Skip to contentSkip to conversation

hw/nvme: reimplement all multi-aio commands with custom aiocbs

RFC PATCH Superseded Klaus Jensen 3 participants 19 messages Jun 4, 2021 – Jun 8, 2021

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover hw/nvme: reimplement all multi-aio commands with custom aiocbs

From: Klaus Jensen <[email protected]>

This series reimplements flush, dsm, copy, zone reset and format nvm to allow cancellation. I posted an RFC back in March ("hw/block/nvme:
convert ad-hoc aio tracking to aiocb") and I've applied some feedback from Stefan and reimplemented the remaining commands.

The basic idea is to define custom AIOCBs for these commands. The custom AIOCB takes care of issuing all the "nested" AIOs one by one instead of blindly sending them off simultaneously without tracking the returned aiocbs.

I've kept the RFC since I'm still new to using the block layer like this. I was hoping that Stefan could find some time to look over this - this is a huge series, so I don't expect non-nvme folks to spend a large amount of time on it, but I would really like feedback on my approach in the reimplementation of flush and format. Those commands are special in that may issue AIOs to multiple namespaces and thus, to multiple block backends. Since this device does not support iothreads, I've opted for simply always returning the main loop aio context, but I wonder if this is acceptable or not. It might be the case that this should contain an assert of some kind, in case someone starts adding iothread support.

Klaus Jensen (11):
hw/nvme: reimplement flush to allow cancellation
hw/nvme: add nvme_block_status_all helper
hw/nvme: reimplement dsm to allow cancellation
hw/nvme: save reftag when generating pi
hw/nvme: remove assert from nvme_get_zone_by_slba hw/nvme: use prinfo directly in nvme_check_prinfo and nvme_dif_check hw/nvme: add dw0/1 to the req completion trace event hw/nvme: reimplement the copy command to allow aio cancellation hw/nvme: reimplement zone reset to allow cancellation hw/nvme: reimplement format nvm to allow cancellation Partially revert "hw/block/nvme: drain namespaces on sq deletion"

 hw/nvme/nvme.h       |   10 +-
 include/block/nvme.h |    8 +
 hw/nvme/ctrl.c       | 1861 ++++++++++++++++++++++++------------------
 hw/nvme/dif.c        |   64 +-
 hw/nvme/trace-events |   21 +-
 5 files changed, 1102 insertions(+), 862 deletions(-)
Signature · 1 line
2.31.1
#1 lore raw

RFC PATCH 1/11 hw/nvme: reimplement flush to allow cancellation

From: Klaus Jensen <[email protected]>

Prior to this patch, a broadcast flush would result in submitting multiple "fire and forget" aios (no reference saved to the aiocbs returned from the blk_aio_flush calls).

Fix this by issuing the flushes one after another.

---
 hw/nvme/nvme.h       |   2 +
 hw/nvme/ctrl.c       | 205 ++++++++++++++++++++++++++-----------------
 hw/nvme/trace-events |   6 +-
 3 files changed, 129 insertions(+), 84 deletions(-)
hw/nvme/nvme.h +2 -0
index 81a35cda142b..95730ff2bec5 100644
@@ -27,6 +27,8 @@
 #define NVME_MAX_CONTROLLERS 32
 #define NVME_MAX_NAMESPACES  256
 
+QEMU_BUILD_BUG_ON(NVME_MAX_NAMESPACES > NVME_NSID_BROADCAST - 1);
+
 typedef struct NvmeCtrl NvmeCtrl;
 typedef struct NvmeNamespace NvmeNamespace;
 
hw/nvme/ctrl.c +124 -81

240 lines not on this page · raw message

hw/nvme/trace-events +3 -3
index ea33d0ccc383..ce6b6ffe9604 100644
@@ -7,16 +7,16 @@ pci_nvme_map_addr(uint64_t addr, uint64_t len) "addr 0x%"PRIx64" len %"PRIu64""
 pci_nvme_map_addr_cmb(uint64_t addr, uint64_t len) "addr 0x%"PRIx64" len %"PRIu64""
 pci_nvme_map_prp(uint64_t trans_len, uint32_t len, uint64_t prp1, uint64_t prp2, int num_prps) "trans_len %"PRIu64" len %"PRIu32" prp1 0x%"PRIx64" prp2 0x%"PRIx64" num_prps %d"
 pci_nvme_map_sgl(uint8_t typ, uint64_t len) "type 0x%"PRIx8" len %"PRIu64""
-pci_nvme_io_cmd(uint16_t cid, uint32_t nsid, uint16_t sqid, uint8_t opcode, const char *opname) "cid %"PRIu16" nsid %"PRIu32" sqid %"PRIu16" opc 0x%"PRIx8" opname '%s'"
+pci_nvme_io_cmd(uint16_t cid, uint32_t nsid, uint16_t sqid, uint8_t opcode, const char *opname) "cid %"PRIu16" nsid 0x%"PRIx32" sqid %"PRIu16" opc 0x%"PRIx8" opname '%s'"
 pci_nvme_admin_cmd(uint16_t cid, uint16_t sqid, uint8_t opcode, const char *opname) "cid %"PRIu16" sqid %"PRIu16" opc 0x%"PRIx8" opname '%s'"
-pci_nvme_flush(uint16_t cid, uint32_t nsid) "cid %"PRIu16" nsid %"PRIu32""
+pci_nvme_flush_ns(uint32_t nsid) "nsid 0x%"PRIx32""
 pci_nvme_format(uint16_t cid, uint32_t nsid, uint8_t lbaf, uint8_t mset, uint8_t pi, uint8_t pil) "cid %"PRIu16" nsid %"PRIu32" lbaf %"PRIu8" mset %"PRIu8" pi %"PRIu8" pil %"PRIu8""
 pci_nvme_format_ns(uint16_t cid, uint32_t nsid, uint8_t lbaf, uint8_t mset, uint8_t pi, uint8_t pil) "cid %"PRIu16" nsid %"PRIu32" lbaf %"PRIu8" mset %"PRIu8" pi %"PRIu8" pil %"PRIu8""
 pci_nvme_format_cb(uint16_t cid, uint32_t nsid) "cid %"PRIu16" nsid %"PRIu32""
 pci_nvme_read(uint16_t cid, uint32_t nsid, uint32_t nlb, uint64_t count, uint64_t lba) "cid %"PRIu16" nsid %"PRIu32" nlb %"PRIu32" count %"PRIu64" lba 0x%"PRIx64""
 pci_nvme_write(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_misc_cb(uint16_t cid, const char *blkname) "cid %"PRIu16" blk '%s'"
+pci_nvme_misc_cb(uint16_t cid) "cid %"PRIu16""
 pci_nvme_dif_rw(uint8_t pract, uint8_t prinfo) "pract 0x%"PRIx8" prinfo 0x%"PRIx8""
 pci_nvme_dif_rw_cb(uint16_t cid, const char *blkname) "cid %"PRIu16" blk '%s'"
 pci_nvme_dif_rw_mdata_in_cb(uint16_t cid, const char *blkname) "cid %"PRIu16" blk '%s'"
Signature · 1 line
2.31.1
#2 lore raw

RFC PATCH 2/11 hw/nvme: add nvme_block_status_all helper

In reply to Klaus Jensen

From: Klaus Jensen <[email protected]>

Pull the gist of nvme_check_dulbe() into a helper function. This is in preparation for dsm refactoring.

---
 hw/nvme/ctrl.c | 41 ++++++++++++++++++++++++++++-------------
 1 file changed, 28 insertions(+), 13 deletions(-)
hw/nvme/ctrl.c +28 -13
index f49268741427..0278f256e534 100644
@@ -1432,18 +1432,15 @@ static inline uint16_t nvme_check_bounds(NvmeNamespace *ns, uint64_t slba,
     return NVME_SUCCESS;
 }
 
-static uint16_t nvme_check_dulbe(NvmeNamespace *ns, uint64_t slba,
-                                 uint32_t nlb)
+static int nvme_block_status_all(NvmeNamespace *ns, uint64_t slba,
+                                 uint32_t nlb, int flags)
 {
     BlockDriverState *bs = blk_bs(ns->blkconf.blk);
 
     int64_t pnum = 0, bytes = nvme_l2b(ns, nlb);
     int64_t offset = nvme_l2b(ns, slba);
-    bool zeroed;
     int ret;
 
-    Error *local_err = NULL;
-
     /*
      * `pnum` holds the number of bytes after offset that shares the same
      * allocation status as the byte at offset. If `pnum` is different from
@@ -1455,23 +1452,41 @@ static uint16_t nvme_check_dulbe(NvmeNamespace *ns, uint64_t slba,
 
         ret = bdrv_block_status(bs, offset, bytes, &pnum, NULL, NULL);
         if (ret < 0) {
-            error_setg_errno(&local_err, -ret, "unable to get block status");
-            error_report_err(local_err);
-
-            return NVME_INTERNAL_DEV_ERROR;
+            return ret;
         }
 
-        zeroed = !!(ret & BDRV_BLOCK_ZERO);
 
-        trace_pci_nvme_block_status(offset, bytes, pnum, ret, zeroed);
+        trace_pci_nvme_block_status(offset, bytes, pnum, ret,
+                                    !!(ret & BDRV_BLOCK_ZERO));
 
-        if (zeroed) {
-            return NVME_DULB;
+        if (!(ret & flags)) {
+            return 1;
         }
 
         offset += pnum;
     } while (pnum != bytes);
 
+    return 0;
+}
+
+static uint16_t nvme_check_dulbe(NvmeNamespace *ns, uint64_t slba,
+                                 uint32_t nlb)
+{
+    int ret;
+    Error *err = NULL;
+
+    ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_DATA);
+    if (ret) {
+        if (ret < 0) {
+            error_setg_errno(&err, -ret, "unable to get block status");
+            error_report_err(err);
+
+            return NVME_INTERNAL_DEV_ERROR;
+        }
+
+        return NVME_DULB;
+    }
+
     return NVME_SUCCESS;
 }
 
Signature · 1 line
2.31.1
#3 lore raw

RFC PATCH 3/11 hw/nvme: reimplement dsm to allow cancellation

In reply to Klaus Jensen

From: Klaus Jensen <[email protected]>

Prior to this patch, a loop was used to issue multiple "fire and forget" aios for each range in the command. Without a reference to the aiocb returned from the blk_aio_pdiscard calls, the aios cannot be canceled.

