Re: bioset changes in 4.18 broke aha1542

[Ondrej Zary]

On Thursday 18 October 2018 22:28:35 Jens Axboe wrote:
> On 10/18/18 2:22 PM, Ondrej Zary wrote:
> > On Thursday 18 October 2018 22:10:31 Jens Axboe wrote:
> >> On 10/18/18 2:04 PM, Ondrej Zary wrote:
> >>> On Thursday 18 October 2018 21:59:09 Jens Axboe wrote:
> >>>> On 10/18/18 1:55 PM, Ondrej Zary wrote:
> >>>>> On Thursday 18 October 2018 20:58:57 Jens Axboe wrote:
> >>>>>> On 10/18/18 12:34 PM, Ondrej Zary wrote:
> >>>>>>> Hello,
> >>>>>>> aha1542 works fine in 4.17 but crashes in 4.18. It's hard to bisect because
> >>>>>>> of many commits that don't compile.
> >>>>>>> # only skipped commits left to test
> >>>>>>> # possible first bad commit: [52190f8abe7f2bf2b4e5f9760cbcc1427ca2136b] fs: convert block_dev.c to bioset_init()
> >>>>>>> # possible first bad commit: [a47a28b74a5c7c27bf621276b85ad6c124651236] target: convert to bioset_init()/mempool_init()
> >>>>>>> # possible first bad commit: [6f1c819c219f7841079f0f43ab62727a55b0d849] dm: convert to bioset_init()/mempool_init()
> >>>>>>> # possible first bad commit: [afeee514ce7f4cab605beedd03be71ebaf0c5fc8] md: convert to bioset_init()/mempool_init()
> >>>>>>> # possible first bad commit: [d19936a26658a7a53edd5619d631ee2c2c3151a2] bcache: convert to bioset_init()/mempool_init()
> >>>>>>> # possible first bad commit: [b906bbb6997785d9ea0bd3f5585537afa6257c43] lightnvm: convert to bioset_init()/mempool_init()
> >>>>>>>
> >>>>>>> Testing manually, a47a28b74a5c7c27bf621276b85ad6c124651236 works.
> >>>>>>> 52190f8abe7f2bf2b4e5f9760cbcc1427ca2136b does not compile
> >>>>>>> 8ac9f7c1fd1d342e82ddf078425423b050652ba0 does not compile
> >>>>>>> e292d7bc63c8f2adb3dfda27910e805f1b6557f9 does not compile
> >>>>>>> dad08527525f9a8ac9c7f278864c65f94bc5e9b3 does not compile
> >>>>>>> 943cf9f3ca16133dbd00f9a4cbfea46512fcb0e8 works
> >>>>>>> ..
> >>>>>>> fedc3abe7bd2dcc4c80bcf3cff8708a3908d8219 works
> >>>>>>> 04c4950d5b373ba712d928592e05e73510785bca crashes
> >>>>>>
> >>>>>> It looks like the ISA bioset pool isn't being initialized. You should
> >>>>>> have messages like this in your dmesg:
> >>>>>>
> >>>>>> isa pool size: 16 pages
> >>>>>>
> >>>>>> (which you do), but also something on the bioset section. Do you have
> >>>>>> this one:
> >>>>>>
> >>>>>> pool size: 64 pages
> >>>>>>
> >>>>>> as well?
> >>>>>>
> >>>>>
> >>>>> No, it's not there.
> >>>>
> >>>> Can you attach your .config? I'm guessing CONFIG_HIGHMEM* isn't set.
> >>>>
> >>>
> >>> It is.
> >>
> >> Puzzled... Does this work?
> >>
> >>
> >> diff --git a/block/bounce.c b/block/bounce.c
> >> index b30071ac4ec6..49564a1bfd22 100644
> >> --- a/block/bounce.c
> >> +++ b/block/bounce.c
> >> @@ -35,10 +35,6 @@ static mempool_t page_pool, isa_page_pool;
> >>  static __init int init_emergency_pool(void)
> >>  {
> >>  	int ret;
> >> -#if defined(CONFIG_HIGHMEM) && !defined(CONFIG_MEMORY_HOTPLUG)
> >> -	if (max_pfn <= max_low_pfn)
> >> -		return 0;
> >> -#endif
> >>  
> >>  	ret = mempool_init_page_pool(&page_pool, POOL_SIZE, 0);
> >>  	BUG_ON(ret);
> >>
> > 
> > Yes, it does! 
> > bounce: pool size: 64 pages
> > and aha1542 works.
> > 
> > Also added printks for pfn:
> > max_pfn=65520, max_low_pfn=65520
> 
> This should be a better fix, though I'm still puzzled why we need
> it now. Can you test this one?
> 
> 
> diff --git a/block/bounce.c b/block/bounce.c
> index b30071ac4ec6..1356a2f4aae2 100644
> --- a/block/bounce.c
> +++ b/block/bounce.c
> @@ -31,6 +31,24 @@
>  static struct bio_set bounce_bio_set, bounce_bio_split;
>  static mempool_t page_pool, isa_page_pool;
>  
> +static __init void init_bounce_bioset(void)
> +{
> +	static bool bounce_bs_setup;
> +	int ret;
> +
> +	if (bounce_bs_setup)
> +		return;
> +
> +	ret = bioset_init(&bounce_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
> +	BUG_ON(ret);
> +	if (bioset_integrity_create(&bounce_bio_set, BIO_POOL_SIZE))
> +		BUG_ON(1);
> +
> +	ret = bioset_init(&bounce_bio_split, BIO_POOL_SIZE, 0, 0);
> +	BUG_ON(ret);
> +	bounce_bs_setup = true;
> +}
> +
>  #if defined(CONFIG_HIGHMEM)
>  static __init int init_emergency_pool(void)
>  {
> @@ -44,14 +62,7 @@ static __init int init_emergency_pool(void)
>  	BUG_ON(ret);
>  	pr_info("pool size: %d pages\n", POOL_SIZE);
>  
> -	ret = bioset_init(&bounce_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
> -	BUG_ON(ret);
> -	if (bioset_integrity_create(&bounce_bio_set, BIO_POOL_SIZE))
> -		BUG_ON(1);
> -
> -	ret = bioset_init(&bounce_bio_split, BIO_POOL_SIZE, 0, 0);
> -	BUG_ON(ret);
> -
> +	init_bounce_bioset();
>  	return 0;
>  }
>  
> @@ -102,6 +113,7 @@ int init_emergency_isa_pool(void)
>  	BUG_ON(ret);
>  
>  	pr_info("isa pool size: %d pages\n", ISA_POOL_SIZE);
> +	init_bounce_bioset();
>  	return 0;
>  }
>  
> 

