Skip to contentSkip to conversation

hw/block/nvme: dulbe and dsm support

PATCH Superseded Klaus Jensen 1 participant 5 messages Oct 27, 2020 – Oct 30, 2020

v7 of 8 Superseded Latest is v8 →

lore
Share & tools
lore
#0 lore raw

v7 cover hw/block/nvme: dulbe and dsm support

From: Klaus Jensen <[email protected]>

This adds support for the Deallocated or Unwritten Logical Block error recovery feature as well as the Dataset Management command.

v7:
- Handle negative return value from bdrv_block_status.
- bdrv_get_info may not be supported on all block drivers, so do not consider it a fatal error.

v6:
- Skip the allocation of the discards integer and just use the opaque value directly (Philippe)
- Split changes to include/block/nvme.h into a separate patch (Philippe)
- Clean up some convoluted checks on the discards value (Philippe)
- Use unambiguous units in the commit messages (Philippe)
- Stack allocate the range array (Keith)

v5:
- Restore status code from callback (Keith)

v4:
- Removed mixed declaration and code (Keith)
- Set NPDG and NPDA and account for the blockdev cluster size.

Klaus Jensen (3):
hw/block/nvme: add dulbe support
nvme: add namespace I/O optimization fields to shared header hw/block/nvme: add the dataset management command

 hw/block/nvme-ns.h    |   4 +
 hw/block/nvme.h       |   2 +
 include/block/nvme.h  |  12 ++-
 hw/block/nvme-ns.c    |  34 ++++++--
 hw/block/nvme.c       | 193 +++++++++++++++++++++++++++++++++++++++++-
 hw/block/trace-events |   4 +
 6 files changed, 240 insertions(+), 9 deletions(-)
Signature · 1 line
2.29.1
#1 lore raw

PATCH v7 1/3 hw/block/nvme: add dulbe support

From: Klaus Jensen <[email protected]>

Add support for reporting the Deallocated or Unwritten Logical Block Error (DULBE).

Rely on the block status flags reported by the block layer and consider any block with the BDRV_BLOCK_ZERO flag to be deallocated.

Multiple factors affect when a Write Zeroes command result in deallocation of blocks.

* the underlying file system block size
* the blockdev format
* the 'discard' and 'logical_block_size' parameters

format | discard | wz (512B) wz (4KiB) wz (64KiB)
-----------------------------------------------------
qcow2 ignore n n y
qcow2 unmap n n y
raw ignore n y y
raw unmap n y y

So, this works best with an image in raw format and 4KiB LBAs, since holes can then be punched on a per-block basis (this assumes a file system with a 4kb block size, YMMV). A qcow2 image, uses a cluster size of 64KiB by default and blocks will only be marked deallocated if a full cluster is zeroed or discarded. However, this *is* consistent with the spec since Write Zeroes "should" deallocate the block if the Deallocate attribute is set and "may" deallocate if the Deallocate attribute is not set. Thus, we always try to deallocate (the BDRV_REQ_MAY_UNMAP flag is always set).

---
 hw/block/nvme-ns.h    |  4 ++
 include/block/nvme.h  |  5 +++
 hw/block/nvme-ns.c    |  8 ++--
 hw/block/nvme.c       | 91 ++++++++++++++++++++++++++++++++++++++++++-
 hw/block/trace-events |  4 ++
 5 files changed, 107 insertions(+), 5 deletions(-)
hw/block/nvme-ns.h +4 -0
index 83734f4606e1..44bf6271b744 100644
@@ -31,6 +31,10 @@ typedef struct NvmeNamespace {
     NvmeIdNs     id_ns;
 
     NvmeNamespaceParams params;
+
+    struct {
+        uint32_t err_rec;
+    } features;
 } NvmeNamespace;
 
 static inline uint32_t nvme_nsid(NvmeNamespace *ns)
include/block/nvme.h +5 -0
index 8a46d9cf015f..966c3bb304bd 100644
@@ -687,6 +687,7 @@ enum NvmeStatusCodes {
     NVME_E2E_REF_ERROR          = 0x0284,
     NVME_CMP_FAILURE            = 0x0285,
     NVME_ACCESS_DENIED          = 0x0286,
+    NVME_DULB                   = 0x0287,
     NVME_MORE                   = 0x2000,
     NVME_DNR                    = 0x4000,
     NVME_NO_COMPLETE            = 0xffff,
@@ -903,6 +904,9 @@ enum NvmeIdCtrlLpa {
 #define NVME_AEC_NS_ATTR(aec)       ((aec >> 8) & 0x1)
 #define NVME_AEC_FW_ACTIVATION(aec) ((aec >> 9) & 0x1)
 
