nvme: scan sequentially only when list scan unsupported
PATCH 1/1 nvme: scan sequentially only when list scan unsupported
Currently, if nvme_scan_ns_list fails, nvme_scan_work will fall back to a sequential scan. nvme_scan_ns_list can fail for a variety of reasons, e.g. transient transport issue. And the resulting sequential scan can be extremely expensive on controllers reporting an NN value close to the maximum allowed (>4 billion). Avoid sequential scans wherever possible by only falling back to them if nvme_scan_ns_list fails due to controller non-support of Identify NS List.
---
This would break devices that claim to support version NVME_VS(1, 1, 0) or above, but don't support Identify NS List. But it looks like we already have NVME_QUIRK_IDENTIFY_CNS to deal with that.
drivers/nvme/host/core.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/nvme/host/core.c +1 -1
@@ -4535,7 +4535,7 @@ static void nvme_scan_work(struct work_struct *work) }mutex_lock (&ctrl->scan_lock); -if (nvme_scan_ns_list (ctrl) !=0 ) +if (nvme_scan_ns_list (ctrl) == -EOPNOTSUPP )nvme_scan_ns_sequential (ctrl);mutex_unlock (&ctrl->scan_lock); }
base-commit: d30a909f9bb5283e701a5fdfffac763ef57a3e7c
Signature · 1 line
2.25.1
Quoted 7 lines
Currently, if
nvme_scan_ns_listfails,nvme_scan_workwill fall back to a sequential scan.nvme_scan_ns_listcan fail for a variety of reasons, e.g. transient transport issue. And the resulting sequential scan can be extremely expensive on controllers reporting an NN value close to the maximum allowed (>4 billion). Avoid sequential scans wherever possible by only falling back to them ifnvme_scan_ns_listfails due to controller non-support of Identify NS List.
Quoted 26 lines
---
This would break devices that claim to support versionNVME_VS(1, 1, 0)or above, but don't support Identify NS List. But it looks like we already haveNVME_QUIRK_IDENTIFY_CNSto deal with that.drivers/nvme/host/core.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)drivers/nvme/host/core.c +1
-1@@ -4535,7 +4535,7 @@ static void nvme_scan_work(struct work_struct *work) }mutex_lock (&ctrl->scan_lock); -if (nvme_scan_ns_list (ctrl) !=0 ) +if (nvme_scan_ns_list (ctrl) == -EOPNOTSUPP )nvme_scan_ns_sequential (ctrl);mutex_unlock (&ctrl->scan_lock); }base-commit:
d30a909f9bb5283e701a5fdfffac763ef57a3e7cSignature · 1 line
2.25.1
Christoph Hellwig
maintainerQuoted 11 lines · Uday Shankar
Currently, if
nvme_scan_ns_listfails,nvme_scan_workwill fall back to a sequential scan.nvme_scan_ns_listcan fail for a variety of reasons, e.g. transient transport issue. And the resulting sequential scan can be extremely expensive on controllers reporting an NN value close to the maximum allowed (>4 billion). Avoid sequential scans wherever possible by only falling back to them ifnvme_scan_ns_listfails due to controller non-support of Identify NS List.
---
This would break devices that claim to support versionNVME_VS(1, 1, 0)or above, but don't support Identify NS List. But it looks like we already haveNVME_QUIRK_IDENTIFY_CNSto deal with that.
Yes. I'm a little worried about changing something this fundamental after such a long time, even if it is the right thing to do. But at least we get a warning when the Identify NS List fails, so between that and the fact tht nvme_scan_ns_list logs an error I think we can cope.
drivers/nvme/host/core.c static void nvme_scan_work(struct work_struct *work) · line 4539
Quoted 4 lines
drivers/nvme/host/core.c +1
-1mutex_lock (&ctrl->scan_lock); -if (nvme_scan_ns_list (ctrl) !=0 ) +if (nvme_scan_ns_list (ctrl) == -EOPNOTSUPP )nvme_scan_ns_sequential (ctrl);
Maybe make this a bit more clear by lifting the nvme_ctrl_limited_cns check into nvme_scan_work, though:
if (nvme_ctrl_limited_cns(ctrl)) nvme_scan_ns_sequential(ctrl); else nvme_scan_ns_list(ctrl);
and drop the now superflous error return from nvme_scan_ns_list.
Quoted 9 lines · Christoph Hellwig
Maybe make this a bit more clear by lifting the
nvme_ctrl_limited_cnscheck intonvme_scan_work, though:if (nvme_ctrl_limited_cns(ctrl)) nvme_scan_ns_sequential(ctrl); else nvme_scan_ns_list(ctrl);and drop the now superflous error return from
nvme_scan_ns_list.
Sure. I will resubmit with that revision.