From: Klaus Jensen <its@irrelevant.dk>
To: qemu-devel@nongnu.org
Cc: Peter Maydell <peter.maydell@linaro.org>,
Alan Adamson <alan.adamson@oracle.com>,
Klaus Jensen <k.jensen@samsung.com>,
Keith Busch <kbusch@kernel.org>, Klaus Jensen <its@irrelevant.dk>,
Jesper Devantier <foss@defmacro.it>,
qemu-block@nongnu.org
Subject: [PULL 5/5] hw/nvme: add atomic write support
Date: Tue, 1 Oct 2024 09:04:17 +0200 [thread overview]
Message-ID: <20241001070418.28737-6-its@irrelevant.dk> (raw)
In-Reply-To: <20241001070418.28737-1-its@irrelevant.dk>
From: Alan Adamson <alan.adamson@oracle.com>
Adds support for the controller atomic parameters: AWUN and AWUPF. Atomic
Compare and Write Unit (ACWU) is not currently supported.
Writes that adhere to the ACWU and AWUPF parameters are guaranteed to be atomic.
New NVMe QEMU Parameters (See NVMe Specification for details):
atomic.dn (default off) - Set the value of Disable Normal.
atomic.awun=UINT16 (default: 0)
atomic.awupf=UINT16 (default: 0)
By default (Disable Normal set to zero), the maximum atomic write size is
set to the AWUN value. If Disable Normal is set, the maximum atomic write
size is set to AWUPF.
Signed-off-by: Alan Adamson <alan.adamson@oracle.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
---
hw/nvme/ctrl.c | 158 ++++++++++++++++++++++++++++++++++++++++++++++++-
hw/nvme/nvme.h | 12 ++++
2 files changed, 169 insertions(+), 1 deletion(-)
diff --git a/hw/nvme/ctrl.c b/hw/nvme/ctrl.c
index 119adec31274..f4e89203c1a6 100644
--- a/hw/nvme/ctrl.c
+++ b/hw/nvme/ctrl.c
@@ -40,6 +40,9 @@
* sriov_vi_flexible=<N[optional]> \
* sriov_max_vi_per_vf=<N[optional]> \
* sriov_max_vq_per_vf=<N[optional]> \
+ * atomic.dn=<on|off[optional]>, \
+ * atomic.awun<N[optional]>, \
+ * atomic.awupf<N[optional]>, \
* subsys=<subsys_id>
* -device nvme-ns,drive=<drive_id>,bus=<bus_name>,nsid=<nsid>,\
* zoned=<true|false[optional]>, \
@@ -254,6 +257,7 @@ static const uint32_t nvme_feature_cap[NVME_FID_MAX] = {
[NVME_ERROR_RECOVERY] = NVME_FEAT_CAP_CHANGE | NVME_FEAT_CAP_NS,
[NVME_VOLATILE_WRITE_CACHE] = NVME_FEAT_CAP_CHANGE,
[NVME_NUMBER_OF_QUEUES] = NVME_FEAT_CAP_CHANGE,
+ [NVME_WRITE_ATOMICITY] = NVME_FEAT_CAP_CHANGE,
[NVME_ASYNCHRONOUS_EVENT_CONF] = NVME_FEAT_CAP_CHANGE,
[NVME_TIMESTAMP] = NVME_FEAT_CAP_CHANGE,
[NVME_HOST_BEHAVIOR_SUPPORT] = NVME_FEAT_CAP_CHANGE,
@@ -6309,8 +6313,10 @@ defaults:
if (ret) {
return ret;
}
- goto out;
+ break;
+ case NVME_WRITE_ATOMICITY:
+ result = n->dn;
break;
default:
result = nvme_feature_default[fid];
@@ -6394,6 +6400,8 @@ static uint16_t nvme_set_feature(NvmeCtrl *n, NvmeRequest *req)
uint8_t save = NVME_SETFEAT_SAVE(dw10);
uint16_t status;
int i;
+ NvmeIdCtrl *id = &n->id_ctrl;
+ NvmeAtomic *atomic = &n->atomic;
trace_pci_nvme_setfeat(nvme_cid(req), nsid, fid, save, dw11);
@@ -6546,6 +6554,22 @@ static uint16_t nvme_set_feature(NvmeCtrl *n, NvmeRequest *req)
return NVME_CMD_SEQ_ERROR | NVME_DNR;
case NVME_FDP_EVENTS:
return nvme_set_feature_fdp_events(n, ns, req);
+ case NVME_WRITE_ATOMICITY:
+
+ n->dn = 0x1 & dw11;
+
+ if (n->dn) {
+ atomic->atomic_max_write_size = le16_to_cpu(id->awupf) + 1;
+ } else {
+ atomic->atomic_max_write_size = le16_to_cpu(id->awun) + 1;
+ }
+
+ if (atomic->atomic_max_write_size == 1) {
+ atomic->atomic_writes = 0;
+ } else {
+ atomic->atomic_writes = 1;
+ }
+ break;
default:
return NVME_FEAT_NOT_CHANGEABLE | NVME_DNR;
}
@@ -7243,6 +7267,81 @@ static void nvme_update_sq_tail(NvmeSQueue *sq)
trace_pci_nvme_update_sq_tail(sq->sqid, sq->tail);
}
+#define NVME_ATOMIC_NO_START 0
+#define NVME_ATOMIC_START_ATOMIC 1
+#define NVME_ATOMIC_START_NONATOMIC 2
+
+static int nvme_atomic_write_check(NvmeCtrl *n, NvmeCmd *cmd,
+ NvmeAtomic *atomic)
+{
+ NvmeRwCmd *rw = (NvmeRwCmd *)cmd;
+ uint64_t slba = le64_to_cpu(rw->slba);
+ uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb);
+ uint64_t elba = slba + nlb;
+ bool cmd_atomic_wr = true;
+ int i;
+
+ if ((cmd->opcode == NVME_CMD_READ) || ((cmd->opcode == NVME_CMD_WRITE) &&
+ ((rw->nlb + 1) > atomic->atomic_max_write_size))) {
+ cmd_atomic_wr = false;
+ }
+
+ /*
+ * Walk the queues to see if there are any atomic conflicts.
