Thread (25 messages) 25 messages, 6 authors, 2021-03-25

Re: [mm, net-next v2] mm: net: memcg accounting for TCP rx zerocopy

From: Arjun Roy <hidden>
Date: 2021-03-24 22:22:12
Also in: cgroups, linux-mm, lkml

On Wed, Mar 24, 2021 at 11:26 AM Shakeel Butt [off-list ref] wrote:
On Tue, Mar 23, 2021 at 11:42 AM Arjun Roy [off-list ref] wrote:
quoted
[...]
quoted
To summarize then, it seems to me that we're on the same page now.
I'll put together a tentative v3 such that:
1. It uses pre-charging, as previously discussed.
2. It uses a page flag to delineate pages of a certain networking sort
(ie. this mechanism).
3. It avails itself of up to 4 words of data inside struct page,
inside the networking specific struct.
4. And it sets up this opt-in lifecycle notification for drivers that
choose to use it, falling back to existing behaviour without.
Arjun, if you don't mind, can you explain how the lifetime of such a
page will look like?

For example:

Driver:
page = dev_alloc_page()
/* page has 1 ref */
Yes, this is the case.
dev_map_page(page)
/* I don't think dev_map_page() takes a ref on page, so the ref remains 1. */
To be clear, do you mean things like DMA setup here? Or specifically
what do you mean by dev_map_page?
On incoming traffic the page goes to skb and which then gets assigned
to a struct sock. Does the kernel increase refcnt of the page on these
operations?
Adding a page to an skb will mean that, when the skb is cleaned up, a
page ref is dropped:
https://github.com/torvalds/linux/blob/master/net/core/skbuff.c#L666

So a driver may bump the refcount for the page, before adding it to the skb:
https://github.com/torvalds/linux/blob/master/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c#L442

The page gets mapped into user space which increments its refcnt.
Yes.
After processing the data, the application unmaps the page and its
refcnt will be decremented.
Yes.

__put_page() will be called when refcnt reaches 0, so, the initial
refcnt which the driver has acquired, has to be transferred to the
next layer. So, I am trying to understand how that will work?
Ah, I see - there was a miscommunication. Johannes mentioned
__put_page() but I read put_page().
That is where I was planning on adding the interposition for these
network pages.

So in put_page(), if it turns out it's a network page, we do our
handling then as I described in prior emails. Sorry for the confusion.

Thanks,
-Arjun
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help