Fix this by processing the ranges one after another.

As a bonus, this fixes how metadata is cleared (i.e. we only zero it out if the data was succesfully discarded).

---
 hw/nvme/ctrl.c       | 233 +++++++++++++++++++++++++++++--------------
 hw/nvme/trace-events |   4 +-
 2 files changed, 162 insertions(+), 75 deletions(-)
hw/nvme/ctrl.c +160 -73
index 0278f256e534..e9423df3a449 100644
@@ -2000,26 +2000,6 @@ out:
     nvme_verify_cb(ctx, ret);
 }
 
-static void nvme_aio_discard_cb(void *opaque, int ret)
-{
-    NvmeRequest *req = opaque;
-    uintptr_t *discards = (uintptr_t *)&req->opaque;
-
-    trace_pci_nvme_aio_discard_cb(nvme_cid(req));
-
-    if (ret) {
-        nvme_aio_err(req, ret);
-    }
-
-    (*discards)--;
-
-    if (*discards) {
-        return;
-    }
-
-    nvme_enqueue_req_completion(nvme_cq(req), req);
-}
-
 struct nvme_zone_reset_ctx {
     NvmeRequest *req;
     NvmeZone    *zone;
@@ -2480,75 +2460,182 @@ out:
     nvme_enqueue_req_completion(nvme_cq(req), req);
 }
 
+typedef struct NvmeDSMAIOCB {
+    BlockAIOCB common;
+    BlockAIOCB *aiocb;
+    NvmeRequest *req;
+    QEMUBH *bh;
+    int ret;
+
+    NvmeDsmRange *range;
+    unsigned int nr;
+    unsigned int idx;
+} NvmeDSMAIOCB;
+
+static void nvme_dsm_cancel(BlockAIOCB *aiocb)
+{
+    NvmeDSMAIOCB *iocb = container_of(aiocb, NvmeDSMAIOCB, common);
+
+    /* break nvme_dsm_cb loop */
+    iocb->idx = iocb->nr;
+    iocb->ret = -ECANCELED;
+
+    if (iocb->aiocb) {
+        blk_aio_cancel_async(iocb->aiocb);
+        iocb->aiocb = NULL;
+    } else {
+        /*
+         * We only reach this if nvme_dsm_cancel() has already been called or
+         * the command ran to completion and nvme_dsm_bh is scheduled to run.
+         */
+        assert(iocb->idx == iocb->nr);
+    }
+}
+
+static const AIOCBInfo nvme_dsm_aiocb_info = {
+    .aiocb_size   = sizeof(NvmeDSMAIOCB),
+    .cancel_async = nvme_dsm_cancel,
+};
+
+static void nvme_dsm_bh(void *opaque)
+{
+    NvmeDSMAIOCB *iocb = opaque;
+
+    iocb->common.cb(iocb->common.opaque, iocb->ret);
+
+    qemu_bh_delete(iocb->bh);
+    iocb->bh = NULL;
+    qemu_aio_unref(iocb);
+}
+
+static void nvme_dsm_cb(void *opaque, int ret);
+
+static void nvme_dsm_md_cb(void *opaque, int ret)
+{
+    NvmeDSMAIOCB *iocb = opaque;
+    NvmeRequest *req = iocb->req;
+    NvmeNamespace *ns = req->ns;
+    NvmeDsmRange *range;
+    uint64_t slba;
+    uint32_t nlb;
+
+    if (ret < 0) {
+        iocb->ret = ret;
+        goto done;
+    }
+
+    if (!ns->lbaf.ms) {
+        nvme_dsm_cb(iocb, 0);
+        return;
+    }
+
+    range = &iocb->range[iocb->idx - 1];
+    slba = le64_to_cpu(range->slba);
+    nlb = le32_to_cpu(range->nlb);
+
+    /*
+     * Check that all block were discarded (zeroed); otherwise we do not zero
+     * the metadata.
+     */
+
+    ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_ZERO);
+    if (ret) {
+        if (ret < 0) {
+            iocb->ret = ret;
+            goto done;
+        }
+
+        nvme_dsm_cb(iocb, 0);
+    }
+
+    iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, nvme_moff(ns, slba),
+                                        nvme_m2b(ns, nlb), BDRV_REQ_MAY_UNMAP,
+                                        nvme_dsm_cb, iocb);
+    return;
+
+done:
+    iocb->aiocb = NULL;
+    qemu_bh_schedule(iocb->bh);
+}
+
+static void nvme_dsm_cb(void *opaque, int ret)
+{
+    NvmeDSMAIOCB *iocb = opaque;
+    NvmeRequest *req = iocb->req;
+    NvmeCtrl *n = nvme_ctrl(req);
+    NvmeNamespace *ns = req->ns;
+    NvmeDsmRange *range;
+    uint64_t slba;
+    uint32_t nlb;
+
+    if (ret < 0) {
+        iocb->ret = ret;
+        goto done;
+    }
+
+next:
+    if (iocb->idx == iocb->nr) {
+        goto done;
+    }
+
+    range = &iocb->range[iocb->idx++];
+    slba = le64_to_cpu(range->slba);
+    nlb = le32_to_cpu(range->nlb);
+
+    trace_pci_nvme_dsm_deallocate(slba, nlb);
+
+    if (nlb > n->dmrsl) {
+        trace_pci_nvme_dsm_single_range_limit_exceeded(nlb, n->dmrsl);
+        goto next;
+    }
+
+    if (nvme_check_bounds(ns, slba, nlb)) {
+        trace_pci_nvme_err_invalid_lba_range(slba, nlb,
+                                             ns->id_ns.nsze);
+        goto next;
+    }
+
+    iocb->aiocb = blk_aio_pdiscard(ns->blkconf.blk, nvme_l2b(ns, slba),
+                                   nvme_l2b(ns, nlb),
+                                   nvme_dsm_md_cb, iocb);
+    return;
+
+done:
+    iocb->aiocb = NULL;
+    qemu_bh_schedule(iocb->bh);
+}
+
 static uint16_t nvme_dsm(NvmeCtrl *n, NvmeRequest *req)
 {
     NvmeNamespace *ns = req->ns;
     NvmeDsmCmd *dsm = (NvmeDsmCmd *) &req->cmd;
-
     uint32_t attr = le32_to_cpu(dsm->attributes);
     uint32_t nr = (le32_to_cpu(dsm->nr) & 0xff) + 1;
-
     uint16_t status = NVME_SUCCESS;
 
-    trace_pci_nvme_dsm(nvme_cid(req), nvme_nsid(ns), nr, attr);
+    trace_pci_nvme_dsm(nr, attr);
 
     if (attr & NVME_DSMGMT_AD) {
-        int64_t offset;
-        size_t len;
-        NvmeDsmRange range[nr];
-        uintptr_t *discards = (uintptr_t *)&req->opaque;
+        NvmeDSMAIOCB *iocb = blk_aio_get(&nvme_dsm_aiocb_info, ns->blkconf.blk,
+                                         nvme_misc_cb, req);
 
-        status = nvme_h2c(n, (uint8_t *)range, sizeof(range), req);
+        iocb->req = req;
+        iocb->bh = qemu_bh_new(nvme_dsm_bh, iocb);
+        iocb->ret = 0;
+        iocb->range = g_new(NvmeDsmRange, nr);
+        iocb->nr = nr;
+        iocb->idx = 0;
+
+        status = nvme_h2c(n, (uint8_t *)iocb->range, sizeof(NvmeDsmRange) * nr,
+                          req);
         if (status) {
             return status;
         }
 
-        /*
-         * AIO callbacks may be called immediately, so initialize discards to 1
-         * to make sure the the callback does not complete the request before
-         * all discards have been issued.
-         */
-        *discards = 1;
+        req->aiocb = &iocb->common;
+        nvme_dsm_cb(iocb, 0);
 
-        for (int i = 0; i < nr; i++) {
-            uint64_t slba = le64_to_cpu(range[i].slba);
-            uint32_t nlb = le32_to_cpu(range[i].nlb);
-
-            if (nvme_check_bounds(ns, slba, nlb)) {
-                continue;
-            }
-
-            trace_pci_nvme_dsm_deallocate(nvme_cid(req), nvme_nsid(ns), slba,
-                                          nlb);
-
-            if (nlb > n->dmrsl) {
-                trace_pci_nvme_dsm_single_range_limit_exceeded(nlb, n->dmrsl);
-            }
-
-            offset = nvme_l2b(ns, slba);
-            len = nvme_l2b(ns, nlb);
-
-            while (len) {
-                size_t bytes = MIN(BDRV_REQUEST_MAX_BYTES, len);
-
-                (*discards)++;
-
-                blk_aio_pdiscard(ns->blkconf.blk, offset, bytes,
-                                 nvme_aio_discard_cb, req);
-
-                offset += bytes;
-                len -= bytes;
-            }
-        }
-
-        /* account for the 1-initialization */
-        (*discards)--;
-
-        if (*discards) {
-            status = NVME_NO_COMPLETE;
-        } else {
-            status = req->status;
-        }
+        return NVME_NO_COMPLETE;
     }
 
     return status;
hw/nvme/trace-events +2 -2
index ce6b6ffe9604..eea4e31e46c4 100644
@@ -37,8 +37,8 @@ pci_nvme_verify_mdata_in_cb(uint16_t cid, const char *blkname) "cid %"PRIu16" bl
 pci_nvme_verify_cb(uint16_t cid, uint8_t prinfo, uint16_t apptag, uint16_t appmask, uint32_t reftag) "cid %"PRIu16" prinfo 0x%"PRIx8" apptag 0x%"PRIx16" appmask 0x%"PRIx16" reftag 0x%"PRIx32""
 pci_nvme_rw_complete_cb(uint16_t cid, const char *blkname) "cid %"PRIu16" blk '%s'"
 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_dsm(uint32_t nr, uint32_t attr) "nr %"PRIu32" attr 0x%"PRIx32""
+pci_nvme_dsm_deallocate(uint64_t slba, uint32_t nlb) "slba %"PRIu64" nlb %"PRIu32""
 pci_nvme_dsm_single_range_limit_exceeded(uint32_t nlb, uint32_t dmrsl) "nlb %"PRIu32" dmrsl %"PRIu32""
 pci_nvme_compare(uint16_t cid, uint32_t nsid, uint64_t slba, uint32_t nlb) "cid %"PRIu16" nsid %"PRIu32" slba 0x%"PRIx64" nlb %"PRIu32""
 pci_nvme_compare_data_cb(uint16_t cid) "cid %"PRIu16""
Signature · 1 line
2.31.1
#4 lore raw

RFC PATCH 4/11 hw/nvme: save reftag when generating pi

In reply to Klaus Jensen

From: Klaus Jensen <[email protected]>

Prepare nvme_dif_pract_generate_dif() and nvme_dif_check() to be callable in smaller increments by making the reftag a pointer parameter updated by the function.