WARNING: vmlinux.o(.text+0x1a5ad7): Section mismatch in reference from the function init_emergency_isa_pool() to the function .init.text:init_bounce_bioset()
The function init_emergency_isa_pool() references
the function __init init_bounce_bioset().
This is often because init_emergency_isa_pool lacks a __init
annotation or the annotation of init_bounce_bioset is wrong.

And it does not work:
[  122.395558] scsi host2: Adaptec AHA-1542 (SCSI-ID 7) at IO 0x330, IRQ 11, DMA 7
[  122.471853] scsi host2: Adaptec 1542
[  122.484207] bounce: isa pool size: 16 pages
[  122.488722] BUG: unable to handle kernel NULL pointer dereference at 00000fff
[  122.492501] *pde = 00000000
[  122.492501] Oops: 0000 [#1] SMP
[  122.492501] CPU: 0 PID: 51 Comm: kworker/u2:1 Not tainted 4.19.0-rc7+ #296
[  122.492501] Hardware name:  /i440ZX-W977TF, BIOS 4.51 PG 07/12/00
[  122.492501] Workqueue: events_unbound async_run_entry_fn
[  122.492501] EIP: init_bounce_bioset+0x2/0x63

Removing __init from init_bounce_bioset() makes it work.



-- 
Ondrej Zary