2020-05-23 16:40:06

by Souptick Joarder

[permalink] [raw]
Subject: [linux-next RFC] mm/gup.c: Convert to use get_user_pages_fast_only()

Renaming the API __get_user_pages_fast() to get_user_pages_
fast_only() to align with pin_user_pages_fast_only().

As part of this we will get rid of write parameter.
Instead caller will pass FOLL_WRITE to get_user_pages_fast_only().
This will not change any existing functionality of the API.

All the callers are changed to pass FOLL_WRITE.

Updated the documentation of the API.

Signed-off-by: Souptick Joarder <[email protected]>
Cc: John Hubbard <[email protected]>
Cc: Matthew Wilcox <[email protected]>
---
arch/powerpc/kvm/book3s_64_mmu_hv.c | 2 +-
arch/powerpc/kvm/book3s_64_mmu_radix.c | 2 +-
arch/powerpc/perf/callchain_64.c | 2 +-
include/linux/mm.h | 4 ++--
kernel/events/core.c | 4 ++--
mm/gup.c | 29 ++++++++++++++++-------------
virt/kvm/kvm_main.c | 6 +++---
7 files changed, 26 insertions(+), 23 deletions(-)

diff --git a/arch/powerpc/kvm/book3s_64_mmu_hv.c b/arch/powerpc/kvm/book3s_64_mmu_hv.c
index 18aed97..34fc5c8 100644
--- a/arch/powerpc/kvm/book3s_64_mmu_hv.c
+++ b/arch/powerpc/kvm/book3s_64_mmu_hv.c
@@ -581,7 +581,7 @@ int kvmppc_book3s_hv_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu,
* We always ask for write permission since the common case
* is that the page is writable.
*/
- if (__get_user_pages_fast(hva, 1, 1, &page) == 1) {
+ if (get_user_pages_fast_only(hva, 1, FOLL_WRITE, &page) == 1) {
write_ok = true;
} else {
/* Call KVM generic code to do the slow-path check */
diff --git a/arch/powerpc/kvm/book3s_64_mmu_radix.c b/arch/powerpc/kvm/book3s_64_mmu_radix.c
index 3248f78..3b6e342 100644
--- a/arch/powerpc/kvm/book3s_64_mmu_radix.c
+++ b/arch/powerpc/kvm/book3s_64_mmu_radix.c
@@ -795,7 +795,7 @@ int kvmppc_book3s_instantiate_page(struct kvm_vcpu *vcpu,
* is that the page is writable.
*/
hva = gfn_to_hva_memslot(memslot, gfn);
- if (!kvm_ro && __get_user_pages_fast(hva, 1, 1, &page) == 1) {
+ if (!kvm_ro && get_user_pages_fast_only(hva, 1, FOLL_WRITE, &page) == 1) {
upgrade_write = true;
} else {
unsigned long pfn;
diff --git a/arch/powerpc/perf/callchain_64.c b/arch/powerpc/perf/callchain_64.c
index 1bff896d..f719a74 100644
--- a/arch/powerpc/perf/callchain_64.c
+++ b/arch/powerpc/perf/callchain_64.c
@@ -32,7 +32,7 @@ int read_user_stack_slow(void __user *ptr, void *buf, int nb)
int nrpages;
void *kaddr;

- nrpages = __get_user_pages_fast(addr, 1, 1, &page);
+ nrpages = get_user_pages_fast_only(addr, 1, FOLL_WRITE, &page);
if (nrpages == 1) {
kaddr = page_address(page);

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 93d93bd..10a6758 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1817,8 +1817,8 @@ extern int mprotect_fixup(struct vm_area_struct *vma,
/*
* doesn't attempt to fault and will return short.
*/
-int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
- struct page **pages);
+int get_user_pages_fast_only(unsigned long start, int nr_pages,
+ unsigned int gup_flags, struct page **pages);
int pin_user_pages_fast_only(unsigned long start, int nr_pages,
unsigned int gup_flags, struct page **pages);
/*
diff --git a/kernel/events/core.c b/kernel/events/core.c
index c94eb27..81d6e73 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -6934,12 +6934,12 @@ static u64 perf_virt_to_phys(u64 virt)
* Walking the pages tables for user address.
* Interrupts are disabled, so it prevents any tear down
* of the page tables.
- * Try IRQ-safe __get_user_pages_fast first.
+ * Try IRQ-safe get_user_pages_fast_only first.
* If failed, leave phys_addr as 0.
*/
if (current->mm != NULL) {
pagefault_disable();
- if (__get_user_pages_fast(virt, 1, 0, &p) == 1)
+ if (get_user_pages_fast_only(virt, 1, 0, &p) == 1)
phys_addr = page_to_phys(p) + virt % PAGE_SIZE;
pagefault_enable();
}
diff --git a/mm/gup.c b/mm/gup.c
index 80f51a36..d8aabc0 100644
--- a/mm/gup.c
+++ b/mm/gup.c
@@ -2278,7 +2278,7 @@ static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
* to be special.
*
* For a futex to be placed on a THP tail page, get_futex_key requires a
- * __get_user_pages_fast implementation that can pin pages. Thus it's still
+ * get_user_pages_fast_only implementation that can pin pages. Thus it's still
* useful to have gup_huge_pmd even if we can't operate on ptes.
*/
static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
@@ -2683,7 +2683,7 @@ static inline void gup_pgd_range(unsigned long addr, unsigned long end,