---
 hw/nvme/nvme.h |  4 ++--
 hw/nvme/ctrl.c | 10 +++++-----
 hw/nvme/dif.c  | 22 +++++++++++-----------
 3 files changed, 18 insertions(+), 18 deletions(-)
hw/nvme/nvme.h +2 -2
index 95730ff2bec5..bfa6632717dc 100644
@@ -538,11 +538,11 @@ uint16_t nvme_dif_mangle_mdata(NvmeNamespace *ns, uint8_t *mbuf, size_t mlen,
                                uint64_t slba);
 void nvme_dif_pract_generate_dif(NvmeNamespace *ns, uint8_t *buf, size_t len,
                                  uint8_t *mbuf, size_t mlen, uint16_t apptag,
-                                 uint32_t reftag);
+                                 uint32_t *reftag);
 uint16_t nvme_dif_check(NvmeNamespace *ns, uint8_t *buf, size_t len,
                         uint8_t *mbuf, size_t mlen, uint16_t ctrl,
                         uint64_t slba, uint16_t apptag,
-                        uint16_t appmask, uint32_t reftag);
+                        uint16_t appmask, uint32_t *reftag);
 uint16_t nvme_dif_rw(NvmeCtrl *n, NvmeRequest *req);
 
 
hw/nvme/ctrl.c +5 -5
index e9423df3a449..0dc86f7b603c 100644
@@ -1953,7 +1953,7 @@ static void nvme_verify_cb(void *opaque, int ret)
 
         req->status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
                                      ctx->mdata.bounce, ctx->mdata.iov.size,
-                                     ctrl, slba, apptag, appmask, reftag);
+                                     ctrl, slba, apptag, appmask, &reftag);
     }
 
 out:
@@ -2189,7 +2189,7 @@ static void nvme_copy_in_complete(NvmeRequest *req)
             reftag = le32_to_cpu(range->reftag);
 
             status = nvme_dif_check(ns, buf, len, mbuf, mlen, prinfor, slba,
-                                    apptag, appmask, reftag);
+                                    apptag, appmask, &reftag);
             if (status) {
                 goto invalid;
             }
@@ -2212,10 +2212,10 @@ static void nvme_copy_in_complete(NvmeRequest *req)
             }
 
             nvme_dif_pract_generate_dif(ns, ctx->bounce, len, ctx->mbounce,
-                                        mlen, apptag, reftag);
+                                        mlen, apptag, &reftag);
         } else {
             status = nvme_dif_check(ns, ctx->bounce, len, ctx->mbounce, mlen,
-                                    prinfow, sdlba, apptag, appmask, reftag);
+                                    prinfow, sdlba, apptag, appmask, &reftag);
             if (status) {
                 goto invalid;
             }
@@ -2355,7 +2355,7 @@ static void nvme_compare_mdata_cb(void *opaque, int ret)
 
         status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
                                 ctx->mdata.bounce, ctx->mdata.iov.size, ctrl,
-                                slba, apptag, appmask, reftag);
+                                slba, apptag, appmask, &reftag);
         if (status) {
             req->status = status;
             goto out;
hw/nvme/dif.c +11 -11
index 88efcbe9bd60..f5f86f1c26ad 100644
@@ -41,7 +41,7 @@ static uint16_t crc_t10dif(uint16_t crc, const unsigned char *buffer,
 
 void nvme_dif_pract_generate_dif(NvmeNamespace *ns, uint8_t *buf, size_t len,
                                  uint8_t *mbuf, size_t mlen, uint16_t apptag,
-                                 uint32_t reftag)
+                                 uint32_t *reftag)
 {
     uint8_t *end = buf + len;
     int16_t pil = 0;
@@ -51,7 +51,7 @@ void nvme_dif_pract_generate_dif(NvmeNamespace *ns, uint8_t *buf, size_t len,
     }
 
     trace_pci_nvme_dif_pract_generate_dif(len, ns->lbasz, ns->lbasz + pil,
-                                          apptag, reftag);
+                                          apptag, *reftag);
 
     for (; buf < end; buf += ns->lbasz, mbuf += ns->lbaf.ms) {
         NvmeDifTuple *dif = (NvmeDifTuple *)(mbuf + pil);
@@ -63,10 +63,10 @@ void nvme_dif_pract_generate_dif(NvmeNamespace *ns, uint8_t *buf, size_t len,
 
         dif->guard = cpu_to_be16(crc);
         dif->apptag = cpu_to_be16(apptag);
-        dif->reftag = cpu_to_be32(reftag);
+        dif->reftag = cpu_to_be32(*reftag);
 
         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) != NVME_ID_NS_DPS_TYPE_3) {
-            reftag++;
+            (*reftag)++;
         }
     }
 }
@@ -132,13 +132,13 @@ static uint16_t nvme_dif_prchk(NvmeNamespace *ns, NvmeDifTuple *dif,
 uint16_t nvme_dif_check(NvmeNamespace *ns, uint8_t *buf, size_t len,
                         uint8_t *mbuf, size_t mlen, uint16_t ctrl,
                         uint64_t slba, uint16_t apptag,
-                        uint16_t appmask, uint32_t reftag)
+                        uint16_t appmask, uint32_t *reftag)
 {
     uint8_t *end = buf + len;
     int16_t pil = 0;
     uint16_t status;
 
-    status = nvme_check_prinfo(ns, ctrl, slba, reftag);
+    status = nvme_check_prinfo(ns, ctrl, slba, *reftag);
     if (status) {
         return status;
     }
@@ -153,13 +153,13 @@ uint16_t nvme_dif_check(NvmeNamespace *ns, uint8_t *buf, size_t len,
         NvmeDifTuple *dif = (NvmeDifTuple *)(mbuf + pil);
 
         status = nvme_dif_prchk(ns, dif, buf, mbuf, pil, ctrl, apptag,
-                                appmask, reftag);
+                                appmask, *reftag);
         if (status) {
             return status;
         }
 
         if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) != NVME_ID_NS_DPS_TYPE_3) {
-            reftag++;
+            (*reftag)++;
         }
     }
 
@@ -270,7 +270,7 @@ static void nvme_dif_rw_check_cb(void *opaque, int ret)
 
     status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
                             ctx->mdata.bounce, ctx->mdata.iov.size, ctrl,
-                            slba, apptag, appmask, reftag);
+                            slba, apptag, appmask, &reftag);
     if (status) {
         req->status = status;
         goto out;
@@ -478,11 +478,11 @@ uint16_t nvme_dif_rw(NvmeCtrl *n, NvmeRequest *req)
         /* splice generated protection information into the buffer */
         nvme_dif_pract_generate_dif(ns, ctx->data.bounce, ctx->data.iov.size,
                                     ctx->mdata.bounce, ctx->mdata.iov.size,
-                                    apptag, reftag);
+                                    apptag, &reftag);
     } else {
         status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
                                 ctx->mdata.bounce, ctx->mdata.iov.size, ctrl,
-                                slba, apptag, appmask, reftag);
+                                slba, apptag, appmask, &reftag);
         if (status) {
             goto err;
         }
Signature · 1 line
2.31.1
#5 lore raw

RFC PATCH 5/11 hw/nvme: remove assert from nvme_get_zone_by_slba

In reply to Klaus Jensen

From: Klaus Jensen <[email protected]>

Make nvme_get_zone_by_slba() return NULL if the slba is out of range. This allows the function to be used without guarding the call with a call to nvme_check_bounds(), in preparation for the next patch.

Add asserts after calling nvme_get_zone_by_slba() instead.

---
 hw/nvme/ctrl.c | 18 ++++++++++++++----
 1 file changed, 14 insertions(+), 4 deletions(-)
hw/nvme/ctrl.c +14 -4
index 0dc86f7b603c..0895faaa0724 100644
@@ -1536,7 +1536,10 @@ static inline NvmeZone *nvme_get_zone_by_slba(NvmeNamespace *ns, uint64_t slba)
 {
     uint32_t zone_idx = nvme_zone_idx(ns, slba);
 
-    assert(zone_idx < ns->num_zones);
+    if (zone_idx >= ns->num_zones) {
+        return NULL;
+    }
+
     return &ns->zone_array[zone_idx];
 }
 
@@ -1613,11 +1616,16 @@ static uint16_t nvme_check_zone_state_for_read(NvmeZone *zone)
 static uint16_t nvme_check_zone_read(NvmeNamespace *ns, uint64_t slba,
                                      uint32_t nlb)
 {
-    NvmeZone *zone = nvme_get_zone_by_slba(ns, slba);
-    uint64_t bndry = nvme_zone_rd_boundary(ns, zone);
-    uint64_t end = slba + nlb;
+    NvmeZone *zone;
+    uint64_t bndry, end;
     uint16_t status;
 
+    zone = nvme_get_zone_by_slba(ns, slba);
+    assert(zone);
+
+    bndry = nvme_zone_rd_boundary(ns, zone);
+    end = slba + nlb;
+
     status = nvme_check_zone_state_for_read(zone);
     if (status) {
         ;
@@ -1780,6 +1788,7 @@ static void nvme_finalize_zoned_write(NvmeNamespace *ns, NvmeRequest *req)
     slba = le64_to_cpu(rw->slba);
     nlb = le16_to_cpu(rw->nlb) + 1;
     zone = nvme_get_zone_by_slba(ns, slba);
+    assert(zone);
 
     nvme_advance_zone_wp(ns, zone, nlb);
 }
@@ -3175,6 +3184,7 @@ static uint16_t nvme_do_write(NvmeCtrl *n, NvmeRequest *req, bool append,
 
     if (ns->params.zoned) {
         zone = nvme_get_zone_by_slba(ns, slba);
+        assert(zone);
 
         if (append) {
             bool piremap = !!(ctrl & NVME_RW_PIREMAP);
Signature · 1 line
2.31.1
#6 lore raw

RFC PATCH 6/11 hw/nvme: use prinfo directly in nvme_check_prinfo and nvme_dif_check

In reply to Klaus Jensen

From: Klaus Jensen <[email protected]>

The nvme_check_prinfo() and nvme_dif_check() functions operate on the 16 bit "control" member of the NvmeCmd. These functions do not otherwise operate on an NvmeCmd or an NvmeRequest, so change them to expect the actual 4 bit PRINFO field and add constants that work on this field as well.

