that we use huge pages at the pmd level.
>
OK, I will keep that in mind for V3.
>> So, should I send another version now, or wait until you walk through
>> the reset of patch set?
>
> Just wait, I'll have a look through all the patches.
>
> BTW, there are several coding style issues, I won't go through them but
> it would help if you have a look at Documentation/CodingStyle
> (especially comment style, if/else brackets).
>
Yes, I forgot to run checkpatch.pl to sanitize this patch set.
Will keep in mind for all next version.
thanks
> Thanks.
>
--
I am a slow learner
but I will keep trying to fight for my dreams!
--bill
^ permalink raw reply [flat|nested] 29+ messages in thread
* [PATCH 1/7] Add various hugetlb arm high level hooks
2012-02-29 10:21 ` bill4carson
@ 2012-02-29 10:23 ` Catalin Marinas
0 siblings, 0 replies; 29+ messages in thread
From: Catalin Marinas @ 2012-02-29 10:23 UTC (permalink / raw)
To: linux-arm-kernel
On 29 February 2012 10:21, bill4carson <bill4carson@gmail.com> wrote:
> On 2012?02?29? 17:39, Catalin Marinas wrote:
>> BTW, there are several coding style issues, I won't go through them but
>> it would help if you have a look at Documentation/CodingStyle
>> (especially comment style, if/else brackets).
>
> Yes, I forgot to run checkpatch.pl to sanitize this patch set.
Don't just rely on checkpatch.pl, have a look at the CodingStyle doc :)
--
Catalin
^ permalink raw reply [flat|nested] 29+ messages in thread
* [PATCH 1/7] Add various hugetlb arm high level hooks
2012-02-13 9:44 ` [PATCH 1/7] Add various hugetlb arm high level hooks Bill Carson
2012-02-28 17:30 ` Catalin Marinas
@ 2012-02-29 10:32 ` Catalin Marinas
2012-02-29 11:28 ` bill4carson
2012-02-29 12:31 ` Arnd Bergmann
2 siblings, 1 reply; 29+ messages in thread
From: Catalin Marinas @ 2012-02-29 10:32 UTC (permalink / raw)
To: linux-arm-kernel
On Mon, Feb 13, 2012 at 09:44:22AM +0000, Bill Carson wrote:
> --- /dev/null
> +++ b/arch/arm/include/asm/hugetlb.h
...
> +#include <asm/page.h>
> +#include <asm/pgtable-2level.h>
Just include asm/pgtable.h, it includes the right 2level.h file
automatically (and I plan to add LPAE support as well).
> +static inline void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
> + pte_t *ptep, pte_t pte)
> +{
> + pgd_t *pgd;
> + pud_t *pud;
> + pmd_t *pmd;
> + pte_t *linuxpte = mm->context.huge_linux_pte;
> +
> + BUG_ON(linuxpte == NULL);
> + BUG_ON(HUGE_LINUX_PTE_INDEX(addr) >= HUGE_LINUX_PTE_COUNT);
> + BUG_ON(ptep != &linuxpte[HUGE_LINUX_PTE_INDEX(addr)]);
> +
> + /* set huge linux pte first */
> + *ptep = pte;
> +
> + /* then set hardware pte */
> + addr &= HPAGE_MASK;
> + pgd = pgd_offset(mm, addr);
> + pud = pud_offset(pgd, addr);
> + pmd = pmd_offset(pud, addr);
> + set_hugepte_at(mm, addr, pmd, pte);
You may want to add a comment here that we only have two levels of page
tables (and there is no need for pud_none() checks).
Also I would say the set_hugepte_at function name is easily confused
with set_huge_pte_at(), maybe change it to __set_huge_pte_at() or
something else.
> +static inline pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
> + unsigned long addr, pte_t *ptep)
> +{
> + pte_t pte = *ptep;
> + pte_t fake = L_PTE_YOUNG;
> + pgd_t *pgd;
> + pud_t *pud;
> + pmd_t *pmd;
> +
> + /* clear linux pte */
> + *ptep = 0;
> +
> + /* let set_hugepte_at clear HW entry */
> + addr &= HPAGE_MASK;
> + pgd = pgd_offset(mm, addr);
> + pud = pud_offset(pgd, addr);
> + pmd = pmd_offset(pud, addr);
> + set_hugepte_at(mm, addr, pmd, fake);
Same here.
> +static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
> + unsigned long addr, pte_t *ptep)
> +{
> + pte_t old_pte = *ptep;
> + set_huge_pte_at(mm, addr, ptep, pte_wrprotect(old_pte));
> +}
You could use the generic ptep_set_wrprotect()
> +static inline int huge_ptep_set_access_flags(struct vm_area_struct *vma,
> + unsigned long addr,
> + pte_t *ptep, pte_t pte,
> + int dirty)
> +{
> + int changed = !pte_same(huge_ptep_get(ptep), pte);
> + if (changed) {
> + set_huge_pte_at(vma->vm_mm, addr, ptep, pte);
> + huge_ptep_clear_flush(vma, addr, &pte);
> + }
> +
> + return changed;
> +}
I could also use the generic ptep_set_access_flags().
> --- /dev/null
> +++ b/arch/arm/mm/hugetlb.c
...
> +pte_t *huge_pte_alloc(struct mm_struct *mm, unsigned long addr,
> + unsigned long sz)
> +{
> + pte_t *linuxpte = mm->context.huge_linux_pte;
> + int index;
> +
> + if (linuxpte == NULL) {
> + linuxpte = kzalloc(HUGE_LINUX_PTE_SIZE, GFP_ATOMIC);
> + if (linuxpte == NULL) {
> + printk(KERN_ERR "Cannot allocate memory for huge linux pte\n");
pr_err()?
> + return NULL;
> + }
> + mm->context.huge_linux_pte = linuxpte;
> + }
> + /* huge page mapping only cover user space address */
> + BUG_ON(HUGE_LINUX_PTE_INDEX(addr) >= HUGE_LINUX_PTE_COUNT);
> + index = HUGE_LINUX_PTE_INDEX(addr);
> + return &linuxpte[HUGE_LINUX_PTE_INDEX(addr)];
> +}
> +
> +pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr)
> +{
> + pgd_t *pgd;
> + pud_t *pud;
> + pmd_t *pmd = NULL;
> + pte_t *linuxpte = mm->context.huge_linux_pte;
> +
> + /* check this mapping exist at pmd level */
> + pgd = pgd_offset(mm, addr);
> + if (pgd_present(*pgd)) {
> + pud = pud_offset(pgd, addr);
> + pmd = pmd_offset(pud, addr);
> + if (!pmd_present(*pmd))
> + return NULL;
> + }
You could add checks for the pud as well, they would be optimised out by
the compiler but it would be easier to add support for LPAE as well. In
my LPAE hugetlb implementation, I have something like this:
pgd = pgd_offset(mm, addr);
if (pgd_present(*pgd)) {
pud = pud_offset(pgd, addr);
if (pud_present(*pud))
pmd = pmd_offset(pud, addr);
}
> + BUG_ON(HUGE_LINUX_PTE_INDEX(addr) >= HUGE_LINUX_PTE_COUNT);
> + BUG_ON((*pmd & PMD_TYPE_MASK) != PMD_TYPE_SECT);
> + return &linuxpte[HUGE_LINUX_PTE_INDEX(addr)];
You could add a macro to make it easier for LPAE:
#define huge_pte(mm, pmd, addr) \
(&mm->context.huge_linux_pte(HUGE_LINUX_PTE_INDEX(addr)))
With LPAE, it would simply be a (pte_t *)pmd cast.
