Re: [PATCH V2 1/2] mm/page_alloc: Replace set_dma_reserve to set_memory_reserve
flat view
From: Vlastimil Babka <hidden>
Date: 2016-08-05 15:32:00
Also in:
linux-mm
On 08/04/2016 07:12 PM, Srikar Dronamraju wrote:
quoted hunk ↗ jump to hunk
Expand the scope of the existing dma_reserve to accommodate other memory reserves too. Accordingly rename variable dma_reserve to nr_memory_reserve. set_memory_reserve also takes a new parameter that helps to identify if the current value needs to be incremented. Suggested-by: Mel Gorman <redacted> Signed-off-by: Srikar Dronamraju <redacted> --- arch/x86/kernel/e820.c | 2 +- include/linux/mm.h | 2 +- mm/page_alloc.c | 20 ++++++++++++-------- 3 files changed, 14 insertions(+), 10 deletions(-)diff --git a/arch/x86/kernel/e820.c b/arch/x86/kernel/e820.c index 621b501..d935983 100644 --- a/arch/x86/kernel/e820.c +++ b/arch/x86/kernel/e820.c@@ -1188,6 +1188,6 @@ void __init memblock_find_dma_reserve(void) nr_free_pages += end_pfn - start_pfn; } - set_dma_reserve(nr_pages - nr_free_pages); + set_memory_reserve(nr_pages - nr_free_pages, false); #endif }diff --git a/include/linux/mm.h b/include/linux/mm.h index 8f468e0..c884ffb 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h@@ -1886,7 +1886,7 @@ extern int __meminit __early_pfn_to_nid(unsigned long pfn, struct mminit_pfnnid_cache *state); #endif -extern void set_dma_reserve(unsigned long new_dma_reserve); +extern void set_memory_reserve(unsigned long nr_reserve, bool inc); extern void memmap_init_zone(unsigned long, int, unsigned long, unsigned long, enum memmap_context); extern void setup_per_zone_wmarks(void);diff --git a/mm/page_alloc.c b/mm/page_alloc.c index c1069ef..a154c2f 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c@@ -253,7 +253,7 @@ int watermark_scale_factor = 10; static unsigned long __meminitdata nr_kernel_pages; static unsigned long __meminitdata nr_all_pages; -static unsigned long __meminitdata dma_reserve; +static unsigned long __meminitdata nr_memory_reserve; #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];@@ -5493,10 +5493,10 @@ static void __paginginit free_area_init_core(struct pglist_data *pgdat) } /* Account for reserved pages */ - if (j == 0 && freesize > dma_reserve) { - freesize -= dma_reserve; + if (j == 0 && freesize > nr_memory_reserve) {
Will this really work (together with patch 2) as intended? This j == 0 means that we are doing this only for the first zone, which is ZONE_DMA (or ZONE_DMA32) on node 0 on many systems. I.e. I don't think it's really true that "dma_reserve has nothing to do with DMA or ZONE_DMA". This zone will have limited amount of memory, so the "freesize > nr_memory_reserve" will easily be false once you set this to many gigabytes, so in fact nothing will get subtracted. On the other hand if the kernel has both CONFIG_ZONE_DMA and CONFIG_ZONE_DMA32 disabled, then j == 0 will be true for ZONE_NORMAL. This zone might be present on multiple nodes (unless they are configured as movable) and then the value intended to be global will be subtracted from several nodes. I don't know what's the exact ppc64 situation here, perhaps there are indeed no DMA/DMA32 zones, and the fadump kernel only uses one node, so it works in the end, but it doesn't seem much robust to me?
quoted hunk ↗ jump to hunk
+ freesize -= nr_memory_reserve; printk(KERN_DEBUG " %s zone: %lu pages reserved\n", - zone_names[0], dma_reserve); + zone_names[0], nr_memory_reserve); } if (!is_highmem_idx(j))@@ -6186,8 +6186,9 @@ void __init mem_init_print_info(const char *str) } /** - * set_dma_reserve - set the specified number of pages reserved in the first zone - * @new_dma_reserve: The number of pages to mark reserved + * set_memory_reserve - set number of pages reserved in the first zone + * @nr_reserve: The number of pages to mark reserved + * @inc: true increment to existing value; false set new value. * * The per-cpu batchsize and zone watermarks are determined by managed_pages. * In the DMA zone, a significant percentage may be consumed by kernel image@@ -6196,9 +6197,12 @@ void __init mem_init_print_info(const char *str) * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and * smaller per-cpu batchsize. */ -void __init set_dma_reserve(unsigned long new_dma_reserve) +void __init set_memory_reserve(unsigned long nr_reserve, bool inc) { - dma_reserve = new_dma_reserve; + if (inc) + nr_memory_reserve += nr_reserve; + else + nr_memory_reserve = nr_reserve; } void __init free_area_init(unsigned long *zones_size)