---
 hw/nvme/nvme.h       |  4 +--
 include/block/nvme.h |  8 ++++++
 hw/nvme/ctrl.c       | 67 +++++++++++++++++---------------------------
 hw/nvme/dif.c        | 44 ++++++++++++++---------------
 4 files changed, 57 insertions(+), 66 deletions(-)
hw/nvme/nvme.h +2 -2
index bfa6632717dc..60685404d351 100644
@@ -532,7 +532,7 @@ static const uint16_t t10_dif_crc_table[256] = {
     0xF0D8, 0x7B6F, 0x6C01, 0xE7B6, 0x42DD, 0xC96A, 0xDE04, 0x55B3
 };
 
-uint16_t nvme_check_prinfo(NvmeNamespace *ns, uint16_t ctrl, uint64_t slba,
+uint16_t nvme_check_prinfo(NvmeNamespace *ns, uint8_t prinfo, uint64_t slba,
                            uint32_t reftag);
 uint16_t nvme_dif_mangle_mdata(NvmeNamespace *ns, uint8_t *mbuf, size_t mlen,
                                uint64_t slba);
@@ -540,7 +540,7 @@ void nvme_dif_pract_generate_dif(NvmeNamespace *ns, uint8_t *buf, size_t len,
                                  uint8_t *mbuf, size_t mlen, uint16_t apptag,
                                  uint32_t *reftag);
 uint16_t nvme_dif_check(NvmeNamespace *ns, uint8_t *buf, size_t len,
-                        uint8_t *mbuf, size_t mlen, uint16_t ctrl,
+                        uint8_t *mbuf, size_t mlen, uint8_t prinfo,
                         uint64_t slba, uint16_t apptag,
                         uint16_t appmask, uint32_t *reftag);
 uint16_t nvme_dif_rw(NvmeCtrl *n, NvmeRequest *req);
include/block/nvme.h +8 -0
index 333affdb8534..0fabe28b7bdd 100644
@@ -708,6 +708,14 @@ enum {
 
 #define NVME_RW_PRINFO(control) ((control >> 10) & 0xf)
 
+enum {
+    NVME_PRINFO_PRACT       = 1 << 3,
+    NVME_PRINFO_PRCHK_GUARD = 1 << 2,
+    NVME_PRINFO_PRCHK_APP   = 1 << 1,
+    NVME_PRINFO_PRCHK_REF   = 1 << 0,
+    NVME_PRINFO_PRCHK_MASK  = 7 << 0,
+};
+
 typedef struct QEMU_PACKED NvmeDsmCmd {
     uint8_t     opcode;
     uint8_t     flags;
hw/nvme/ctrl.c +25 -42

194 lines not on this page · raw message

hw/nvme/dif.c +22 -22

158 lines not on this page · raw message

Signature · 1 line
2.31.1
#7 lore raw

RFC PATCH 7/11 hw/nvme: add dw0/1 to the req completion trace event

In reply to Klaus Jensen

From: Klaus Jensen <[email protected]>

Some commands report additional useful information in dw0 and dw1 of the completion queue entry.

Add them to the trace.

---
 hw/nvme/ctrl.c       | 2 ++
 hw/nvme/trace-events | 2 +-
 2 files changed, 3 insertions(+), 1 deletion(-)
hw/nvme/ctrl.c +2 -0
index 527c38c0cf82..04027c80ad90 100644
@@ -1284,6 +1284,8 @@ static void nvme_enqueue_req_completion(NvmeCQueue *cq, NvmeRequest *req)
 {
     assert(cq->cqid == req->sq->cqid);
     trace_pci_nvme_enqueue_req_completion(nvme_cid(req), cq->cqid,
+                                          le32_to_cpu(req->cqe.result),
+                                          le32_to_cpu(req->cqe.dw1),
                                           req->status);
 
     if (req->status) {
hw/nvme/trace-events +1 -1
index eea4e31e46c4..a3e11346865e 100644
@@ -80,7 +80,7 @@ pci_nvme_enqueue_event(uint8_t typ, uint8_t info, uint8_t log_page) "type 0x%"PR
 pci_nvme_enqueue_event_noqueue(int queued) "queued %d"
 pci_nvme_enqueue_event_masked(uint8_t typ) "type 0x%"PRIx8""
 pci_nvme_no_outstanding_aers(void) "ignoring event; no outstanding AERs"
-pci_nvme_enqueue_req_completion(uint16_t cid, uint16_t cqid, uint16_t status) "cid %"PRIu16" cqid %"PRIu16" status 0x%"PRIx16""
+pci_nvme_enqueue_req_completion(uint16_t cid, uint16_t cqid, uint32_t dw0, uint32_t dw1, uint16_t status) "cid %"PRIu16" cqid %"PRIu16" dw0 0x%"PRIx32" dw1 0x%"PRIx32" status 0x%"PRIx16""
 pci_nvme_mmio_read(uint64_t addr, unsigned size) "addr 0x%"PRIx64" size %d"
 pci_nvme_mmio_write(uint64_t addr, uint64_t data, unsigned size) "addr 0x%"PRIx64" data 0x%"PRIx64" size %d"
 pci_nvme_mmio_doorbell_cq(uint16_t cqid, uint16_t new_head) "cqid %"PRIu16" new_head %"PRIu16""
Signature · 1 line
2.31.1
#8 lore raw

RFC PATCH 8/11 hw/nvme: reimplement the copy command to allow aio cancellation

In reply to Klaus Jensen

From: Klaus Jensen <[email protected]>

Before this patch the code would issue several aios simultaneously without saving a reference to the aiocb. Without the aiocb reference the individual copies cannot be canceled.

Fix this by issuing copies of the ranges one after another.

---
 hw/nvme/ctrl.c       | 705 +++++++++++++++++++++++--------------------
 hw/nvme/trace-events |   3 +-
 2 files changed, 373 insertions(+), 335 deletions(-)
hw/nvme/ctrl.c +372 -333
index 04027c80ad90..fb65a4f12d1f 100644
@@ -2078,226 +2078,6 @@ out:
     nvme_aio_zone_reset_complete_cb(opaque, ret);
 }
 