+ */
+ for (i = 1; i < n->params.max_ioqpairs + 1; i++) {
+ NvmeSQueue *sq;
+ NvmeRequest *req;
+ NvmeRwCmd *req_rw;
+ uint64_t req_slba;
+ uint32_t req_nlb;
+ uint64_t req_elba;
+
+ sq = n->sq[i];
+ if (!sq) {
+ continue;
+ }
+
+ /*
+ * Walk all the requests on a given queue.
+ */
+ QTAILQ_FOREACH(req, &sq->out_req_list, entry) {
+ req_rw = (NvmeRwCmd *)&req->cmd;
+
+ if (((req_rw->opcode == NVME_CMD_WRITE) ||
+ (req_rw->opcode == NVME_CMD_READ)) &&
+ (cmd->nsid == req->ns->params.nsid)) {
+ req_slba = le64_to_cpu(req_rw->slba);
+ req_nlb = (uint32_t)le16_to_cpu(req_rw->nlb);
+ req_elba = req_slba + req_nlb;
+
+ if (cmd_atomic_wr) {
+ if ((elba >= req_slba) && (slba <= req_elba)) {
+ return NVME_ATOMIC_NO_START;
+ }
+ } else {
+ if (req->atomic_write && ((elba >= req_slba) &&
+ (slba <= req_elba))) {
+ return NVME_ATOMIC_NO_START;
+ }
+ }
+ }
+ }
+ }
+ if (cmd_atomic_wr) {
+ return NVME_ATOMIC_START_ATOMIC;
+ }
+ return NVME_ATOMIC_START_NONATOMIC;
+}
+
+static NvmeAtomic *nvme_get_atomic(NvmeCtrl *n, NvmeCmd *cmd)
+{
+ if (n->atomic.atomic_writes) {
+ return &n->atomic;
+ }
+ return NULL;
+}
+
static void nvme_process_sq(void *opaque)
{
NvmeSQueue *sq = opaque;
@@ -7259,6 +7358,9 @@ static void nvme_process_sq(void *opaque)
}
while (!(nvme_sq_empty(sq) || QTAILQ_EMPTY(&sq->req_list))) {
+ NvmeAtomic *atomic;
+ bool cmd_is_atomic;
+
addr = sq->dma_addr + (sq->head << NVME_SQES);
if (nvme_addr_read(n, addr, (void *)&cmd, sizeof(cmd))) {
trace_pci_nvme_err_addr_read(addr);
@@ -7266,6 +7368,26 @@ static void nvme_process_sq(void *opaque)
stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
break;
}
+
+ atomic = nvme_get_atomic(n, &cmd);
+
+ cmd_is_atomic = false;
+ if (sq->sqid && atomic) {
+ int ret;
+
+ ret = nvme_atomic_write_check(n, &cmd, atomic);
+ switch (ret) {
+ case NVME_ATOMIC_NO_START:
+ qemu_bh_schedule(sq->bh);
+ return;
+ case NVME_ATOMIC_START_ATOMIC:
+ cmd_is_atomic = true;
+ break;
+ case NVME_ATOMIC_START_NONATOMIC:
+ default:
+ break;
+ }
+ }
nvme_inc_sq_head(sq);
req = QTAILQ_FIRST(&sq->req_list);
@@ -7275,6 +7397,10 @@ static void nvme_process_sq(void *opaque)
req->cqe.cid = cmd.cid;
memcpy(&req->cmd, &cmd, sizeof(NvmeCmd));
+ if (sq->sqid && atomic) {
+ req->atomic_write = cmd_is_atomic;
+ }
+
status = sq->sqid ? nvme_io_cmd(n, req) :
nvme_admin_cmd(n, req);
if (status != NVME_NO_COMPLETE) {
@@ -7378,6 +7504,8 @@ static void nvme_ctrl_reset(NvmeCtrl *n, NvmeResetType rst)
n->outstanding_aers = 0;
n->qs_created = false;
+ n->dn = n->params.atomic_dn; /* Set Disable Normal */
+
nvme_update_msixcap_ts(pci_dev, n->conf_msix_qsize);
if (pci_is_vf(pci_dev)) {
@@ -8154,6 +8282,8 @@ static void nvme_init_state(NvmeCtrl *n)
NvmeSecCtrlEntry *list = n->sec_ctrl_list;
NvmeSecCtrlEntry *sctrl;
PCIDevice *pci = PCI_DEVICE(n);
+ NvmeAtomic *atomic = &n->atomic;
+ NvmeIdCtrl *id = &n->id_ctrl;
uint8_t max_vfs;
int i;
@@ -8211,6 +8341,29 @@ static void nvme_init_state(NvmeCtrl *n)
cpu_to_le16(n->params.sriov_max_vi_per_vf) :
