remove block layer bounce buffering v2

remove block layer bounce buffering v2

[Christoph Hellwig]

Hi all,

the block layer bounce buffering from the early days of highmem keeps
being a wart in the block layer despite usage by only four drivers,
all through the SCSI layer.  One of them is an old PIO-only ISA
card, two are parallel port adapters, and the fourth is the usb-storage
driver.

This series makes the first three depend on !HIGHMEM and for the fourth
rejects the probe only when used on highmem system and the HCD is one
of the few annoying one that does not support DMA.

This is not the most pretty way to do things, but none of the driver
maintainers stepped up to convert them to properly kmapping buffers
before doing PIO or MMIO operations it does shift the burden to these
drivers and might get them to actually move that way if there are any
users of the drivers, or in case of usb-storage configurations left.

A nice benefit is that we remove the per-numa node stat used for the
number of bounce buffer operations.

Changes since v1:
 - return -EINVAL from usb-storage ->probe
 - improve the error message in usb-storage
 - improve comments in usb-storage

Diffstat:
 b/arch/mips/configs/gcw0_defconfig  |    1 
 b/block/Makefile                    |    1 
 b/block/blk-map.c                   |    5 
 b/block/blk-mq.c                    |    2 
 b/block/blk-settings.c              |    5 
 b/block/blk.h                       |   17 --
 b/drivers/base/node.c               |    2 
 b/drivers/scsi/Kconfig              |    3 
 b/drivers/scsi/aha152x.c            |    1 
 b/drivers/scsi/imm.c                |    1 
 b/drivers/scsi/ppa.c                |    1 
 b/drivers/scsi/scsi_lib.c           |    3 
 b/drivers/usb/storage/usb.c         |   20 +-
 b/fs/proc/meminfo.c                 |    3 
 b/include/linux/blk_types.h         |    1 
 b/include/linux/blkdev.h            |    5 
 b/include/linux/mmzone.h            |    1 
 b/include/scsi/scsi_host.h          |    2 
 b/include/trace/events/block.h      |   15 --
 b/include/uapi/linux/blktrace_api.h |    2 
 b/kernel/trace/blktrace.c           |    9 -
 b/mm/show_mem.c                     |    4 
 block/bounce.c                      |  267 ------------------------------------
 23 files changed, 24 insertions(+), 347 deletions(-)

[PATCH 1/7] scsi: make aha152x depend on !HIGHMEM

[Christoph Hellwig]

This is one of the last drivers depending on the block layer bounce
buffering code.  Restrict it to run on non-highmem configs so that the
bounce buffering code can be removed.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
---
 drivers/scsi/Kconfig   | 1 +
 drivers/scsi/aha152x.c | 1 -
 2 files changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/scsi/Kconfig b/drivers/scsi/Kconfig
index 5a3c670aec27..c89e6991e934 100644
--- a/drivers/scsi/Kconfig
+++ b/drivers/scsi/Kconfig
@@ -403,6 +403,7 @@ config SCSI_ACARD
 config SCSI_AHA152X
 	tristate "Adaptec AHA152X/2825 support"
 	depends on ISA && SCSI
+	depends on !HIGHMEM
 	select SCSI_SPI_ATTRS
 	select CHECK_SIGNATURE
 	help
diff --git a/drivers/scsi/aha152x.c b/drivers/scsi/aha152x.c
index 4276f868cd91..e94c0a19c435 100644
--- a/drivers/scsi/aha152x.c
+++ b/drivers/scsi/aha152x.c
@@ -746,7 +746,6 @@ struct Scsi_Host *aha152x_probe_one(struct aha152x_setup *setup)
 	/* need to have host registered before triggering any interrupt */
 	list_add_tail(&HOSTDATA(shpnt)->host_list, &aha152x_host_list);
 
-	shpnt->no_highmem = true;
 	shpnt->io_port   = setup->io_port;
 	shpnt->n_io_port = IO_RANGE;
 	shpnt->irq       = setup->irq;
-- 
2.47.2

Re: [PATCH 1/7] scsi: make aha152x depend on !HIGHMEM

[John Garry]

On 05/05/2025 09:11, Christoph Hellwig wrote:
> This is one of the last drivers depending on the block layer bounce
> buffering code.  Restrict it to run on non-highmem configs so that the
> bounce buffering code can be removed.
> 
> Signed-off-by: Christoph Hellwig<hch@lst.de>
> Reviewed-by: Hannes Reinecke<hare@suse.de>

Reviewed-by: John Garry <john.g.garry@oracle.com>

[PATCH 2/7] scsi: make imm depend on !HIGHMEM

[Christoph Hellwig]

This is one of the last drivers depending on the block layer bounce
buffering code.  Restrict it to run on non-highmem configs so that the
bounce buffering code can be removed.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
---
 drivers/scsi/Kconfig | 1 +
 drivers/scsi/imm.c   | 1 -
 2 files changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/scsi/Kconfig b/drivers/scsi/Kconfig
index c89e6991e934..82e0b7eaed4c 100644
--- a/drivers/scsi/Kconfig
+++ b/drivers/scsi/Kconfig
@@ -823,6 +823,7 @@ config SCSI_PPA
 config SCSI_IMM
 	tristate "IOMEGA parallel port (imm - newer drives)"
 	depends on SCSI && PARPORT_PC
+	depends on !HIGHMEM
 	help
 	  This driver supports newer versions of IOMEGA's parallel port ZIP
 	  drive (a 100 MB removable media device).
diff --git a/drivers/scsi/imm.c b/drivers/scsi/imm.c
index 1d4c7310f1a6..0821cf994b98 100644
--- a/drivers/scsi/imm.c
+++ b/drivers/scsi/imm.c
@@ -1224,7 +1224,6 @@ static int __imm_attach(struct parport *pb)
 	host = scsi_host_alloc(&imm_template, sizeof(imm_struct *));
 	if (!host)
 		goto out1;
-	host->no_highmem = true;
 	host->io_port = pb->base;
 	host->n_io_port = ports;
 	host->dma_channel = -1;
-- 
2.47.2

