[PATCH v2 0/2] hw/nvme: add monitor commands for inspecting state

[PATCH v2 0/2] hw/nvme: add monitor commands for inspecting state

[Mateusz Nowicki]

Add two HMP commands for inspecting emulated NVMe controllers from
the QEMU monitor without attaching gdb to the QEMU process:

  - 'info nvme'        - per-controller summary (PCI, identify
                         fields, CC/CSTS/AQA, queue counts)
  - 'info nvme-queues' - per-queue listing of admin and I/O SQ/CQ
                         (size, head/tail, PRP1, doorbell offset,
                         phase tag)

Useful for verifying queue setup, doorbell rings, AERs held in the
admin SQ and similar driver/controller interaction details from a
running QEMU monitor.

Changes in v2:
  - hw/nvme/meson.build: add the missing trailing newline (Markus).
  - Pick up Acked-by tags from Dr. David Alan Gilbert and Markus
    Armbruster on both patches.

v1: https://lore.kernel.org/qemu-devel/cover.1778409416.git.mateusz.nowicki@posteo.net/

Mateusz Nowicki (2):
  hw/nvme: add 'info nvme' HMP command
  hw/nvme: add 'info nvme-queues' HMP command

 hmp-commands-info.hx |  28 +++++++++
 hw/nvme/meson.build  |   2 +-
 hw/nvme/monitor.c    | 139 +++++++++++++++++++++++++++++++++++++++++++
 qapi/machine.json    |  34 +++++++++++
 4 files changed, 202 insertions(+), 1 deletion(-)
 create mode 100644 hw/nvme/monitor.c

-- 
2.53.0

[PATCH v2 1/2] hw/nvme: add 'info nvme' HMP command

[Mateusz Nowicki]

Add an 'info nvme' HMP command for inspecting emulated NVMe
controllers from the QEMU monitor.

For each NVMe controller in the QOM tree the command prints:

  - PCI BDF, vendor/device ID and BAR0 base address;
  - Identify Controller fields visible to the host driver: SN, MN, FR
    and CNTLID;
  - the CC, CSTS and AQA registers (raw 32-bit values);
  - the number of admin and active I/O submission/completion queues.

Useful when debugging the Linux NVMe driver against the QEMU emulation
without attaching gdb to the QEMU process.

Example:

  (qemu) info nvme
  /machine/peripheral-anon/device[0]
    PCI:    BDF 00:04.0  VID=8086 DID=5845  BAR0=0x00000000feb50000
    ID:     SN=NVME0001  MN=QEMU NVMe Ctrl  FR=...  CNTLID=0x0000
    CC:     0x00460001
    CSTS:   0x00000001
    AQA:    0x001f001f
    Queues: 1 admin + 4 IO SQ / 4 IO CQ

Signed-off-by: Mateusz Nowicki <mateusz.nowicki@posteo.net>
Acked-by: Dr. David Alan Gilbert <dave@treblig.org>
Acked-by: Markus Armbruster <armbru@redhat.com>
---
 hmp-commands-info.hx | 13 ++++++++
 hw/nvme/meson.build  |  2 +-
 hw/nvme/monitor.c    | 76 ++++++++++++++++++++++++++++++++++++++++++++
 qapi/machine.json    | 17 ++++++++++
 4 files changed, 107 insertions(+), 1 deletion(-)
 create mode 100644 hw/nvme/monitor.c

diff --git a/hmp-commands-info.hx b/hmp-commands-info.hx
index 82134eb6c21..b984691c3ce 100644
--- a/hmp-commands-info.hx
+++ b/hmp-commands-info.hx
@@ -340,6 +340,19 @@ SRST
     Show guest USB devices.
 ERST
 