> +int pud_huge(pud_t pud)
> +{
> + return 0; } struct page * follow_huge_pmd(struct mm_struct *mm, unsigned long address, pmd_t *pmd, int write)
Something went wrong around here.
> +{
> + struct page *page = NULL;
You don't need to initialise page here.
> + unsigned long pfn;
> +
> + BUG_ON((pmd_val(*pmd) & PMD_TYPE_MASK) != PMD_TYPE_SECT);
> + pfn = ((pmd_val(*pmd) & HPAGE_MASK) >> PAGE_SHIFT);
> + page = pfn_to_page(pfn);
> + return page;
> +}
> +
> +static int __init add_huge_page_size(unsigned long long size)
> +{
> + int shift = __ffs(size);
> + u32 mmfr3 = 0;
> +
> + /* Check that it is a page size supported by the hardware and
> + * that it fits within pagetable and slice limits. */
> + if (!is_power_of_2(size) || (shift != HPAGE_SHIFT))
> + return -EINVAL;
You could use get_order() instead of __ffs(), the latter just finds the
first bit set.
But here you should have set hpage_shift.
--
Catalin
^ permalink raw reply [flat|nested] 29+ messages in thread* [PATCH 1/7] Add various hugetlb arm high level hooks
2012-02-29 10:32 ` Catalin Marinas
@ 2012-02-29 11:28 ` bill4carson
2012-02-29 11:36 ` Catalin Marinas
2012-02-29 15:38 ` Catalin Marinas
0 siblings, 2 replies; 29+ messages in thread
From: bill4carson @ 2012-02-29 11:28 UTC (permalink / raw)
To: linux-arm-kernel
On 2012?02?29? 18:32, Catalin Marinas wrote:
> On Mon, Feb 13, 2012 at 09:44:22AM +0000, Bill Carson wrote:
>> --- /dev/null
>> +++ b/arch/arm/include/asm/hugetlb.h
> ...
>> +#include<asm/page.h>
>> +#include<asm/pgtable-2level.h>
>
> Just include asm/pgtable.h, it includes the right 2level.h file
> automatically (and I plan to add LPAE support as well).
>
>> +static inline void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
>> + pte_t *ptep, pte_t pte)
>> +{
>> + pgd_t *pgd;
>> + pud_t *pud;
>> + pmd_t *pmd;
>> + pte_t *linuxpte = mm->context.huge_linux_pte;
>> +
>> + BUG_ON(linuxpte == NULL);
>> + BUG_ON(HUGE_LINUX_PTE_INDEX(addr)>= HUGE_LINUX_PTE_COUNT);
>> + BUG_ON(ptep !=&linuxpte[HUGE_LINUX_PTE_INDEX(addr)]);
>> +
>> + /* set huge linux pte first */
>> + *ptep = pte;
>> +
>> + /* then set hardware pte */
>> + addr&= HPAGE_MASK;
>> + pgd = pgd_offset(mm, addr);
>> + pud = pud_offset(pgd, addr);
>> + pmd = pmd_offset(pud, addr);
>> + set_hugepte_at(mm, addr, pmd, pte);
>
> You may want to add a comment here that we only have two levels of page
> tables (and there is no need for pud_none() checks).
>
I will add a comment to clarify this.
> Also I would say the set_hugepte_at function name is easily confused
> with set_huge_pte_at(), maybe change it to __set_huge_pte_at() or
> something else.
>
yes, I will make that change.
>> +static inline pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
>> + unsigned long addr, pte_t *ptep)
>> +{
>> + pte_t pte = *ptep;
>> + pte_t fake = L_PTE_YOUNG;
>> + pgd_t *pgd;
>> + pud_t *pud;
>> + pmd_t *pmd;
>> +
>> + /* clear linux pte */
>> + *ptep = 0;
>> +
>> + /* let set_hugepte_at clear HW entry */
>> + addr&= HPAGE_MASK;
>> + pgd = pgd_offset(mm, addr);
>> + pud = pud_offset(pgd, addr);
>> + pmd = pmd_offset(pud, addr);
>> + set_hugepte_at(mm, addr, pmd, fake);
>
> Same here.
>
>> +static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
>> + unsigned long addr, pte_t *ptep)
>> +{
>> + pte_t old_pte = *ptep;
>> + set_huge_pte_at(mm, addr, ptep, pte_wrprotect(old_pte));
>> +}
>
> You could use the generic ptep_set_wrprotect()
I'm a bit of confused about this.
generic ptep_set_wrprotect() can not set huge pte, that's why
set_huge_pte_at is used instead here.
>
>> +static inline int huge_ptep_set_access_flags(struct vm_area_struct *vma,
>> + unsigned long addr,
>> + pte_t *ptep, pte_t pte,
>> + int dirty)
>> +{
>> + int changed = !pte_same(huge_ptep_get(ptep), pte);
>> + if (changed) {
>> + set_huge_pte_at(vma->vm_mm, addr, ptep, pte);
>> + huge_ptep_clear_flush(vma, addr,&pte);
>> + }
>> +
>> + return changed;
>> +}
>
> I could also use the generic ptep_set_access_flags().
Same as above.
IMHO, cannot use generic hooks here, cause we are setting huge pte
with a different set pte API than set_pte_at.
>
>> --- /dev/null
>> +++ b/arch/arm/mm/hugetlb.c
> ...
>> +pte_t *huge_pte_alloc(struct mm_struct *mm, unsigned long addr,
>> + unsigned long sz)
>> +{
>> + pte_t *linuxpte = mm->context.huge_linux_pte;
>> + int index;
>> +
>> + if (linuxpte == NULL) {
>> + linuxpte = kzalloc(HUGE_LINUX_PTE_SIZE, GFP_ATOMIC);
>> + if (linuxpte == NULL) {
>> + printk(KERN_ERR "Cannot allocate memory for huge linux pte\n");
>
> pr_err()?