Re: [PATCH 2/7] scsi: make imm depend on !HIGHMEM

[John Garry]

On 05/05/2025 09:11, Christoph Hellwig wrote:
> This is one of the last drivers depending on the block layer bounce
> buffering code.  Restrict it to run on non-highmem configs so that the
> bounce buffering code can be removed.
> 
> Signed-off-by: Christoph Hellwig<hch@lst.de>
> Reviewed-by: Hannes Reinecke<hare@suse.de>
Reviewed-by: John Garry <john.g.garry@oracle.com>

[PATCH 3/7] scsi: make ppa depend on !HIGHMEM

[Christoph Hellwig]

This is one of the last drivers depending on the block layer bounce
buffering code.  Restrict it to run on non-highmem configs so that the
bounce buffering code can be removed.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
---
 drivers/scsi/Kconfig | 1 +
 drivers/scsi/ppa.c   | 1 -
 2 files changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/scsi/Kconfig b/drivers/scsi/Kconfig
index 82e0b7eaed4c..5522310bab8d 100644
--- a/drivers/scsi/Kconfig
+++ b/drivers/scsi/Kconfig
@@ -796,6 +796,7 @@ config SCSI_PPA
 	tristate "IOMEGA parallel port (ppa - older drives)"
 	depends on SCSI && PARPORT_PC
 	depends on HAS_IOPORT
+	depends on !HIGHMEM
 	help
 	  This driver supports older versions of IOMEGA's parallel port ZIP
 	  drive (a 100 MB removable media device).
diff --git a/drivers/scsi/ppa.c b/drivers/scsi/ppa.c
index a06329b47851..1ed3171f1797 100644
--- a/drivers/scsi/ppa.c
+++ b/drivers/scsi/ppa.c
@@ -1104,7 +1104,6 @@ static int __ppa_attach(struct parport *pb)
 	host = scsi_host_alloc(&ppa_template, sizeof(ppa_struct *));
 	if (!host)
 		goto out1;
-	host->no_highmem = true;
 	host->io_port = pb->base;
 	host->n_io_port = ports;
 	host->dma_channel = -1;
-- 
2.47.2

Re: [PATCH 3/7] scsi: make ppa depend on !HIGHMEM

[John Garry]

On 05/05/2025 09:11, Christoph Hellwig wrote:
> This is one of the last drivers depending on the block layer bounce
> buffering code.  Restrict it to run on non-highmem configs so that the
> bounce buffering code can be removed.
> 
> Signed-off-by: Christoph Hellwig<hch@lst.de>
> Reviewed-by: Hannes Reinecke<hare@suse.de>
Reviewed-by: John Garry <john.g.garry@oracle.com>

[PATCH 4/7] usb-storage: reject probe of device one non-DMA HCDs when using highmem

[Christoph Hellwig]

usb-storage is the last user of the block layer bounce buffering now,
and only uses it for HCDs that do not support DMA on highmem configs.

Remove this support and fail the probe so that the block layer bounce
buffering can go away.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
---
 drivers/usb/storage/usb.c | 20 ++++++++++++++------
 1 file changed, 14 insertions(+), 6 deletions(-)

diff --git a/drivers/usb/storage/usb.c b/drivers/usb/storage/usb.c
index d36f3b6992bb..152ee3376550 100644
--- a/drivers/usb/storage/usb.c
+++ b/drivers/usb/storage/usb.c
@@ -1056,13 +1056,20 @@ int usb_stor_probe1(struct us_data **pus,
 		goto BadDevice;
 
 	/*
-	 * Some USB host controllers can't do DMA; they have to use PIO.
-	 * For such controllers we need to make sure the block layer sets
-	 * up bounce buffers in addressable memory.
+	 * Some USB host controllers can't do DMA: They have to use PIO, or they
+	 * have to use a small dedicated local memory area, or they have other
+	 * restrictions on addressable memory.
+	 *
+	 * We can't support these controllers on highmem systems as we don't
+	 * kmap or bounce buffer.
 	 */
-	if (!hcd_uses_dma(bus_to_hcd(us->pusb_dev->bus)) ||
-	    bus_to_hcd(us->pusb_dev->bus)->localmem_pool)
-		host->no_highmem = true;
+	if (IS_ENABLED(CONFIG_HIGHMEM) &&
+	    (!hcd_uses_dma(bus_to_hcd(us->pusb_dev->bus)) ||
+	     bus_to_hcd(us->pusb_dev->bus)->localmem_pool)) {
+		dev_warn(&intf->dev, "USB Mass Storage not supported on this host controller\n");
+		result = -EINVAL;
+		goto release;
+	}
 
 	/* Get the unusual_devs entries and the descriptors */
 	result = get_device_info(us, id, unusual_dev);
@@ -1081,6 +1088,7 @@ int usb_stor_probe1(struct us_data **pus,
 
