linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Lorenzo Stoakes <lstoakes@gmail.com>
To: Baoquan He <bhe@redhat.com>
Cc: linux-kernel@vger.kernel.org, linux-mm@kvack.org,
	urezki@gmail.com, stephen.s.brennan@oracle.com,
	willy@infradead.org, akpm@linux-foundation.org,
	hch@infradead.org
Subject: Re: [PATCH v2 3/7] mm/vmalloc.c: allow vread() to read out vm_map_ram areas
Date: Mon, 9 Jan 2023 07:12:53 +0000	[thread overview]
Message-ID: <Y7u+dQfq3ZbDcf/d@lucifer> (raw)
In-Reply-To: <Y7uZeBj56VcnJhzJ@fedora>

On Mon, Jan 09, 2023 at 12:35:04PM +0800, Baoquan He wrote:
> Sorry for late reply, just come back from vacation.

Hope you had a great time! :)

>
> Lei + mutt sounds like a good idea. I relied too much on mbsync in the
> past.
>

Yeah I'm finding it works well,
https://josefbacik.github.io/kernel/2021/10/18/lei-and-b4.html is a handy guide!

[snip]
> > Maybe let me rephrase:-
> >
> > - We want to read `count` bytes from `addr` into `buf`
> > - We iterate over _used_ blocks, placing the start/end of each block in `rs`, `re`
> >   respectively.
> > - If we hit a block whose start address is above the one in which we are interested then:-
> >   - Place a zero byte in the buffer
> >   - Increment `addr` by 1 byte
> >   - Decrement the `count` by 1 byte
> >   - Carry on
> >
> > I am seriously confused as to why we do this? Surely we should be checking
> > whether the range [addr, addr + count) overlaps this block at all, and only then
> > copying the relevant region?
>
> I guessed this could be your concern, but not very sure. That
> code block is copied from vread(), and my considerations are:
> 1) We could starting read from any position of kcore file. /proc/kcore
> is a elf file logically, it's allowed to read from anywhere, right? We
> don't have to read the entire file always. So the vmap_block reading is
> not necessarily page aligned. It's very similar with the empty area
> filling in vread().
> 2) memset() is doing the byte by byte reading. We can
> change code as below. While we don't save the effort very much, and we
> need introduce an extra local variable to store the value of
> (start - end).
>
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index b054081aa66b..dce4a843a9e8 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -3576,6 +3576,15 @@ static void vmap_ram_vread(char *buf, char *addr, int count, unsigned long flags
> +		if (addr < start) {
> +			int num = min(count, (start - add));
> +			memset(buf, 0, count);
> +			count -= num;
> +			if (count == 0)
> +				break;
> +			buf -= num;
> +			addr -= num;
> +		}
>  		/*it could start reading from the middle of used region*/
>  		offset = offset_in_page(addr);
>  		n = ((re - rs + 1) << PAGE_SHIFT) - offset;
>

The difference with vread() is that uses a while loop rather than an if clause
so operates over the whole region byte-by-byte, your original would only do this
for 1 byte so now things make a lot more sense!

This approach makes sense though I'd put the count == 0 check first and nit
'add' should be 'addr'.

I am happy with either this or a while loop instead of an if which it seems is
what the original issue was!

> void *memset(void *s, int c, size_t count)
> {
>         char *xs = s;
>
>         while (count--)
>                 *xs++ = c;
>         return s;
> }
>
> >
> > It's the fact that blocks are at base page granularity but then this condition
> > is at byte granularity that is confusing to me (again it's _very_ possible I am
> > just being dumb here and missing something, just really want to understand this
> > better :)
>
> I like this kind of reviewing with careful checking and deep thinking.
> For above code block, I think it's a very great point. From my point of
> view, I like the memset version better, it's easier to understand. If we
> all agree, we can change it to take memset way. When I made patches,
> several issues related to patches were hovering in my mind at the same
> time, I did not consider this one so deeply.
>

Thanks :) I have a particular interest in vmalloc so am happy to dive in with
reviews here!