+    {
+        .name       = "nvme",
+        .args_type  = "",
+        .params     = "",
+        .help       = "show emulated NVMe controllers",
+        .cmd_info_hrt = qmp_x_query_nvme,
+    },
+
+SRST
+  ``info nvme``
+    Show emulated NVMe controllers.
+ERST
+
     {
         .name       = "usbhost",
         .args_type  = "",
diff --git a/hw/nvme/meson.build b/hw/nvme/meson.build
index 7d5caa53c28..33dee048b0a 100644
--- a/hw/nvme/meson.build
+++ b/hw/nvme/meson.build
@@ -1 +1 @@
-system_ss.add(when: 'CONFIG_NVME_PCI', if_true: files('ctrl.c', 'dif.c', 'ns.c', 'subsys.c', 'nguid.c'))
\ No newline at end of file
+system_ss.add(when: 'CONFIG_NVME_PCI', if_true: files('ctrl.c', 'dif.c', 'ns.c', 'subsys.c', 'nguid.c', 'monitor.c'))
diff --git a/hw/nvme/monitor.c b/hw/nvme/monitor.c
new file mode 100644
index 00000000000..95a67544376
--- /dev/null
+++ b/hw/nvme/monitor.c
@@ -0,0 +1,76 @@
+/*
+ * QEMU NVMe Controller monitor commands
+ *
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
+
+#include "qemu/osdep.h"
+#include "qapi/error.h"
+#include "qapi/qapi-commands-machine.h"
+#include "qapi/type-helpers.h"
+#include "hw/pci/pci.h"
+
+#include "nvme.h"
+
+static int collect_one(Object *obj, void *opaque)
+{
+    GString *buf = opaque;
+    NvmeCtrl *n;
+    PCIDevice *pci;
+    pcibus_t bar0;
+    unsigned io_sq = 0, io_cq = 0;
+
+    if (!object_dynamic_cast(obj, TYPE_NVME)) {
+        return 0;
+    }
+    n = NVME(obj);
+    pci = PCI_DEVICE(n);
+    bar0 = pci_get_bar_addr(pci, 0);
+
+    for (unsigned i = 1; i <= n->params.max_ioqpairs; i++) {
+        if (n->sq && n->sq[i]) {
+            io_sq++;
+        }
+        if (n->cq && n->cq[i]) {
+            io_cq++;
+        }
+    }
+
+    g_string_append_printf(buf, "%s\n", object_get_canonical_path(obj));
+    g_string_append_printf(buf,
+        "  PCI:    BDF %02x:%02x.%x  VID=%04x DID=%04x  ",
+        pci_dev_bus_num(pci), PCI_SLOT(pci->devfn), PCI_FUNC(pci->devfn),
+        pci_get_word(pci->config + PCI_VENDOR_ID),
+        pci_get_word(pci->config + PCI_DEVICE_ID));
+    if (bar0 == PCI_BAR_UNMAPPED) {
+        g_string_append(buf, "BAR0=unmapped\n");
+    } else {
+        g_string_append_printf(buf, "BAR0=0x%016" PRIx64 "\n",
+                               (uint64_t)bar0);
+    }
+    g_string_append_printf(buf,
+        "  ID:     SN=%.20s  MN=%.40s  FR=%.8s  CNTLID=0x%04x\n",
+        n->id_ctrl.sn, n->id_ctrl.mn, n->id_ctrl.fr, n->cntlid);
+    g_string_append_printf(buf, "  CC:     0x%08x\n",
+                           ldl_le_p(&n->bar.cc));
+    g_string_append_printf(buf, "  CSTS:   0x%08x\n",
+                           ldl_le_p(&n->bar.csts));
+    g_string_append_printf(buf, "  AQA:    0x%08x\n",
+                           ldl_le_p(&n->bar.aqa));
+    g_string_append_printf(buf,
+        "  Queues: 1 admin + %u IO SQ / %u IO CQ\n", io_sq, io_cq);
+    return 0;
+}
+
+HumanReadableText *qmp_x_query_nvme(Error **errp)
+{
+    g_autoptr(GString) buf = g_string_new("");
+
+    object_child_foreach_recursive(object_get_root(), collect_one, buf);
+
+    if (buf->len == 0) {
+        g_string_append(buf, "no NVMe controllers\n");
+    }
+
+    return human_readable_text_from_str(buf);
+}
diff --git a/qapi/machine.json b/qapi/machine.json
index 685e4e29b87..d4a589e7681 100644
--- a/qapi/machine.json
+++ b/qapi/machine.json
@@ -1772,6 +1772,23 @@
   'returns': 'HumanReadableText',
   'features': [ 'unstable' ] }
 