 BadDevice:
 	usb_stor_dbg(us, "storage_probe() failed\n");
+release:
 	release_everything(us);
 	return result;
 }
-- 
2.47.2

Re: [PATCH 4/7] usb-storage: reject probe of device one non-DMA HCDs when using highmem

[Alan Stern]

On Mon, May 05, 2025 at 10:11:23AM +0200, Christoph Hellwig wrote:
> usb-storage is the last user of the block layer bounce buffering now,
> and only uses it for HCDs that do not support DMA on highmem configs.
> 
> Remove this support and fail the probe so that the block layer bounce
> buffering can go away.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Reviewed-by: Hannes Reinecke <hare@suse.de>

Reviewed-by: Alan Stern <stern@rowland.harvard.edu>

> ---
>  drivers/usb/storage/usb.c | 20 ++++++++++++++------
>  1 file changed, 14 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/usb/storage/usb.c b/drivers/usb/storage/usb.c
> index d36f3b6992bb..152ee3376550 100644
> --- a/drivers/usb/storage/usb.c
> +++ b/drivers/usb/storage/usb.c
> @@ -1056,13 +1056,20 @@ int usb_stor_probe1(struct us_data **pus,
>  		goto BadDevice;
>  
>  	/*
> -	 * Some USB host controllers can't do DMA; they have to use PIO.
> -	 * For such controllers we need to make sure the block layer sets
> -	 * up bounce buffers in addressable memory.
> +	 * Some USB host controllers can't do DMA: They have to use PIO, or they
> +	 * have to use a small dedicated local memory area, or they have other
> +	 * restrictions on addressable memory.
> +	 *
> +	 * We can't support these controllers on highmem systems as we don't
> +	 * kmap or bounce buffer.
>  	 */
> -	if (!hcd_uses_dma(bus_to_hcd(us->pusb_dev->bus)) ||
> -	    bus_to_hcd(us->pusb_dev->bus)->localmem_pool)
> -		host->no_highmem = true;
> +	if (IS_ENABLED(CONFIG_HIGHMEM) &&
> +	    (!hcd_uses_dma(bus_to_hcd(us->pusb_dev->bus)) ||
> +	     bus_to_hcd(us->pusb_dev->bus)->localmem_pool)) {
> +		dev_warn(&intf->dev, "USB Mass Storage not supported on this host controller\n");
> +		result = -EINVAL;
> +		goto release;
> +	}
>  
>  	/* Get the unusual_devs entries and the descriptors */
>  	result = get_device_info(us, id, unusual_dev);
> @@ -1081,6 +1088,7 @@ int usb_stor_probe1(struct us_data **pus,
>  
>  BadDevice:
>  	usb_stor_dbg(us, "storage_probe() failed\n");
> +release:
>  	release_everything(us);
>  	return result;
>  }

[PATCH 5/7] scsi: remove the no_highmem flag in the host

[Christoph Hellwig]

All users are gone now.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
---
 drivers/scsi/scsi_lib.c  | 3 ---
 include/scsi/scsi_host.h | 2 --
 2 files changed, 5 deletions(-)

diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index 0d29470e86b0..39320dccbdd5 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -2000,9 +2000,6 @@ void scsi_init_limits(struct Scsi_Host *shost, struct queue_limits *lim)
 	lim->dma_alignment = max_t(unsigned int,
 		shost->dma_alignment, dma_get_cache_alignment() - 1);
 
-	if (shost->no_highmem)
-		lim->features |= BLK_FEAT_BOUNCE_HIGH;
-
 	/*
 	 * Propagate the DMA formation properties to the dma-mapping layer as
 	 * a courtesy service to the LLDDs.  This needs to check that the buses
diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h
index 26bc23419cfd..c53812b9026f 100644
--- a/include/scsi/scsi_host.h
+++ b/include/scsi/scsi_host.h
@@ -670,8 +670,6 @@ struct Scsi_Host {
 	/* The transport requires the LUN bits NOT to be stored in CDB[1] */
 	unsigned no_scsi2_lun_in_cdb:1;
 
-	unsigned no_highmem:1;
-
 	/*
 	 * Optional work queue to be utilized by the transport
 	 */
-- 
2.47.2

Re: [PATCH 5/7] scsi: remove the no_highmem flag in the host

[John Garry]

On 05/05/2025 09:11, Christoph Hellwig wrote:
> All users are gone now.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> Reviewed-by: Hannes Reinecke <hare@suse.de>

Reviewed-by: John Garry <john.g.garry@oracle.com>

> ---
>   drivers/scsi/scsi_lib.c  | 3 ---
>   include/scsi/scsi_host.h | 2 --
>   2 files changed, 5 deletions(-)
> 
> diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
> index 0d29470e86b0..39320dccbdd5 100644
> --- a/drivers/scsi/scsi_lib.c
> +++ b/drivers/scsi/scsi_lib.c
> @@ -2000,9 +2000,6 @@ void scsi_init_limits(struct Scsi_Host *shost, struct queue_limits *lim)
>   	lim->dma_alignment = max_t(unsigned int,
>   		shost->dma_alignment, dma_get_cache_alignment() - 1);
>   
> -	if (shost->no_highmem)
> -		lim->features |= BLK_FEAT_BOUNCE_HIGH;
> -
>   	/*
>   	 * Propagate the DMA formation properties to the dma-mapping layer as
>   	 * a courtesy service to the LLDDs.  This needs to check that the buses
> diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h
> index 26bc23419cfd..c53812b9026f 100644
> --- a/include/scsi/scsi_host.h
> +++ b/include/scsi/scsi_host.h
> @@ -670,8 +670,6 @@ struct Scsi_Host {
>   	/* The transport requires the LUN bits NOT to be stored in CDB[1] */
>   	unsigned no_scsi2_lun_in_cdb:1;
>   
> -	unsigned no_highmem:1;

I note that there are some ancient references to highmem_io around. I 
guess that this was the previous name.

