From: sheng.zhao@bytedance.com
To: mst@redhat.com, jasowang@redhat.com, xuanzhuo@linux.alibaba.com,
eperezma@redhat.com
Cc: virtualization@lists.linux.dev, linux-kernel@vger.kernel.org,
xieyongji@bytedance.com, Sheng Zhao <sheng.zhao@bytedance.com>
Subject: [PATCH] vduse: Use fixed 4KB bounce pages for arm64 64KB page size
Date: Mon, 15 Sep 2025 15:34:29 +0800 [thread overview]
Message-ID: <20250915073429.54027-1-sheng.zhao@bytedance.com> (raw)
From: Sheng Zhao <sheng.zhao@bytedance.com>
The allocation granularity of bounce pages is PAGE_SIZE. This may cause
even small IO requests to occupy an entire bounce page exclusively. The
kind of memory waste will be more significant on arm64 with 64KB pages.
So, optimize it by using fixed 4KB bounce pages.
Signed-off-by: Sheng Zhao <sheng.zhao@bytedance.com>
---
drivers/vdpa/vdpa_user/iova_domain.c | 120 +++++++++++++++++----------
drivers/vdpa/vdpa_user/iova_domain.h | 5 ++
2 files changed, 83 insertions(+), 42 deletions(-)
diff --git a/drivers/vdpa/vdpa_user/iova_domain.c b/drivers/vdpa/vdpa_user/iova_domain.c
index 58116f89d8da..768313c80b62 100644
--- a/drivers/vdpa/vdpa_user/iova_domain.c
+++ b/drivers/vdpa/vdpa_user/iova_domain.c
@@ -103,19 +103,26 @@ void vduse_domain_clear_map(struct vduse_iova_domain *domain,
static int vduse_domain_map_bounce_page(struct vduse_iova_domain *domain,
u64 iova, u64 size, u64 paddr)
{
- struct vduse_bounce_map *map;
+ struct vduse_bounce_map *map, *head_map;
+ struct page *tmp_page;
u64 last = iova + size - 1;
while (iova <= last) {
- map = &domain->bounce_maps[iova >> PAGE_SHIFT];
+ map = &domain->bounce_maps[iova >> BOUNCE_PAGE_SHIFT];
if (!map->bounce_page) {
- map->bounce_page = alloc_page(GFP_ATOMIC);
- if (!map->bounce_page)
- return -ENOMEM;
+ head_map = &domain->bounce_maps[(iova & PAGE_MASK) >> BOUNCE_PAGE_SHIFT];
+ if (!head_map->bounce_page) {
+ tmp_page = alloc_page(GFP_ATOMIC);
+ if (!tmp_page)
+ return -ENOMEM;
+ if (cmpxchg(&head_map->bounce_page, NULL, tmp_page))
+ __free_page(tmp_page);
+ }
+ map->bounce_page = head_map->bounce_page;
}
map->orig_phys = paddr;
- paddr += PAGE_SIZE;
- iova += PAGE_SIZE;
+ paddr += BOUNCE_PAGE_SIZE;
+ iova += BOUNCE_PAGE_SIZE;
}
return 0;
}
@@ -127,12 +134,17 @@ static void vduse_domain_unmap_bounce_page(struct vduse_iova_domain *domain,
u64 last = iova + size - 1;
while (iova <= last) {
- map = &domain->bounce_maps[iova >> PAGE_SHIFT];
+ map = &domain->bounce_maps[iova >> BOUNCE_PAGE_SHIFT];
map->orig_phys = INVALID_PHYS_ADDR;
- iova += PAGE_SIZE;
+ iova += BOUNCE_PAGE_SIZE;
}
}
+static unsigned int offset_in_bounce_page(dma_addr_t addr)
+{
+ return (addr & ~BOUNCE_PAGE_MASK);
+}
+
static void do_bounce(phys_addr_t orig, void *addr, size_t size,
enum dma_data_direction dir)
{
@@ -163,7 +175,7 @@ static void vduse_domain_bounce(struct vduse_iova_domain *domain,
{
struct vduse_bounce_map *map;
struct page *page;
- unsigned int offset;
+ unsigned int offset, head_offset;
void *addr;
size_t sz;
@@ -171,9 +183,10 @@ static void vduse_domain_bounce(struct vduse_iova_domain *domain,
return;
while (size) {
- map = &domain->bounce_maps[iova >> PAGE_SHIFT];
- offset = offset_in_page(iova);
- sz = min_t(size_t, PAGE_SIZE - offset, size);
+ map = &domain->bounce_maps[iova >> BOUNCE_PAGE_SHIFT];
+ head_offset = offset_in_page(iova);
+ offset = offset_in_bounce_page(iova);
+ sz = min_t(size_t, BOUNCE_PAGE_SIZE - offset, size);