> >
> > > > > -		vm = va->vm;
> > > > > -		vaddr = (char *) vm->addr;
> > > > > -		if (addr >= vaddr + get_vm_area_size(vm))
> > > > > +		vaddr = (char *) va->va_start;
> > > > > +		size = flags ? va_size(va) : get_vm_area_size(vm);
> > > >
> > > > For example here, I feel that this ternary should be reversed and based on
> > > > whether vm is null, unles we expect vm to ever be non-null _and_ flags to be
> > > > set?
> > >
> > > Now only vm_map_ram area sets flags, all other types has vm not null.
> > > Since those temporary state, e.g vm==NULL, flags==0 case has been
> > > filtered out. Is below you suggested?
> > >
> > > 		size = (!vm&&flags)? va_size(va) : get_vm_area_size(vm);
> > > 		or
> > > 		size = (vm&&!flags)? get_vm_area_size(vm):va_size(va);
> > >
> >
> > Sorry I didn't phrase this very well, my point is that the key thing you're
> > relying on here is whether vm exists in order to use it so I simply meant:-
> >
> > size = vm ? get_vm_area_size(vm) : va_size(va);
> >
> > This just makes it really explicit that you need vm to be non-NULL, and you've
> > already done the flags check before so this should suffice.
>
> Sounds reasonable, I will copy above line you pasted. Thanks a lot.
>

Cheers!

  reply	other threads:[~2023-01-09  7:15 UTC|newest]

Thread overview: 32+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-12-17  1:54 [PATCH v2 0/7] mm/vmalloc.c: allow vread() to read out vm_map_ram areas Baoquan He
2022-12-17  1:54 ` [PATCH v2 1/7] mm/vmalloc.c: add used_map into vmap_block to track space of vmap_block Baoquan He
2022-12-17  1:54 ` [PATCH v2 2/7] mm/vmalloc.c: add flags to mark vm_map_ram area Baoquan He
2022-12-17 11:44   ` Lorenzo Stoakes
2022-12-19  8:01     ` Baoquan He
2022-12-19  9:09       ` Lorenzo Stoakes
2022-12-19 12:24         ` Baoquan He
2022-12-19 13:01           ` Lorenzo Stoakes
2022-12-20 12:14             ` Baoquan He
2022-12-20 12:42               ` Lorenzo Stoakes
2022-12-20 16:55   ` Uladzislau Rezki
2022-12-23  4:14     ` Baoquan He
2023-01-13  3:55       ` Baoquan He
2023-01-16 17:54         ` Uladzislau Rezki
2023-01-18  3:09           ` Baoquan He
2023-01-18 12:20             ` Uladzislau Rezki
2022-12-17  1:54 ` [PATCH v2 3/7] mm/vmalloc.c: allow vread() to read out vm_map_ram areas Baoquan He
2022-12-17  4:10   ` kernel test robot
2022-12-17  6:41   ` kernel test robot
2022-12-17  9:46     ` Baoquan He
2022-12-17 12:06   ` Lorenzo Stoakes
2023-01-04  8:01     ` Baoquan He
2023-01-04 20:20       ` Lorenzo Stoakes
2023-01-09  4:35         ` Baoquan He
2023-01-09  7:12           ` Lorenzo Stoakes [this message]
2023-01-09 12:49             ` Baoquan He
2022-12-17  1:54 ` [PATCH v2 4/7] mm/vmalloc: explicitly identify vm_map_ram area when shown in /proc/vmcoreinfo Baoquan He
2022-12-17  1:54 ` [PATCH v2 5/7] mm/vmalloc: skip the uninitilized vmalloc areas Baoquan He
2022-12-17 12:07   ` Lorenzo Stoakes
2022-12-19  7:16     ` Baoquan He
2022-12-17  1:54 ` [PATCH v2 6/7] powerpc: mm: add VM_IOREMAP flag to the vmalloc area Baoquan He
2022-12-17  1:54 ` [PATCH v2 7/7] sh: mm: set " Baoquan He

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=Y7u+dQfq3ZbDcf/d@lucifer \
    --to=lstoakes@gmail.com \
    --cc=akpm@linux-foundation.org \
    --cc=bhe@redhat.com \
    --cc=hch@infradead.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=stephen.s.brennan@oracle.com \
    --cc=urezki@gmail.com \
    --cc=willy@infradead.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).