> -
>   	/*
>   	 * Optional work queue to be utilized by the transport
>   	 */

[PATCH 6/7] block: remove bounce buffering support

[Christoph Hellwig]

The block layer bounce buffering support is unused now, remove it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
---
 arch/mips/configs/gcw0_defconfig  |   1 -
 block/Makefile                    |   1 -
 block/blk-map.c                   |   5 +-
 block/blk-mq.c                    |   2 -
 block/blk-settings.c              |   5 -
 block/blk.h                       |  17 --
 block/bounce.c                    | 267 ------------------------------
 include/linux/blk_types.h         |   1 -
 include/linux/blkdev.h            |   5 +-
 include/trace/events/block.h      |  15 --
 include/uapi/linux/blktrace_api.h |   2 +-
 kernel/trace/blktrace.c           |   9 -
 12 files changed, 3 insertions(+), 327 deletions(-)
 delete mode 100644 block/bounce.c

diff --git a/arch/mips/configs/gcw0_defconfig b/arch/mips/configs/gcw0_defconfig
index bc1ef66e3999..8b7ad877e07a 100644
--- a/arch/mips/configs/gcw0_defconfig
+++ b/arch/mips/configs/gcw0_defconfig
@@ -13,7 +13,6 @@ CONFIG_MIPS_CMDLINE_DTB_EXTEND=y
 CONFIG_MODULES=y
 CONFIG_MODULE_UNLOAD=y
 # CONFIG_BLK_DEV_BSG is not set
-# CONFIG_BOUNCE is not set
 CONFIG_NET=y
 CONFIG_PACKET=y
 CONFIG_UNIX=y
diff --git a/block/Makefile b/block/Makefile
index 3a941dc0d27f..36033c0f07bc 100644
--- a/block/Makefile
+++ b/block/Makefile
@@ -11,7 +11,6 @@ obj-y		:= bdev.o fops.o bio.o elevator.o blk-core.o blk-sysfs.o \
 			genhd.o ioprio.o badblocks.o partitions/ blk-rq-qos.o \
 			disk-events.o blk-ia-ranges.o early-lookup.o
 
-obj-$(CONFIG_BOUNCE)		+= bounce.o
 obj-$(CONFIG_BLK_DEV_BSG_COMMON) += bsg.o
 obj-$(CONFIG_BLK_DEV_BSGLIB)	+= bsg-lib.o
 obj-$(CONFIG_BLK_CGROUP)	+= blk-cgroup.o
diff --git a/block/blk-map.c b/block/blk-map.c
index d2f22744b3d1..cadbf11b50a3 100644
--- a/block/blk-map.c
+++ b/block/blk-map.c
@@ -556,8 +556,6 @@ int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
 
 	if (map_data)
 		copy = true;
-	else if (blk_queue_may_bounce(q))
-		copy = true;
 	else if (iov_iter_alignment(iter) & align)
 		copy = true;
 	else if (iov_iter_is_bvec(iter))
@@ -713,8 +711,7 @@ int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
 	if (!len || !kbuf)
 		return -EINVAL;
 
-	if (!blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf) ||
-	    blk_queue_may_bounce(q))
+	if (!blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf))
 		bio = bio_copy_kern(q, kbuf, len, gfp_mask, reading);
 	else
 		bio = bio_map_kern(q, kbuf, len, gfp_mask);
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 796baeccd37b..83c651a7facd 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -3144,8 +3144,6 @@ void blk_mq_submit_bio(struct bio *bio)
 		goto new_request;
 	}
 
-	bio = blk_queue_bounce(bio, q);
-
 	/*
 	 * The cached request already holds a q_usage_counter reference and we
 	 * don't have to acquire a new one if we use it.
diff --git a/block/blk-settings.c b/block/blk-settings.c
index 4817e7ca03f8..a000daafbfb4 100644
--- a/block/blk-settings.c
+++ b/block/blk-settings.c
@@ -124,11 +124,6 @@ static int blk_validate_integrity_limits(struct queue_limits *lim)
 		return 0;
 	}
 
-	if (lim->features & BLK_FEAT_BOUNCE_HIGH) {
-		pr_warn("no bounce buffer support for integrity metadata\n");
-		return -EINVAL;
-	}
-
 	if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)) {
 		pr_warn("integrity support disabled.\n");
 		return -EINVAL;
diff --git a/block/blk.h b/block/blk.h
index 328075787814..ff21f234b62f 100644
--- a/block/blk.h
+++ b/block/blk.h
@@ -443,23 +443,6 @@ static inline void ioc_clear_queue(struct request_queue *q)
 }
 #endif /* CONFIG_BLK_ICQ */
 
