Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1757589AbZJBKoV (ORCPT ); Fri, 2 Oct 2009 06:44:21 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1757580AbZJBKoU (ORCPT ); Fri, 2 Oct 2009 06:44:20 -0400 Received: from gir.skynet.ie ([193.1.99.77]:37960 "EHLO gir.skynet.ie" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757579AbZJBKoS (ORCPT ); Fri, 2 Oct 2009 06:44:18 -0400 Date: Fri, 2 Oct 2009 11:44:25 +0100 From: Mel Gorman To: Ingo Molnar , Badari Pulavarty , Brian King , Benjamin Herrenschmidt , Paul Mackerras , Martin Schwidefsky , linux-kernel@vger.kernel.org, linux-mm@kvack.org, linuxppc-dev@ozlabs.org Subject: Re: [PATCH 1/2] mm: add notifier in pageblock isolation for balloon drivers Message-ID: <20091002104425.GN21906@csn.ul.ie> References: <20091001195311.GA16667@austin.ibm.com> MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-15 Content-Disposition: inline In-Reply-To: <20091001195311.GA16667@austin.ibm.com> User-Agent: Mutt/1.5.17+20080114 (2008-01-14) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 8215 Lines: 238 On Thu, Oct 01, 2009 at 02:53:11PM -0500, Robert Jennings wrote: > Memory balloon drivers can allocate a large amount of memory which > is not movable but could be freed to accommodate memory hotplug remove. > > Prior to calling the memory hotplug notifier chain the memory in the > pageblock is isolated. If the migrate type is not MIGRATE_MOVABLE the > isolation will not proceed, causing the memory removal for that page > range to fail. > > Rather than immediately failing pageblock isolation if the the > migrateteype is not MIGRATE_MOVABLE, this patch checks if all of the > pages in the pageblock are owned by a registered balloon driver using a > notifier chain. If all of the non-movable pages are owned by a balloon, > they can be freed later through the memory notifier chain and the range > can still be isolated in set_migratetype_isolate(). > > Signed-off-by: Robert Jennings > > --- > drivers/base/memory.c | 19 +++++++++++++++++++ > include/linux/memory.h | 22 ++++++++++++++++++++++ > mm/page_alloc.c | 49 +++++++++++++++++++++++++++++++++++++++++-------- > 3 files changed, 82 insertions(+), 8 deletions(-) > > Index: b/drivers/base/memory.c > =================================================================== > --- a/drivers/base/memory.c > +++ b/drivers/base/memory.c > @@ -63,6 +63,20 @@ void unregister_memory_notifier(struct n > } > EXPORT_SYMBOL(unregister_memory_notifier); > > +static BLOCKING_NOTIFIER_HEAD(memory_isolate_chain); > + > +int register_memory_isolate_notifier(struct notifier_block *nb) > +{ > + return blocking_notifier_chain_register(&memory_isolate_chain, nb); > +} > +EXPORT_SYMBOL(register_memory_isolate_notifier); > + > +void unregister_memory_isolate_notifier(struct notifier_block *nb) > +{ > + blocking_notifier_chain_unregister(&memory_isolate_chain, nb); > +} > +EXPORT_SYMBOL(unregister_memory_isolate_notifier); > + > /* > * register_memory - Setup a sysfs device for a memory block > */ > @@ -157,6 +171,11 @@ int memory_notify(unsigned long val, voi > return blocking_notifier_call_chain(&memory_chain, val, v); > } > > +int memory_isolate_notify(unsigned long val, void *v) > +{ > + return blocking_notifier_call_chain(&memory_isolate_chain, val, v); > +} > + > /* > * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is > * OK to have direct references to sparsemem variables in here. > Index: b/include/linux/memory.h > =================================================================== > --- a/include/linux/memory.h > +++ b/include/linux/memory.h > @@ -50,6 +50,14 @@ struct memory_notify { > int status_change_nid; > }; > > +#define MEM_ISOLATE_COUNT (1<<0) > + This needs a comment explaining that that this is an action to count the number of pages within a range that have been isolated within a range of pages and not a default value for "nr_pages" in the next structure. > +struct memory_isolate_notify { > + unsigned long start_addr; > + unsigned int nr_pages; > + unsigned int pages_found; > +}; Is there any particular reason you used virtual address of the mapped page instead of PFN? I am guessing at this point that the balloon driver is based on addresses but the code that populates the structure more commonly deals with PFNs. Outside of debugging code, page_address is rarely used in mm/page_alloc.c . It's picky but it feels more natural to me to have the structure have start_pfn and nr_pages or start_addr and end_addr but not a mix of both. > + > struct notifier_block; > struct mem_section; > > @@ -76,14 +84,28 @@ static inline int memory_notify(unsigned > { > return 0; > } > +static inline int register_memory_isolate_notifier(struct notifier_block *nb) > +{ > + return 0; > +} > +static inline void unregister_memory_isolate_notifier(struct notifier_block *nb) > +{ > +} > +static inline int memory_isolate_notify(unsigned long val, void *v) > +{ > + return 0; > +} > #else > extern int register_memory_notifier(struct notifier_block *nb); > extern void unregister_memory_notifier(struct notifier_block *nb); > +extern int register_memory_isolate_notifier(struct notifier_block *nb); > +extern void unregister_memory_isolate_notifier(struct notifier_block *nb); > extern int register_new_memory(int, struct mem_section *); > extern int unregister_memory_section(struct mem_section *); > extern int memory_dev_init(void); > extern int remove_memory_block(unsigned long, struct mem_section *, int); > extern int memory_notify(unsigned long val, void *v); > +extern int memory_isolate_notify(unsigned long val, void *v); > extern struct memory_block *find_memory_block(struct mem_section *); > #define CONFIG_MEM_BLOCK_SIZE (PAGES_PER_SECTION< enum mem_add_context { BOOT, HOTPLUG }; > Index: b/mm/page_alloc.c > =================================================================== > --- a/mm/page_alloc.c > +++ b/mm/page_alloc.c > @@ -48,6 +48,7 @@ > #include > #include > #include > +#include > #include > > #include > @@ -4985,23 +4986,55 @@ void set_pageblock_flags_group(struct pa > int set_migratetype_isolate(struct page *page) > { > struct zone *zone; > - unsigned long flags; > + unsigned long flags, pfn, iter; > + long immobile = 0; So, the count in the structure is unsigned long, but long here. Why the difference in types? > + struct memory_isolate_notify arg; > + int notifier_ret; > int ret = -EBUSY; > int zone_idx; > > zone = page_zone(page); > zone_idx = zone_idx(zone); > + > + pfn = page_to_pfn(page); > + arg.start_addr = (unsigned long)page_address(page); > + arg.nr_pages = pageblock_nr_pages; > + arg.pages_found = 0; > + > spin_lock_irqsave(&zone->lock, flags); > /* > * In future, more migrate types will be able to be isolation target. > */ > - if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE && > - zone_idx != ZONE_MOVABLE) > - goto out; > - set_pageblock_migratetype(page, MIGRATE_ISOLATE); > - move_freepages_block(zone, page, MIGRATE_ISOLATE); > - ret = 0; > -out: > + do { > + if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE && > + zone_idx == ZONE_MOVABLE) { So, this condition requires the zone be MOVABLE and the migrate type be movable. That prevents MIGRATE_RESERVE regions in ZONE_MOVABLE being off-lined even though they can likely be off-lined. It also prevents MIGRATE_MOVABLE sections in other zones being off-lined. Did you mean || here instead of && ? Might want to expand the comment explaining this condition instead of leaving it in the old location which is confusing. > + ret = 0; > + break; > + } Why do you wrap all this in a do {} while(0) instead of preserving the out: label and using goto? > + > + /* > + * If all of the pages in a zone are used by a balloon, > + * the range can be still be isolated. The balloon will > + * free these pages from the memory notifier chain. > + */ > + notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg); > + notifier_ret = notifier_to_errno(ret); > + if (notifier_ret || !arg.pages_found) > + break; > + > + for (iter = pfn; iter < (pfn + pageblock_nr_pages); iter++) > + if (page_count(pfn_to_page(iter))) > + immobile++; > + > + if (arg.pages_found == immobile) and here you compare a signed with an unsigned type. Probably harmless but why do it? > + ret = 0; > + } while (0); > + So the out label would go here and you'd get rid of the do {} while(0) loop. > + if (!ret) { > + set_pageblock_migratetype(page, MIGRATE_ISOLATE); > + move_freepages_block(zone, page, MIGRATE_ISOLATE); > + } > + > spin_unlock_irqrestore(&zone->lock, flags); > if (!ret) > drain_all_pages(); > -- Mel Gorman Part-time Phd Student Linux Technology Center University of Limerick IBM Dublin Software Lab -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/