[dm-devel] [PATCH] block: Set stacked device's bounce_gfp same as that of the underlying device

[Malahal Naineni]

Ops, my bad for missing the header. Here is the header:

Stacked device (dm-multipath etc) doen't get its bounce_gfp set to that
of the bottom device. The stacked device needs call
blk_queue_bounce_limit() with correct limits.bounce_pfn or move the
bounce_gfp field from the request_queue to request_limits structure.
This patch acomplishes the latter. The manifestation of this issue is
system panic with multipath devices attached to HBA's with
BLK_BOUNCE_ISA

Signed-off-by: Malahal Naineni (malahal at us.ibm.com)


Malahal Naineni [malahal at us.ibm.com] wrote:
> diff -r 5d97698a7bc5 -r 8305d08f6ecf block/blk-settings.c
> --- a/block/blk-settings.c	Wed Aug 25 13:44:33 2010 -0700
> +++ b/block/blk-settings.c	Wed Aug 25 13:44:33 2010 -0700
> @@ -126,6 +126,7 @@ void blk_set_default_limits(struct queue
>  	lim->io_opt = 0;
>  	lim->misaligned = 0;
>  	lim->no_cluster = 0;
> +	lim->bounce_gfp = 0;
>  }
>  EXPORT_SYMBOL(blk_set_default_limits);
> 
> @@ -204,7 +205,7 @@ void blk_queue_bounce_limit(struct reque
>  	unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
>  	int dma = 0;
> 
> -	q->bounce_gfp = GFP_NOIO;
> +	q->limits.bounce_gfp = GFP_NOIO;
>  #if BITS_PER_LONG == 64
>  	/*
>  	 * Assume anything <= 4GB can be handled by IOMMU.  Actually
> @@ -221,7 +222,7 @@ void blk_queue_bounce_limit(struct reque
>  #endif
>  	if (dma) {
>  		init_emergency_isa_pool();
> -		q->bounce_gfp = GFP_NOIO | GFP_DMA;
> +		q->limits.bounce_gfp = GFP_NOIO | GFP_DMA;
>  		q->limits.bounce_pfn = b_pfn;
>  	}
>  }
> @@ -549,6 +550,7 @@ int blk_stack_limits(struct queue_limits
> 
>  	t->no_cluster |= b->no_cluster;
>  	t->discard_zeroes_data &= b->discard_zeroes_data;
> +	t->bounce_gfp |= b->bounce_gfp;
> 
>  	/* Physical block size a multiple of the logical block size? */
>  	if (t->physical_block_size & (t->logical_block_size - 1)) {
> diff -r 5d97698a7bc5 -r 8305d08f6ecf block/scsi_ioctl.c
> --- a/block/scsi_ioctl.c	Wed Aug 25 13:44:33 2010 -0700
> +++ b/block/scsi_ioctl.c	Wed Aug 25 13:44:33 2010 -0700
> @@ -436,7 +436,7 @@ int sg_scsi_ioctl(struct request_queue *
> 
>  	bytes = max(in_len, out_len);
>  	if (bytes) {
> -		buffer = kzalloc(bytes, q->bounce_gfp | GFP_USER| __GFP_NOWARN);
> +		buffer = kzalloc(bytes, q->limits.bounce_gfp | GFP_USER| __GFP_NOWARN);
>  		if (!buffer)
>  			return -ENOMEM;
> 
> diff -r 5d97698a7bc5 -r 8305d08f6ecf drivers/ata/libata-scsi.c
> --- a/drivers/ata/libata-scsi.c	Wed Aug 25 13:44:33 2010 -0700
> +++ b/drivers/ata/libata-scsi.c	Wed Aug 25 13:44:33 2010 -0700
> @@ -1140,7 +1140,7 @@ static int ata_scsi_dev_config(struct sc
>  					 ATA_DMA_PAD_SZ - 1);
> 
>  		/* configure draining */
> -		buf = kmalloc(ATAPI_MAX_DRAIN, q->bounce_gfp | GFP_KERNEL);