+#define NVME_ERR_REC_TLER(err_rec)  (err_rec & 0xffff)
+#define NVME_ERR_REC_DULBE(err_rec) (err_rec & 0x10000)
+
 enum NvmeFeatureIds {
     NVME_ARBITRATION                = 0x1,
     NVME_POWER_MANAGEMENT           = 0x2,
@@ -1023,6 +1027,7 @@ enum NvmeNsIdentifierType {
 
 
 #define NVME_ID_NS_NSFEAT_THIN(nsfeat)      ((nsfeat & 0x1))
+#define NVME_ID_NS_NSFEAT_DULBE(nsfeat)     ((nsfeat >> 2) & 0x1)
 #define NVME_ID_NS_FLBAS_EXTENDED(flbas)    ((flbas >> 4) & 0x1)
 #define NVME_ID_NS_FLBAS_INDEX(flbas)       ((flbas & 0xf))
 #define NVME_ID_NS_MC_SEPARATE(mc)          ((mc >> 1) & 0x1)
hw/block/nvme-ns.c +5 -3
index 31c80cdf5b5f..f1cc734c60f5 100644
@@ -33,9 +33,7 @@ static void nvme_ns_init(NvmeNamespace *ns)
     NvmeIdNs *id_ns = &ns->id_ns;
     int lba_index = NVME_ID_NS_FLBAS_INDEX(ns->id_ns.flbas);
 
-    if (blk_get_flags(ns->blkconf.blk) & BDRV_O_UNMAP) {
-        ns->id_ns.dlfeat = 0x9;
-    }
+    ns->id_ns.dlfeat = 0x9;
 
     id_ns->lbaf[lba_index].ds = 31 - clz32(ns->blkconf.logical_block_size);
 
@@ -44,6 +42,9 @@ static void nvme_ns_init(NvmeNamespace *ns)
     /* no thin provisioning */
     id_ns->ncap = id_ns->nsze;
     id_ns->nuse = id_ns->ncap;
+
+    /* support DULBE */
+    id_ns->nsfeat |= 0x4;
 }
 
 static int nvme_ns_init_blk(NvmeCtrl *n, NvmeNamespace *ns, Error **errp)
@@ -92,6 +93,7 @@ int nvme_ns_setup(NvmeCtrl *n, NvmeNamespace *ns, Error **errp)
     }
 
     nvme_ns_init(ns);
+
     if (nvme_register_namespace(n, ns, errp)) {
         return -1;
     }
hw/block/nvme.c +89 -2

143 lines not on this page · raw message

hw/block/trace-events +4 -0
index c1537e3ac0b0..1ffe0b3f76b5 100644
@@ -43,6 +43,10 @@ pci_nvme_admin_cmd(uint16_t cid, uint16_t sqid, uint8_t opcode, const char *opna
 pci_nvme_rw(uint16_t cid, const char *verb, uint32_t nsid, uint32_t nlb, uint64_t count, uint64_t lba) "cid %"PRIu16" opname '%s' nsid %"PRIu32" nlb %"PRIu32" count %"PRIu64" lba 0x%"PRIx64""
 pci_nvme_rw_cb(uint16_t cid, const char *blkname) "cid %"PRIu16" blk '%s'"
 pci_nvme_write_zeroes(uint16_t cid, uint32_t nsid, uint64_t slba, uint32_t nlb) "cid %"PRIu16" nsid %"PRIu32" slba %"PRIu64" nlb %"PRIu32""
+pci_nvme_block_status(int64_t offset, int64_t bytes, int64_t pnum, int ret, bool zeroed) "offset %"PRId64" bytes %"PRId64" pnum %"PRId64" ret 0x%x zeroed %d"
+pci_nvme_dsm(uint16_t cid, uint32_t nsid, uint32_t nr, uint32_t attr) "cid %"PRIu16" nsid %"PRIu32" nr %"PRIu32" attr 0x%"PRIx32""
+pci_nvme_dsm_deallocate(uint16_t cid, uint32_t nsid, uint64_t slba, uint32_t nlb) "cid %"PRIu16" nsid %"PRIu32" slba %"PRIu64" nlb %"PRIu32""
+pci_nvme_aio_discard_cb(uint16_t cid) "cid %"PRIu16""
 pci_nvme_create_sq(uint64_t addr, uint16_t sqid, uint16_t cqid, uint16_t qsize, uint16_t qflags) "create submission queue, addr=0x%"PRIx64", sqid=%"PRIu16", cqid=%"PRIu16", qsize=%"PRIu16", qflags=%"PRIu16""
 pci_nvme_create_cq(uint64_t addr, uint16_t cqid, uint16_t vector, uint16_t size, uint16_t qflags, int ien) "create completion queue, addr=0x%"PRIx64", cqid=%"PRIu16", vector=%"PRIu16", qsize=%"PRIu16", qflags=%"PRIu16", ien=%d"
 pci_nvme_del_sq(uint16_t qid) "deleting submission queue sqid=%"PRIu16""
Signature · 1 line
2.29.1
#2 lore raw

