shm: scroll: implement offset wrap-around

* Impose a maximum memfd size limit. In theory, this can be
  2GB (wl_shm_create_pool() is the limiting factor - its size argument
  is an int32_t). For now, use 256MB.

  This is mainly to reduce the amount of virtual address space used by
  the compositor, which keeps at least one mmapping (of the entire
  memfd) around. One mmapping *per terminal window* that is.

  Given that we have 128TB with 48-bit virtual addresses, we could
  probably bump this to 2GB without any issues. However, 256MB should
  be enough.

  TODO: check how much we typically move the offset when scrolling in
  a fullscreen window on a 4K monitor. 256MB may turn out to be too
  small.

  On 32-bit shm_scroll() is completely disabled. There simply isn't
  enough address space.

* Wrapping is done by moving the offset to "the other end" of the
  memfd, and copying the buffer contents to the new, wrapped offset.

  The "normal" scrolling code then does the actual scrolling. This
  means we'll re-instantiate all objects twice when wrapping.
This commit is contained in:
Daniel Eklöf 2020-03-24 17:46:48 +01:00
parent dc393bd5d7
commit 5c5f1d096c
No known key found for this signature in database
GPG key ID: 5BBD4992C116573F
2 changed files with 154 additions and 42 deletions

193
shm.c
View file

@ -23,13 +23,33 @@
#define TIME_SCROLL 0
/*
* Maximum memfd size allowed.
*
* On 64-bit, we could in theory use up to 2GB (wk_shm_create_pool()
* is limited to int32_t), since we never mmap() the entire region.
*
* The compositor is different matter - it needs to mmap() the entire
* range, and *keep* the mapping for as long as is has buffers
* referencing it (thus - always). And if we open multiple terminals,
* then the required address space multiples...
*
* That said, 128TB (the total amount of available user address space
* on 64-bit) is *a lot*; we can fit 67108864 2GB memfds into
* that. But, let's be conservative for now.
*
* On 32-bit the available address space is too small and SHM
* scrolling is disabled.
*/
static const off_t max_pool_size = 256 * 1024 * 1024;
static tll(struct buffer) buffers;
static bool can_punch_hole = false;
static bool can_punch_hole_initialized = false;
static void
buffer_destroy(struct buffer *buf)
buffer_destroy_dont_close(struct buffer *buf)
{
if (buf->pix != NULL)
pixman_image_unref(buf->pix);
@ -37,8 +57,20 @@ buffer_destroy(struct buffer *buf)
wl_buffer_destroy(buf->wl_buf);
if (buf->real_mmapped != MAP_FAILED)
munmap(buf->real_mmapped, buf->mmap_size);
buf->pix = NULL;
buf->wl_buf = NULL;
buf->real_mmapped = NULL;
buf->mmapped = NULL;
}
static void
buffer_destroy(struct buffer *buf)
{
buffer_destroy_dont_close(buf);
if (buf->fd >= 0)
close(buf->fd);
buf->fd = -1;
}
static void
@ -55,24 +87,30 @@ static const struct wl_buffer_listener buffer_listener = {
.release = &buffer_release,
};
static size_t
page_size(void)
{
static size_t size = 0;
if (size == 0) {
size = sysconf(_SC_PAGE_SIZE);
if (size < 0) {
LOG_ERRNO("failed to get page size");
size = 4096;
}
}
assert(size > 0);
return size;
}
static bool
instantiate_offset(struct wl_shm *shm, struct buffer *buf, size_t new_offset)
instantiate_offset(struct wl_shm *shm, struct buffer *buf, off_t new_offset)
{
assert(buf->fd >= 0);
assert(buf->mmapped == NULL);
assert(buf->real_mmapped == NULL);
assert(buf->wl_buf == NULL);
assert(buf->pix == NULL);
static size_t page_size = 0;
if (page_size == 0) {
page_size = sysconf(_SC_PAGE_SIZE);
if (page_size < 0) {
