This patch fixes an issue that the following error happens on
swiotlb environment:

        xhci-hcd ee000000.usb: swiotlb buffer is full (sz: 524288 bytes), total 
32768 (slots), used 1338 (slots)

On the kernel v5.1, block settings of a usb-storage with SuperSpeed
were the following so that the block layer will allocate buffers
up to 64 KiB, and then the issue didn't happen.

        max_segment_size = 65536
        max_hw_sectors_kb = 1024

After the commit 09324d32d2a0 ("block: force an unlimited segment
size on queues with a virt boundary") is applied, the block settings
are the following. So, the block layer will allocate buffers up to
1024 KiB, and then the issue happens:

        max_segment_size = 4294967295
        max_hw_sectors_kb = 1024

To fix the issue, the usb-storage driver checks the maximum size of
a mapping for the device and then adjusts the max_hw_sectors_kb
if required. After this patch is applied, the block settings will
be the following, and then the issue doesn't happen.

        max_segment_size = 4294967295
        max_hw_sectors_kb = 256

Fixes: 09324d32d2a0 ("block: force an unlimited segment size on queues with a 
virt boundary")
Signed-off-by: Yoshihiro Shimoda <[email protected]>
---
 We investigated this issue on the following ML:
 https://marc.info/?l=linux-block&m=156033909218970&w=2

 drivers/usb/storage/scsiglue.c | 23 +++++++++++++++++------
 1 file changed, 17 insertions(+), 6 deletions(-)

diff --git a/drivers/usb/storage/scsiglue.c b/drivers/usb/storage/scsiglue.c
index 59190d8..89c3640 100644
--- a/drivers/usb/storage/scsiglue.c
+++ b/drivers/usb/storage/scsiglue.c
@@ -28,6 +28,8 @@
  * status of a command.
  */
 
+#include <linux/blkdev.h>
+#include <linux/dma-mapping.h>
 #include <linux/module.h>
 #include <linux/mutex.h>
 
@@ -99,6 +101,7 @@ static int slave_alloc (struct scsi_device *sdev)
 static int slave_configure(struct scsi_device *sdev)
 {
        struct us_data *us = host_to_us(sdev->host);
+       unsigned int max_sectors = 0;   /* 0 means no update required */
 
        /*
         * Many devices have trouble transferring more than 32KB at a time,
@@ -106,26 +109,34 @@ static int slave_configure(struct scsi_device *sdev)
         * are limiting both to 32K (64 sectores).
         */
        if (us->fflags & (US_FL_MAX_SECTORS_64 | US_FL_MAX_SECTORS_MIN)) {
-               unsigned int max_sectors = 64;
+               max_sectors = 64;
 
                if (us->fflags & US_FL_MAX_SECTORS_MIN)
                        max_sectors = PAGE_SIZE >> 9;
-               if (queue_max_hw_sectors(sdev->request_queue) > max_sectors)
-                       blk_queue_max_hw_sectors(sdev->request_queue,
-                                             max_sectors);
+               if (queue_max_hw_sectors(sdev->request_queue) <= max_sectors)
+                       max_sectors = 0;
        } else if (sdev->type == TYPE_TAPE) {
                /*
                 * Tapes need much higher max_sector limits, so just
                 * raise it to the maximum possible (4 GB / 512) and
                 * let the queue segment size sort out the real limit.
                 */
-               blk_queue_max_hw_sectors(sdev->request_queue, 0x7FFFFF);
+               max_sectors = 0x7FFFFF;
        } else if (us->pusb_dev->speed >= USB_SPEED_SUPER) {
                /*
                 * USB3 devices will be limited to 2048 sectors. This gives us
                 * better throughput on most devices.
                 */
-               blk_queue_max_hw_sectors(sdev->request_queue, 2048);
+               max_sectors = 2048;
+       }
+
+       if (max_sectors > 0) {
+               struct device *dev = us->pusb_dev->bus->sysdev;
+               size_t max_dma_sectors = dma_max_mapping_size(dev) >>
+                                        SECTOR_SHIFT;
+
+               max_sectors = min_t(size_t, max_sectors, max_dma_sectors);
+               blk_queue_max_hw_sectors(sdev->request_queue, max_sectors);
        }
 
        /*
-- 
2.7.4

Reply via email to