scan_work improvement
Subsystems (2)
v3 cover scan_work improvement
By the scan_work ns_list is scanned then updated to latest NS state. For the ns_attach and ns_mgmt commands ns_list update needed always. Then initialize known effects to set ns_mgmt NCC and ns_attach NIC.
Tokunori Ikegami (2):
nvme: delete unnecessary empty lines
nvme: initialize known effects to set ns_mgmt NCC and ns_attach NIC
drivers/nvme/host/apple.c | 3 +-- drivers/nvme/host/core.c | 20 +++++++++++++++----- drivers/nvme/host/ioctl.c | 1 - 3 files changed, 16 insertions(+), 8 deletions(-)
Signature · 1 line
2.51.0
PATCH v3 1/2 nvme: delete unnecessary empty lines
Also fix incorrect indentation errors.
---
Changes since v1:
- Fix the incorrect indentation for the continuation of the condition.
drivers/nvme/host/apple.c | 3 +-- drivers/nvme/host/core.c | 5 +---- drivers/nvme/host/ioctl.c | 1 - 3 files changed, 2 insertions(+), 7 deletions(-)
drivers/nvme/host/apple.c +1 -2
@@ -1060,8 +1060,7 @@ static void apple_nvme_reset_work(struct work_struct *work)* or the previous stage shut it down cleanly. */ if (!(readl (anv->mmio_coproc +APPLE_ANS_COPROC_CPU_CONTROL ) & -APPLE_ANS_COPROC_CPU_CONTROL_RUN )) { - +APPLE_ANS_COPROC_CPU_CONTROL_RUN )) { ret =reset_control_assert (anv->reset);if (ret) goto out;
drivers/nvme/host/core.c +1 -4
@@ -1774,7 +1774,6 @@ static void nvme_enable_aen(struct nvme_ctrl *ctrl)static int nvme_ns_open (struct nvme_ns *ns) { -/* should never be called due to GENHD_FL_HIDDEN */ if (WARN_ON_ONCE (nvme_ns_head_multipath (ns->head))) goto fail;
@@ -1793,7 +1792,6 @@ static int nvme_ns_open(struct nvme_ns *ns)static void nvme_ns_release (struct nvme_ns *ns) { -module_put (ns->ctrl->ops->module);nvme_put_ns (ns); }
@@ -4684,14 +4682,13 @@ static void nvme_async_event_work(struct work_struct *work)* The transport drivers must guarantee AER submission here is safe by * flushing ctrl async_event_work after changing the controller state * from LIVE and before freeing the admin queue. -*/ +*/ if (nvme_ctrl_state (ctrl) ==NVME_CTRL_LIVE ) ctrl->ops->submit_async_event (ctrl); }static bool nvme_ctrl_pp_status (struct nvme_ctrl *ctrl) { -u32 csts;if (ctrl->ops->reg_read32 (ctrl,NVME_REG_CSTS , &csts))
drivers/nvme/host/ioctl.c +0 -1
@@ -638,7 +638,6 @@ long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg)static int nvme_uring_cmd_checks (unsigned int issue_flags) { -/* NVMe passthrough requires big SQE/CQE support */ if ((issue_flags & (IO_URING_F_SQE128 |IO_URING_F_CQE32 )) != (IO_URING_F_SQE128 |IO_URING_F_CQE32 ))
Signature · 1 line
2.51.0
PATCH v3 2/2 nvme: initialize known effects to set ns_mgmt NCC and ns_attach NIC
In reply to Tokunori IkegamiThis is to make sure scan_work done for the commands. Since nvme_passthru_end called when effects set. Also scan_work is flushed when NIC or NCC effects set.
The change is to improvement the scan_work to make sure. Always do scan for the ns_mgmt and ns_attach commands. All drives supporting ns_mgmt and ns_attach should have the entries. But no need to be depended on it as always do scan.
---
Changes since v2:
- Document the change explanation in both the commit message and code comments.