Yes, pr_err should be used in here.
thanks
>
>> + return NULL;
>> + }
>> + mm->context.huge_linux_pte = linuxpte;
>> + }
>> + /* huge page mapping only cover user space address */
>> + BUG_ON(HUGE_LINUX_PTE_INDEX(addr)>= HUGE_LINUX_PTE_COUNT);
>> + index = HUGE_LINUX_PTE_INDEX(addr);
>> + return&linuxpte[HUGE_LINUX_PTE_INDEX(addr)];
>> +}
>> +
>> +pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr)
>> +{
>> + pgd_t *pgd;
>> + pud_t *pud;
>> + pmd_t *pmd = NULL;
>> + pte_t *linuxpte = mm->context.huge_linux_pte;
>> +
>> + /* check this mapping exist at pmd level */
>> + pgd = pgd_offset(mm, addr);
>> + if (pgd_present(*pgd)) {
>> + pud = pud_offset(pgd, addr);
>> + pmd = pmd_offset(pud, addr);
>> + if (!pmd_present(*pmd))
>> + return NULL;
>> + }
>
> You could add checks for the pud as well, they would be optimised out by
> the compiler but it would be easier to add support for LPAE as well. In
> my LPAE hugetlb implementation, I have something like this:
>
> pgd = pgd_offset(mm, addr);
> if (pgd_present(*pgd)) {
> pud = pud_offset(pgd, addr);
> if (pud_present(*pud))
> pmd = pmd_offset(pud, addr);
> }
>
Ok, I will add the pud checks as per your comment.
>> + BUG_ON(HUGE_LINUX_PTE_INDEX(addr)>= HUGE_LINUX_PTE_COUNT);
>> + BUG_ON((*pmd& PMD_TYPE_MASK) != PMD_TYPE_SECT);
>> + return&linuxpte[HUGE_LINUX_PTE_INDEX(addr)];
>
> You could add a macro to make it easier for LPAE:
>
> #define huge_pte(mm, pmd, addr) \
> (&mm->context.huge_linux_pte(HUGE_LINUX_PTE_INDEX(addr)))
>
Nice, I will keep this in V3.
> With LPAE, it would simply be a (pte_t *)pmd cast.
>
>> +int pud_huge(pud_t pud)
>> +{
>> + return 0; } struct page * follow_huge_pmd(struct mm_struct *mm, unsigned long address, pmd_t *pmd, int write)
>
> Something went wrong around here.
>
crap! I will make it cleaner next time. I promise!
>> +{
>> + struct page *page = NULL;
>
> You don't need to initialise page here.
OK, I will drop the "NULL".
>
>> + unsigned long pfn;
>> +
>> + BUG_ON((pmd_val(*pmd)& PMD_TYPE_MASK) != PMD_TYPE_SECT);
>> + pfn = ((pmd_val(*pmd)& HPAGE_MASK)>> PAGE_SHIFT);
>> + page = pfn_to_page(pfn);
>> + return page;
>> +}
>> +
>> +static int __init add_huge_page_size(unsigned long long size)
>> +{
>> + int shift = __ffs(size);
>> + u32 mmfr3 = 0;
>> +
>> + /* Check that it is a page size supported by the hardware and
>> + * that it fits within pagetable and slice limits. */
>> + if (!is_power_of_2(size) || (shift != HPAGE_SHIFT))
>> + return -EINVAL;
>
> You could use get_order() instead of __ffs(), the latter just finds the
> first bit set.
With all due respect, I'm afraid I can't agree with you on this.
here, we should use __ffs to return this "shift" not the order.
For "hugepagesz=2M", hpage_shift/HPAGE_SHIFT should be set to 21,
*not* the order 9(21-12), that's what HUGETLB_PAGE_ORDER for.
>
> But here you should have set hpage_shift.
>
--
I am a slow learner
but I will keep trying to fight for my dreams!
--bill
^ permalink raw reply [flat|nested] 29+ messages in thread* [PATCH 1/7] Add various hugetlb arm high level hooks
2012-02-29 11:28 ` bill4carson
@ 2012-02-29 11:36 ` Catalin Marinas
2012-02-29 15:38 ` Catalin Marinas
1 sibling, 0 replies; 29+ messages in thread
From: Catalin Marinas @ 2012-02-29 11:36 UTC (permalink / raw)
To: linux-arm-kernel
On Wed, Feb 29, 2012 at 11:28:30AM +0000, bill4carson wrote:
> On 2012?02?29? 18:32, Catalin Marinas wrote:
> > On Mon, Feb 13, 2012 at 09:44:22AM +0000, Bill Carson wrote:
> >> +static int __init add_huge_page_size(unsigned long long size)
> >> +{
> >> + int shift = __ffs(size);
> >> + u32 mmfr3 = 0;
> >> +
> >> + /* Check that it is a page size supported by the hardware and
> >> + * that it fits within pagetable and slice limits. */
> >> + if (!is_power_of_2(size) || (shift != HPAGE_SHIFT))
> >> + return -EINVAL;
> >
> > You could use get_order() instead of __ffs(), the latter just finds the
> > first bit set.
>
> With all due respect, I'm afraid I can't agree with you on this.
> here, we should use __ffs to return this "shift" not the order.
>
> For "hugepagesz=2M", hpage_shift/HPAGE_SHIFT should be set to 21,
> *not* the order 9(21-12), that's what HUGETLB_PAGE_ORDER for.
I agree (I got confused by get_order() and log2()).
--
Catalin
^ permalink raw reply [flat|nested] 29+ messages in thread* [PATCH 1/7] Add various hugetlb arm high level hooks
2012-02-29 11:28 ` bill4carson
2012-02-29 11:36 ` Catalin Marinas
@ 2012-02-29 15:38 ` Catalin Marinas
2012-03-08 0:35 ` bill4carson
1 sibling, 1 reply; 29+ messages in thread
From: Catalin Marinas @ 2012-02-29 15:38 UTC (permalink / raw)
To: linux-arm-kernel
On Wed, Feb 29, 2012 at 11:28:30AM +0000, bill4carson wrote:
> On 2012?02?29? 18:32, Catalin Marinas wrote:
> > On Mon, Feb 13, 2012 at 09:44:22AM +0000, Bill Carson wrote:
> >> +static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
> >> + unsigned long addr, pte_t *ptep)
> >> +{
> >> + pte_t old_pte = *ptep;
> >> + set_huge_pte_at(mm, addr, ptep, pte_wrprotect(old_pte));
> >> +}
> >
> > You could use the generic ptep_set_wrprotect()
>
> I'm a bit of confused about this.
>
> generic ptep_set_wrprotect() can not set huge pte, that's why
> set_huge_pte_at is used instead here.
Ah, the generic one can only work with with LPAE where set_huge_pte_at()
is just a set_pte_at(). So this part looks good.
--
Catalin
^ permalink raw reply [flat|nested] 29+ messages in thread* [PATCH 1/7] Add various hugetlb arm high level hooks
2012-02-29 15:38 ` Catalin Marinas
@ 2012-03-08 0:35 ` bill4carson
2012-03-08 9:21 ` Catalin Marinas
0 siblings, 1 reply; 29+ messages in thread
From: bill4carson @ 2012-03-08 0:35 UTC (permalink / raw)
To: linux-arm-kernel
On 2012?02?29? 23:38, Catalin Marinas wrote:
> On Wed, Feb 29, 2012 at 11:28:30AM +0000, bill4carson wrote:
>> On 2012?02?29? 18:32, Catalin Marinas wrote:
>>> On Mon, Feb 13, 2012 at 09:44:22AM +0000, Bill Carson wrote:
>>>> +static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
>>>> + unsigned long addr, pte_t *ptep)
>>>> +{
>>>> + pte_t old_pte = *ptep;
>>>> + set_huge_pte_at(mm, addr, ptep, pte_wrprotect(old_pte));
>>>> +}
>>>
>>> You could use the generic ptep_set_wrprotect()
>>
>> I'm a bit of confused about this.
