public-mirrors/linux
1
0
Fork 0
mirror of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git synced 2025-04-13 09:59:31 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

io_uring/kbuf: vmap pinned buffer ring

Browse source 
This avoids needing to care about HIGHMEM, and it makes the buffer
indexing easier as both ring provided buffer methods are now virtually
mapped in a contigious fashion.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
This commit is contained in:
Jens Axboe 2024-03-12 10:42:27 -06:00
parent 1943f96b38
commit e270bfd22a
1 changed files with 15 additions and 24 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

39
io_uring/kbuf.c
View file

@ -7,6 +7,7 @@
#include <linux/slab.h> #include <linux/slab.h>
#include <linux/namei.h> #include <linux/namei.h>
#include <linux/poll.h> #include <linux/poll.h>
#include <linux/vmalloc.h>
#include <linux/io_uring.h> #include <linux/io_uring.h>
#include <uapi/linux/io_uring.h> #include <uapi/linux/io_uring.h>
@ -146,15 +147,7 @@ static void __user *io_ring_buffer_select(struct io_kiocb *req, size_t *len,
req->flags |= REQ_F_BL_EMPTY; req->flags |= REQ_F_BL_EMPTY;
head &= bl->mask; head &= bl->mask;
/* mmaped buffers are always contig */ buf = &br->bufs[head];
if (bl->is_mmap || head < IO_BUFFER_LIST_BUF_PER_PAGE) {
buf = &br->bufs[head];
} else {
int off = head & (IO_BUFFER_LIST_BUF_PER_PAGE - 1);
int index = head / IO_BUFFER_LIST_BUF_PER_PAGE;
buf = page_address(bl->buf_pages[index]);
buf += off;
}
if (*len == 0 || *len > buf->len) if (*len == 0 || *len > buf->len)
*len = buf->len; *len = buf->len;
req->flags |= REQ_F_BUFFER_RING; req->flags |= REQ_F_BUFFER_RING;
@ -241,6 +234,7 @@ static int __io_remove_buffers(struct io_ring_ctx *ctx,
for (j = 0; j < bl->buf_nr_pages; j++) for (j = 0; j < bl->buf_nr_pages; j++)
unpin_user_page(bl->buf_pages[j]); unpin_user_page(bl->buf_pages[j]);
kvfree(bl->buf_pages); kvfree(bl->buf_pages);
vunmap(bl->buf_ring);
bl->buf_pages = NULL; bl->buf_pages = NULL;
bl->buf_nr_pages = 0; bl->buf_nr_pages = 0;
} }
@ -498,9 +492,9 @@ err:
static int io_pin_pbuf_ring(struct io_uring_buf_reg *reg, static int io_pin_pbuf_ring(struct io_uring_buf_reg *reg,
struct io_buffer_list *bl) struct io_buffer_list *bl)
{ {
struct io_uring_buf_ring *br; struct io_uring_buf_ring *br = NULL;
int nr_pages, ret, i;
struct page **pages; struct page **pages;
int i, nr_pages;
pages = io_pin_pages(reg->ring_addr, pages = io_pin_pages(reg->ring_addr,
flex_array_size(br, bufs, reg->ring_entries), flex_array_size(br, bufs, reg->ring_entries),
@ -508,18 +502,12 @@ static int io_pin_pbuf_ring(struct io_uring_buf_reg *reg,
if (IS_ERR(pages)) if (IS_ERR(pages))
return PTR_ERR(pages); return PTR_ERR(pages);
/* br = vmap(pages, nr_pages, VM_MAP, PAGE_KERNEL);
* Apparently some 32-bit boxes (ARM) will return highmem pages, if (!br) {
* which then need to be mapped. We could support that, but it'd ret = -ENOMEM;
* complicate the code and slowdown the common cases quite a bit. goto error_unpin;
* So just error out, returning -EINVAL just like we did on kernels }
* that didn't support mapped buffer rings.
*/
for (i = 0; i < nr_pages; i++)
if (PageHighMem(pages[i]))
goto error_unpin;
br = page_address(pages[0]);
#ifdef SHM_COLOUR #ifdef SHM_COLOUR
/* /*
* On platforms that have specific aliasing requirements, SHM_COLOUR * On platforms that have specific aliasing requirements, SHM_COLOUR
@ -530,8 +518,10 @@ static int io_pin_pbuf_ring(struct io_uring_buf_reg *reg,
* should use IOU_PBUF_RING_MMAP instead, and liburing will handle * should use IOU_PBUF_RING_MMAP instead, and liburing will handle
* this transparently. * this transparently.
*/ */
if ((reg->ring_addr | (unsigned long) br) & (SHM_COLOUR - 1)) if ((reg->ring_addr | (unsigned long) br) & (SHM_COLOUR - 1)) {
ret = -EINVAL;
goto error_unpin; goto error_unpin;
}
#endif #endif
bl->buf_pages = pages; bl->buf_pages = pages;
bl->buf_nr_pages = nr_pages; bl->buf_nr_pages = nr_pages;
@ -543,7 +533,8 @@ error_unpin:
for (i = 0; i < nr_pages; i++) for (i = 0; i < nr_pages; i++)
unpin_user_page(pages[i]); unpin_user_page(pages[i]);
kvfree(pages); kvfree(pages);
return -EINVAL; vunmap(br);
return ret;
} }
/* /*