drivers/nvme/host/core.c | 15 ++++++++++++++- 1 file changed, 14 insertions(+), 1 deletion(-)
drivers/nvme/host/core.c +14 -1
@@ -3448,7 +3448,7 @@ static int nvme_init_effects_log(struct nvme_ctrl *ctrl,static void nvme_init_known_nvm_effects (struct nvme_ctrl *ctrl) { -struct nvme_effects_log *log = ctrl->effects; +struct nvme_effects_log *log = ctrl->effects; log->acs[nvme_admin_format_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
@@ -3473,6 +3473,19 @@ static void nvme_init_known_nvm_effects(struct nvme_ctrl *ctrl)*/ log->acs[nvme_admin_security_recv] &=cpu_to_le32 (~NVME_CMD_EFFECTS_CSE_MASK ); +/* +* This is to make sure scan_work done for the commands. +* Since nvme_passthru_end called when effects set. +* Also scan_work is flushed when NIC or NCC effects set. +* +* The change is to improvement the scan_work to make sure. +* Always do scan for the ns_mgmt and ns_attach commands. +* All drives supporting ns_mgmt and ns_attach should have the entries. +* But no need to be depended on it as always do scan. +*/ + log->acs[nvme_admin_ns_mgmt] |=cpu_to_le32 (NVME_CMD_EFFECTS_NCC ); + log->acs[nvme_admin_ns_attach] |=cpu_to_le32 (NVME_CMD_EFFECTS_NIC ); + log->iocs[nvme_cmd_write] |=cpu_to_le32 (NVME_CMD_EFFECTS_LBCC ); log->iocs[nvme_cmd_write_zeroes] |=cpu_to_le32 (NVME_CMD_EFFECTS_LBCC ); log->iocs[nvme_cmd_write_uncor] |=cpu_to_le32 (NVME_CMD_EFFECTS_LBCC );
Signature · 1 line
2.51.0
Keith Busch
maintainerRe: [PATCH v3 2/2] nvme: initialize known effects to set ns_mgmt NCC and ns_attach NIC
Quoted 8 lines · Tokunori Ikegami
This is to make sure
scan_workdone for the commands. Sincenvme_passthru_endcalled when effects set. Alsoscan_workis flushed when NIC or NCC effects set.The change is to improvement the
scan_workto make sure. Always do scan for thens_mgmtandns_attachcommands. All drives supportingns_mgmtandns_attachshould have the entries. But no need to be depended on it as always do scan.
I suppose that's why the "known_effects" exists, but it'd be dissappointing if we really need this. The other opcodes in there existed before the Command Effects Log was created, so understandable that controllers may not implement it. But NS management/attach were introduced after the log, and there is also the AEN that the driver could rescan on receipt, so it's doubly bad if a controller implementation really needs this.
Re: [PATCH v3 2/2] nvme: initialize known effects to set ns_mgmt NCC and ns_attach NIC
Quoted 7 lines · Keith Busch
I suppose that's why the "
known_effects" exists, but it'd be dissappointing if we really need this. The other opcodes in there existed before the Command Effects Log was created, so understandable that controllers may not implement it. But NS management/attach were introduced after the log, and there is also the AEN that the driver could rescan on receipt, so it's doubly bad if a controller implementation really needs this.
Just rechecked the specification again then the log page identifier: commands supported and effects mentioned is optional for NVM express revision 1.4 and earlier then for the older controller the changes not bad I think. By the way about the scan_work executions for the command effects log and AEN I thought it is better to be handled only once but I do not have an actual idea for the exclusive control. Thank you.
Keith Busch
maintainerRe: [PATCH v3 2/2] nvme: initialize known effects to set ns_mgmt NCC and ns_attach NIC
Quoted 9 lines · Tokunori Ikegami
Quoted 7 lines · Keith Busch
I suppose that's why the "
known_effects" exists, but it'd be dissappointing if we really need this. The other opcodes in there existed before the Command Effects Log was created, so understandable that controllers may not implement it. But NS management/attach were introduced after the log, and there is also the AEN that the driver could rescan on receipt, so it's doubly bad if a controller implementation really needs this.
Quoted 4 lines
Just rechecked the specification again then the log page identifier:
commands supported and effects mentioned is optional for NVM express revision 1.4 and earlier then for the older controller the changes not bad I think.
Yes, but the namespace management commands are also optional, so it sounds weird to me for an implementation to choose to support the optional command that's more difficult to do and skip the easy log page that advertises support for it.
Quoted 3 lines
By the way about the
scan_workexecutions for the command effects log and AEN I thought it is better to be handled only once but I do not have an actual idea for the exclusive control.
You can attach namespaces to controllers driven by a different host, so the AEN is the only way to trigger a rescan in such a scenario without admin intervention. An extra scan_work is otherwise harmless, I don't see a particular need to introduce a mechanism to avoid it.
Re: [PATCH v3 2/2] nvme: initialize known effects to set ns_mgmt NCC and ns_attach NIC
Quoted 4 lines · Keith Busch
Yes, but the namespace management commands are also optional, so it sounds weird to me for an implementation to choose to support the optional command that's more difficult to do and skip the easy log page that advertises support for it.
Okay understood as the changes not necessary.
Quoted 4 lines
You can attach namespaces to controllers driven by a different host, so the AEN is the only way to trigger a rescan in such a scenario without admin intervention. An extra
scan_workis otherwise harmless, I don't see a particular need to introduce a mechanism to avoid it.
Noted this also. Thank you.