>>
>> generic ptep_set_wrprotect() can not set huge pte, that's why
>> set_huge_pte_at is used instead here.
>
> Ah, the generic one can only work with with LPAE where set_huge_pte_at()
> is just a set_pte_at(). So this part looks good.
>
Hi, Catalin
Thanks for your time to review PATCH1/7 :)
I'm wondering if reset of this patch set is ok,
I would like to start working on V3 as per your comments.
--
I am a slow learner
but I will keep trying to fight for my dreams!
--bill
^ permalink raw reply [flat|nested] 29+ messages in thread* [PATCH 1/7] Add various hugetlb arm high level hooks
2012-03-08 0:35 ` bill4carson
@ 2012-03-08 9:21 ` Catalin Marinas
0 siblings, 0 replies; 29+ messages in thread
From: Catalin Marinas @ 2012-03-08 9:21 UTC (permalink / raw)
To: linux-arm-kernel
On Thu, Mar 08, 2012 at 12:35:36AM +0000, bill4carson wrote:
> On 2012?02?29? 23:38, Catalin Marinas wrote:
> > On Wed, Feb 29, 2012 at 11:28:30AM +0000, bill4carson wrote:
> >> On 2012?02?29? 18:32, Catalin Marinas wrote:
> >>> On Mon, Feb 13, 2012 at 09:44:22AM +0000, Bill Carson wrote:
> >>>> +static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
> >>>> + unsigned long addr, pte_t *ptep)
> >>>> +{
> >>>> + pte_t old_pte = *ptep;
> >>>> + set_huge_pte_at(mm, addr, ptep, pte_wrprotect(old_pte));
> >>>> +}
> >>>
> >>> You could use the generic ptep_set_wrprotect()
> >>
> >> I'm a bit of confused about this.
> >>
> >> generic ptep_set_wrprotect() can not set huge pte, that's why
> >> set_huge_pte_at is used instead here.
> >
> > Ah, the generic one can only work with with LPAE where set_huge_pte_at()
> > is just a set_pte_at(). So this part looks good.
>
> Thanks for your time to review PATCH1/7 :)
>
> I'm wondering if reset of this patch set is ok,
> I would like to start working on V3 as per your comments.
I haven't finished looking through the patches. Probably I will by the
end of this week.
--
Catalin
^ permalink raw reply [flat|nested] 29+ messages in thread
* [PATCH 1/7] Add various hugetlb arm high level hooks
2012-02-13 9:44 ` [PATCH 1/7] Add various hugetlb arm high level hooks Bill Carson
2012-02-28 17:30 ` Catalin Marinas
2012-02-29 10:32 ` Catalin Marinas
@ 2012-02-29 12:31 ` Arnd Bergmann
2012-02-29 14:22 ` Catalin Marinas
2 siblings, 1 reply; 29+ messages in thread
From: Arnd Bergmann @ 2012-02-29 12:31 UTC (permalink / raw)
To: linux-arm-kernel
On Monday 13 February 2012, Bill Carson wrote:
> +
> +/* 2M and 16M hugepage linux ptes are stored in mmu_context_t->huge_linux_pte
> + *
> + * 2M hugepage
> + * ===========
> + * one huge linux pte caters to two HW ptes,
> + *
I think this needs more explanation. Why do you use a 2MB hugepage with two ptes here?
Wouldn't it be more logical to use 1MB hugepages?
Arnd
^ permalink raw reply [flat|nested] 29+ messages in thread
* [PATCH 1/7] Add various hugetlb arm high level hooks
2012-02-29 12:31 ` Arnd Bergmann
@ 2012-02-29 14:22 ` Catalin Marinas
0 siblings, 0 replies; 29+ messages in thread
From: Catalin Marinas @ 2012-02-29 14:22 UTC (permalink / raw)
To: linux-arm-kernel
On Wed, Feb 29, 2012 at 12:31:51PM +0000, Arnd Bergmann wrote:
> On Monday 13 February 2012, Bill Carson wrote:
> > +
> > +/* 2M and 16M hugepage linux ptes are stored in mmu_context_t->huge_linux_pte
> > + *
> > + * 2M hugepage
> > + * ===========
> > + * one huge linux pte caters to two HW ptes,
> > + *
>
> I think this needs more explanation. Why do you use a 2MB hugepage with two ptes here?
> Wouldn't it be more logical to use 1MB hugepages?
That's because how pgd_t is defined for the classic MMU (PMD_SHIFT also
being 21). I think Bill explained in the first series of the patch why
using 1MB for huge pages doesn't go that well. But I agree, the
explanation should go in here as well.
--
Catalin
^ permalink raw reply [flat|nested] 29+ messages in thread
* [PATCH 2/7] Add various hugetlb page table fix
2012-02-13 9:44 [RFC-PATCH V2] ARM hugetlb support Bill Carson
2012-02-13 9:44 ` [PATCH 1/7] Add various hugetlb arm high level hooks Bill Carson
@ 2012-02-13 9:44 ` Bill Carson
2012-03-01 10:13 ` Catalin Marinas
2012-02-13 9:44 ` [PATCH 3/7] Introduce set_hugepte_ext api to setup huge hardware pmds Bill Carson
` (4 subsequent siblings)
6 siblings, 1 reply; 29+ messages in thread
From: Bill Carson @ 2012-02-13 9:44 UTC (permalink / raw)
To: linux-arm-kernel
Signed-off-by: Bill Carson <bill4carson@gmail.com>
---
arch/arm/include/asm/pgtable-2level.h | 3 +++
arch/arm/include/asm/pgtable.h | 9 +++++++++
2 files changed, 12 insertions(+), 0 deletions(-)
diff --git a/arch/arm/include/asm/pgtable-2level.h b/arch/arm/include/asm/pgtable-2level.h
index 2317a71..031c50e 100644
--- a/arch/arm/include/asm/pgtable-2level.h
+++ b/arch/arm/include/asm/pgtable-2level.h
@@ -178,6 +178,9 @@ static inline pmd_t *pmd_offset(pud_t *pud, unsigned long addr)
#define pmd_addr_end(addr,end) (end)
#define set_pte_ext(ptep,pte,ext) cpu_set_pte_ext(ptep,pte,ext)
+#ifdef CONFIG_ARM_HUGETLB_SUPPORT
+#define set_hugepte_ext(ptep,pte,ext) cpu_set_hugepte_ext(ptep,pte,ext)
+#endif
#endif /* __ASSEMBLY__ */
diff --git a/arch/arm/include/asm/pgtable.h b/arch/arm/include/asm/pgtable.h
index f66626d..a119ac6 100644
--- a/arch/arm/include/asm/pgtable.h
+++ b/arch/arm/include/asm/pgtable.h