#ifndef gup_fast_permitted
/*
- * Check if it's allowed to use __get_user_pages_fast() for the range, or
+ * Check if it's allowed to use get_user_pages_fast_only() for the range, or
* we need to fall back to the slow version:
*/
static bool gup_fast_permitted(unsigned long start, unsigned long end)
@@ -2776,8 +2776,14 @@ static int internal_get_user_pages_fast(unsigned long start, int nr_pages,

return ret;
}
-
-/*
+/**
+ * get_user_pages_fast_only() - pin user pages in memory
+ * @start: starting user address
+ * @nr_pages: number of pages from start to pin
+ * @gup_flags: flags modifying pin behaviour
+ * @pages: array that receives pointers to the pages pinned.
+ * Should be at least nr_pages long.
+ *
* Like get_user_pages_fast() except it's IRQ-safe in that it won't fall back to
* the regular GUP.
* Note a difference with get_user_pages_fast: this always returns the
@@ -2786,8 +2792,8 @@ static int internal_get_user_pages_fast(unsigned long start, int nr_pages,
* If the architecture does not support this function, simply return with no
* pages pinned.
*/
-int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
- struct page **pages)
+int get_user_pages_fast_only(unsigned long start, int nr_pages,
+ unsigned int gup_flags, struct page **pages)
{
int nr_pinned;
/*
@@ -2797,10 +2803,7 @@ int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
* FOLL_FAST_ONLY is required in order to match the API description of
* this routine: no fall back to regular ("slow") GUP.
*/
- unsigned int gup_flags = FOLL_GET | FOLL_FAST_ONLY;
-
- if (write)
- gup_flags |= FOLL_WRITE;
+ gup_flags = FOLL_GET | FOLL_FAST_ONLY;

nr_pinned = internal_get_user_pages_fast(start, nr_pages, gup_flags,
pages);
@@ -2815,7 +2818,7 @@ int __get_user_pages_fast(unsigned long start, int nr_pages, int write,

return nr_pinned;
}
-EXPORT_SYMBOL_GPL(__get_user_pages_fast);
+EXPORT_SYMBOL_GPL(get_user_pages_fast_only);

/**
* get_user_pages_fast() - pin user pages in memory
@@ -2886,8 +2889,8 @@ int pin_user_pages_fast(unsigned long start, int nr_pages,
EXPORT_SYMBOL_GPL(pin_user_pages_fast);

/*
- * This is the FOLL_PIN equivalent of __get_user_pages_fast(). Behavior is the
- * same, except that this one sets FOLL_PIN instead of FOLL_GET.
+ * This is the FOLL_PIN equivalent of get_user_pages_fast_only(). Behavior
+ * is the same, except that this one sets FOLL_PIN instead of FOLL_GET.
*
* The API rules are the same, too: no negative values may be returned.
*/
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index fc38d63..cec7919 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -1750,7 +1750,7 @@ static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
if (!(write_fault || writable))
return false;

- npages = __get_user_pages_fast(addr, 1, 1, page);
+ npages = get_user_pages_fast_only(addr, 1, FOLL_WRITE, page);
if (npages == 1) {
*pfn = page_to_pfn(page[0]);

@@ -1791,7 +1791,7 @@ static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
if (unlikely(!write_fault) && writable) {
struct page *wpage;

- if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
+ if (get_user_pages_fast_only(addr, 1, FOLL_WRITE, &wpage) == 1) {
*writable = true;
put_page(page);
page = wpage;
@@ -1998,7 +1998,7 @@ int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
if (entry < nr_pages)
return 0;

- return __get_user_pages_fast(addr, nr_pages, 1, pages);
+ return get_user_pages_fast(addr, nr_pages, FOLL_WRITE, pages);
}
EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);

--
1.9.1


2020-05-23 17:28:24

by Matthew Wilcox

[permalink] [raw]
Subject: Re: [linux-next RFC] mm/gup.c: Convert to use get_user_pages_fast_only()

On Sat, May 23, 2020 at 10:11:12PM +0530, Souptick Joarder wrote:
> Renaming the API __get_user_pages_fast() to get_user_pages_
> fast_only() to align with pin_user_pages_fast_only().