-struct nvme_copy_ctx {
-    int copies;
-    uint8_t *bounce;
-    uint8_t *mbounce;
-    uint32_t nlb;
-    NvmeCopySourceRange *ranges;
-};
-
-struct nvme_copy_in_ctx {
-    NvmeRequest *req;
-    QEMUIOVector iov;
-    NvmeCopySourceRange *range;
-};
-
-static void nvme_copy_complete_cb(void *opaque, int ret)
-{
-    NvmeRequest *req = opaque;
-    NvmeNamespace *ns = req->ns;
-    struct nvme_copy_ctx *ctx = req->opaque;
-
-    if (ret) {
-        block_acct_failed(blk_get_stats(ns->blkconf.blk), &req->acct);
-        nvme_aio_err(req, ret);
-        goto out;
-    }
-
-    block_acct_done(blk_get_stats(ns->blkconf.blk), &req->acct);
-
-out:
-    if (ns->params.zoned) {
-        NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
-        uint64_t sdlba = le64_to_cpu(copy->sdlba);
-        NvmeZone *zone = nvme_get_zone_by_slba(ns, sdlba);
-
-        nvme_advance_zone_wp(ns, zone, ctx->nlb);
-    }
-
-    g_free(ctx->bounce);
-    g_free(ctx->mbounce);
-    g_free(ctx);
-
-    nvme_enqueue_req_completion(nvme_cq(req), req);
-}
-
-static void nvme_copy_cb(void *opaque, int ret)
-{
-    NvmeRequest *req = opaque;
-    NvmeNamespace *ns = req->ns;
-    struct nvme_copy_ctx *ctx = req->opaque;
-
-    trace_pci_nvme_copy_cb(nvme_cid(req));
-
-    if (ret) {
-        goto out;
-    }
-
-    if (ns->lbaf.ms) {
-        NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
-        uint64_t sdlba = le64_to_cpu(copy->sdlba);
-        int64_t offset = nvme_moff(ns, sdlba);
-
-        qemu_iovec_reset(&req->sg.iov);
-        qemu_iovec_add(&req->sg.iov, ctx->mbounce, nvme_m2b(ns, ctx->nlb));
-
-        req->aiocb = blk_aio_pwritev(ns->blkconf.blk, offset, &req->sg.iov, 0,
-                                     nvme_copy_complete_cb, req);
-        return;
-    }
-
-out:
-    nvme_copy_complete_cb(opaque, ret);
-}
-
-static void nvme_copy_in_complete(NvmeRequest *req)
-{
-    NvmeNamespace *ns = req->ns;
-    NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
-    struct nvme_copy_ctx *ctx = req->opaque;
-    uint64_t sdlba = le64_to_cpu(copy->sdlba);
-    uint16_t status;
-
-    trace_pci_nvme_copy_in_complete(nvme_cid(req));
-
-    block_acct_done(blk_get_stats(ns->blkconf.blk), &req->acct);
-
-    if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
-        uint8_t prinfor = (copy->control[0] >> 4) & 0xf;
-        uint8_t prinfow = (copy->control[2] >> 2) & 0xf;
-        uint16_t nr = copy->nr + 1;
-        NvmeCopySourceRange *range;
-        uint64_t slba;
-        uint32_t nlb;
-        uint16_t apptag, appmask;
-        uint32_t reftag;
-        uint8_t *buf = ctx->bounce, *mbuf = ctx->mbounce;
-        size_t len, mlen;
-        int i;
-
-        for (i = 0; i < nr; i++) {
-            range = &ctx->ranges[i];
-            slba = le64_to_cpu(range->slba);
-            nlb = le16_to_cpu(range->nlb) + 1;
-            len = nvme_l2b(ns, nlb);
-            mlen = nvme_m2b(ns, nlb);
-            apptag = le16_to_cpu(range->apptag);
-            appmask = le16_to_cpu(range->appmask);
-            reftag = le32_to_cpu(range->reftag);
-
-            status = nvme_dif_check(ns, buf, len, mbuf, mlen, prinfor, slba,
-                                    apptag, appmask, &reftag);
-            if (status) {
-                goto invalid;
-            }
-
-            buf += len;
-            mbuf += mlen;
-        }
-
-        apptag = le16_to_cpu(copy->apptag);
-        appmask = le16_to_cpu(copy->appmask);
-        reftag = le32_to_cpu(copy->reftag);
-
-        if (prinfow & NVME_RW_PRINFO_PRACT) {
-            size_t len = nvme_l2b(ns, ctx->nlb);
-            size_t mlen = nvme_m2b(ns, ctx->nlb);
-
-            status = nvme_check_prinfo(ns, prinfow, sdlba, reftag);
-            if (status) {
-                goto invalid;
-            }
-
-            nvme_dif_pract_generate_dif(ns, ctx->bounce, len, ctx->mbounce,
-                                        mlen, apptag, &reftag);
-        } else {
-            status = nvme_dif_check(ns, ctx->bounce, len, ctx->mbounce, mlen,
-                                    prinfow, sdlba, apptag, appmask, &reftag);
-            if (status) {
-                goto invalid;
-            }
-        }
-    }
-
-    status = nvme_check_bounds(ns, sdlba, ctx->nlb);
-    if (status) {
-        goto invalid;
-    }
-
-    if (ns->params.zoned) {
-        NvmeZone *zone = nvme_get_zone_by_slba(ns, sdlba);
-
-        status = nvme_check_zone_write(ns, zone, sdlba, ctx->nlb);
-        if (status) {
-            goto invalid;
-        }
-
-        status = nvme_zrm_auto(ns, zone);
-        if (status) {
-            goto invalid;
-        }
-
-        zone->w_ptr += ctx->nlb;
-    }
-
-    qemu_iovec_init(&req->sg.iov, 1);
-    qemu_iovec_add(&req->sg.iov, ctx->bounce, nvme_l2b(ns, ctx->nlb));
-
-    block_acct_start(blk_get_stats(ns->blkconf.blk), &req->acct, 0,
-                     BLOCK_ACCT_WRITE);
-
-    req->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_l2b(ns, sdlba),
-                                 &req->sg.iov, 0, nvme_copy_cb, req);
-
-    return;
-
-invalid:
-    req->status = status;
-
-    g_free(ctx->bounce);
-    g_free(ctx);
-
-    nvme_enqueue_req_completion(nvme_cq(req), req);
-}
-
-static void nvme_aio_copy_in_cb(void *opaque, int ret)
-{
-    struct nvme_copy_in_ctx *in_ctx = opaque;
-    NvmeRequest *req = in_ctx->req;
-    NvmeNamespace *ns = req->ns;
-    struct nvme_copy_ctx *ctx = req->opaque;
-
-    qemu_iovec_destroy(&in_ctx->iov);
-    g_free(in_ctx);
-
-    trace_pci_nvme_aio_copy_in_cb(nvme_cid(req));
-
-    if (ret) {
-        nvme_aio_err(req, ret);
-    }
-
-    ctx->copies--;
-
-    if (ctx->copies) {
-        return;
-    }
-
-    if (req->status) {
-        block_acct_failed(blk_get_stats(ns->blkconf.blk), &req->acct);
-
-        g_free(ctx->bounce);
-        g_free(ctx->mbounce);
-        g_free(ctx);
-
-        nvme_enqueue_req_completion(nvme_cq(req), req);
-
-        return;
-    }
-
-    nvme_copy_in_complete(req);
-}
-
 struct nvme_compare_ctx {
     struct {
         QEMUIOVector iov;
@@ -2698,153 +2478,412 @@ static uint16_t nvme_verify(NvmeCtrl *n, NvmeRequest *req)
     return NVME_NO_COMPLETE;
 }
 