-struct bio *__blk_queue_bounce(struct bio *bio, struct request_queue *q);
-
-static inline bool blk_queue_may_bounce(struct request_queue *q)
-{
-	return IS_ENABLED(CONFIG_BOUNCE) &&
-		(q->limits.features & BLK_FEAT_BOUNCE_HIGH) &&
-		max_low_pfn >= max_pfn;
-}
-
-static inline struct bio *blk_queue_bounce(struct bio *bio,
-		struct request_queue *q)
-{
-	if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio)))
-		return __blk_queue_bounce(bio, q);
-	return bio;
-}
-
 #ifdef CONFIG_BLK_DEV_ZONED
 void disk_init_zone_resources(struct gendisk *disk);
 void disk_free_zone_resources(struct gendisk *disk);
diff --git a/block/bounce.c b/block/bounce.c
deleted file mode 100644
index 09a9616cf209..000000000000
--- a/block/bounce.c
+++ /dev/null
@@ -1,267 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0
-/* bounce buffer handling for block devices
- *
- * - Split from highmem.c
- */
-
-#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
-
-#include <linux/mm.h>
-#include <linux/export.h>
-#include <linux/swap.h>
-#include <linux/gfp.h>
-#include <linux/bio-integrity.h>
-#include <linux/pagemap.h>
-#include <linux/mempool.h>
-#include <linux/blkdev.h>
-#include <linux/backing-dev.h>
-#include <linux/init.h>
-#include <linux/hash.h>
-#include <linux/highmem.h>
-#include <linux/printk.h>
-#include <asm/tlbflush.h>
-
-#include <trace/events/block.h>
-#include "blk.h"
-#include "blk-cgroup.h"
-
-#define POOL_SIZE	64
-#define ISA_POOL_SIZE	16
-
-static struct bio_set bounce_bio_set, bounce_bio_split;
-static mempool_t page_pool;
-
-static 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);
-
-	ret = bioset_init(&bounce_bio_split, BIO_POOL_SIZE, 0, 0);
-	BUG_ON(ret);
-	bounce_bs_setup = true;
-}
-
-static __init int init_emergency_pool(void)
-{
-	int ret;
-
-#ifndef 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);
-	pr_info("pool size: %d pages\n", POOL_SIZE);
-
-	init_bounce_bioset();
-	return 0;
-}
-
-__initcall(init_emergency_pool);
-
-/*
- * Simple bounce buffer support for highmem pages. Depending on the
- * queue gfp mask set, *to may or may not be a highmem page. kmap it
- * always, it will do the Right Thing
- */
-static void copy_to_high_bio_irq(struct bio *to, struct bio *from)
-{
-	struct bio_vec tovec, fromvec;
-	struct bvec_iter iter;
-	/*
-	 * The bio of @from is created by bounce, so we can iterate
-	 * its bvec from start to end, but the @from->bi_iter can't be
-	 * trusted because it might be changed by splitting.
-	 */
-	struct bvec_iter from_iter = BVEC_ITER_ALL_INIT;
-
-	bio_for_each_segment(tovec, to, iter) {
-		fromvec = bio_iter_iovec(from, from_iter);
-		if (tovec.bv_page != fromvec.bv_page) {
-			/*
-			 * fromvec->bv_offset and fromvec->bv_len might have
-			 * been modified by the block layer, so use the original
-			 * copy, bounce_copy_vec already uses tovec->bv_len
-			 */
-			memcpy_to_bvec(&tovec, page_address(fromvec.bv_page) +
-				       tovec.bv_offset);
-		}
-		bio_advance_iter(from, &from_iter, tovec.bv_len);
-	}
-}
-
-static void bounce_end_io(struct bio *bio)
-{
-	struct bio *bio_orig = bio->bi_private;
-	struct bio_vec *bvec, orig_vec;
-	struct bvec_iter orig_iter = bio_orig->bi_iter;
-	struct bvec_iter_all iter_all;
-
-	/*
-	 * free up bounce indirect pages used
-	 */
-	bio_for_each_segment_all(bvec, bio, iter_all) {
-		orig_vec = bio_iter_iovec(bio_orig, orig_iter);
-		if (bvec->bv_page != orig_vec.bv_page) {
-			dec_zone_page_state(bvec->bv_page, NR_BOUNCE);
-			mempool_free(bvec->bv_page, &page_pool);
-		}
-		bio_advance_iter(bio_orig, &orig_iter, orig_vec.bv_len);
-	}
-
-	bio_orig->bi_status = bio->bi_status;
-	bio_endio(bio_orig);
-	bio_put(bio);
-}
-
-static void bounce_end_io_write(struct bio *bio)
-{
-	bounce_end_io(bio);
-}
-
-static void bounce_end_io_read(struct bio *bio)
-{
-	struct bio *bio_orig = bio->bi_private;
-
-	if (!bio->bi_status)
-		copy_to_high_bio_irq(bio_orig, bio);
-
-	bounce_end_io(bio);
-}
-
-static struct bio *bounce_clone_bio(struct bio *bio_src)
-{
-	struct bvec_iter iter;
-	struct bio_vec bv;
-	struct bio *bio;
-
-	/*
-	 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
-	 * bio_src->bi_io_vec to bio->bi_io_vec.
-	 *
-	 * We can't do that anymore, because:
-	 *