Please don't split a function name across lines. That messes
up people who are grepping for the function name in the changelog.

> As part of this we will get rid of write parameter.
> Instead caller will pass FOLL_WRITE to get_user_pages_fast_only().
> This will not change any existing functionality of the API.
>
> All the callers are changed to pass FOLL_WRITE.
>
> Updated the documentation of the API.

Everything you have done here is an improvement, and I'd be happy to
see it go in (after fixing the bug I note below).

But in reading through it, I noticed almost every user ...

> - if (__get_user_pages_fast(hva, 1, 1, &page) == 1) {
> + if (get_user_pages_fast_only(hva, 1, FOLL_WRITE, &page) == 1) {

passes '1' as the second parameter. So do we want to add:

static inline bool get_user_page_fast_only(unsigned long addr,
unsigned int gup_flags, struct page **pagep)
{
return get_user_pages_fast_only(addr, 1, gup_flags, pagep) == 1;
}

> @@ -2797,10 +2803,7 @@ int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
> * FOLL_FAST_ONLY is required in order to match the API description of
> * this routine: no fall back to regular ("slow") GUP.
> */
> - unsigned int gup_flags = FOLL_GET | FOLL_FAST_ONLY;
> -
> - if (write)
> - gup_flags |= FOLL_WRITE;
> + gup_flags = FOLL_GET | FOLL_FAST_ONLY;

Er ... gup_flags |=, surely?

2020-05-23 19:31:57

by Souptick Joarder

[permalink] [raw]
Subject: Re: [linux-next RFC] mm/gup.c: Convert to use get_user_pages_fast_only()

On Sat, May 23, 2020 at 10:55 PM Matthew Wilcox <[email protected]> wrote:
>
> On Sat, May 23, 2020 at 10:11:12PM +0530, Souptick Joarder wrote:
> > Renaming the API __get_user_pages_fast() to get_user_pages_
> > fast_only() to align with pin_user_pages_fast_only().
>
> Please don't split a function name across lines. That messes
> up people who are grepping for the function name in the changelog.

Ok.

>
> > As part of this we will get rid of write parameter.
> > Instead caller will pass FOLL_WRITE to get_user_pages_fast_only().
> > This will not change any existing functionality of the API.
> >
> > All the callers are changed to pass FOLL_WRITE.
> >
> > Updated the documentation of the API.
>
> Everything you have done here is an improvement, and I'd be happy to
> see it go in (after fixing the bug I note below).
>
> But in reading through it, I noticed almost every user ...
>
> > - if (__get_user_pages_fast(hva, 1, 1, &page) == 1) {
> > + if (get_user_pages_fast_only(hva, 1, FOLL_WRITE, &page) == 1) {
>
> passes '1' as the second parameter. So do we want to add:
>
> static inline bool get_user_page_fast_only(unsigned long addr,
> unsigned int gup_flags, struct page **pagep)
> {
> return get_user_pages_fast_only(addr, 1, gup_flags, pagep) == 1;
> }
>
Yes, this can be added. Does get_user_page_fast_only() naming is fine ?


> > @@ -2797,10 +2803,7 @@ int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
> > * FOLL_FAST_ONLY is required in order to match the API description of
> > * this routine: no fall back to regular ("slow") GUP.
> > */
> > - unsigned int gup_flags = FOLL_GET | FOLL_FAST_ONLY;
> > -
> > - if (write)
> > - gup_flags |= FOLL_WRITE;
> > + gup_flags = FOLL_GET | FOLL_FAST_ONLY;
>
> Er ... gup_flags |=, surely?

Poor mistake.


@@ -1998,7 +1998,7 @@ int gfn_to_page_many_atomic(struct
kvm_memory_slot *slot, gfn_t gfn,
if (entry < nr_pages)
return 0;

- return __get_user_pages_fast(addr, nr_pages, 1, pages);
+ return get_user_pages_fast(addr, nr_pages, FOLL_WRITE, pages);

Also this needs to be corrected.