+##
+# @x-query-nvme:
+#
+# Query state of emulated NVMe controllers.
+#
+# Features:
+#
+# @unstable: This command is meant for debugging.
+#
+# Returns: NVMe controller state as human-readable text
+#
+# Since: 11.1
+##
+{ 'command': 'x-query-nvme',
+  'returns': 'HumanReadableText',
+  'features': [ 'unstable' ] }
+
 ##
 # @SmbiosEntryPointType:
 #
-- 
2.53.0

[PATCH v2 2/2] hw/nvme: add 'info nvme-queues' HMP command

[Mateusz Nowicki]

Add an 'info nvme-queues' HMP command that lists, for every emulated
NVMe controller, the admin SQ/CQ and every active I/O SQ/CQ.

For each queue the command prints:

  - the ring size, head and tail indices;
  - the associated CQ id (for SQ) or interrupt vector (for CQ);
  - the queue's PRP1 (DMA base address of the ring);
  - the BAR0-relative doorbell offset (SQyTDBL / CQyHDBL);
  - the current CQ phase tag.

The doorbell offsets are computed from CAP.DSTRD so they remain
correct if the emulation ever exposes a non-zero stride.

Useful when debugging the Linux NVMe driver against the QEMU
emulation - queue setup, doorbell rings, AERs held in the admin SQ
etc. - without attaching gdb to the QEMU process.

Example, with the long PRP1 ring address and BAR0-relative doorbell
offset replaced by '...'; phase tag column omitted for brevity:

  (qemu) info nvme-queues
  /machine/peripheral-anon/device[0]
  SQ 0  size=32    head=9     tail=9     cqid=0    prp1=...  SQTDBL=...
  CQ 0  size=32    head=8     tail=8     iv=0      prp1=...  CQHDBL=...
  SQ 1  size=1024  head=2     tail=2     cqid=1    prp1=...  SQTDBL=...
  CQ 1  size=1024  head=2     tail=2     iv=1      prp1=...  CQHDBL=...

Signed-off-by: Mateusz Nowicki <mateusz.nowicki@posteo.net>
Acked-by: Dr. David Alan Gilbert <dave@treblig.org>
Acked-by: Markus Armbruster <armbru@redhat.com>
---
 hmp-commands-info.hx | 15 +++++++++++
 hw/nvme/monitor.c    | 63 ++++++++++++++++++++++++++++++++++++++++++++
 qapi/machine.json    | 17 ++++++++++++
 3 files changed, 95 insertions(+)

diff --git a/hmp-commands-info.hx b/hmp-commands-info.hx
index b984691c3ce..874c1e19707 100644
--- a/hmp-commands-info.hx
+++ b/hmp-commands-info.hx
@@ -353,6 +353,21 @@ SRST
     Show emulated NVMe controllers.
 ERST
 
+    {
+        .name       = "nvme-queues",
+        .args_type  = "",
+        .params     = "",
+        .help       = "show active NVMe queues and their doorbells",
+        .cmd_info_hrt = qmp_x_query_nvme_queues,
+    },
+
+SRST
+  ``info nvme-queues``
+    Show all active NVMe submission and completion queues, including
+    head/tail, DMA address of the ring and BAR0-relative doorbell
+    offset.
+ERST
+
     {
         .name       = "usbhost",
         .args_type  = "",
diff --git a/hw/nvme/monitor.c b/hw/nvme/monitor.c
index 95a67544376..c8161943c08 100644
--- a/hw/nvme/monitor.c
+++ b/hw/nvme/monitor.c
@@ -74,3 +74,66 @@ HumanReadableText *qmp_x_query_nvme(Error **errp)
 