LOG_ERRNO("failed to get page size");
page_size = 4096;
}
}
assert(page_size > 0);
assert(new_offset + buf->size <= max_pool_size);
void *real_mmapped = MAP_FAILED;
void *mmapped = MAP_FAILED;
@ -81,8 +119,8 @@ instantiate_offset(struct wl_shm *shm, struct buffer *buf, size_t new_offset)
pixman_image_t *pix = NULL;
/* mmap offset must be page aligned */
size_t aligned_offset = new_offset & ~(page_size - 1);
size_t page_offset = new_offset & (page_size - 1);
off_t aligned_offset = new_offset & ~(page_size() - 1);
size_t page_offset = new_offset & (page_size() - 1);
size_t mmap_size = buf->size + page_offset;
assert(aligned_offset <= new_offset);
@ -222,7 +260,17 @@ shm_get_buffer(struct wl_shm *shm, int width, int height, unsigned long cookie)
goto err;
}
if (ftruncate(pool_fd, size) == -1) {
/*
* If we can figure our if we can punch holes *before* this, we
* could set the initial offset to somewhere in the middle of the
* avaiable address space. This would allow both backward and
* forward scrolling without immediately needing to wrap.
*/
//off_t initial_offset = (max_pool_size / 4) & ~(page_size() - 1);
off_t initial_offset = 0;
off_t memfd_size = initial_offset + size;
if (ftruncate(pool_fd, memfd_size) == -1) {
LOG_ERRNO("failed to truncate SHM pool");
goto err;
}
@ -256,7 +304,7 @@ shm_get_buffer(struct wl_shm *shm, int width, int height, unsigned long cookie)
);
struct buffer *ret = &tll_back(buffers);
if (!instantiate_offset(shm, ret, 0))
if (!instantiate_offset(shm, ret, initial_offset))
goto err;
return ret;
@ -281,7 +329,36 @@ shm_fini(void)
bool
shm_can_scroll(void)
{
#if defined(__i386__)
/* Not enough virtual address space in 32-bit */
return false;
#else
return can_punch_hole;
#endif
}
static bool
wrap_buffer(struct wl_shm *shm, struct buffer *buf, off_t new_offset)
{
off_t aligned_offset = new_offset & ~(page_size() - 1);
size_t page_offset = new_offset & (page_size() - 1);
size_t mmap_size = buf->size + page_offset;
assert(aligned_offset <= new_offset);
assert(mmap_size >= buf->size);
uint8_t *m = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_UNINITIALIZED, buf->fd, aligned_offset);
if (m == MAP_FAILED) {
LOG_ERRNO("failed to mmap");
return false;
}
memcpy(m + page_offset, buf->mmapped, buf->size);
munmap(m, mmap_size);
/* Re-instantiate pixman+wl_buffer+raw pointersw */
buffer_destroy_dont_close(buf);
return instantiate_offset(shm, buf, new_offset);
}
static bool
@ -303,7 +380,22 @@ shm_scroll_forward(struct wl_shm *shm, struct buffer *buf, int rows,
assert(rows > 0);
assert(rows * buf->stride < buf->size);
const size_t new_offset = buf->offset + rows * buf->stride;
if (buf->offset + rows * buf->stride + buf->size > max_pool_size) {
LOG_INFO("memfd offset wrap around");
assert(buf->offset > buf->size);
/*
* Wrap around by moving the offset to the beginning of the
* memfd. The ftruncate() we do below takes care of trimming
* down the size.
*/
if (!wrap_buffer(shm, buf, 0))
goto err;
}
off_t new_offset = buf->offset + rows * buf->stride;
assert(new_offset + buf->size <= max_pool_size);
#if TIME_SCROLL
struct timeval time0;
@ -312,9 +404,9 @@ shm_scroll_forward(struct wl_shm *shm, struct buffer *buf, int rows,
/* Increase file size */
if (ftruncate(buf->fd, new_offset + buf->size) < 0) {
LOG_ERRNO("failed increase memfd size from %zu -> %zu",
LOG_ERRNO("failed to resize memfd from %zu -> %zu",
buf->offset + buf->size, new_offset + buf->size);
return false;
goto err;
}
#if TIME_SCROLL
