linux/drivers/infiniband/core/umem.c
Parav Pandit 3994586f4d RDMA/core: Acquire and release mmap_sem on page range
Currently mmap_sem is read locked while pinning the memory.  In a
multi-threaded application of a process, holding mmap_sem lock creates
contention with other threads who might be either registering memory,
creating QPs or simply doing mmap() as such operations also require to
hold the mmap_sem write lock.

All such operation cannot make forward progress until one memory pin
operation is completed.  It becomes more worse if the memory is unpinned
and/or memory registration is large (in GB range).

Therefore, instead of holding mmap_sem for too long (for whole region
pinning), acquire and release the lock for every few pages.  For example
on x86 with 4K page size, acquire and release mmap_sem for every 2Mbytes
memory chunk.

This allows other competing threads to make progress who might wish to
hold mmap_sem for shorter duration.

When memory registration latency is measured using [1] for memory sizes
ranging from 4K to 48GB, <= 1% or 0.5% degradation is noticed. In many
runs no difference is seen other than run-to-run variance.

In other targeted tests of users with large memory, desired improvements
are seen due to reduced contention of mmap_sem.

[1] https://github.com/paravmellanox/rtool

$ rdma_resource_lat -c 1 -s 48G -a -u L -i 500 -A

It registers pinned memory from 4K to 48GB size with 500 iterations for
each memory size.

$ rdma_resource_lat -c 1 -s 12G -a -u L -i 500 -t 4

4 competing threads pin memory, each of 12GB size with 500 iterations.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-09-27 12:40:20 -06:00

