Re: [PATCH] fadump: Register the memory reserved by fadump

[Mel Gorman <mgorman@techsingularity.net>]

On Thu, Aug 04, 2016 at 07:12:45PM +0530, Srikar Dronamraju wrote:
> Fadump kernel reserves large chunks of memory even before the pages are
> initialized. This could mean memory that corresponds to several nodes might
> fall in memblock reserved regions.
> 
> Kernels compiled with CONFIG_DEFERRED_STRUCT_PAGE_INIT will initialize
> only certain size memory per node. The certain size takes into account
> the dentry and inode cache sizes. Currently the cache sizes are
> calculated based on the total system memory including the reserved
> memory. However such a kernel when booting the same kernel as fadump
> kernel will not be able to allocate the required amount of memory to
> suffice for the dentry and inode caches. This results in crashes like
> the below on large systems such as 32 TB systems.
> 
> Dentry cache hash table entries: 536870912 (order: 16, 4294967296 bytes)
> vmalloc: allocation failure, allocated 4097114112 of 17179934720 bytes
> swapper/0: page allocation failure: order:0, mode:0x2080020(GFP_ATOMIC)
> CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.6-master+ #3
> Call Trace:
> [c00000000108fb10] [c0000000007fac88] dump_stack+0xb0/0xf0 (unreliable)
> [c00000000108fb50] [c000000000235264] warn_alloc_failed+0x114/0x160
> [c00000000108fbf0] [c000000000281484] __vmalloc_node_range+0x304/0x340
> [c00000000108fca0] [c00000000028152c] __vmalloc+0x6c/0x90
> [c00000000108fd40] [c000000000aecfb0]
> alloc_large_system_hash+0x1b8/0x2c0
> [c00000000108fe00] [c000000000af7240] inode_init+0x94/0xe4
> [c00000000108fe80] [c000000000af6fec] vfs_caches_init+0x8c/0x13c
> [c00000000108ff00] [c000000000ac4014] start_kernel+0x50c/0x578
> [c00000000108ff90] [c000000000008c6c] start_here_common+0x20/0xa8
> 
> Register the memory reserved by fadump, so that the cache sizes are
> calculated based on the free memory (i.e Total memory - reserved
> memory).
> 
> Suggested-by: Mel Gorman <mgorman@techsingularity.net>

I didn't suggest this specifically. While it happens to be safe on ppc64,
it potentially overwrites any future caller of set_dma_reserve. While the
only other one is for the e820 map, it may be better to change the API
to inc_dma_reserve?

It's also unfortunate that it's called dma_reserve because it has
nothing to do with DMA or ZONE_DMA. inc_kernel_reserve may be more
appropriate?

-- 
Mel Gorman
SUSE Labs