PATCH v7 2/3 nvme: add namespace I/O optimization fields to shared header

In reply to Klaus Jensen

From: Klaus Jensen <[email protected]>

This adds the NPWG, NPWA, NPDG, NPDA and NOWS family of fields to the shared nvme.h header for use by later patches.

---
 include/block/nvme.h | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
include/block/nvme.h +6 -1
index 966c3bb304bd..e95ff6ca9b37 100644
@@ -990,7 +990,12 @@ typedef struct QEMU_PACKED NvmeIdNs {
     uint16_t    nabspf;
     uint16_t    noiob;
     uint8_t     nvmcap[16];
-    uint8_t     rsvd64[40];
+    uint16_t    npwg;
+    uint16_t    npwa;
+    uint16_t    npdg;
+    uint16_t    npda;
+    uint16_t    nows;
+    uint8_t     rsvd74[30];
     uint8_t     nguid[16];
     uint64_t    eui64;
     NvmeLBAF    lbaf[16];
Signature · 1 line
2.29.1
#3 lore raw

PATCH v7 3/3 hw/block/nvme: add the dataset management command

In reply to Klaus Jensen

From: Klaus Jensen <[email protected]>

Add support for the Dataset Management command and the Deallocate attribute. Deallocation results in discards being sent to the underlying block device. Whether of not the blocks are actually deallocated is affected by the same factors as Write Zeroes (see previous commit).

format | discard | dsm (512B) dsm (4KiB) dsm (64KiB)
--------------------------------------------------------
qcow2 ignore n n n
qcow2 unmap n n y
raw ignore n n n
raw unmap n y y

Again, a raw format and 4KiB LBAs are preferable.

In order to set the Namespace Preferred Deallocate Granularity and Alignment fields (NPDG and NPDA), choose a sane minimum discard granularity of 4KiB. If we are using a passthru device supporting discard at a 512B granularity, user should set the discard_granularity property explicitly. NPDG and NPDA will also account for the cluster_size of the block driver if required (i.e. for QCOW2).

See NVM Express 1.3d, Section 6.7 ("Dataset Management command").

---
 hw/block/nvme.h    |   2 +
 hw/block/nvme-ns.c |  30 +++++++++++--
 hw/block/nvme.c    | 102 ++++++++++++++++++++++++++++++++++++++++++++-
 3 files changed, 129 insertions(+), 5 deletions(-)
hw/block/nvme.h +2 -0
index e080a2318a50..574333caa3f9 100644
@@ -28,6 +28,7 @@ typedef struct NvmeRequest {
     struct NvmeNamespace    *ns;
     BlockAIOCB              *aiocb;
     uint16_t                status;
+    void                    *opaque;
     NvmeCqe                 cqe;
     NvmeCmd                 cmd;
     BlockAcctCookie         acct;
@@ -60,6 +61,7 @@ static inline const char *nvme_io_opc_str(uint8_t opc)
     case NVME_CMD_WRITE:            return "NVME_NVM_CMD_WRITE";
     case NVME_CMD_READ:             return "NVME_NVM_CMD_READ";
     case NVME_CMD_WRITE_ZEROES:     return "NVME_NVM_CMD_WRITE_ZEROES";
+    case NVME_CMD_DSM:              return "NVME_NVM_CMD_DSM";
     default:                        return "NVME_NVM_CMD_UNKNOWN";
     }
 }
hw/block/nvme-ns.c +26 -4
index f1cc734c60f5..2d69b5177b51 100644
@@ -28,10 +28,14 @@
 #include "nvme.h"
 #include "nvme-ns.h"
 
