Skip to contentSkip to conversation

NVMe: Initial commit

PATCH In review Keith Busch 3 participants 3 messages Mar 19, 2013 – Apr 3, 2013

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 NVMe: Initial commit

Initial commit for emulated nvme pci device. Implements the minimum from the specification to work with existing drivers.

Just going for the simplest patch possible for this device. If this is okay, the remainining features implementing the full specification won't be far behind.

I've incorporated all the comments received. The only thing I didn't see how to make use of was the new initialization APIs; it doesn't look like there is anything for me to do with this device in "instance_init".

Thanks for all the feedback!

---
 MAINTAINERS             |    5 +
 default-configs/pci.mak |    1 +
 hw/Makefile.objs        |    2 +
 hw/nvme.c               |  930 +++++++++++++++++++++++++++++++++++++++++++++++
 hw/nvme.h               |  713 ++++++++++++++++++++++++++++++++++++
 hw/pci/pci_ids.h        |    1 +
 6 files changed, 1652 insertions(+), 0 deletions(-)
 create mode 100644 hw/nvme.c
 create mode 100644 hw/nvme.h
MAINTAINERS +5 -0
index 0ca7e1d..9bd8956 100644
@@ -585,6 +585,11 @@ S: Supported
 F: hw/virtio-serial*
 F: hw/virtio-console*
 
+nvme
+M: Keith Busch <[email protected]>
+S: Supported
+F: hw/nvme*
+
 Xilinx EDK
 M: Peter Crosthwaite <[email protected]>
 M: Edgar E. Iglesias <[email protected]>
default-configs/pci.mak +1 -0
index ee2d18d..56565e7 100644
@@ -11,6 +11,7 @@ CONFIG_PCNET_PCI=y
 CONFIG_PCNET_COMMON=y
 CONFIG_LSI_SCSI_PCI=y
 CONFIG_MEGASAS_SCSI_PCI=y
+CONFIG_NVME_PCI=y
 CONFIG_RTL8139_PCI=y
 CONFIG_E1000_PCI=y
 CONFIG_IDE_CORE=y
hw/Makefile.objs +2 -0
index 11812c6..b92d96f 100644
@@ -46,6 +46,8 @@ common-obj-$(CONFIG_I8259) += i8259_common.o i8259.o
 common-obj-$(CONFIG_SDHCI) += sdhci.o
 common-obj-y += pam.o
 
+common-obj-$(CONFIG_NVME_PCI) += nvme.o
+
 # PPC devices
 common-obj-$(CONFIG_PREP_PCI) += prep_pci.o
 common-obj-$(CONFIG_I82378) += i82378.o
hw/nvme.c new file +930 -0

930 lines not on this page · raw message

hw/nvme.h new file +713 -0

713 lines not on this page · raw message

hw/pci/pci_ids.h +1 -0
index d8dc2f1..08f8161 100644
@@ -19,6 +19,7 @@
 #define PCI_CLASS_STORAGE_IDE            0x0101
 #define PCI_CLASS_STORAGE_RAID           0x0104
 #define PCI_CLASS_STORAGE_SATA           0x0106
+#define PCI_CLASS_STORAGE_EXPRESS        0x0108
 #define PCI_CLASS_STORAGE_OTHER          0x0180
 
 #define PCI_CLASS_NETWORK_ETHERNET       0x0200
Signature · 1 line
1.7.0.4
#1 lore raw
Quoted 14 lines · Keith Busch
hw/nvme.c new file +14 -0
+static int nvme_init(PCIDevice *pci_dev)
+{
+    NvmeCtrl *n = NVME(pci_dev);
+    NvmeIdCtrl *id = &n->id_ctrl;
+
+    int i;
+    int64_t bs_size;
+    char serial[sizeof(id->sn)];
+    uint8_t *pci_conf;
+
+    if (!(n->conf.bs)) {
+        return -1;
+    }
+

hw/nvme.c line 802

Quoted 10 lines
hw/nvme.c new file +10 -0
+    bs_size =  bdrv_getlength(n->conf.bs);
+    if (bs_size <= 0) {
+        return -1;
+    }
+
+    n->instance = instance++;
+    blkconf_serial(&n->conf, &n->serial);
+    if (!n->serial) {
+        snprintf((char *)serial, sizeof(serial), "NVMeQx10%02x", n->instance);
+    }

A patch that adds serial autogeneration to virtio-blk was recently posted. Here were my concerns:

Autogenerated IDs have been proposed (for other devices?) before and I think we should avoid them.

The autogenerated serial is not persistent. It can change between QEMU invocations - due to internal changes in QEMU or because the management tool reordered -device options.

Users will be confused and their guests may stop working if they depend on an ID like this.

The solution is to do persistent naming either by really passing -drive serial= or with udev inside the guest using the bus address (PCI devfn) like the new persistent network interface naming for Linux.

Stefan

#2 lore raw
Quoted 30 lines · Stefan Hajnoczi
Quoted 23 lines · Keith Busch
hw/nvme.c new file +23 -0
+static int nvme_init(PCIDevice *pci_dev)
+{
+    NvmeCtrl *n = NVME(pci_dev);
+    NvmeIdCtrl *id = &n->id_ctrl;
+
+    int i;
+    int64_t bs_size;
+    char serial[sizeof(id->sn)];
+    uint8_t *pci_conf;
+
+    if (!(n->conf.bs)) {
+        return -1;
+    }
+
+    bs_size =  bdrv_getlength(n->conf.bs);
+    if (bs_size <= 0) {
+        return -1;
+    }
+
+    n->instance = instance++;
+    blkconf_serial(&n->conf, &n->serial);
+    if (!n->serial) {
+ snprintf((char *)serial, sizeof(serial), "NVMeQx10%02x",
Quoted 2 lines

n->instance);
+ }

A patch that adds serial autogeneration to virtio-blk was recently posted. Here were my concerns:

Quoted 2 lines

Autogenerated IDs have been proposed (for other devices?) before and I think we should avoid them.

ACK

Quoted 6 lines

The autogenerated serial is not persistent. It can change between QEMU invocations - due to internal changes in QEMU or because the management tool reordered -device options.

Users will be confused and their guests may stop working if they depend on an ID like this.

Only IDE devices auto-generate their serial number. Can't fix that mistake, because it's ABI[*]. All other devices don't. Let's keep it that way.

Quoted 3 lines

The solution is to do persistent naming either by really passing -drive serial= or with udev inside the guest using the bus address (PCI devfn) like the new persistent network interface naming for Linux.

ACK

[*] I could have avoided the mistake for ide-hd and ide-cd, but I didn't see the problem then. Perhaps we could've killed it for AHCI.