[PATCH v3 0/3] Account reserved memory when allocating system hash

[Andrew Morton]

On Mon, 29 Aug 2016 18:36:47 +0530 Srikar Dronamraju <srikar at linux.vnet.ibm.com> wrote:

> Fadump kernel reserves large chunks of memory even before the pages are
> initialised. This could mean memory that corresponds to several nodes might
> fall in memblock reserved regions.
> 
> Kernels compiled with CONFIG_DEFERRED_STRUCT_PAGE_INIT will initialise
> only certain size memory per node. The certain size takes into account
> the dentry and inode cache sizes. However such a kernel when booting a
> secondary 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
> 
> This patchset solves this problem by accounting the size of reserved memory
> when calculating the size of large system hashes.
> 
> While this patchset applies on v4.8-rc3, it cannot be tested on v4.8-rc3
> because of http://lkml.kernel.org/r/20160829093844.GA2592@linux.vnet.ibm.com
> However it has been tested on v4.7/v4.6 and v4.4

That looks like a pretty serious regression.

I'll grab the patchset anyway.  It will come good when we fix that kswapd
thing.