@@ -213,6 +213,14 @@ static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
set_pte_ext(ptep, pteval, PTE_EXT_NG);
}
}
+#ifdef CONFIG_ARM_HUGETLB_SUPPORT
+static inline void set_hugepte_at(struct mm_struct *mm, unsigned long addr,
+ pte_t *ptep, pte_t pteval)
+{
+ __sync_icache_dcache(pteval);
+ set_hugepte_ext(ptep, pteval, PTE_EXT_NG);
+}
+#endif
#define pte_none(pte) (!pte_val(pte))
#define pte_present(pte) (pte_val(pte) & L_PTE_PRESENT)
@@ -236,6 +244,7 @@ PTE_BIT_FUNC(mkdirty, |= L_PTE_DIRTY);
PTE_BIT_FUNC(mkold, &= ~L_PTE_YOUNG);
PTE_BIT_FUNC(mkyoung, |= L_PTE_YOUNG);
+static inline pte_t pte_mkhuge(pte_t pte) { return pte; }
static inline pte_t pte_mkspecial(pte_t pte) { return pte; }
static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
--
1.7.1
^ permalink raw reply related [flat|nested] 29+ messages in thread* [PATCH 3/7] Introduce set_hugepte_ext api to setup huge hardware pmds
2012-02-13 9:44 [RFC-PATCH V2] ARM hugetlb support Bill Carson
2012-02-13 9:44 ` [PATCH 1/7] Add various hugetlb arm high level hooks Bill Carson
2012-02-13 9:44 ` [PATCH 2/7] Add various hugetlb page table fix Bill Carson
@ 2012-02-13 9:44 ` Bill Carson
2012-02-13 9:44 ` [PATCH 4/7] Store huge page linux pte in mmu_context_t Bill Carson
` (3 subsequent siblings)
6 siblings, 0 replies; 29+ messages in thread
From: Bill Carson @ 2012-02-13 9:44 UTC (permalink / raw)
To: linux-arm-kernel
Signed-off-by: Bill Carson <bill4carson@gmail.com>
---
arch/arm/include/asm/glue-proc.h | 3 +
arch/arm/include/asm/proc-fns.h | 3 +
arch/arm/mm/proc-v7-2level.S | 87 ++++++++++++++++++++++++++++++++++++++
3 files changed, 93 insertions(+), 0 deletions(-)
diff --git a/arch/arm/include/asm/glue-proc.h b/arch/arm/include/asm/glue-proc.h
index e2be7f1..2bbd452 100644
--- a/arch/arm/include/asm/glue-proc.h
+++ b/arch/arm/include/asm/glue-proc.h
@@ -256,6 +256,9 @@
#define cpu_dcache_clean_area __glue(CPU_NAME,_dcache_clean_area)
#define cpu_do_switch_mm __glue(CPU_NAME,_switch_mm)
#define cpu_set_pte_ext __glue(CPU_NAME,_set_pte_ext)
+#ifdef CONFIG_ARM_HUGETLB_SUPPORT
+#define cpu_set_hugepte_ext __glue(CPU_NAME,_set_hugepte_ext)
+#endif
#define cpu_suspend_size __glue(CPU_NAME,_suspend_size)
#define cpu_do_suspend __glue(CPU_NAME,_do_suspend)
#define cpu_do_resume __glue(CPU_NAME,_do_resume)
diff --git a/arch/arm/include/asm/proc-fns.h b/arch/arm/include/asm/proc-fns.h
index f3628fb..75bd755 100644
--- a/arch/arm/include/asm/proc-fns.h
+++ b/arch/arm/include/asm/proc-fns.h
@@ -87,6 +87,9 @@ extern void cpu_do_switch_mm(unsigned long pgd_phys, struct mm_struct *mm);
extern void cpu_set_pte_ext(pte_t *ptep, pte_t pte);
#else
extern void cpu_set_pte_ext(pte_t *ptep, pte_t pte, unsigned int ext);
+#ifdef CONFIG_ARM_HUGETLB_SUPPORT
+extern void cpu_set_hugepte_ext(pte_t *ptep, pte_t pte, unsigned int ext);
+#endif
#endif
extern void cpu_reset(unsigned long addr) __attribute__((noreturn));
diff --git a/arch/arm/mm/proc-v7-2level.S b/arch/arm/mm/proc-v7-2level.S
index 3a4b3e7..4dc1554 100644
--- a/arch/arm/mm/proc-v7-2level.S
+++ b/arch/arm/mm/proc-v7-2level.S
@@ -111,6 +111,93 @@ ENTRY(cpu_v7_set_pte_ext)
mov pc, lr
ENDPROC(cpu_v7_set_pte_ext)
+#ifdef CONFIG_ARM_HUGETLB_SUPPORT
+ENTRY(cpu_v7_set_hugepte_ext)
+ @mask out AP[2:0] TEX[2:0] in first level section descriptor
+ bic r3, r1, #0x0000fc00
+ @clear NX/IMP
+ bic r3, r3,#0x210
+
+ @clear BIT1:0
+ bic r3, r3, #PMD_TYPE_MASK
+
+ @set extension bit
+ orr r3, r3, #PMD_SECT_nG @HUGEPAGE always non-global
+
+ @set SECT mapping,1M section or 16M supersection
+ orr r3, r3, #PMD_SECT_AP_WRITE
+ orr r3, r3, #PMD_TYPE_SECT
+
+ @BIT18 1: 16M supersection 0: 1M section
+ bic r3, r3, #PMD_SECT_SUPER
+
+ @ shared bit
+ tst r1,#L_PTE_SHARED
+ orrne r3,r3,#PMD_SECT_S
+
+ @ shared device ?
+ tst r1, #1 << 4
+ orrne r3, r3, #PMD_SECT_TEX(1)
+
+ eor r1, r1, #L_PTE_DIRTY
+ tst r1, #L_PTE_RDONLY | L_PTE_DIRTY
+ orrne r3, r3, #PMD_SECT_APX
+
+ tst r1, #L_PTE_USER
+ orrne r3, r3, #PMD_SECT_AP_READ
+#ifdef CONFIG_CPU_USE_DOMAINS
+ tstne r3, #PMD_SECT_APX
+ bicne r3, r3, #PMD_SECT_APX | PMD_SECT_AP_WRITE
+#endif
+
+ @set domain
+ bic r3, r3, #(0xf << 5)
+ orr r3, r3, #PMD_DOMAIN(0x1)
+
+#ifdef CONFIG_HUGEPAGE_SIZE_16MB
+ @ for supersection mapping
+ @ clear domain setting and extend addr
+ @ set BIT18 to denote supersection
+ ldr r2, =((0xf << 5) | (0xf << 20))
+ bic r3, r3, r2
+ orr r3, r3, #PMD_SECT_SUPER
+#endif
+
+ tst r1, #L_PTE_XN
+ orrne r3, r3, #PMD_SECT_XN
+
+ tst r1, #L_PTE_YOUNG
+ tstne r1, #L_PTE_PRESENT
+ moveq r3, #0
+
+#ifdef CONFIG_HUGEPAGE_SIZE_16MB
+ @ set 16M huge page
+ .rept 16
+ str r3, [r0]
+ mcr p15, 0, r0, c7, c10, 1 @ flush_pte
+ add r0, r0, #4
+ .endr
+#endif
+
+#ifdef CONFIG_HUGEPAGE_SIZE_2MB
+ @ set 2M huge page
+ @ 1st 1MB mapping
+ str r3, [r0]
+ mcr p15, 0, r0, c7, c10, 1 @ flush_pte
+ @ 2st 1MB mapping
+ cmp r3,#0
+ movne r2,#0x100000
+ addne r3, r3, r2
+
+ add r0, r0, #4
+ str r3, [r0]
+ mcr p15, 0, r0, c7, c10, 1 @ flush_pte
+#endif
+ mov pc, lr
+#endif /*COFNIG_ARM_HUGETLB_SUPPORT*/
+
+ENDPROC(cpu_v7_set_hugepte_ext)
+
/*
* Memory region attributes with SCTLR.TRE=1
*
--
1.7.1
^ permalink raw reply related [flat|nested] 29+ messages in thread* [PATCH 4/7] Store huge page linux pte in mmu_context_t
2012-02-13 9:44 [RFC-PATCH V2] ARM hugetlb support Bill Carson
` (2 preceding siblings ...)