-static void nvme_ns_init(NvmeNamespace *ns)
+#define MIN_DISCARD_GRANULARITY (4 * KiB)
+
+static int nvme_ns_init(NvmeNamespace *ns, Error **errp)
 {
+    BlockDriverInfo bdi;
     NvmeIdNs *id_ns = &ns->id_ns;
     int lba_index = NVME_ID_NS_FLBAS_INDEX(ns->id_ns.flbas);
+    int npdg;
 
     ns->id_ns.dlfeat = 0x9;
 
@@ -43,8 +47,19 @@ static void nvme_ns_init(NvmeNamespace *ns)
     id_ns->ncap = id_ns->nsze;
     id_ns->nuse = id_ns->ncap;
 
-    /* support DULBE */
-    id_ns->nsfeat |= 0x4;
+    /* support DULBE and I/O optimization fields */
+    id_ns->nsfeat |= (0x4 | 0x10);
+
+    npdg = ns->blkconf.discard_granularity / ns->blkconf.logical_block_size;
+
+    if (bdrv_get_info(blk_bs(ns->blkconf.blk), &bdi) >= 0 &&
+        bdi.cluster_size > ns->blkconf.discard_granularity) {
+        npdg = bdi.cluster_size / ns->blkconf.logical_block_size;
+    }
+
+    id_ns->npda = id_ns->npdg = npdg - 1;
+
+    return 0;
 }
 
 static int nvme_ns_init_blk(NvmeCtrl *n, NvmeNamespace *ns, Error **errp)
@@ -59,6 +74,11 @@ static int nvme_ns_init_blk(NvmeCtrl *n, NvmeNamespace *ns, Error **errp)
         return -1;
     }
 
+    if (ns->blkconf.discard_granularity == -1) {
+        ns->blkconf.discard_granularity =
+            MAX(ns->blkconf.logical_block_size, MIN_DISCARD_GRANULARITY);
+    }
+
     ns->size = blk_getlength(ns->blkconf.blk);
     if (ns->size < 0) {
         error_setg_errno(errp, -ns->size, "could not get blockdev size");
@@ -92,7 +112,9 @@ int nvme_ns_setup(NvmeCtrl *n, NvmeNamespace *ns, Error **errp)
         return -1;
     }
 
-    nvme_ns_init(ns);
+    if (nvme_ns_init(ns, errp)) {
+        return -1;
+    }
 
     if (nvme_register_namespace(n, ns, errp)) {
         return -1;
hw/block/nvme.c +101 -1

120 lines not on this page · raw message

Signature · 1 line
2.29.1
#4 lore raw
In reply to Klaus Jensen
Quoted 31 lines · Klaus Jensen

From: Klaus Jensen <[email protected]>

This adds support for the Deallocated or Unwritten Logical Block error recovery feature as well as the Dataset Management command.

v7:
- Handle negative return value from bdrv_block_status.
- bdrv_get_info may not be supported on all block drivers, so do not consider it a fatal error.

v6:
- Skip the allocation of the discards integer and just use the opaque value directly (Philippe)
- Split changes to include/block/nvme.h into a separate patch (Philippe)
- Clean up some convoluted checks on the discards value (Philippe)
- Use unambiguous units in the commit messages (Philippe)
- Stack allocate the range array (Keith)

v5:
- Restore status code from callback (Keith)

v4:
- Removed mixed declaration and code (Keith)
- Set NPDG and NPDA and account for the blockdev cluster size.

Klaus Jensen (3):
hw/block/nvme: add dulbe support
nvme: add namespace I/O optimization fields to shared header hw/block/nvme: add the dataset management command

Quoted 11 lines
 hw/block/nvme-ns.h    |   4 +
 hw/block/nvme.h       |   2 +
 include/block/nvme.h  |  12 ++-
 hw/block/nvme-ns.c    |  34 ++++++--
 hw/block/nvme.c       | 193 +++++++++++++++++++++++++++++++++++++++++-
 hw/block/trace-events |   4 +
 6 files changed, 240 insertions(+), 9 deletions(-)
Signature · 1 line
2.29.1

Keith, I cleared your R-b's from both patch 1 and 3 - please re-review. The diff from v6 is very small, but it does include functional changes.