cap->vifrt / MAX(max_vfs, 1);
}
+
+ /* Atomic Write */
+ id->awun = cpu_to_le16(n->params.atomic_awun);
+ id->awupf = cpu_to_le16(n->params.atomic_awupf);
+ n->dn = n->params.atomic_dn;
+
+ if (id->awun || id->awupf) {
+ if (id->awupf > id->awun) {
+ id->awupf = 0;
+ }
+
+ if (n->dn) {
+ atomic->atomic_max_write_size = id->awupf + 1;
+ } else {
+ atomic->atomic_max_write_size = id->awun + 1;
+ }
+
+ if (atomic->atomic_max_write_size == 1) {
+ atomic->atomic_writes = 0;
+ } else {
+ atomic->atomic_writes = 1;
+ }
+ }
}
static void nvme_init_cmb(NvmeCtrl *n, PCIDevice *pci_dev)
@@ -8756,6 +8909,9 @@ static Property nvme_props[] = {
DEFINE_PROP_UINT16("mqes", NvmeCtrl, params.mqes, 0x7ff),
DEFINE_PROP_UINT16("spdm_port", PCIDevice, spdm_port, 0),
DEFINE_PROP_BOOL("ctratt.mem", NvmeCtrl, params.ctratt.mem, false),
+ DEFINE_PROP_BOOL("atomic.dn", NvmeCtrl, params.atomic_dn, 0),
+ DEFINE_PROP_UINT16("atomic.awun", NvmeCtrl, params.atomic_awun, 0),
+ DEFINE_PROP_UINT16("atomic.awupf", NvmeCtrl, params.atomic_awupf, 0),
DEFINE_PROP_END_OF_LIST(),
};
diff --git a/hw/nvme/nvme.h b/hw/nvme/nvme.h
index bd3c6ba33a67..7566b316d12d 100644
--- a/hw/nvme/nvme.h
+++ b/hw/nvme/nvme.h
@@ -220,6 +220,11 @@ typedef struct NvmeNamespaceParams {
} fdp;
} NvmeNamespaceParams;
+typedef struct NvmeAtomic {
+ uint32_t atomic_max_write_size;
+ bool atomic_writes;
+} NvmeAtomic;
+
typedef struct NvmeNamespace {
DeviceState parent_obj;
BlockConf blkconf;
@@ -421,6 +426,7 @@ typedef struct NvmeRequest {
NvmeCmd cmd;
BlockAcctCookie acct;
NvmeSg sg;
+ bool atomic_write;
QTAILQ_ENTRY(NvmeRequest)entry;
} NvmeRequest;
@@ -542,6 +548,10 @@ typedef struct NvmeParams {
struct {
bool mem;
} ctratt;
+
+ uint16_t atomic_awun;
+ uint16_t atomic_awupf;
+ bool atomic_dn;
} NvmeParams;
typedef struct NvmeCtrl {
@@ -623,6 +633,8 @@ typedef struct NvmeCtrl {
uint16_t vqrfap;
uint16_t virfap;
} next_pri_ctrl_cap; /* These override pri_ctrl_cap after reset */
+ uint32_t dn; /* Disable Normal */
+ NvmeAtomic atomic;
} NvmeCtrl;
typedef enum NvmeResetType {
--
2.45.2
next prev parent reply other threads:[~2024-10-01 14:13 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-10-01 7:04 [PULL 0/5] nvme queue Klaus Jensen
2024-10-01 7:04 ` [PULL 1/5] hw/nvme: report id controller metadata sgl support Klaus Jensen
2024-10-01 7:04 ` [PULL 2/5] hw/nvme: clear masked events from the aer queue Klaus Jensen
2024-10-01 7:04 ` [PULL 3/5] hw/nvme: support CTRATT.MEM Klaus Jensen
2024-10-01 7:04 ` [PULL 4/5] hw/nvme: add knob for CTRATT.MEM Klaus Jensen
2024-10-01 7:04 ` Klaus Jensen [this message]
2024-10-01 12:09 ` [PULL 0/5] nvme queue Peter Maydell
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20241001070418.28737-6-its@irrelevant.dk \
--to=its@irrelevant.dk \
--cc=alan.adamson@oracle.com \
--cc=foss@defmacro.it \
--cc=k.jensen@samsung.com \
--cc=kbusch@kernel.org \
--cc=peter.maydell@linaro.org \
--cc=qemu-block@nongnu.org \
--cc=qemu-devel@nongnu.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).