> +		buf = kmalloc(ATAPI_MAX_DRAIN, q->limits.bounce_gfp | GFP_KERNEL);
>  		if (!buf) {
>  			ata_dev_printk(dev, KERN_ERR,
>  				       "drain buffer allocation failed\n");
> diff -r 5d97698a7bc5 -r 8305d08f6ecf fs/bio-integrity.c
> --- a/fs/bio-integrity.c	Wed Aug 25 13:44:33 2010 -0700
> +++ b/fs/bio-integrity.c	Wed Aug 25 13:44:33 2010 -0700
> @@ -413,7 +413,7 @@ int bio_integrity_prep(struct bio *bio)
> 
>  	/* Allocate kernel buffer for protection data */
>  	len = sectors * blk_integrity_tuple_size(bi);
> -	buf = kmalloc(len, GFP_NOIO | __GFP_NOFAIL | q->bounce_gfp);
> +	buf = kmalloc(len, GFP_NOIO | __GFP_NOFAIL | q->limits.bounce_gfp);
>  	if (unlikely(buf == NULL)) {
>  		printk(KERN_ERR "could not allocate integrity buffer\n");
>  		return -EIO;
> diff -r 5d97698a7bc5 -r 8305d08f6ecf fs/bio.c
> --- a/fs/bio.c	Wed Aug 25 13:44:33 2010 -0700
> +++ b/fs/bio.c	Wed Aug 25 13:44:33 2010 -0700
> @@ -871,7 +871,7 @@ struct bio *bio_copy_user_iov(struct req
> 
>  			i++;
>  		} else {
> -			page = alloc_page(q->bounce_gfp | gfp_mask);
> +			page = alloc_page(q->limits.bounce_gfp | gfp_mask);
>  			if (!page) {
>  				ret = -ENOMEM;
>  				break;
> diff -r 5d97698a7bc5 -r 8305d08f6ecf include/linux/blkdev.h
> --- a/include/linux/blkdev.h	Wed Aug 25 13:44:33 2010 -0700
> +++ b/include/linux/blkdev.h	Wed Aug 25 13:44:33 2010 -0700
> @@ -248,6 +248,11 @@ struct queue_limits {
>  	unsigned char		discard_misaligned;
>  	unsigned char		no_cluster;
>  	signed char		discard_zeroes_data;
> +
> +	/*
> +	 * queue needs bounce pages for pages above this limit
> +	 */
> +	gfp_t			bounce_gfp;
>  };
> 
>  struct request_queue
> @@ -298,11 +303,6 @@ struct request_queue
>  	void			*queuedata;
> 
>  	/*
> -	 * queue needs bounce pages for pages above this limit
> -	 */
> -	gfp_t			bounce_gfp;
> -
> -	/*
>  	 * various queue flags, see QUEUE_* below
>  	 */
>  	unsigned long		queue_flags;
> diff -r 5d97698a7bc5 -r 8305d08f6ecf mm/bounce.c
> --- a/mm/bounce.c	Wed Aug 25 13:44:33 2010 -0700
> +++ b/mm/bounce.c	Wed Aug 25 13:44:33 2010 -0700
> @@ -207,7 +207,7 @@ static void __blk_queue_bounce(struct re
> 
>  		to = bio->bi_io_vec + i;
> 
> -		to->bv_page = mempool_alloc(pool, q->bounce_gfp);
> +		to->bv_page = mempool_alloc(pool, q->limits.bounce_gfp);
>  		to->bv_len = from->bv_len;
>  		to->bv_offset = from->bv_offset;
>  		inc_zone_page_state(to->bv_page, NR_BOUNCE);
> @@ -282,7 +282,7 @@ void blk_queue_bounce(struct request_que
>  	 * to or bigger than the highest pfn in the system -- in that case,
>  	 * don't waste time iterating over bio segments
>  	 */
> -	if (!(q->bounce_gfp & GFP_DMA)) {
> +	if (!(q->limits.bounce_gfp & GFP_DMA)) {
>  		if (queue_bounce_pfn(q) >= blk_max_pfn)
>  			return;
>  		pool = page_pool;
> 
> --
> dm-devel mailing list
> dm-devel at redhat.com
> https://www.redhat.com/mailman/listinfo/dm-devel