Re: [PATCH v2] block: Increase BLK_DEF_MAX_SECTORS_CAP

[Keith Busch <kbusch@kernel.org>]

On Thu, Apr 02, 2026 at 04:33:47PM +0200, Friedrich Weber wrote:
> We only have limited access to the test machine, so testing this is not
> trivial. If I understand correctly, there is a lead pointing in the direction
> of mpt3sas [1], so I'd postpone this test for now. But if needed, we're happy
> to look into it.

Yeah, the mpt3sas driver isn't using an appropriate sized buffer for
nvme prp handling. The easy option is just force the block layer to
split requests so the driver never sees anything bigger than what it can
currently handle. This should do it. I don't have any such device to
test on, though.

---
diff --git a/drivers/scsi/mpt3sas/mpt3sas_scsih.c b/drivers/scsi/mpt3sas/mpt3sas_scsih.c
index 6ff7885572942..c76f5b958c56f 100644
--- a/drivers/scsi/mpt3sas/mpt3sas_scsih.c
+++ b/drivers/scsi/mpt3sas/mpt3sas_scsih.c
@@ -2739,7 +2739,10 @@ scsih_sdev_configure(struct scsi_device *sdev, struct queue_limits *lim)
 				pcie_device->connector_name);
 
 		if (pcie_device->nvme_mdts)
-			lim->max_hw_sectors = pcie_device->nvme_mdts / 512;
+			lim->max_hw_sectors = min(pcie_device->nvme_mdts / 512,
+						(SZ_2M / 512) - 8);
+		else
+			lim->max_hw_sectors = (SZ_2M / 512) - 8;
 
 		pcie_device_put(pcie_device);
 		spin_unlock_irqrestore(&ioc->pcie_device_lock, flags);
--