     return human_readable_text_from_str(buf);
 }
+
+static void append_sq(GString *buf, NvmeSQueue *sq, unsigned stride)
+{
+    hwaddr db_off = 0x1000 + 2 * sq->sqid * stride;
+
+    g_string_append_printf(buf,
+        "  SQ %u  size=%-5u head=%-5u tail=%-5u cqid=%-3u  "
+        "prp1=0x%016" PRIx64 "  SQTDBL=BAR0+0x%03" HWADDR_PRIx "\n",
+        sq->sqid, sq->size, sq->head, sq->tail, sq->cqid,
+        sq->dma_addr, db_off);
+}
+
+static void append_cq(GString *buf, NvmeCQueue *cq, unsigned stride)
+{
+    hwaddr db_off = 0x1000 + (2 * cq->cqid + 1) * stride;
+
+    g_string_append_printf(buf,
+        "  CQ %u  size=%-5u head=%-5u tail=%-5u iv=%-5u  "
+        "prp1=0x%016" PRIx64 "  CQHDBL=BAR0+0x%03" HWADDR_PRIx
+        "  phaseTag=%u\n",
+        cq->cqid, cq->size, cq->head, cq->tail, cq->vector,
+        cq->dma_addr, db_off, cq->phase);
+}
+
+static int collect_queues(Object *obj, void *opaque)
+{
+    GString *buf = opaque;
+    NvmeCtrl *n;
+    unsigned stride;
+
+    if (!object_dynamic_cast(obj, TYPE_NVME)) {
+        return 0;
+    }
+    n = NVME(obj);
+    stride = 4u << NVME_CAP_DSTRD(ldq_le_p(&n->bar.cap));
+
+    g_string_append_printf(buf, "%s\n", object_get_canonical_path(obj));
+    append_sq(buf, &n->admin_sq, stride);
+    append_cq(buf, &n->admin_cq, stride);
+
+    for (unsigned i = 1; i <= n->params.max_ioqpairs; i++) {
+        if (n->sq && n->sq[i]) {
+            append_sq(buf, n->sq[i], stride);
+        }
+        if (n->cq && n->cq[i]) {
+            append_cq(buf, n->cq[i], stride);
+        }
+    }
+    return 0;
+}
+
+HumanReadableText *qmp_x_query_nvme_queues(Error **errp)
+{
+    g_autoptr(GString) buf = g_string_new("");
+
+    object_child_foreach_recursive(object_get_root(), collect_queues, buf);
+
+    if (buf->len == 0) {
+        g_string_append(buf, "no NVMe controllers\n");
+    }
+
+    return human_readable_text_from_str(buf);
+}
diff --git a/qapi/machine.json b/qapi/machine.json
index d4a589e7681..c21b128e0a9 100644
--- a/qapi/machine.json
+++ b/qapi/machine.json
@@ -1789,6 +1789,23 @@
   'returns': 'HumanReadableText',
   'features': [ 'unstable' ] }
 
+##
+# @x-query-nvme-queues:
+#
+# Query state of all active SQ/CQ for emulated NVMe controllers.
+#
+# Features:
+#
+# @unstable: This command is meant for debugging.
+#
+# Returns: per-queue state as human-readable text
+#
+# Since: 11.1
+##
+{ 'command': 'x-query-nvme-queues',
+  'returns': 'HumanReadableText',
+  'features': [ 'unstable' ] }
+
 ##
 # @SmbiosEntryPointType:
 #
-- 
2.53.0

Re: [PATCH v2 0/2] hw/nvme: add monitor commands for inspecting state

[Markus Armbruster]

Mateusz Nowicki <mateusz.nowicki@posteo.net> writes:

> Add two HMP commands for inspecting emulated NVMe controllers from
> the QEMU monitor without attaching gdb to the QEMU process:
>
>   - 'info nvme'        - per-controller summary (PCI, identify
>                          fields, CC/CSTS/AQA, queue counts)
>   - 'info nvme-queues' - per-queue listing of admin and I/O SQ/CQ
>                          (size, head/tail, PRP1, doorbell offset,
>                          phase tag)
>
> Useful for verifying queue setup, doorbell rings, AERs held in the
> admin SQ and similar driver/controller interaction details from a
> running QEMU monitor.

I can take this, but I'd like at least an Acked-by from an NVME
maintainer.