@ -343,19 +435,17 @@ shm_scroll_forward(struct wl_shm *shm, struct buffer *buf, int rows,
}
/* Destroy old objects (they point to the old offset) */
pixman_image_unref(buf->pix);
wl_buffer_destroy(buf->wl_buf);
munmap(buf->real_mmapped, buf->mmap_size);
buf->pix = NULL;
buf->wl_buf = NULL;
buf->real_mmapped = buf->mmapped = NULL;
buffer_destroy_dont_close(buf);
/* Free unused memory */
if (fallocate(buf->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 0, new_offset) < 0) {
abort();
if (new_offset > 0 &&
fallocate(buf->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 0, new_offset) < 0)
{
LOG_ERRNO("fallocate(FALLOC_FL_PUNCH_HOLE, 0, %lu) failed", new_offset);
goto err;
}
#if TIME_SCROLL
struct timeval time3;
gettimeofday(&time3, NULL);
@ -363,6 +453,7 @@ shm_scroll_forward(struct wl_shm *shm, struct buffer *buf, int rows,
LOG_INFO("PUNCH HOLE: %lds %ldus", tot.tv_sec, tot.tv_usec);
#endif
/* Re-instantiate pixman+wl_buffer+raw pointersw */
bool ret = instantiate_offset(shm, buf, new_offset);
if (ret && bottom_keep_rows > 0) {
@ -382,6 +473,10 @@ shm_scroll_forward(struct wl_shm *shm, struct buffer *buf, int rows,
}
return ret;
err:
abort();
return false;
}
static bool
@ -391,12 +486,27 @@ shm_scroll_reverse(struct wl_shm *shm, struct buffer *buf, int rows,
{
assert(rows > 0);
size_t diff = rows * buf->stride;
if (diff > buf->offset)
return false;
off_t diff = rows * buf->stride;
if (diff > buf->offset) {
LOG_INFO("memfd offset reverse wrap-around");
const size_t new_offset = buf->offset - rows * buf->stride;
assert(new_offset < buf->offset);
/*
* Wrap around by resizing the memfd and moving the offset to
* the end of the file, taking care the new offset is aligned.
*/
if (ftruncate(buf->fd, max_pool_size) < 0) {
LOG_ERRNO("failed to resize memfd from %zu -> %zu",
buf->offset + buf->size, max_pool_size - buf->size);
goto err;
}
if (!wrap_buffer(shm, buf, (max_pool_size - buf->size) & ~(page_size() - 1)))
goto err;
}
off_t new_offset = buf->offset - diff;
assert(new_offset <= max_pool_size);
#if TIME_SCROLL
struct timeval time0;
@ -421,16 +531,12 @@ shm_scroll_reverse(struct wl_shm *shm, struct buffer *buf, int rows,
}
/* Destroy old objects (they point to the old offset) */
pixman_image_unref(buf->pix);
wl_buffer_destroy(buf->wl_buf);
munmap(buf->real_mmapped, buf->mmap_size);
buf->pix = NULL;
buf->wl_buf = NULL;
buf->real_mmapped = buf->mmapped = NULL;
buffer_destroy_dont_close(buf);
if (ftruncate(buf->fd, new_offset + buf->size) < 0) {
abort();
LOG_ERRNO("failed to resize memfd from %zu -> %zu",
buf->offset + buf->size, new_offset + buf->size);
goto err;
}
#if TIME_SCROLL
@ -440,6 +546,7 @@ shm_scroll_reverse(struct wl_shm *shm, struct buffer *buf, int rows,
LOG_INFO("ftruncate: %lds %ldus", tot.tv_sec, tot.tv_usec);
#endif
/* Re-instantiate pixman+wl_buffer+raw pointers */
bool ret = instantiate_offset(shm, buf, new_offset);
if (ret && top_keep_rows > 0) {
@ -458,6 +565,10 @@ shm_scroll_reverse(struct wl_shm *shm, struct buffer *buf, int rows,
}
return ret;
err:
abort();
return false;
}
bool

3
shm.h
View file

@ -2,6 +2,7 @@
#include <stdbool.h>
#include <stddef.h>
#include <sys/types.h>
#include <pixman.h>
#include <wayland-client.h>
@ -24,7 +25,7 @@ struct buffer {
int fd; /* memfd */
void *real_mmapped; /* Address returned from mmap */
size_t mmap_size; /* Size of mmap (>= size) */
size_t offset; /* Offset into memfd where data begins */
off_t offset; /* Offset into memfd where data begins */
bool purge; /* True if this buffer should be destroyed */
};