lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <8e46b1c5-f52b-0155-4d4f-e3bbdea95384@redhat.com>
Date:   Tue, 12 Nov 2019 12:12:33 +0100
From:   David Hildenbrand <david@...hat.com>
To:     Michael Ellerman <mpe@...erman.id.au>, linux-kernel@...r.kernel.org
Cc:     linux-mm@...ck.org, linuxppc-dev@...ts.ozlabs.org,
        Andrew Morton <akpm@...ux-foundation.org>,
        Benjamin Herrenschmidt <benh@...nel.crashing.org>,
        Paul Mackerras <paulus@...ba.org>,
        Thiago Jung Bauermann <bauerman@...ux.ibm.com>,
        Geert Uytterhoeven <geert@...ux-m68k.org>,
        Anshuman Khandual <khandual@...ux.vnet.ibm.com>,
        Oliver O'Halloran <oohall@...il.com>,
        Alexey Kardashevskiy <aik@...abs.ru>,
        "Enrico Weigelt, metux IT consult" <info@...ux.net>,
        Thomas Gleixner <tglx@...utronix.de>,
        Allison Randal <allison@...utok.net>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Pavel Tatashin <pasha.tatashin@...een.com>,
        Arun KS <arunks@...eaurora.org>, Todd Kjos <tkjos@...gle.com>,
        Christian Brauner <christian@...uner.io>,
        Gao Xiang <xiang@...nel.org>,
        Greg Hackmann <ghackmann@...gle.com>,
        David Howells <dhowells@...hat.com>
Subject: Re: [PATCH v1 08/12] powerpc/pseries: CMM: Implement balloon
 compaction

On 12.11.19 11:46, Michael Ellerman wrote:
> David Hildenbrand <david@...hat.com> writes:
>> On 31.10.19 15:29, David Hildenbrand wrote:
>>> We can now get rid of the cmm_lock and completely rely on the balloon
>>> compaction internals, which now also manage the page list and the lock.
> ...
>>> +
>>> +static int cmm_migratepage(struct balloon_dev_info *b_dev_info,
>>> +			   struct page *newpage, struct page *page,
>>> +			   enum migrate_mode mode)
>>> +{
>>> +	unsigned long flags;
>>> +
>>> +	/*
>>> +	 * loan/"inflate" the newpage first.
>>> +	 *
>>> +	 * We might race against the cmm_thread who might discover after our
>>> +	 * loan request that another page is to be unloaned. However, once
>>> +	 * the cmm_thread runs again later, this error will automatically
>>> +	 * be corrected.
>>> +	 */
>>> +	if (plpar_page_set_loaned(newpage)) {
>>> +		/* Unlikely, but possible. Tell the caller not to retry now. */
>>> +		pr_err_ratelimited("%s: Cannot set page to loaned.", __func__);
>>> +		return -EBUSY;
>>> +	}
>>> +
>>> +	/* balloon page list reference */
>>> +	get_page(newpage);
>>> +
>>> +	spin_lock_irqsave(&b_dev_info->pages_lock, flags);
>>> +	balloon_page_insert(b_dev_info, newpage);
>>> +	balloon_page_delete(page);
>>
>> I think I am missing a b_dev_info->isolated_pages-- here.
> 
> I don't know this code at all, but looking at other balloon drivers they
> do seem to do that in roughly the same spot.
> 
> I'll add it, how can we test that it's correct?

It's certainly correct. We increment when we isolate 
(balloon_page_isolate()) and decrement when we un-isolate.

Un-isolate happens when we putback a isolated page 
(balloon_page_putback() - migration aborted) or when we successfully 
migrate it (via balloon_page_migrate()).

The issue is that we cannot decrement in balloon_page_migrate(), as we 
have to hold the b_dev_info->pages_lock. That's why we have to do it in 
the registered callback under lock.

Please note that b_dev_info->isolated_pages is only needed for a sanity 
check in balloon_page_dequeue(). That's why I didn't notice during 
testing. I wonder if we should at some point rip out that sanity check ...

Thanks and cheers!

> 
> cheers
> 


-- 

Thanks,

David / dhildenb

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