2012-02-13 9:44 ` [PATCH 3/7] Introduce set_hugepte_ext api to setup huge hardware pmds Bill Carson
@ 2012-02-13 9:44 ` Bill Carson
2012-02-13 9:44 ` [PATCH 5/7] Using do_page_fault for section fault handling Bill Carson
` (2 subsequent siblings)
6 siblings, 0 replies; 29+ messages in thread
From: Bill Carson @ 2012-02-13 9:44 UTC (permalink / raw)
To: linux-arm-kernel
Signed-off-by: Bill Carson <bill4carson@gmail.com>
---
arch/arm/include/asm/mmu.h | 4 ++++
arch/arm/mm/pgd.c | 11 +++++++++++
2 files changed, 15 insertions(+), 0 deletions(-)
diff --git a/arch/arm/include/asm/mmu.h b/arch/arm/include/asm/mmu.h
index 1496565..579bf82 100644
--- a/arch/arm/include/asm/mmu.h
+++ b/arch/arm/include/asm/mmu.h
@@ -9,6 +9,10 @@ typedef struct {
raw_spinlock_t id_lock;
#endif
unsigned int kvm_seq;
+#ifdef CONFIG_ARM_HUGETLB_SUPPORT
+ /* we place hugepage linux pte at mm_struct.context */
+ pte_t *huge_linux_pte;
+#endif
} mm_context_t;
#ifdef CONFIG_CPU_HAS_ASID
diff --git a/arch/arm/mm/pgd.c b/arch/arm/mm/pgd.c
index a3e78cc..895bb3f 100644
--- a/arch/arm/mm/pgd.c
+++ b/arch/arm/mm/pgd.c
@@ -91,6 +91,12 @@ pgd_t *pgd_alloc(struct mm_struct *mm)
pte_unmap(new_pte);
}
+#ifdef CONFIG_ARM_HUGETLB_SUPPORT
+ /* reset the hugepage linux pte pointer
+ * for new mm_struct when we do the fork
+ */
+ mm->context.huge_linux_pte = NULL;
+#endif
return new_pgd;
no_pte:
@@ -135,6 +141,11 @@ no_pud:
pgd_clear(pgd);
pud_free(mm, pud);
no_pgd:
+#ifdef CONFIG_ARM_HUGETLB_SUPPORT
+ /* free huge linux pte table */
+ if (mm->context.huge_linux_pte != NULL)
+ kfree(mm->context.huge_linux_pte);
+#endif
#ifdef CONFIG_ARM_LPAE
/*
* Free modules/pkmap or identity pmd tables.
--
1.7.1
^ permalink raw reply related [flat|nested] 29+ messages in thread* [PATCH 5/7] Using do_page_fault for section fault handling
2012-02-13 9:44 [RFC-PATCH V2] ARM hugetlb support Bill Carson
` (3 preceding siblings ...)
2012-02-13 9:44 ` [PATCH 4/7] Store huge page linux pte in mmu_context_t Bill Carson
@ 2012-02-13 9:44 ` Bill Carson
2012-02-13 9:44 ` [PATCH 6/7] Add hugetlb Kconfig option Bill Carson
2012-02-13 9:44 ` [PATCH 7/7] Minor compiling fix Bill Carson
6 siblings, 0 replies; 29+ messages in thread
From: Bill Carson @ 2012-02-13 9:44 UTC (permalink / raw)
To: linux-arm-kernel
Signed-off-by: Bill Carson <bill4carson@gmail.com>
---
arch/arm/mm/fault.c | 15 +++++++++++++++
1 files changed, 15 insertions(+), 0 deletions(-)
diff --git a/arch/arm/mm/fault.c b/arch/arm/mm/fault.c
index bb7eac3..af6703d 100644
--- a/arch/arm/mm/fault.c
+++ b/arch/arm/mm/fault.c
@@ -19,6 +19,7 @@
#include <linux/sched.h>
#include <linux/highmem.h>
#include <linux/perf_event.h>
+#include <linux/hugetlb.h>
#include <asm/exception.h>
#include <asm/system.h>
@@ -485,6 +486,7 @@ do_translation_fault(unsigned long addr, unsigned int fsr,
}
#endif /* CONFIG_MMU */
+#ifndef CONFIG_ARM_HUGETLB_SUPPORT
/*
* Some section permission faults need to be handled gracefully.
* They can happen due to a __{get,put}_user during an oops.
@@ -496,6 +498,19 @@ do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
return 0;
}
+#else
+
+/* Since normal 4K page based vma will never fault into section traps,
+ * This will enable us to use do_page_fault for section permission fault.
+ */
+static int
+do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
+{
+ do_page_fault(addr, fsr, regs);
+ return 0;
+}
+#endif
+
/*
* This abort handler always returns "fault".
*/
--
1.7.1
^ permalink raw reply related [flat|nested] 29+ messages in thread* [PATCH 6/7] Add hugetlb Kconfig option
2012-02-13 9:44 [RFC-PATCH V2] ARM hugetlb support Bill Carson
` (4 preceding siblings ...)
