Re: linux-next: powerpc le qemu boot failure after merge of the akpm tree

[Mike Rapoport]

(added Andrey Konovalov)

On Thu, Jan 31, 2019 at 07:15:26AM +0100, Christophe Leroy wrote:
> 
> Le 31/01/2019 à 07:06, Stephen Rothwell a écrit :
> >Hi all,
> >
> >On Thu, 31 Jan 2019 16:38:54 +1100 Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx> wrote:
> >>
> >>[I am guessing that is is something in Andrew's tree that has caused
> >>this.]
> >>
> >>My qemu boot of the powerpc pseries_le_defconfig config failed like this:
> >>
> >>htab_hash_mask    = 0x1ffff
> >>-----------------------------------------------------
> >>numa:   NODE_DATA [mem 0x7ffe7000-0x7ffebfff]
> >>Kernel panic - not syncing: sparse_buffer_init: Failed to allocate 2147483648 bytes align=0x10000 nid=0 from=fffffffffffffff
> >>CPU: 0 PID: 0 Comm: swapper Not tainted 5.0.0-rc4 #2
> >>Call Trace:
> >>[c00000000105bbd0] [c000000000b1345c] dump_stack+0xb0/0xf4 (unreliable)
> >>[c00000000105bc10] [c000000000111120] panic+0x168/0x3b8
> >>[c00000000105bcb0] [c000000000e701c8] sparse_init_nid+0x178/0x550
> >>[c00000000105bd70] [c000000000e709b4] sparse_init+0x210/0x238
> >>[c00000000105bdb0] [c000000000e468f4] initmem_init+0x1e0/0x260
> >>[c00000000105be80] [c000000000e3b9b0] setup_arch+0x354/0x3d4
> >>[c00000000105bef0] [c000000000e33afc] start_kernel+0x98/0x648
> >>[c00000000105bf90] [c00000000000b270] start_here_common+0x1c/0x52c
> >
> >A quick bisect leads to this:
> >
> >1c3c9328cde027eb875ba4692f0a5d66b0afe862 is the first bad commit
> >commit 1c3c9328cde027eb875ba4692f0a5d66b0afe862
> >Author: Mike Rapoport <rppt@xxxxxxxxxxxxx>
> >Date:   Thu Jan 31 10:51:32 2019 +1100
> >
> >     treewide: add checks for the return value of memblock_alloc*()
> >     Add check for the return value of memblock_alloc*() functions and call
> >     panic() in case of error.  The panic message repeats the one used by
> >     panicing memblock allocators with adjustment of parameters to include only
> >     relevant ones.
> >
> >Which is just adding the panic we hit.  So, presumably, the bug is in a
> >preceding patch :-(
> >
> >I have left the kernel not booting for today.
> >
> 
> No I think the error is really in that patch, see my other mail.
> 
> See https://elixir.bootlin.com/linux/v5.0-rc4/source/mm/memblock.c#L1455,
> memblock_alloc_try_nid_raw() is not supposed to panic, so the last hunk of
> this patch should be reverted.
> 
> Found in total three problematic hunks in that patch:
> 
> @@ -48,6 +53,11 @@ static phys_addr_t __init kasan_alloc_raw_page(int node)
>  	void *p = memblock_alloc_try_nid_raw(PAGE_SIZE, PAGE_SIZE,
>  						__pa(MAX_DMA_ADDRESS),
>  						MEMBLOCK_ALLOC_KASAN, node);
> +	if (!p)
> +		panic("%s: Failed to allocate %lu bytes align=0x%lx nid=%d from=%llx\n",
> +		      __func__, PAGE_SIZE, PAGE_SIZE, node,
> +		      __pa(MAX_DMA_ADDRESS));
> +
>  	return __pa(p);
>  }
 
I've looked more closely to the code that uses this function and it does
not seem to handle allocation error.
I can replace the panic with WARN(), but I think that panic() here is
appropriate.

Andrey, can you comment?


> @@ -211,6 +211,9 @@ static int __init iob_init(struct device_node *dn)
>  	iob_l2_base = memblock_alloc_try_nid_raw(1UL << 21, 1UL << 21,
>  					MEMBLOCK_LOW_LIMIT, 0x80000000,
>  					NUMA_NO_NODE);
> +	if (!iob_l2_base)
> +		panic("%s: Failed to allocate %lu bytes align=0x%lx max_addr=%x\n",
> +		      __func__, 1UL << 21, 1UL << 21, 0x80000000);
> 
>  	pr_info("IOBMAP L2 allocated at: %p\n", iob_l2_base);
 
This one is actually fixes my own mistake from one of the previous patches
that converted memblock_alloc_base() to memblock_alloc_try_nid_raw() without
adding the panic() (commit 47e382eb08cfa0199c4ea9f9cc73f1b48a3a4b1d
"powerpc: prefer memblock APIs returning virtual address")
 
> @@ -425,6 +436,10 @@ static void __init sparse_buffer_init(unsigned long
> size, int nid)
>  		memblock_alloc_try_nid_raw(size, PAGE_SIZE,
>  						__pa(MAX_DMA_ADDRESS),
>  						MEMBLOCK_ALLOC_ACCESSIBLE, nid);
> +	if (!sparsemap_buf)
> +		panic("%s: Failed to allocate %lu bytes align=0x%lx nid=%d from=%lx\n",
> +		      __func__, size, PAGE_SIZE, nid, __pa(MAX_DMA_ADDRESS));
> +
>  	sparsemap_buf_end = sparsemap_buf + size;
>  }
 
This hunk was not needed as sparse can deal with this allocation failure.

Andrew, can you please add the below patch to as a fixup to "treewide: add
checks for the return value of memblock_alloc*()"?