-	 *  - The point of cloning the biovec is to produce a bio with a biovec
-	 *    the caller can modify: bi_idx and bi_bvec_done should be 0.
-	 *
-	 *  - The original bio could've had more than BIO_MAX_VECS biovecs; if
-	 *    we tried to clone the whole thing bio_alloc_bioset() would fail.
-	 *    But the clone should succeed as long as the number of biovecs we
-	 *    actually need to allocate is fewer than BIO_MAX_VECS.
-	 *
-	 *  - Lastly, bi_vcnt should not be looked at or relied upon by code
-	 *    that does not own the bio - reason being drivers don't use it for
-	 *    iterating over the biovec anymore, so expecting it to be kept up
-	 *    to date (i.e. for clones that share the parent biovec) is just
-	 *    asking for trouble and would force extra work.
-	 */
-	bio = bio_alloc_bioset(bio_src->bi_bdev, bio_segments(bio_src),
-			       bio_src->bi_opf, GFP_NOIO, &bounce_bio_set);
-	if (bio_flagged(bio_src, BIO_REMAPPED))
-		bio_set_flag(bio, BIO_REMAPPED);
-	bio->bi_ioprio		= bio_src->bi_ioprio;
-	bio->bi_write_hint	= bio_src->bi_write_hint;
-	bio->bi_iter.bi_sector	= bio_src->bi_iter.bi_sector;
-	bio->bi_iter.bi_size	= bio_src->bi_iter.bi_size;
-
-	switch (bio_op(bio)) {
-	case REQ_OP_DISCARD:
-	case REQ_OP_SECURE_ERASE:
-	case REQ_OP_WRITE_ZEROES:
-		break;
-	default:
-		bio_for_each_segment(bv, bio_src, iter)
-			bio->bi_io_vec[bio->bi_vcnt++] = bv;
-		break;
-	}
-
-	if (bio_crypt_clone(bio, bio_src, GFP_NOIO) < 0)
-		goto err_put;
-
-	if (bio_integrity(bio_src) &&
-	    bio_integrity_clone(bio, bio_src, GFP_NOIO) < 0)
-		goto err_put;
-
-	bio_clone_blkg_association(bio, bio_src);
-
-	return bio;
-
-err_put:
-	bio_put(bio);
-	return NULL;
-}
-
-struct bio *__blk_queue_bounce(struct bio *bio_orig, struct request_queue *q)
-{
-	struct bio *bio;
-	int rw = bio_data_dir(bio_orig);
-	struct bio_vec *to, from;
-	struct bvec_iter iter;
-	unsigned i = 0, bytes = 0;
-	bool bounce = false;
-	int sectors;
-
-	bio_for_each_segment(from, bio_orig, iter) {
-		if (i++ < BIO_MAX_VECS)
-			bytes += from.bv_len;
-		if (PageHighMem(from.bv_page))
-			bounce = true;
-	}
-	if (!bounce)
-		return bio_orig;
-
-	/*
-	 * Individual bvecs might not be logical block aligned. Round down
-	 * the split size so that each bio is properly block size aligned,
-	 * even if we do not use the full hardware limits.
-	 */
-	sectors = ALIGN_DOWN(bytes, queue_logical_block_size(q)) >>
-			SECTOR_SHIFT;
-	if (sectors < bio_sectors(bio_orig)) {
-		bio = bio_split(bio_orig, sectors, GFP_NOIO, &bounce_bio_split);
-		bio_chain(bio, bio_orig);
-		submit_bio_noacct(bio_orig);
-		bio_orig = bio;
-	}
-	bio = bounce_clone_bio(bio_orig);
-
-	/*
-	 * Bvec table can't be updated by bio_for_each_segment_all(),
-	 * so retrieve bvec from the table directly. This way is safe
-	 * because the 'bio' is single-page bvec.
-	 */
-	for (i = 0, to = bio->bi_io_vec; i < bio->bi_vcnt; to++, i++) {
-		struct page *bounce_page;
-
-		if (!PageHighMem(to->bv_page))
-			continue;
-
-		bounce_page = mempool_alloc(&page_pool, GFP_NOIO);
-		inc_zone_page_state(bounce_page, NR_BOUNCE);
-
-		if (rw == WRITE) {
-			flush_dcache_page(to->bv_page);
-			memcpy_from_bvec(page_address(bounce_page), to);
-		}
-		to->bv_page = bounce_page;
-	}
-
-	trace_block_bio_bounce(bio_orig);
-
-	bio->bi_flags |= (1 << BIO_BOUNCED);
-
-	if (rw == READ)
-		bio->bi_end_io = bounce_end_io_read;
-	else
-		bio->bi_end_io = bounce_end_io_write;
-
-	bio->bi_private = bio_orig;
-	return bio;
-}
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index dce7615c35e7..5a46067e85b1 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -286,7 +286,6 @@ struct bio {
 enum {
 	BIO_PAGE_PINNED,	/* Unpin pages in bio_release_pages() */
 	BIO_CLONED,		/* doesn't own data */
-	BIO_BOUNCED,		/* bio is a bounce bio */
 	BIO_QUIET,		/* Make BIO Quiet */
 	BIO_CHAIN,		/* chained bio, ->bi_remaining in effect */
 	BIO_REFFED,		/* bio has elevated ->bi_cnt */
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index f3d74f9dae8e..5ccb961ee2ae 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -331,9 +331,6 @@ typedef unsigned int __bitwise blk_features_t;
 /* skip this queue in blk_mq_(un)quiesce_tagset */
 #define BLK_FEAT_SKIP_TAGSET_QUIESCE	((__force blk_features_t)(1u << 13))
 
