lkml.org 
[lkml]   [2017]   [Mar]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH kernel v8 3/4] mm: add inerface to offer info about unused pages
On Thu, 16 Mar 2017 15:08:46 +0800 Wei Wang <wei.w.wang@intel.com> wrote:

> From: Liang Li <liang.z.li@intel.com>
>
> This patch adds a function to provides a snapshot of the present system
> unused pages. An important usage of this function is to provide the
> unsused pages to the Live migration thread, which skips the transfer of
> thoses unused pages. Newly used pages can be re-tracked by the dirty
> page logging mechanisms.

I don't think this will be useful for anything other than
virtio-balloon. I guess it would be better to keep this code in the
virtio-balloon driver if possible, even though that's rather a layering
violation :( What would have to be done to make that possible? Perhaps
we can put some *small* helpers into page_alloc.c to prevent things
from becoming too ugly.

> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -4498,6 +4498,120 @@ void show_free_areas(unsigned int filter)
> show_swap_cache_info();
> }
>
> +static int __record_unused_pages(struct zone *zone, int order,
> + __le64 *buf, unsigned int size,
> + unsigned int *offset, bool part_fill)
> +{
> + unsigned long pfn, flags;
> + int t, ret = 0;
> + struct list_head *curr;
> + __le64 *chunk;
> +
> + if (zone_is_empty(zone))
> + return 0;
> +
> + spin_lock_irqsave(&zone->lock, flags);
> +
> + if (*offset + zone->free_area[order].nr_free > size && !part_fill) {
> + ret = -ENOSPC;
> + goto out;
> + }
> + for (t = 0; t < MIGRATE_TYPES; t++) {
> + list_for_each(curr, &zone->free_area[order].free_list[t]) {
> + pfn = page_to_pfn(list_entry(curr, struct page, lru));
> + chunk = buf + *offset;
> + if (*offset + 2 > size) {
> + ret = -ENOSPC;
> + goto out;
> + }
> + /* Align to the chunk format used in virtio-balloon */
> + *chunk = cpu_to_le64(pfn << 12);
> + *(chunk + 1) = cpu_to_le64((1 << order) << 12);
> + *offset += 2;
> + }
> + }
> +
> +out:
> + spin_unlock_irqrestore(&zone->lock, flags);
> +
> + return ret;
> +}

This looks like it could disable interrupts for a long time. Too long?

> +/*
> + * The record_unused_pages() function is used to record the system unused
> + * pages. The unused pages can be skipped to transfer during live migration.
> + * Though the unused pages are dynamically changing, dirty page logging
> + * mechanisms are able to capture the newly used pages though they were
> + * recorded as unused pages via this function.
> + *
> + * This function scans the free page list of the specified order to record
> + * the unused pages, and chunks those continuous pages following the chunk
> + * format below:
> + * --------------------------------------
> + * | Base (52-bit) | Rsvd (12-bit) |
> + * --------------------------------------
> + * --------------------------------------
> + * | Size (52-bit) | Rsvd (12-bit) |
> + * --------------------------------------
> + *
> + * @start_zone: zone to start the record operation.
> + * @order: order of the free page list to record.
> + * @buf: buffer to record the unused page info in chunks.
> + * @size: size of the buffer in __le64 to record
> + * @offset: offset in the buffer to record.
> + * @part_fill: indicate if partial fill is used.
> + *
> + * return -EINVAL if parameter is invalid
> + * return -ENOSPC when the buffer is too small to record all the unsed pages
> + * return 0 when sccess
> + */

It's a strange thing - it returns information which will instantly
become incorrect.

> +int record_unused_pages(struct zone **start_zone, int order,
> + __le64 *buf, unsigned int size,
> + unsigned int *offset, bool part_fill)
> +{
> + struct zone *zone;
> + int ret = 0;
> + bool skip_check = false;
> +
> + /* Make sure all the parameters are valid */
> + if (buf == NULL || offset == NULL || order >= MAX_ORDER)
> + return -EINVAL;
> +
> + if (*start_zone != NULL) {
> + bool found = false;
> +
> + for_each_populated_zone(zone) {
> + if (zone != *start_zone)
> + continue;
> + found = true;
> + break;
> + }
> + if (!found)
> + return -EINVAL;
> + } else
> + skip_check = true;
> +
> + for_each_populated_zone(zone) {
> + /* Start from *start_zone if it's not NULL */
> + if (!skip_check) {
> + if (*start_zone != zone)
> + continue;
> + else
> + skip_check = true;
> + }
> + ret = __record_unused_pages(zone, order, buf, size,
> + offset, part_fill);
> + if (ret < 0) {
> + /* record the failed zone */
> + *start_zone = zone;
> + break;
> + }
> + }
> +
> + return ret;
> +}
> +EXPORT_SYMBOL(record_unused_pages);

\
 
 \ /
  Last update: 2017-03-16 22:30    [W:1.971 / U:1.036 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site