+typedef struct NvmeCopyAIOCB {
+    BlockAIOCB common;
+    BlockAIOCB *aiocb;
+    NvmeRequest *req;
+    QEMUBH *bh;
+    int ret;
+
+    NvmeCopySourceRange *ranges;
+    int nr;
+    int idx;
+
+    uint8_t *bounce;
+    QEMUIOVector iov;
+    struct {
+        BlockAcctCookie read;
+        BlockAcctCookie write;
+    } acct;
+
+    uint32_t reftag;
+    uint64_t slba;
+
+    NvmeZone *zone;
+} NvmeCopyAIOCB;
+
+static void nvme_copy_cancel(BlockAIOCB *aiocb)
+{
+    NvmeCopyAIOCB *iocb = container_of(aiocb, NvmeCopyAIOCB, common);
+
+    iocb->ret = -ECANCELED;
+    iocb->idx = iocb->nr;
+
+    if (iocb->aiocb) {
+        blk_aio_cancel_async(iocb->aiocb);
+        iocb->aiocb = NULL;
+    }
+}
+
+static const AIOCBInfo nvme_copy_aiocb_info = {
+    .aiocb_size   = sizeof(NvmeCopyAIOCB),
+    .cancel_async = nvme_copy_cancel,
+};
+
+static void nvme_copy_bh(void *opaque)
+{
+    NvmeCopyAIOCB *iocb = opaque;
+    NvmeRequest *req = iocb->req;
+    NvmeNamespace *ns = req->ns;
+    BlockAcctStats *stats = blk_get_stats(ns->blkconf.blk);
+
+    if (iocb->idx != iocb->nr) {
+        req->cqe.result = cpu_to_le32(iocb->idx);
+    }
+
+    qemu_iovec_destroy(&iocb->iov);
+    g_free(iocb->bounce);
+
+    qemu_bh_delete(iocb->bh);
+    iocb->bh = NULL;
+
+    if (iocb->ret < 0) {
+        block_acct_failed(stats, &iocb->acct.read);
+        block_acct_failed(stats, &iocb->acct.write);
+    } else {
+        block_acct_done(stats, &iocb->acct.read);
+        block_acct_done(stats, &iocb->acct.write);
+    }
+
+    iocb->common.cb(iocb->common.opaque, iocb->ret);
+    qemu_aio_unref(iocb);
+}
+
+static void nvme_copy_cb(void *opaque, int ret);
+
+static void nvme_copy_out_completed_cb(void *opaque, int ret)
+{
+    NvmeCopyAIOCB *iocb = opaque;
+    NvmeRequest *req = iocb->req;
+    NvmeNamespace *ns = req->ns;
+    NvmeCopySourceRange *range = &iocb->ranges[iocb->idx];
+    uint32_t nlb = le32_to_cpu(range->nlb) + 1;
+
+    if (ret < 0) {
+        nvme_copy_cb(iocb, ret);
+        return;
+    }
+
+    if (ns->params.zoned) {
+        nvme_advance_zone_wp(ns, iocb->zone, nlb);
+    }
+
+    iocb->idx++;
+    iocb->slba += nlb;
+    nvme_copy_cb(iocb, 0);
+}
+
+static void nvme_copy_out_cb(void *opaque, int ret)
+{
+    NvmeCopyAIOCB *iocb = opaque;
+    NvmeRequest *req = iocb->req;
+    NvmeNamespace *ns = req->ns;
+    NvmeCopySourceRange *range;
+    uint32_t nlb;
+    size_t mlen;
+    uint8_t *mbounce;
+
+    if (ret < 0) {
+        nvme_copy_cb(iocb, ret);
+        return;
+    }
+
+    if (!ns->lbaf.ms) {
+        nvme_copy_out_completed_cb(iocb, 0);
+        return;
+    }
+
+    range = &iocb->ranges[iocb->idx];
+    nlb = le32_to_cpu(range->nlb) + 1;
+
+    mlen = nvme_m2b(ns, nlb);
+    mbounce = iocb->bounce + nvme_l2b(ns, nlb);
+
+    qemu_iovec_reset(&iocb->iov);
+    qemu_iovec_add(&iocb->iov, mbounce, mlen);
+
+    iocb->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_moff(ns, iocb->slba),
+                                  &iocb->iov, 0, nvme_copy_out_completed_cb,
+                                  iocb);
+
+    return;
+}
+
+static void nvme_copy_in_completed_cb(void *opaque, int ret)
+{
+    NvmeCopyAIOCB *iocb = opaque;
+    NvmeRequest *req = iocb->req;
+    NvmeNamespace *ns = req->ns;
+    NvmeCopySourceRange *range;
+    uint32_t nlb;
+    size_t len;
+    uint16_t status;
+
+    if (ret < 0) {
+        nvme_copy_cb(iocb, ret);
+        return;
+    }
+
+    range = &iocb->ranges[iocb->idx];
+    nlb = le32_to_cpu(range->nlb) + 1;
+    len = nvme_l2b(ns, nlb);
+
+    trace_pci_nvme_copy_out(iocb->slba, nlb);
+
+    if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
+        NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
+
+        uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
+        uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
+
+        uint16_t apptag = le16_to_cpu(range->apptag);
+        uint16_t appmask = le16_to_cpu(range->appmask);
+        uint32_t reftag = le32_to_cpu(range->reftag);
+
+        uint64_t slba = le64_to_cpu(range->slba);
+        size_t mlen = nvme_m2b(ns, nlb);
+        uint8_t *mbounce = iocb->bounce + nvme_l2b(ns, nlb);
+
+        status = nvme_dif_check(ns, iocb->bounce, len, mbounce, mlen, prinfor,
+                                slba, apptag, appmask, &reftag);
+        if (status) {
+            goto invalid;
+        }
+
+        apptag = le16_to_cpu(copy->apptag);
+        appmask = le16_to_cpu(copy->appmask);
+
+        if (prinfow & NVME_PRINFO_PRACT) {
+            status = nvme_check_prinfo(ns, prinfow, iocb->slba, iocb->reftag);
+            if (status) {
+                goto invalid;
+            }
+
+            nvme_dif_pract_generate_dif(ns, iocb->bounce, len, mbounce, mlen,
+                                        apptag, &iocb->reftag);
+        } else {
+            status = nvme_dif_check(ns, iocb->bounce, len, mbounce, mlen,
+                                    prinfow, iocb->slba, apptag, appmask,
+                                    &iocb->reftag);
+            if (status) {
+                goto invalid;
+            }
+        }
+    }
+
+    status = nvme_check_bounds(ns, iocb->slba, nlb);
+    if (status) {
+        goto invalid;
+    }
+
+    if (ns->params.zoned) {
+        status = nvme_check_zone_write(ns, iocb->zone, iocb->slba, nlb);
+        if (status) {
+            goto invalid;
+        }
+
+        iocb->zone->w_ptr += nlb;
+    }
+
+    qemu_iovec_reset(&iocb->iov);
+    qemu_iovec_add(&iocb->iov, iocb->bounce, len);
+
+    iocb->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_l2b(ns, iocb->slba),
+                                  &iocb->iov, 0, nvme_copy_out_cb, iocb);
+
+    return;
+
+invalid:
+    req->status = status;
+    iocb->aiocb = NULL;
+    if (iocb->bh) {
+        qemu_bh_schedule(iocb->bh);
+    }
+}
+
+static void nvme_copy_in_cb(void *opaque, int ret)
+{
+    NvmeCopyAIOCB *iocb = opaque;
+    NvmeRequest *req = iocb->req;
+    NvmeNamespace *ns = req->ns;
+    NvmeCopySourceRange *range;
+    uint64_t slba;
+    uint32_t nlb;
+
+    if (ret < 0) {
+        nvme_copy_cb(iocb, ret);
+        return;
+    }
+
+    if (!ns->lbaf.ms) {
+        nvme_copy_in_completed_cb(iocb, 0);
+        return;
+    }
+
+    range = &iocb->ranges[iocb->idx];
+    slba = le64_to_cpu(range->slba);
+    nlb = le32_to_cpu(range->nlb) + 1;
+
+    qemu_iovec_reset(&iocb->iov);
+    qemu_iovec_add(&iocb->iov, iocb->bounce + nvme_l2b(ns, nlb),
+                   nvme_m2b(ns, nlb));
+
+    iocb->aiocb = blk_aio_preadv(ns->blkconf.blk, nvme_moff(ns, slba),
+                                 &iocb->iov, 0, nvme_copy_in_completed_cb,
+                                 iocb);
+    return;
+}
+
+static void nvme_copy_cb(void *opaque, int ret)
+{
+    NvmeCopyAIOCB *iocb = opaque;
+    NvmeRequest *req = iocb->req;
+    NvmeNamespace *ns = req->ns;
+    NvmeCopySourceRange *range;
+    uint64_t slba;
+    uint32_t nlb;
+    size_t len;
+    uint16_t status;
+
+    if (ret < 0) {
+        iocb->ret = ret;
+        goto done;
+    }
+
+    if (iocb->idx == iocb->nr) {
+        goto done;
+    }
+
+    range = &iocb->ranges[iocb->idx];
+    slba = le64_to_cpu(range->slba);
+    nlb = le32_to_cpu(range->nlb) + 1;
+    len = nvme_l2b(ns, nlb);
+
+    trace_pci_nvme_copy_source_range(slba, nlb);
+
+    if (nlb > le16_to_cpu(ns->id_ns.mssrl)) {
+        status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
+        goto invalid;
+    }
+
+    status = nvme_check_bounds(ns, slba, nlb);
+    if (status) {
+        goto invalid;
+    }
+
+    if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
+        status = nvme_check_dulbe(ns, slba, nlb);
+        if (status) {
+            goto invalid;
+        }
+    }
+
+    if (ns->params.zoned) {
+        status = nvme_check_zone_read(ns, slba, nlb);
+        if (status) {
+            goto invalid;
+        }
+    }
+
+    qemu_iovec_reset(&iocb->iov);
+    qemu_iovec_add(&iocb->iov, iocb->bounce, len);
+
+    iocb->aiocb = blk_aio_preadv(ns->blkconf.blk, nvme_l2b(ns, slba),
+                                 &iocb->iov, 0, nvme_copy_in_cb, iocb);
+    return;
+
+invalid:
+    req->status = status;
+done:
+    iocb->aiocb = NULL;
+    if (iocb->bh) {
+        qemu_bh_schedule(iocb->bh);
+    }
+}
+
+
 static uint16_t nvme_copy(NvmeCtrl *n, NvmeRequest *req)
 {
     NvmeNamespace *ns = req->ns;
     NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
-
+    NvmeCopyAIOCB *iocb = blk_aio_get(&nvme_copy_aiocb_info, ns->blkconf.blk,
+                                      nvme_misc_cb, req);
     uint16_t nr = copy->nr + 1;
     uint8_t format = copy->control[0] & 0xf;
-    uint8_t prinfor = (copy->control[0] >> 4) & 0xf;
-    uint8_t prinfow = (copy->control[2] >> 2) & 0xf;
+    uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
+    uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
 
-    uint32_t nlb = 0;
-    uint8_t *bounce = NULL, *bouncep = NULL;
-    uint8_t *mbounce = NULL, *mbouncep = NULL;
-    struct nvme_copy_ctx *ctx;
     uint16_t status;
-    int i;
 
     trace_pci_nvme_copy(nvme_cid(req), nvme_nsid(ns), nr, format);
 
+    iocb->ranges = NULL;
+    iocb->zone = NULL;
+
     if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) &&
         ((prinfor & NVME_PRINFO_PRACT) != (prinfow & NVME_PRINFO_PRACT))) {
-        return NVME_INVALID_FIELD | NVME_DNR;
+        status = NVME_INVALID_FIELD | NVME_DNR;
+        goto invalid;
     }
 
     if (!(n->id_ctrl.ocfs & (1 << format))) {
         trace_pci_nvme_err_copy_invalid_format(format);
-        return NVME_INVALID_FIELD | NVME_DNR;
+        status = NVME_INVALID_FIELD | NVME_DNR;
+        goto invalid;
     }
 
     if (nr > ns->id_ns.msrc + 1) {
-        return NVME_CMD_SIZE_LIMIT | NVME_DNR;
-    }
-
-    ctx = g_new(struct nvme_copy_ctx, 1);
-    ctx->ranges = g_new(NvmeCopySourceRange, nr);
-
-    status = nvme_h2c(n, (uint8_t *)ctx->ranges,
-                      nr * sizeof(NvmeCopySourceRange), req);
-    if (status) {
-        goto out;
-    }
-
-    for (i = 0; i < nr; i++) {
-        uint64_t slba = le64_to_cpu(ctx->ranges[i].slba);
-        uint32_t _nlb = le16_to_cpu(ctx->ranges[i].nlb) + 1;
-
-        if (_nlb > le16_to_cpu(ns->id_ns.mssrl)) {
-            status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
-            goto out;
-        }
-
-        status = nvme_check_bounds(ns, slba, _nlb);
-        if (status) {
-            goto out;
-        }
-
-        if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
-            status = nvme_check_dulbe(ns, slba, _nlb);
-            if (status) {
-                goto out;
-            }
-        }
-
-        if (ns->params.zoned) {
-            status = nvme_check_zone_read(ns, slba, _nlb);
-            if (status) {
-                goto out;
-            }
-        }
-
-        nlb += _nlb;
-    }
-
-    if (nlb > le32_to_cpu(ns->id_ns.mcl)) {
         status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
-        goto out;
+        goto invalid;
     }
 
-    bounce = bouncep = g_malloc(nvme_l2b(ns, nlb));
-    if (ns->lbaf.ms) {
-        mbounce = mbouncep = g_malloc(nvme_m2b(ns, nlb));
+    iocb->ranges = g_new(NvmeCopySourceRange, nr);
+
+    status = nvme_h2c(n, (uint8_t *)iocb->ranges,
+                      sizeof(NvmeCopySourceRange) * nr, req);
+    if (status) {
+        goto invalid;
     }
 
-    block_acct_start(blk_get_stats(ns->blkconf.blk), &req->acct, 0,
-                     BLOCK_ACCT_READ);
+    iocb->slba = le64_to_cpu(copy->sdlba);
 
-    ctx->bounce = bounce;
-    ctx->mbounce = mbounce;
-    ctx->nlb = nlb;
-    ctx->copies = 1;
+    if (ns->params.zoned) {
+        iocb->zone = nvme_get_zone_by_slba(ns, iocb->slba);
+        if (!iocb->zone) {
+            status = NVME_LBA_RANGE | NVME_DNR;
+            goto invalid;
+        }
 
