[PATCH blktests v1] nvme/029: reserve hugepages for lager allocations

[PATCH blktests v1] nvme/029: reserve hugepages for lager allocations

[Daniel Wagner]

The test is issuing larger IO workload. This depends on being able to
allocate larger chunks of linear memory. nvme-cli used to use libhugetlb
to automatically allocate the HugeTLB pool. Though nvme-cli dropped the
dependency on the library, thus the test should try to provision the
system accordingly.

Link: https://github.com/linux-nvme/nvme-cli/issues/2218
Reported-by: Yi Zhang <yi.zhang@redhat.com>
Tested-by: Yi Zhang <yi.zhang@redhat.com>
Signed-off-by: Daniel Wagner <dwagner@suse.de>
---
 tests/nvme/029 | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/tests/nvme/029 b/tests/nvme/029
index db6e8b91f707..a4f0cb1402b1 100755
--- a/tests/nvme/029
+++ b/tests/nvme/029
@@ -54,6 +54,7 @@ test() {
 	_setup_nvmet
 
 	local nvmedev
+	local reset_nr_hugepages=false
 
 	_nvmet_target_setup
 
@@ -62,6 +63,15 @@ test() {
 	nvmedev=$(_find_nvme_dev "${def_subsysnqn}")
 	_check_uuid "${nvmedev}"
 
+	# nvme-cli may fail to allocate linear memory for rather large IO buffers.
+	# Increase nr_hugepages to allow nvme-cli to try the linear memory allocation
+	# from HugeTLB pool.
+	if [[  -r /proc/sys/vm/nr_hugepages &&
+		       "$(cat /proc/sys/vm/nr_hugepages)" -eq 0 ]]; then
+		echo 20 > /proc/sys/vm/nr_hugepages
+		reset_nr_hugepages=true
+	fi
+
 	local dev="/dev/${nvmedev}n1"
 	test_user_io "$dev" 1 512 > "$FULL" 2>&1 || echo FAIL
 	test_user_io "$dev" 1 511 > "$FULL" 2>&1 || echo FAIL
@@ -70,6 +80,10 @@ test() {
 	test_user_io "$dev" 511 1023 > "$FULL" 2>&1 || echo FAIL
 	test_user_io "$dev" 511 1025 > "$FULL" 2>&1 || echo FAIL
 
+	if [[ ${reset_nr_hugepages} = true ]]; then
+		echo 0 > /proc/sys/vm/nr_hugepages
+	fi
+
 	_nvme_disconnect_subsys "${def_subsysnqn}" >> "$FULL" 2>&1
 
 	_nvmet_target_cleanup
-- 
2.43.1

Re: [PATCH blktests v1] nvme/029: reserve hugepages for lager allocations

[Chaitanya Kulkarni]

On 2/20/24 23:43, Daniel Wagner wrote:
> The test is issuing larger IO workload. This depends on being able to
> allocate larger chunks of linear memory. nvme-cli used to use libhugetlb
> to automatically allocate the HugeTLB pool. Though nvme-cli dropped the
> dependency on the library, thus the test should try to provision the
> system accordingly.
>
> Link: https://github.com/linux-nvme/nvme-cli/issues/2218
> Reported-by: Yi Zhang <yi.zhang@redhat.com>
> Tested-by: Yi Zhang <yi.zhang@redhat.com>
> Signed-off-by: Daniel Wagner <dwagner@suse.de>
> ---
>   


Looks good.

Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>

-ck

Re: [PATCH blktests v1] nvme/029: reserve hugepages for lager allocations

[Shinichiro Kawasaki]

On Feb 21, 2024 / 08:43, Daniel Wagner wrote:
> The test is issuing larger IO workload. This depends on being able to
> allocate larger chunks of linear memory. nvme-cli used to use libhugetlb
> to automatically allocate the HugeTLB pool. Though nvme-cli dropped the
> dependency on the library, thus the test should try to provision the
> system accordingly.
> 
> Link: https://github.com/linux-nvme/nvme-cli/issues/2218
> Reported-by: Yi Zhang <yi.zhang@redhat.com>
> Tested-by: Yi Zhang <yi.zhang@redhat.com>
> Signed-off-by: Daniel Wagner <dwagner@suse.de>

Thanks, applied.

I note here that an opportunity is left to improve stability of nvme-cli linear
memory allocation in the test case as discussed [1].

[1] https://lore.kernel.org/linux-block/99bba6e6-1ae3-49d2-842b-680257cedbad@kernel.org/