-/* bounce all highmem pages */
-#define BLK_FEAT_BOUNCE_HIGH		((__force blk_features_t)(1u << 14))
-
 /* undocumented magic for bcache */
 #define BLK_FEAT_RAID_PARTIAL_STRIPES_EXPENSIVE \
 	((__force blk_features_t)(1u << 15))
@@ -347,7 +344,7 @@ typedef unsigned int __bitwise blk_features_t;
  */
 #define BLK_FEAT_INHERIT_MASK \
 	(BLK_FEAT_WRITE_CACHE | BLK_FEAT_FUA | BLK_FEAT_ROTATIONAL | \
-	 BLK_FEAT_STABLE_WRITES | BLK_FEAT_ZONED | BLK_FEAT_BOUNCE_HIGH | \
+	 BLK_FEAT_STABLE_WRITES | BLK_FEAT_ZONED | \
 	 BLK_FEAT_RAID_PARTIAL_STRIPES_EXPENSIVE)
 
 /* internal flags in queue_limits.flags */
diff --git a/include/trace/events/block.h b/include/trace/events/block.h
index bd0ea07338eb..ad36e73b8579 100644
--- a/include/trace/events/block.h
+++ b/include/trace/events/block.h
@@ -360,21 +360,6 @@ DECLARE_EVENT_CLASS(block_bio,
 		  __entry->nr_sector, __entry->comm)
 );
 
-/**
- * block_bio_bounce - used bounce buffer when processing block operation
- * @bio: block operation
- *
- * A bounce buffer was used to handle the block operation @bio in @q.
- * This occurs when hardware limitations prevent a direct transfer of
- * data between the @bio data memory area and the IO device.  Use of a
- * bounce buffer requires extra copying of data and decreases
- * performance.
- */
-DEFINE_EVENT(block_bio, block_bio_bounce,
-	TP_PROTO(struct bio *bio),
-	TP_ARGS(bio)
-);
-
 /**
  * block_bio_backmerge - merging block operation to the end of an existing operation
  * @bio: new block operation to merge
diff --git a/include/uapi/linux/blktrace_api.h b/include/uapi/linux/blktrace_api.h
index 690621b610e5..1bfb635e309b 100644
--- a/include/uapi/linux/blktrace_api.h
+++ b/include/uapi/linux/blktrace_api.h
@@ -49,7 +49,7 @@ enum blktrace_act {
 	__BLK_TA_UNPLUG_TIMER,		/* queue was unplugged by timer */
 	__BLK_TA_INSERT,		/* insert request */
 	__BLK_TA_SPLIT,			/* bio was split */
-	__BLK_TA_BOUNCE,		/* bio was bounced */
+	__BLK_TA_BOUNCE,		/* unused, was: bio was bounced */
 	__BLK_TA_REMAP,			/* bio was remapped */
 	__BLK_TA_ABORT,			/* request aborted */
 	__BLK_TA_DRV_DATA,		/* driver-specific binary data */
diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
index 3679a6d18934..d5f36f415d9d 100644
--- a/kernel/trace/blktrace.c
+++ b/kernel/trace/blktrace.c
@@ -893,11 +893,6 @@ static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 	rcu_read_unlock();
 }
 
-static void blk_add_trace_bio_bounce(void *ignore, struct bio *bio)
-{
-	blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_BOUNCE, 0);
-}
-
 static void blk_add_trace_bio_complete(void *ignore,
 				       struct request_queue *q, struct bio *bio)
 {
@@ -1089,8 +1084,6 @@ static void blk_register_tracepoints(void)
 	WARN_ON(ret);
 	ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
 	WARN_ON(ret);
-	ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
-	WARN_ON(ret);
 	ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
 	WARN_ON(ret);
 	ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
@@ -1125,7 +1118,6 @@ static void blk_unregister_tracepoints(void)
 	unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
 	unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
 	unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
-	unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
 	unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
 	unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
 	unregister_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
@@ -1462,7 +1454,6 @@ static const struct {
 	[__BLK_TA_UNPLUG_TIMER]	= {{ "UT", "unplug_timer" }, blk_log_unplug },
 	[__BLK_TA_INSERT]	= {{  "I", "insert" },	   blk_log_generic },
 	[__BLK_TA_SPLIT]	= {{  "X", "split" },	   blk_log_split },
-	[__BLK_TA_BOUNCE]	= {{  "B", "bounce" },	   blk_log_generic },
 	[__BLK_TA_REMAP]	= {{  "A", "remap" },	   blk_log_remap },
 };
 
-- 
2.47.2

Re: [PATCH 6/7] block: remove bounce buffering support

[John Garry]

On 05/05/2025 09:11, Christoph Hellwig wrote:
> The block layer bounce buffering support is unused now, remove it.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> Reviewed-by: Hannes Reinecke <hare@suse.de>

Reviewed-by: John Garry <john.g.garry@oracle.com>

>   	unregister_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
> @@ -1462,7 +1454,6 @@ static const struct {
>   	[__BLK_TA_UNPLUG_TIMER]	= {{ "UT", "unplug_timer" }, blk_log_unplug },
>   	[__BLK_TA_INSERT]	= {{  "I", "insert" },	   blk_log_generic },
>   	[__BLK_TA_SPLIT]	= {{  "X", "split" },	   blk_log_split },
> -	[__BLK_TA_BOUNCE]	= {{  "B", "bounce" },	   blk_log_generic },


I don't expect any problem if someone re-cycles "B" in future for 
another action

 >   	[__BLK_TA_REMAP]	= {{  "A", "remap" },	   blk_log_remap },
 >   };
 >

[PATCH 7/7] mm: remove NR_BOUNCE zone stat

[Christoph Hellwig]