-    req->opaque = ctx;
-
-    for (i = 0; i < nr; i++) {
-        uint64_t slba = le64_to_cpu(ctx->ranges[i].slba);
-        uint32_t nlb = le16_to_cpu(ctx->ranges[i].nlb) + 1;
-
-        size_t len = nvme_l2b(ns, nlb);
-        int64_t offset = nvme_l2b(ns, slba);
-
-        trace_pci_nvme_copy_source_range(slba, nlb);
-
-        struct nvme_copy_in_ctx *in_ctx = g_new(struct nvme_copy_in_ctx, 1);
-        in_ctx->req = req;
-
-        qemu_iovec_init(&in_ctx->iov, 1);
-        qemu_iovec_add(&in_ctx->iov, bouncep, len);
-
-        ctx->copies++;
-
-        blk_aio_preadv(ns->blkconf.blk, offset, &in_ctx->iov, 0,
-                       nvme_aio_copy_in_cb, in_ctx);
-
-        bouncep += len;
-
-        if (ns->lbaf.ms) {
-            len = nvme_m2b(ns, nlb);
-            offset = nvme_moff(ns, slba);
-
-            in_ctx = g_new(struct nvme_copy_in_ctx, 1);
-            in_ctx->req = req;
-
-            qemu_iovec_init(&in_ctx->iov, 1);
-            qemu_iovec_add(&in_ctx->iov, mbouncep, len);
-
-            ctx->copies++;
-
-            blk_aio_preadv(ns->blkconf.blk, offset, &in_ctx->iov, 0,
-                           nvme_aio_copy_in_cb, in_ctx);
-
-            mbouncep += len;
+        status = nvme_zrm_auto(ns, iocb->zone);
+        if (status) {
+            goto invalid;
         }
     }
 
-    /* account for the 1-initialization */
-    ctx->copies--;
+    iocb->req = req;
+    iocb->bh = qemu_bh_new(nvme_copy_bh, iocb);
+    iocb->ret = 0;
+    iocb->nr = nr;
+    iocb->idx = 0;
+    iocb->reftag = le32_to_cpu(copy->reftag);
+    iocb->bounce = g_malloc_n(le16_to_cpu(ns->id_ns.mssrl),
+                              ns->lbasz + ns->lbaf.ms);
 
-    if (!ctx->copies) {
-        nvme_copy_in_complete(req);
-    }
+    qemu_iovec_init(&iocb->iov, 1);
+
+    block_acct_start(blk_get_stats(ns->blkconf.blk), &iocb->acct.read, 0,
+                     BLOCK_ACCT_READ);
+    block_acct_start(blk_get_stats(ns->blkconf.blk), &iocb->acct.write, 0,
+                     BLOCK_ACCT_WRITE);
+
+    req->aiocb = &iocb->common;
+    nvme_copy_cb(iocb, 0);
 
     return NVME_NO_COMPLETE;
 
-out:
-    g_free(ctx->ranges);
-    g_free(ctx);
-
+invalid:
+    g_free(iocb->ranges);
+    qemu_aio_unref(iocb);
     return status;
 }
 
hw/nvme/trace-events +1 -2
index a3e11346865e..cd65f8b28895 100644
@@ -30,8 +30,7 @@ pci_nvme_dif_prchk_apptag(uint16_t apptag, uint16_t elbat, uint16_t elbatm) "app
 pci_nvme_dif_prchk_reftag(uint32_t reftag, uint32_t elbrt) "reftag 0x%"PRIx32" elbrt 0x%"PRIx32""
 pci_nvme_copy(uint16_t cid, uint32_t nsid, uint16_t nr, uint8_t format) "cid %"PRIu16" nsid %"PRIu32" nr %"PRIu16" format 0x%"PRIx8""
 pci_nvme_copy_source_range(uint64_t slba, uint32_t nlb) "slba 0x%"PRIx64" nlb %"PRIu32""
-pci_nvme_copy_in_complete(uint16_t cid) "cid %"PRIu16""
-pci_nvme_copy_cb(uint16_t cid) "cid %"PRIu16""
+pci_nvme_copy_out(uint64_t slba, uint32_t nlb) "slba 0x%"PRIx64" nlb %"PRIu32""
 pci_nvme_verify(uint16_t cid, uint32_t nsid, uint64_t slba, uint32_t nlb) "cid %"PRIu16" nsid %"PRIu32" slba 0x%"PRIx64" nlb %"PRIu32""
 pci_nvme_verify_mdata_in_cb(uint16_t cid, const char *blkname) "cid %"PRIu16" blk '%s'"
 pci_nvme_verify_cb(uint16_t cid, uint8_t prinfo, uint16_t apptag, uint16_t appmask, uint32_t reftag) "cid %"PRIu16" prinfo 0x%"PRIx8" apptag 0x%"PRIx16" appmask 0x%"PRIx16" reftag 0x%"PRIx32""
Signature · 1 line
2.31.1
#9 lore raw

RFC PATCH 9/11 hw/nvme: reimplement zone reset to allow cancellation

In reply to Klaus Jensen

From: Klaus Jensen <[email protected]>

Prior to this patch, the aios associated with zone reset are submitted anonymously (no reference saved to the aiocb from the blk_aio call).

Fix this by resetting the zones one after another, saving a reference to the aiocb for each reset.

---
 hw/nvme/ctrl.c       | 288 +++++++++++++++++++++++++------------------
 hw/nvme/trace-events |   2 +-
 2 files changed, 169 insertions(+), 121 deletions(-)
hw/nvme/ctrl.c +168 -120
index fb65a4f12d1f..f430a75a3fc5 100644
@@ -1685,6 +1685,29 @@ static uint16_t nvme_zrm_close(NvmeNamespace *ns, NvmeZone *zone)
     }
 }
 
+static uint16_t nvme_zrm_reset(NvmeNamespace *ns, NvmeZone *zone)
+{
+    switch (nvme_get_zone_state(zone)) {
+    case NVME_ZONE_STATE_EXPLICITLY_OPEN:
+    case NVME_ZONE_STATE_IMPLICITLY_OPEN:
+        nvme_aor_dec_open(ns);
+        /* fallthrough */
+    case NVME_ZONE_STATE_CLOSED:
+        nvme_aor_dec_active(ns);
+        /* fallthrough */
+    case NVME_ZONE_STATE_FULL:
+        zone->w_ptr = zone->d.zslba;
+        zone->d.wp = zone->w_ptr;
+        nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EMPTY);
+        /* fallthrough */
+    case NVME_ZONE_STATE_EMPTY:
+        return NVME_SUCCESS;
+
+    default:
+        return NVME_ZONE_INVAL_TRANSITION;
+    }
+}
+
 static void nvme_zrm_auto_transition_zone(NvmeNamespace *ns)
 {
     NvmeZone *zone;
@@ -2005,79 +2028,6 @@ out:
     nvme_verify_cb(ctx, ret);
 }
 
-struct nvme_zone_reset_ctx {
-    NvmeRequest *req;
-    NvmeZone    *zone;
-};
-
-static void nvme_aio_zone_reset_complete_cb(void *opaque, int ret)
-{
-    struct nvme_zone_reset_ctx *ctx = opaque;
-    NvmeRequest *req = ctx->req;
-    NvmeNamespace *ns = req->ns;
-    NvmeZone *zone = ctx->zone;
-    uintptr_t *resets = (uintptr_t *)&req->opaque;
-
-    if (ret) {
-        nvme_aio_err(req, ret);
-        goto out;
-    }
-
-    switch (nvme_get_zone_state(zone)) {
-    case NVME_ZONE_STATE_EXPLICITLY_OPEN:
-    case NVME_ZONE_STATE_IMPLICITLY_OPEN:
-        nvme_aor_dec_open(ns);
-        /* fall through */
-    case NVME_ZONE_STATE_CLOSED:
-        nvme_aor_dec_active(ns);
-        /* fall through */
-    case NVME_ZONE_STATE_FULL:
-        zone->w_ptr = zone->d.zslba;
-        zone->d.wp = zone->w_ptr;
-        nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EMPTY);
-        /* fall through */
-    default:
-        break;
-    }
-
-out:
-    g_free(ctx);
-
-    (*resets)--;
-
-    if (*resets) {
-        return;
-    }
-
-    nvme_enqueue_req_completion(nvme_cq(req), req);
-}
-
-static void nvme_aio_zone_reset_cb(void *opaque, int ret)
-{
-    struct nvme_zone_reset_ctx *ctx = opaque;
-    NvmeRequest *req = ctx->req;
-    NvmeNamespace *ns = req->ns;
-    NvmeZone *zone = ctx->zone;
-
-    trace_pci_nvme_aio_zone_reset_cb(nvme_cid(req), zone->d.zslba);
-
-    if (ret) {
-        goto out;
-    }
-
-    if (ns->lbaf.ms) {
-        int64_t offset = nvme_moff(ns, zone->d.zslba);
-
-        blk_aio_pwrite_zeroes(ns->blkconf.blk, offset,
-                              nvme_m2b(ns, ns->zone_size), BDRV_REQ_MAY_UNMAP,
-                              nvme_aio_zone_reset_complete_cb, ctx);
-        return;
-    }
-
-out:
-    nvme_aio_zone_reset_complete_cb(opaque, ret);
-}
-
 struct nvme_compare_ctx {
     struct {
         QEMUIOVector iov;
@@ -3363,41 +3313,6 @@ static uint16_t nvme_finish_zone(NvmeNamespace *ns, NvmeZone *zone,
     return nvme_zrm_finish(ns, zone);
 }
 
-static uint16_t nvme_reset_zone(NvmeNamespace *ns, NvmeZone *zone,
-                                NvmeZoneState state, NvmeRequest *req)
-{
-    uintptr_t *resets = (uintptr_t *)&req->opaque;
-    struct nvme_zone_reset_ctx *ctx;
-
-    switch (state) {
-    case NVME_ZONE_STATE_EMPTY:
-        return NVME_SUCCESS;
-    case NVME_ZONE_STATE_EXPLICITLY_OPEN:
-    case NVME_ZONE_STATE_IMPLICITLY_OPEN:
-    case NVME_ZONE_STATE_CLOSED:
-    case NVME_ZONE_STATE_FULL:
-        break;
-    default:
-        return NVME_ZONE_INVAL_TRANSITION;
-    }
-
-    /*
-     * The zone reset aio callback needs to know the zone that is being reset
-     * in order to transition the zone on completion.
-     */
-    ctx = g_new(struct nvme_zone_reset_ctx, 1);
-    ctx->req = req;
-    ctx->zone = zone;
-
-    (*resets)++;
-
-    blk_aio_pwrite_zeroes(ns->blkconf.blk, nvme_l2b(ns, zone->d.zslba),
-                          nvme_l2b(ns, ns->zone_size), BDRV_REQ_MAY_UNMAP,
-                          nvme_aio_zone_reset_cb, ctx);
-
-    return NVME_NO_COMPLETE;
-}
-
 static uint16_t nvme_offline_zone(NvmeNamespace *ns, NvmeZone *zone,
                                   NvmeZoneState state, NvmeRequest *req)
 {
@@ -3526,12 +3441,144 @@ out:
     return status;
 }
 
