A drive with atomic write support should have awun / nawun defined,
for these drives it should be possible to play with and experiment
safely with LBS support up to awun / nawun settings if you are
completely ignoring power failure situations. Add support to
experiment with this. The rationale to limit to awun / nawun is
to avoid races with other on flight commands which otherwise
could cause unexpected results.
This also means this debug module parameter feature is not supported
if your drive does not support atomics / awun / nawun.
Suggested-by: Dan Helmick <dan.helmick@xxxxxxxxxxx>
Signed-off-by: Luis Chamberlain <mcgrof@xxxxxxxxxx>
---
drivers/nvme/host/core.c | 18 ++++++++++++++++++
1 file changed, 18 insertions(+)
diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 7a3c51ac13bd..c1f9d8e3ea93 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -92,6 +92,10 @@ static bool debug_large_lbas;
module_param(debug_large_lbas, bool, 0644);
MODULE_PARM_DESC(debug_large_lbas, "allow LBAs > PAGE_SIZE");
+static unsigned int debug_large_atomics;
+module_param(debug_large_atomics, uint, 0644);
+MODULE_PARM_DESC(debug_large_atomics, "allow large atomics <= awun or nawun <= mdts");
+
/*
* nvme_wq - hosts nvme related works that are not reset or delete
* nvme_reset_wq - hosts nvme reset works
@@ -1958,6 +1962,20 @@ static void nvme_update_disk_info(struct gendisk *disk,
* be aware of out of order reads/writes as npwg and nows
* are purely performance optimizations.
*/
+
+ /*
+ * If you're not concerned about power failure, in theory,
+ * you should be able to experiment up to awun rather safely.
+ *
+ * Ignore qemu awun value of 1.
+ */
+ if (debug_large_atomics && awun != 1) {
+ debug_large_atomics = min(awun_bs, debug_large_atomics);
+ phys_bs = atomic_bs = debug_large_atomics;
+ dev_info(ns->ctrl->device,
+ "Forcing large atomic: %u (awun_bs: %u awun: %u)\n",
+ debug_large_atomics, awun_bs, awun);
+ }
}
blk_queue_logical_block_size(disk->queue, bs);
--
2.39.2