if (WARN_ON(!map->bounce_page ||
map->orig_phys == INVALID_PHYS_ADDR))
@@ -183,7 +196,7 @@ static void vduse_domain_bounce(struct vduse_iova_domain *domain,
map->user_bounce_page : map->bounce_page;
addr = kmap_local_page(page);
- do_bounce(map->orig_phys + offset, addr + offset, sz, dir);
+ do_bounce(map->orig_phys + offset, addr + head_offset, sz, dir);
kunmap_local(addr);
size -= sz;
iova += sz;
@@ -218,7 +231,7 @@ vduse_domain_get_bounce_page(struct vduse_iova_domain *domain, u64 iova)
struct page *page = NULL;
read_lock(&domain->bounce_lock);
- map = &domain->bounce_maps[iova >> PAGE_SHIFT];
+ map = &domain->bounce_maps[iova >> BOUNCE_PAGE_SHIFT];
if (domain->user_bounce_pages || !map->bounce_page)
goto out;
@@ -236,7 +249,7 @@ vduse_domain_free_kernel_bounce_pages(struct vduse_iova_domain *domain)
struct vduse_bounce_map *map;
unsigned long pfn, bounce_pfns;
- bounce_pfns = domain->bounce_size >> PAGE_SHIFT;
+ bounce_pfns = domain->bounce_size >> BOUNCE_PAGE_SHIFT;
for (pfn = 0; pfn < bounce_pfns; pfn++) {
map = &domain->bounce_maps[pfn];
@@ -246,7 +259,8 @@ vduse_domain_free_kernel_bounce_pages(struct vduse_iova_domain *domain)
if (!map->bounce_page)
continue;
- __free_page(map->bounce_page);
+ if (!((pfn << BOUNCE_PAGE_SHIFT) & ~PAGE_MASK))
+ __free_page(map->bounce_page);
map->bounce_page = NULL;
}
}
@@ -254,8 +268,12 @@ vduse_domain_free_kernel_bounce_pages(struct vduse_iova_domain *domain)
int vduse_domain_add_user_bounce_pages(struct vduse_iova_domain *domain,
struct page **pages, int count)
{
- struct vduse_bounce_map *map;
- int i, ret;
+ struct vduse_bounce_map *map, *head_map;
+ int i, j, ret;
+ int inner_pages = PAGE_SIZE / BOUNCE_PAGE_SIZE;
+ int bounce_pfns = domain->bounce_size >> BOUNCE_PAGE_SHIFT;
+ struct page *head_page = NULL;
+ bool need_copy;
/* Now we don't support partial mapping */
if (count != (domain->bounce_size >> PAGE_SHIFT))
@@ -267,16 +285,23 @@ int vduse_domain_add_user_bounce_pages(struct vduse_iova_domain *domain,
goto out;
for (i = 0; i < count; i++) {
- map = &domain->bounce_maps[i];
- if (map->bounce_page) {
+ need_copy = false;
+ head_map = &domain->bounce_maps[(i * inner_pages)];
+ head_page = head_map->bounce_page;
+ for (j = 0; j < inner_pages; j++) {
+ if ((i * inner_pages + j) >= bounce_pfns)
+ break;
+ map = &domain->bounce_maps[(i * inner_pages + j)];
/* Copy kernel page to user page if it's in use */
- if (map->orig_phys != INVALID_PHYS_ADDR)
- memcpy_to_page(pages[i], 0,
- page_address(map->bounce_page),
- PAGE_SIZE);
+ if ((head_page) && (map->orig_phys != INVALID_PHYS_ADDR))
+ need_copy = true;
+ map->user_bounce_page = pages[i];
}
- map->user_bounce_page = pages[i];
get_page(pages[i]);
+ if ((head_page) && (need_copy))
+ memcpy_to_page(pages[i], 0,
+ page_address(head_page),
+ PAGE_SIZE);
}
domain->user_bounce_pages = true;
ret = 0;
@@ -288,8 +313,12 @@ int vduse_domain_add_user_bounce_pages(struct vduse_iova_domain *domain,
void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain)
{
- struct vduse_bounce_map *map;
- unsigned long i, count;
+ struct vduse_bounce_map *map, *head_map;
+ unsigned long i, j, count;
+ int inner_pages = PAGE_SIZE / BOUNCE_PAGE_SIZE;
+ int bounce_pfns = domain->bounce_size >> BOUNCE_PAGE_SHIFT;
+ struct page *head_page = NULL;
+ bool need_copy;
write_lock(&domain->bounce_lock);
if (!domain->user_bounce_pages)