2012-02-13 9:44 ` [PATCH 5/7] Using do_page_fault for section fault handling Bill Carson
@ 2012-02-13 9:44 ` Bill Carson
2012-02-13 11:00 ` Sergei Shtylyov
2012-02-13 9:44 ` [PATCH 7/7] Minor compiling fix Bill Carson
6 siblings, 1 reply; 29+ messages in thread
From: Bill Carson @ 2012-02-13 9:44 UTC (permalink / raw)
To: linux-arm-kernel
Signed-off-by: Bill Carson <bill4carson@gmail.com>
---
arch/arm/Kconfig | 29 +++++++++++++++++++++++++++++
1 files changed, 29 insertions(+), 0 deletions(-)
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index a48aecc..161bca6 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -985,6 +985,35 @@ config ARCH_ZYNQ
Support for Xilinx Zynq ARM Cortex A9 Platform
endchoice
+config SYS_SUPPORTS_HUGETLBFS
+ def_bool n
+
+config ARM_HUGETLB_SUPPORT
+ bool "Support HUGETLB for ARMv7 (EXPERIMENTAL)"
+ depends on CPU_V7 && EXPERIMENTAL
+ select SYS_SUPPORTS_HUGETLBFS
+ select HUGETLBFS
+ default y
+
+choice
+ prompt "Huge Page Size"
+ depends on ARM_HUGETLB_SUPPORT
+ default HUGEPAGE_SIZE_2MB
+
+config HUGEPAGE_SIZE_2MB
+ bool "2MB"
+ depends on ARM_HUGETLB_SUPPORT
+ help
+ This option select huge page size in 2MB units
+
+config HUGEPAGE_SIZE_16MB
+ bool "16MB"
+ depends on ARM_HUGETLB_SUPPORT
+ help
+ This option select huge page size in 16MB units
+
+endchoice
+
#
# This is sorted alphabetically by mach-* pathname. However, plat-*
# Kconfigs may be included either alphabetically (according to the
--
1.7.1
^ permalink raw reply related [flat|nested] 29+ messages in thread* [PATCH 6/7] Add hugetlb Kconfig option
2012-02-13 9:44 ` [PATCH 6/7] Add hugetlb Kconfig option Bill Carson
@ 2012-02-13 11:00 ` Sergei Shtylyov
2012-02-15 2:50 ` bill4carson
0 siblings, 1 reply; 29+ messages in thread
From: Sergei Shtylyov @ 2012-02-13 11:00 UTC (permalink / raw)
To: linux-arm-kernel
Hello.
On 13-02-2012 13:44, Bill Carson wrote:
> Signed-off-by: Bill Carson<bill4carson@gmail.com>
> ---
> arch/arm/Kconfig | 29 +++++++++++++++++++++++++++++
> 1 files changed, 29 insertions(+), 0 deletions(-)
> diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
> index a48aecc..161bca6 100644
> --- a/arch/arm/Kconfig
> +++ b/arch/arm/Kconfig
> @@ -985,6 +985,35 @@ config ARCH_ZYNQ
> Support for Xilinx Zynq ARM Cortex A9 Platform
> endchoice
>
> +config SYS_SUPPORTS_HUGETLBFS
> + def_bool n
This is probably overindented.
WBR, Sergei
^ permalink raw reply [flat|nested] 29+ messages in thread* [PATCH 6/7] Add hugetlb Kconfig option
2012-02-13 11:00 ` Sergei Shtylyov
@ 2012-02-15 2:50 ` bill4carson
0 siblings, 0 replies; 29+ messages in thread
From: bill4carson @ 2012-02-15 2:50 UTC (permalink / raw)
To: linux-arm-kernel
On 2012?02?13? 19:00, Sergei Shtylyov wrote:
> Hello.
>
> On 13-02-2012 13:44, Bill Carson wrote:
>
>> Signed-off-by: Bill Carson<bill4carson@gmail.com>
>> ---
>> arch/arm/Kconfig | 29 +++++++++++++++++++++++++++++
>> 1 files changed, 29 insertions(+), 0 deletions(-)
>
>> diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
>> index a48aecc..161bca6 100644
>> --- a/arch/arm/Kconfig
>> +++ b/arch/arm/Kconfig
>> @@ -985,6 +985,35 @@ config ARCH_ZYNQ
>> Support for Xilinx Zynq ARM Cortex A9 Platform
>> endchoice
>>
>> +config SYS_SUPPORTS_HUGETLBFS
>> + def_bool n
>
> This is probably overindented.
>
Hi, Sergei
Thanks for your point :)
I'm sure there are other coding style issues like this, I'll pay
attention to it in the next review round.
> WBR, Sergei
>
--
I am a slow learner
but I will keep trying to fight for my dreams!
--bill
^ permalink raw reply [flat|nested] 29+ messages in thread
* [PATCH 7/7] Minor compiling fix
2012-02-13 9:44 [RFC-PATCH V2] ARM hugetlb support Bill Carson
` (5 preceding siblings ...)
2012-02-13 9:44 ` [PATCH 6/7] Add hugetlb Kconfig option Bill Carson
@ 2012-02-13 9:44 ` Bill Carson
2012-02-29 12:35 ` Arnd Bergmann
6 siblings, 1 reply; 29+ messages in thread
From: Bill Carson @ 2012-02-13 9:44 UTC (permalink / raw)
To: linux-arm-kernel
Signed-off-by: Bill Carson <bill4carson@gmail.com>
---
arch/arm/mm/Makefile | 1 +
arch/arm/mm/dma-mapping.c | 3 ---
2 files changed, 1 insertions(+), 3 deletions(-)
diff --git a/arch/arm/mm/Makefile b/arch/arm/mm/Makefile
index bca7e61..9348e1e 100644
--- a/arch/arm/mm/Makefile
+++ b/arch/arm/mm/Makefile
@@ -99,4 +99,5 @@ AFLAGS_proc-v7.o :=-Wa,-march=armv7-a
obj-$(CONFIG_CACHE_FEROCEON_L2) += cache-feroceon-l2.o
obj-$(CONFIG_CACHE_L2X0) += cache-l2x0.o
obj-$(CONFIG_CACHE_XSC3L2) += cache-xsc3l2.o
+obj-$(CONFIG_HUGETLB_PAGE) += hugetlb.o
obj-$(CONFIG_CACHE_TAUROS2) += cache-tauros2.o
diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
index 1aa664a..8dc5fb4 100644
--- a/arch/arm/mm/dma-mapping.c
+++ b/arch/arm/mm/dma-mapping.c
@@ -152,9 +152,6 @@ static struct arm_vmregion_head consistent_head = {
.vm_end = CONSISTENT_END,
};
-#ifdef CONFIG_HUGETLB_PAGE
-#error ARM Coherent DMA allocator does not (yet) support huge TLB
-#endif
/*
* Initialise the consistent memory allocation.
--
1.7.1
^ permalink raw reply related [flat|nested] 29+ messages in thread* [PATCH 7/7] Minor compiling fix
2012-02-13 9:44 ` [PATCH 7/7] Minor compiling fix Bill Carson
@ 2012-02-29 12:35 ` Arnd Bergmann
2012-03-01 8:44 ` bill4carson
0 siblings, 1 reply; 29+ messages in thread
From: Arnd Bergmann @ 2012-02-29 12:35 UTC (permalink / raw)
To: linux-arm-kernel
On Monday 13 February 2012, Bill Carson wrote:
> --- a/arch/arm/mm/dma-mapping.c
> +++ b/arch/arm/mm/dma-mapping.c
> @@ -152,9 +152,6 @@ static struct arm_vmregion_head consistent_head = {
> .vm_end = CONSISTENT_END,
> };
>
> -#ifdef CONFIG_HUGETLB_PAGE
> -#error ARM Coherent DMA allocator does not (yet) support huge TLB
> -#endif
This one needs an explanation, too. What has changed that now allows
dma_alloc_coherent to work with CONFIG_HUGETLB_PAGE set that did not
work before?