357 lines
9.0 KiB
C

/*
* Copyright (c) 2005 Topspin Communications. All rights reserved.
* Copyright (c) 2005 Cisco Systems. All rights reserved.
* Copyright (c) 2005 Mellanox Technologies. All rights reserved.
*
* This software is available to you under a choice of one of two
* licenses. You may choose to be licensed under the terms of the GNU
* General Public License (GPL) Version 2, available from the file
* COPYING in the main directory of this source tree, or the
* OpenIB.org BSD license below:
*
* Redistribution and use in source and binary forms, with or
* without modification, are permitted provided that the following
* conditions are met:
*
* - Redistributions of source code must retain the above
* copyright notice, this list of conditions and the following
* disclaimer.
*
* - Redistributions in binary form must reproduce the above
* copyright notice, this list of conditions and the following
* disclaimer in the documentation and/or other materials
* provided with the distribution.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include <linux/mm.h>
#include <linux/dma-mapping.h>
#include <linux/sched/signal.h>
#include <linux/sched/mm.h>
#include <linux/export.h>
#include <linux/hugetlb.h>
#include <linux/slab.h>
#include <rdma/ib_umem_odp.h>
#include "uverbs.h"
static void __ib_umem_release(struct ib_device *dev, struct ib_umem *umem, int dirty)
{
struct scatterlist *sg;
struct page *page;
int i;
if (umem->nmap > 0)
ib_dma_unmap_sg(dev, umem->sg_head.sgl,
umem->npages,
DMA_BIDIRECTIONAL);
for_each_sg(umem->sg_head.sgl, sg, umem->npages, i) {
page = sg_page(sg);
if (!PageDirty(page) && umem->writable && dirty)
set_page_dirty_lock(page);
put_page(page);
}
sg_free_table(&umem->sg_head);
}
/**
* ib_umem_get - Pin and DMA map userspace memory.
*
* If access flags indicate ODP memory, avoid pinning. Instead, stores
* the mm for future page fault handling in conjunction with MMU notifiers.
*
* @context: userspace context to pin memory for
* @addr: userspace virtual address to start at
* @size: length of region to pin
* @access: IB_ACCESS_xxx flags for memory being pinned
* @dmasync: flush in-flight DMA when the memory region is written
*/
struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
size_t size, int access, int dmasync)
{
struct ib_umem *umem;
struct page **page_list;
struct vm_area_struct **vma_list;
unsigned long lock_limit;
unsigned long new_pinned;
unsigned long cur_base;
struct mm_struct *mm;
unsigned long npages;
int ret;
int i;
unsigned long dma_attrs = 0;
struct scatterlist *sg, *sg_list_start;
unsigned int gup_flags = FOLL_WRITE;
if (dmasync)
dma_attrs |= DMA_ATTR_WRITE_BARRIER;
/*
* If the combination of the addr and size requested for this memory
* region causes an integer overflow, return error.
*/
if (((addr + size) < addr) ||
PAGE_ALIGN(addr + size) < (addr + size))
return ERR_PTR(-EINVAL);
if (!can_do_mlock())
return ERR_PTR(-EPERM);
if (access & IB_ACCESS_ON_DEMAND) {
umem = kzalloc(sizeof(struct ib_umem_odp), GFP_KERNEL);
if (!umem)
return ERR_PTR(-ENOMEM);
umem->is_odp = 1;
} else {
umem = kzalloc(sizeof(*umem), GFP_KERNEL);
if (!umem)
return ERR_PTR(-ENOMEM);
}
umem->context = context;
umem->length = size;
umem->address = addr;
umem->page_shift = PAGE_SHIFT;
umem->writable = ib_access_writable(access);
umem->owning_mm = mm = current->mm;
mmgrab(mm);
if (access & IB_ACCESS_ON_DEMAND) {
ret = ib_umem_odp_get(to_ib_umem_odp(umem), access);
if (ret)
goto umem_kfree;
return umem;
}
/* We assume the memory is from hugetlb until proved otherwise */
umem->hugetlb = 1;
page_list = (struct page **) __get_free_page(GFP_KERNEL);
if (!page_list) {
ret = -ENOMEM;
goto umem_kfree;
}
/*
* if we can't alloc the vma_list, it's not so bad;
* just assume the memory is not hugetlb memory
*/
vma_list = (struct vm_area_struct **) __get_free_page(GFP_KERNEL);
if (!vma_list)
umem->hugetlb = 0;
npages = ib_umem_num_pages(umem);
if (npages == 0 || npages > UINT_MAX) {
ret = -EINVAL;
goto out;
}
lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
down_write(&mm->mmap_sem);
if (check_add_overflow(mm->pinned_vm, npages, &new_pinned) ||