The stat is always 0 now, so remove it and hardwire the user visible
output to 0.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
---
 drivers/base/node.c    | 2 +-
 fs/proc/meminfo.c      | 3 +--
 include/linux/mmzone.h | 1 -
 mm/show_mem.c          | 4 ++--
 4 files changed, 4 insertions(+), 6 deletions(-)

diff --git a/drivers/base/node.c b/drivers/base/node.c
index cd13ef287011..618712071a1e 100644
--- a/drivers/base/node.c
+++ b/drivers/base/node.c
@@ -468,7 +468,7 @@ static ssize_t node_read_meminfo(struct device *dev,
 			     nid, K(node_page_state(pgdat, NR_PAGETABLE)),
 			     nid, K(node_page_state(pgdat, NR_SECONDARY_PAGETABLE)),
 			     nid, 0UL,
-			     nid, K(sum_zone_node_page_state(nid, NR_BOUNCE)),
+			     nid, 0UL,
 			     nid, K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
 			     nid, K(sreclaimable +
 				    node_page_state(pgdat, NR_KERNEL_MISC_RECLAIMABLE)),
diff --git a/fs/proc/meminfo.c b/fs/proc/meminfo.c
index 83be312159c9..bc2bc60c36cc 100644
--- a/fs/proc/meminfo.c
+++ b/fs/proc/meminfo.c
@@ -120,8 +120,7 @@ static int meminfo_proc_show(struct seq_file *m, void *v)
 		    global_node_page_state(NR_SECONDARY_PAGETABLE));
 
 	show_val_kb(m, "NFS_Unstable:   ", 0);
-	show_val_kb(m, "Bounce:         ",
-		    global_zone_page_state(NR_BOUNCE));
+	show_val_kb(m, "Bounce:         ", 0);
 	show_val_kb(m, "WritebackTmp:   ",
 		    global_node_page_state(NR_WRITEBACK_TEMP));
 	show_val_kb(m, "CommitLimit:    ", vm_commit_limit());
diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index 6ccec1bf2896..b1c459f7a485 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -148,7 +148,6 @@ enum zone_stat_item {
 	NR_ZONE_WRITE_PENDING,	/* Count of dirty, writeback and unstable pages */
 	NR_MLOCK,		/* mlock()ed pages found and moved off LRU */
 	/* Second 128 byte cacheline */
-	NR_BOUNCE,
 #if IS_ENABLED(CONFIG_ZSMALLOC)
 	NR_ZSPAGES,		/* allocated in zsmalloc */
 #endif
diff --git a/mm/show_mem.c b/mm/show_mem.c
index 6af13bcd2ab3..5acb51a9fc49 100644
--- a/mm/show_mem.c
+++ b/mm/show_mem.c
@@ -223,7 +223,7 @@ static void show_free_areas(unsigned int filter, nodemask_t *nodemask, int max_z
 		global_node_page_state(NR_SHMEM),
 		global_node_page_state(NR_PAGETABLE),
 		global_node_page_state(NR_SECONDARY_PAGETABLE),
-		global_zone_page_state(NR_BOUNCE),
+		0UL,
 		global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE),
 		global_zone_page_state(NR_FREE_PAGES),
 		free_pcp,
@@ -341,7 +341,7 @@ static void show_free_areas(unsigned int filter, nodemask_t *nodemask, int max_z
 			K(zone->present_pages),
 			K(zone_managed_pages(zone)),
 			K(zone_page_state(zone, NR_MLOCK)),
-			K(zone_page_state(zone, NR_BOUNCE)),
+			0UL,
 			K(free_pcp),
 			K(this_cpu_read(zone->per_cpu_pageset->count)),
 			K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
-- 
2.47.2

Re: remove block layer bounce buffering v2

[Johannes Thumshirn]

On Mon, May 05, 2025 at 10:11:19AM +0200, Christoph Hellwig wrote:
> Hi all,
> Changes since v1:
>  - return -EINVAL from usb-storage ->probe
>  - improve the error message in usb-storage
>  - improve comments in usb-storage

Sorry for sending it to v1:
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>

Re: remove block layer bounce buffering v2

[Jens Axboe]

On Mon, 05 May 2025 10:11:19 +0200, Christoph Hellwig wrote:
> the block layer bounce buffering from the early days of highmem keeps
> being a wart in the block layer despite usage by only four drivers,
> all through the SCSI layer.  One of them is an old PIO-only ISA
> card, two are parallel port adapters, and the fourth is the usb-storage
> driver.
> 
> This series makes the first three depend on !HIGHMEM and for the fourth
> rejects the probe only when used on highmem system and the HCD is one
> of the few annoying one that does not support DMA.
> 
> [...]

Applied, thanks!

[1/7] scsi: make aha152x depend on !HIGHMEM
      commit: 7b32cb540bff6d6c8a1659babf930e9f66283c2c
[2/7] scsi: make imm depend on !HIGHMEM
      commit: bf69bd3fc26a107611e76b342027bb60b2411d4e
[3/7] scsi: make ppa depend on !HIGHMEM
      commit: 27a0918d4b701d4825e191448e44b9f14dc0a3b3
[4/7] usb-storage: reject probe of device one non-DMA HCDs when using highmem
      commit: 48610ec22f0cf7ee5b5658b2b3bab27a8f2ef78b
[5/7] scsi: remove the no_highmem flag in the host
      commit: a9437f6a1d8d0b3787fe6ff03d9aab4d3fe9b940
[6/7] block: remove bounce buffering support
      commit: eeadd68e2a5f6bfe0bf1038ec49e3a8d99eb5fe8
[7/7] mm: remove NR_BOUNCE zone stat
      commit: 194df9f66db8d6f74f03c78c2ad47b74a5a8b886

Best regards,
-- 
Jens Axboe