In general, please add a description with each patch you send, following
the instructions in Documentation/SubmittingPatches.
Arnd
^ permalink raw reply [flat|nested] 29+ messages in thread* [PATCH 7/7] Minor compiling fix
2012-02-29 12:35 ` Arnd Bergmann
@ 2012-03-01 8:44 ` bill4carson
2012-03-01 11:40 ` Arnd Bergmann
0 siblings, 1 reply; 29+ messages in thread
From: bill4carson @ 2012-03-01 8:44 UTC (permalink / raw)
To: linux-arm-kernel
On 2012?02?29? 20:35, Arnd Bergmann wrote:
> On Monday 13 February 2012, Bill Carson wrote:
>> --- a/arch/arm/mm/dma-mapping.c
>> +++ b/arch/arm/mm/dma-mapping.c
>> @@ -152,9 +152,6 @@ static struct arm_vmregion_head consistent_head = {
>> .vm_end = CONSISTENT_END,
>> };
>>
>> -#ifdef CONFIG_HUGETLB_PAGE
>> -#error ARM Coherent DMA allocator does not (yet) support huge TLB
>> -#endif
>
> This one needs an explanation, too. What has changed that now allows
> dma_alloc_coherent to work with CONFIG_HUGETLB_PAGE set that did not
> work before?
>
AFAIK, __dma_alloc has the finest granularity at 4k page size even when
user requests 1M bytes memory.
but current hugetlb low level set_hugepte_at *ONLY*/*MUST* support 2M
page mapping, so it can't be used to create huge dma mapping in a more
general way.
comments this error out doesn't mean that it could support huge page dma
mapping now with this patch, just shut up gcc when compiling.
Adding a single low level hook to create 1M page mapping can satisfy
the needs, but apparently that's out of hugetlb scope.
And it's on my TODO list :)
> In general, please add a description with each patch you send, following
> the instructions in Documentation/SubmittingPatches.
>
Thanks, I will follow the rules.
> Arnd
>
--
I am a slow learner
but I will keep trying to fight for my dreams!
--bill
^ permalink raw reply [flat|nested] 29+ messages in thread* [PATCH 7/7] Minor compiling fix
2012-03-01 8:44 ` bill4carson
@ 2012-03-01 11:40 ` Arnd Bergmann
2012-03-01 12:42 ` carson bill
0 siblings, 1 reply; 29+ messages in thread
From: Arnd Bergmann @ 2012-03-01 11:40 UTC (permalink / raw)
To: linux-arm-kernel
On Thursday 01 March 2012, bill4carson wrote:
> On 2012?02?29? 20:35, Arnd Bergmann wrote:
> > On Monday 13 February 2012, Bill Carson wrote:
> >> --- a/arch/arm/mm/dma-mapping.c
> >> +++ b/arch/arm/mm/dma-mapping.c
> >> @@ -152,9 +152,6 @@ static struct arm_vmregion_head consistent_head = {
> >> .vm_end = CONSISTENT_END,
> >> };
> >>
> >> -#ifdef CONFIG_HUGETLB_PAGE
> >> -#error ARM Coherent DMA allocator does not (yet) support huge TLB
> >> -#endif
> >
> > This one needs an explanation, too. What has changed that now allows
> > dma_alloc_coherent to work with CONFIG_HUGETLB_PAGE set that did not
> > work before?
> >
>
> AFAIK, __dma_alloc has the finest granularity at 4k page size even when
> user requests 1M bytes memory.
>
> but current hugetlb low level set_hugepte_at *ONLY*/*MUST* support 2M
> page mapping, so it can't be used to create huge dma mapping in a more
> general way.
>
> comments this error out doesn't mean that it could support huge page dma
> mapping now with this patch, just shut up gcc when compiling.
>
> Adding a single low level hook to create 1M page mapping can satisfy
> the needs, but apparently that's out of hugetlb scope.
> And it's on my TODO list :)
Hmm, I'm still not sure I'm following. The coherent DMA area is currently
limited to 2MB or a small multiple of that, so you would not be able
to allocate coherent hugepages anyway.
Further, I don't think there are any device drivers that call
dma_alloc_coherent and expect to map the resulting page into
user space, or that we even have an interface for that.
This entire chunk of code is also getting changed with the addition of CMA.
Arnd
^ permalink raw reply [flat|nested] 29+ messages in thread* [PATCH 7/7] Minor compiling fix
2012-03-01 11:40 ` Arnd Bergmann
@ 2012-03-01 12:42 ` carson bill
0 siblings, 0 replies; 29+ messages in thread
From: carson bill @ 2012-03-01 12:42 UTC (permalink / raw)
To: linux-arm-kernel
2012/3/1 Arnd Bergmann <arnd@arndb.de>
>
> On Thursday 01 March 2012, bill4carson wrote:
> > On 2012?02?29? 20:35, Arnd Bergmann wrote:
> > > On Monday 13 February 2012, Bill Carson wrote:
> > >> --- a/arch/arm/mm/dma-mapping.c
> > >> +++ b/arch/arm/mm/dma-mapping.c
> > >> @@ -152,9 +152,6 @@ static struct arm_vmregion_head consistent_head = {
> > >> .vm_end = CONSISTENT_END,
> > >> };
> > >>
> > >> -#ifdef CONFIG_HUGETLB_PAGE
> > >> -#error ARM Coherent DMA allocator does not (yet) support huge TLB
> > >> -#endif
> > >
> > > This one needs an explanation, too. What has changed that now allows
> > > dma_alloc_coherent to work with CONFIG_HUGETLB_PAGE set that did not
> > > work before?
> > >
> >
> > AFAIK, __dma_alloc has the finest granularity at 4k page size even when
> > user requests 1M bytes memory.
> >
> > but current hugetlb low level set_hugepte_at *ONLY*/*MUST* support 2M
> > page mapping, so it can't be used to create huge dma mapping in a more
> > general way.
> >
> > comments this error out doesn't mean that it could support huge page dma
> > mapping now with this patch, just shut up gcc when compiling.
> >
> > Adding a single low level hook to create 1M page mapping can satisfy
> > the needs, but apparently that's out of hugetlb scope.
> > And it's on my TODO list :)
>
> Hmm, I'm still not sure I'm following. The coherent DMA area is currently
> limited to 2MB or a small multiple of that, so you would not be able
> to allocate coherent hugepages anyway.
>
>
> Further, I don't think there are any device drivers that call
> dma_alloc_coherent and expect to map the resulting page into
> user space, or that we even have an interface for that.
>
This "ifdef" is used to remind dma implementation should use hugetlb low
API to set up huge page mappings one day hugetlb is supported, just it!
For now, I don't see any direct relationship dma alloc has to do with hugetlb
in a more general way other than both could share low level page table
management.
> This entire chunk of code is also getting changed with the addition of CMA.
>
> Arnd
^ permalink raw reply [flat|nested] 29+ messages in thread