(new_pinned > lock_limit && !capable(CAP_IPC_LOCK))) {
up_write(&mm->mmap_sem);
ret = -ENOMEM;
goto out;
}
mm->pinned_vm = new_pinned;
up_write(&mm->mmap_sem);
cur_base = addr & PAGE_MASK;
ret = sg_alloc_table(&umem->sg_head, npages, GFP_KERNEL);
if (ret)
goto vma;
if (!umem->writable)
gup_flags |= FOLL_FORCE;
sg_list_start = umem->sg_head.sgl;
while (npages) {
down_read(&mm->mmap_sem);
ret = get_user_pages_longterm(cur_base,
min_t(unsigned long, npages,
PAGE_SIZE / sizeof (struct page *)),
gup_flags, page_list, vma_list);
if (ret < 0) {
up_read(&mm->mmap_sem);
goto umem_release;
}
umem->npages += ret;
cur_base += ret * PAGE_SIZE;
npages -= ret;
/* Continue to hold the mmap_sem as vma_list access
* needs to be protected.
*/
for_each_sg(sg_list_start, sg, ret, i) {
if (vma_list && !is_vm_hugetlb_page(vma_list[i]))
umem->hugetlb = 0;
sg_set_page(sg, page_list[i], PAGE_SIZE, 0);
}
up_read(&mm->mmap_sem);
/* preparing for next loop */
sg_list_start = sg;
}
umem->nmap = ib_dma_map_sg_attrs(context->device,
umem->sg_head.sgl,
umem->npages,
DMA_BIDIRECTIONAL,
dma_attrs);
if (!umem->nmap) {
ret = -ENOMEM;
goto umem_release;
}
ret = 0;
goto out;
umem_release:
__ib_umem_release(context->device, umem, 0);
vma:
down_write(&mm->mmap_sem);
mm->pinned_vm -= ib_umem_num_pages(umem);
up_write(&mm->mmap_sem);
out:
if (vma_list)
free_page((unsigned long) vma_list);
free_page((unsigned long) page_list);
umem_kfree:
if (ret) {
mmdrop(umem->owning_mm);
kfree(umem);
}
return ret ? ERR_PTR(ret) : umem;
}
EXPORT_SYMBOL(ib_umem_get);
static void __ib_umem_release_tail(struct ib_umem *umem)
{
mmdrop(umem->owning_mm);
if (umem->is_odp)
kfree(to_ib_umem_odp(umem));
else
kfree(umem);
}
static void ib_umem_release_defer(struct work_struct *work)
{
struct ib_umem *umem = container_of(work, struct ib_umem, work);
down_write(&umem->owning_mm->mmap_sem);
umem->owning_mm->pinned_vm -= ib_umem_num_pages(umem);
up_write(&umem->owning_mm->mmap_sem);
__ib_umem_release_tail(umem);
}
/**
* ib_umem_release - release memory pinned with ib_umem_get
* @umem: umem struct to release
*/
void ib_umem_release(struct ib_umem *umem)
{
struct ib_ucontext *context = umem->context;
if (umem->is_odp) {
ib_umem_odp_release(to_ib_umem_odp(umem));
__ib_umem_release_tail(umem);
return;
}
__ib_umem_release(umem->context->device, umem, 1);
/*
* We may be called with the mm's mmap_sem already held. This
* can happen when a userspace munmap() is the call that drops
* the last reference to our file and calls our release
* method. If there are memory regions to destroy, we'll end
* up here and not be able to take the mmap_sem. In that case
* we defer the vm_locked accounting a workqueue.
*/
if (context->closing) {
if (!down_write_trylock(&umem->owning_mm->mmap_sem)) {
INIT_WORK(&umem->work, ib_umem_release_defer);
queue_work(ib_wq, &umem->work);
return;
}
} else {
down_write(&umem->owning_mm->mmap_sem);
}
umem->owning_mm->pinned_vm -= ib_umem_num_pages(umem);
up_write(&umem->owning_mm->mmap_sem);
__ib_umem_release_tail(umem);
}
EXPORT_SYMBOL(ib_umem_release);
int ib_umem_page_count(struct ib_umem *umem)
{
int i;
int n;
struct scatterlist *sg;
if (umem->is_odp)
return ib_umem_num_pages(umem);
n = 0;
for_each_sg(umem->sg_head.sgl, sg, umem->nmap, i)
n += sg_dma_len(sg) >> umem->page_shift;
return n;
}
EXPORT_SYMBOL(ib_umem_page_count);
/*
* Copy from the given ib_umem's pages to the given buffer.
*
* umem - the umem to copy from
* offset - offset to start copying from
* dst - destination buffer
* length - buffer length
*
* Returns 0 on success, or an error code.
*/
int ib_umem_copy_from(void *dst, struct ib_umem *umem, size_t offset,
size_t length)
{
size_t end = offset + length;
int ret;
if (offset > umem->length || length > umem->length - offset) {
pr_err("ib_umem_copy_from not in range. offset: %zd umem length: %zd end: %zd\n",
offset, umem->length, end);
return -EINVAL;
}
ret = sg_pcopy_to_buffer(umem->sg_head.sgl, umem->npages, dst, length,
offset + ib_umem_offset(umem));
if (ret < 0)
return ret;
else if (ret != length)
return -EINVAL;
else
return 0;
}
EXPORT_SYMBOL(ib_umem_copy_from);