@@ -297,20 +326,27 @@ void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain)
count = domain->bounce_size >> PAGE_SHIFT;
for (i = 0; i < count; i++) {
- struct page *page = NULL;
-
- map = &domain->bounce_maps[i];
- if (WARN_ON(!map->user_bounce_page))
+ need_copy = false;
+ head_map = &domain->bounce_maps[(i * inner_pages)];
+ if (WARN_ON(!head_map->user_bounce_page))
continue;
-
- /* Copy user page to kernel page if it's in use */
- if (map->orig_phys != INVALID_PHYS_ADDR) {
- page = map->bounce_page;
- memcpy_from_page(page_address(page),
- map->user_bounce_page, 0, PAGE_SIZE);
+ head_page = head_map->user_bounce_page;
+
+ for (j = 0; j < inner_pages; j++) {
+ if ((i * inner_pages + j) >= bounce_pfns)
+ break;
+ map = &domain->bounce_maps[(i * inner_pages + j)];
+ if (WARN_ON(!map->user_bounce_page))
+ continue;
+ /* Copy user page to kernel page if it's in use */
+ if ((map->orig_phys != INVALID_PHYS_ADDR) && (head_map->bounce_page))
+ need_copy = true;
+ map->user_bounce_page = NULL;
}
- put_page(map->user_bounce_page);
- map->user_bounce_page = NULL;
+ if (need_copy)
+ memcpy_from_page(page_address(head_map->bounce_page),
+ head_page, 0, PAGE_SIZE);
+ put_page(head_page);
}
domain->user_bounce_pages = false;
out:
@@ -581,7 +617,7 @@ vduse_domain_create(unsigned long iova_limit, size_t bounce_size)
unsigned long pfn, bounce_pfns;
int ret;
- bounce_pfns = PAGE_ALIGN(bounce_size) >> PAGE_SHIFT;
+ bounce_pfns = PAGE_ALIGN(bounce_size) >> BOUNCE_PAGE_SHIFT;
if (iova_limit <= bounce_size)
return NULL;
@@ -613,7 +649,7 @@ vduse_domain_create(unsigned long iova_limit, size_t bounce_size)
rwlock_init(&domain->bounce_lock);
spin_lock_init(&domain->iotlb_lock);
init_iova_domain(&domain->stream_iovad,
- PAGE_SIZE, IOVA_START_PFN);
+ BOUNCE_PAGE_SIZE, IOVA_START_PFN);
ret = iova_domain_init_rcaches(&domain->stream_iovad);
if (ret)
goto err_iovad_stream;
diff --git a/drivers/vdpa/vdpa_user/iova_domain.h b/drivers/vdpa/vdpa_user/iova_domain.h
index 7f3f0928ec78..23139a2eaf5c 100644
--- a/drivers/vdpa/vdpa_user/iova_domain.h
+++ b/drivers/vdpa/vdpa_user/iova_domain.h
@@ -19,6 +19,11 @@
#define INVALID_PHYS_ADDR (~(phys_addr_t)0)
+#define BOUNCE_PAGE_SHIFT 12
+#define BOUNCE_PAGE_SIZE (1 << BOUNCE_PAGE_SHIFT)
+#define BOUNCE_PAGE_MASK (~(BOUNCE_PAGE_SIZE - 1))
+#define BOUNCE_PAGE_ALIGN(addr) (((addr) + BOUNCE_PAGE_SIZE - 1) & ~(BOUNCE_PAGE_SIZE - 1))
+
struct vduse_bounce_map {
struct page *bounce_page;
struct page *user_bounce_page;
--
2.20.1
next reply other threads:[~2025-09-15 7:34 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-09-15 7:34 sheng.zhao [this message]
2025-09-15 8:21 ` Jason Wang
2025-09-15 11:07 ` [External] " Sheng Zhao
2025-09-16 7:34 ` Jason Wang
2025-09-16 13:54 ` Sheng Zhao
2025-09-17 8:16 ` Jason Wang
2025-09-23 12:37 ` Sheng Zhao
2025-09-24 0:57 ` Jason Wang
2025-09-24 4:05 ` Sheng Zhao
2025-09-24 4:15 ` Jason Wang
2025-09-24 6:38 ` Sheng Zhao
2025-09-25 0:20 ` Jason Wang
2025-09-25 3:13 ` Sheng Zhao
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20250915073429.54027-1-sheng.zhao@bytedance.com \
--to=sheng.zhao@bytedance.com \
--cc=eperezma@redhat.com \
--cc=jasowang@redhat.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mst@redhat.com \
--cc=virtualization@lists.linux.dev \
--cc=xieyongji@bytedance.com \
--cc=xuanzhuo@linux.alibaba.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®