+typedef struct NvmeZoneResetAIOCB {
+    BlockAIOCB common;
+    BlockAIOCB *aiocb;
+    NvmeRequest *req;
+    QEMUBH *bh;
+    int ret;
+
+    bool all;
+    int idx;
+    NvmeZone *zone;
+} NvmeZoneResetAIOCB;
+
+static void nvme_zone_reset_cancel(BlockAIOCB *aiocb)
+{
+    NvmeZoneResetAIOCB *iocb = container_of(aiocb, NvmeZoneResetAIOCB, common);
+    NvmeRequest *req = iocb->req;
+    NvmeNamespace *ns = req->ns;
+
+    iocb->idx = ns->num_zones;
+
+    iocb->ret = -ECANCELED;
+
+    if (iocb->aiocb) {
+        blk_aio_cancel_async(iocb->aiocb);
+        iocb->aiocb = NULL;
+    }
+}
+
+static const AIOCBInfo nvme_zone_reset_aiocb_info = {
+    .aiocb_size = sizeof(NvmeZoneResetAIOCB),
+    .cancel_async = nvme_zone_reset_cancel,
+};
+
+static void nvme_zone_reset_bh(void *opaque)
+{
+    NvmeZoneResetAIOCB *iocb = opaque;
+
+    iocb->common.cb(iocb->common.opaque, iocb->ret);
+
+    qemu_bh_delete(iocb->bh);
+    iocb->bh = NULL;
+    qemu_aio_unref(iocb);
+}
+
+static void nvme_zone_reset_cb(void *opaque, int ret);
+
+static void nvme_zone_reset_epilogue_cb(void *opaque, int ret)
+{
+    NvmeZoneResetAIOCB *iocb = opaque;
+    NvmeRequest *req = iocb->req;
+    NvmeNamespace *ns = req->ns;
+    int64_t moff;
+    int count;
+
+    if (ret < 0) {
+        nvme_zone_reset_cb(iocb, ret);
+        return;
+    }
+
+    if (!ns->lbaf.ms) {
+        nvme_zone_reset_cb(iocb, 0);
+        return;
+    }
+
+    moff = nvme_moff(ns, iocb->zone->d.zslba);
+    count = nvme_m2b(ns, ns->zone_size);
+
+    iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, moff, count,
+                                        BDRV_REQ_MAY_UNMAP,
+                                        nvme_zone_reset_cb, iocb);
+    return;
+}
+
+static void nvme_zone_reset_cb(void *opaque, int ret)
+{
+    NvmeZoneResetAIOCB *iocb = opaque;
+    NvmeRequest *req = iocb->req;
+    NvmeNamespace *ns = req->ns;
+
+    if (ret < 0) {
+        iocb->ret = ret;
+        goto done;
+    }
+
+    if (iocb->zone) {
+        nvme_zrm_reset(ns, iocb->zone);
+
+        if (!iocb->all) {
+            goto done;
+        }
+    }
+
+    while (iocb->idx < ns->num_zones) {
+        NvmeZone *zone = &ns->zone_array[iocb->idx++];
+
+        switch (nvme_get_zone_state(zone)) {
+        case NVME_ZONE_STATE_EMPTY:
+            if (!iocb->all) {
+                goto done;
+            }
+
+            continue;
+
+        case NVME_ZONE_STATE_EXPLICITLY_OPEN:
+        case NVME_ZONE_STATE_IMPLICITLY_OPEN:
+        case NVME_ZONE_STATE_CLOSED:
+        case NVME_ZONE_STATE_FULL:
+            iocb->zone = zone;
+            break;
+
+        default:
+            continue;
+        }
+
+        trace_pci_nvme_zns_zone_reset(zone->d.zslba);
+
+        iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk,
+                                            nvme_l2b(ns, zone->d.zslba),
+                                            nvme_l2b(ns, ns->zone_size),
+                                            BDRV_REQ_MAY_UNMAP,
+                                            nvme_zone_reset_epilogue_cb,
+                                            iocb);
+        return;
+    }
+
+done:
+    iocb->aiocb = NULL;
+    if (iocb->bh) {
+        qemu_bh_schedule(iocb->bh);
+    }
+}
+
 static uint16_t nvme_zone_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
 {
     NvmeCmd *cmd = (NvmeCmd *)&req->cmd;
     NvmeNamespace *ns = req->ns;
     NvmeZone *zone;
-    uintptr_t *resets;
+    NvmeZoneResetAIOCB *iocb;
     uint8_t *zd_ext;
     uint32_t dw13 = le32_to_cpu(cmd->cdw13);
     uint64_t slba = 0;
@@ -3542,7 +3589,7 @@ static uint16_t nvme_zone_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
     enum NvmeZoneProcessingMask proc_mask = NVME_PROC_CURRENT_ZONE;
 
     action = dw13 & 0xff;
-    all = dw13 & 0x100;
+    all = !!(dw13 & 0x100);
 
     req->status = NVME_SUCCESS;
 
@@ -3586,21 +3633,22 @@ static uint16_t nvme_zone_mgmt_send(NvmeCtrl *n, NvmeRequest *req)
         break;
 
     case NVME_ZONE_ACTION_RESET:
-        resets = (uintptr_t *)&req->opaque;
-
-        if (all) {
-            proc_mask = NVME_PROC_OPENED_ZONES | NVME_PROC_CLOSED_ZONES |
-                NVME_PROC_FULL_ZONES;
-        }
         trace_pci_nvme_reset_zone(slba, zone_idx, all);
 
-        *resets = 1;
+        iocb = blk_aio_get(&nvme_zone_reset_aiocb_info, ns->blkconf.blk,
+                           nvme_misc_cb, req);
 
-        status = nvme_do_zone_op(ns, zone, proc_mask, nvme_reset_zone, req);
+        iocb->req = req;
+        iocb->bh = qemu_bh_new(nvme_zone_reset_bh, iocb);
+        iocb->ret = 0;
+        iocb->all = all;
+        iocb->idx = zone_idx;
+        iocb->zone = NULL;
 
-        (*resets)--;
+        req->aiocb = &iocb->common;
+        nvme_zone_reset_cb(iocb, 0);
 
-        return *resets ? NVME_NO_COMPLETE : req->status;
+        return NVME_NO_COMPLETE;
 
     case NVME_ZONE_ACTION_OFFLINE:
         if (all) {
hw/nvme/trace-events +1 -1
index cd65f8b28895..dc00c2860db7 100644
@@ -44,7 +44,6 @@ pci_nvme_compare_data_cb(uint16_t cid) "cid %"PRIu16""
 pci_nvme_compare_mdata_cb(uint16_t cid) "cid %"PRIu16""
 pci_nvme_aio_discard_cb(uint16_t cid) "cid %"PRIu16""
 pci_nvme_aio_copy_in_cb(uint16_t cid) "cid %"PRIu16""
-pci_nvme_aio_zone_reset_cb(uint16_t cid, uint64_t zslba) "cid %"PRIu16" zslba 0x%"PRIx64""
 pci_nvme_aio_flush_cb(uint16_t cid, const char *blkname) "cid %"PRIu16" blk '%s'"
 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"
@@ -100,6 +99,7 @@ pci_nvme_open_zone(uint64_t slba, uint32_t zone_idx, int all) "open zone, slba=%
 pci_nvme_close_zone(uint64_t slba, uint32_t zone_idx, int all) "close zone, slba=%"PRIu64", idx=%"PRIu32", all=%"PRIi32""
 pci_nvme_finish_zone(uint64_t slba, uint32_t zone_idx, int all) "finish zone, slba=%"PRIu64", idx=%"PRIu32", all=%"PRIi32""
 pci_nvme_reset_zone(uint64_t slba, uint32_t zone_idx, int all) "reset zone, slba=%"PRIu64", idx=%"PRIu32", all=%"PRIi32""
+pci_nvme_zns_zone_reset(uint64_t zslba) "zslba 0x%"PRIx64""
 pci_nvme_offline_zone(uint64_t slba, uint32_t zone_idx, int all) "offline zone, slba=%"PRIu64", idx=%"PRIu32", all=%"PRIi32""
 pci_nvme_set_descriptor_extension(uint64_t slba, uint32_t zone_idx) "set zone descriptor extension, slba=%"PRIu64", idx=%"PRIu32""
 pci_nvme_zd_extension_set(uint32_t zone_idx) "set descriptor extension for zone_idx=%"PRIu32""
Signature · 1 line
2.31.1
#10 lore raw

RFC PATCH 10/11 hw/nvme: reimplement format nvm to allow cancellation

In reply to Klaus Jensen

From: Klaus Jensen <[email protected]>

Prior to this patch, the aios associated with broadcast format are submitted anonymously (no aiocb reference saved from the blk_aio call).

Fix this by formatting the namespaces one after another, saving a reference to the aiocb for each.

---
 hw/nvme/ctrl.c       | 301 +++++++++++++++++++++++--------------------
 hw/nvme/trace-events |   4 +-
 2 files changed, 162 insertions(+), 143 deletions(-)
hw/nvme/ctrl.c +161 -140
index f430a75a3fc5..ee77abd19661 100644
@@ -1909,42 +1909,6 @@ out:
     nvme_rw_complete_cb(req, ret);
 }
 
-struct nvme_aio_format_ctx {
-    NvmeRequest   *req;
-    NvmeNamespace *ns;
-
-    /* number of outstanding write zeroes for this namespace */
-    int *count;
-};
-
-static void nvme_aio_format_cb(void *opaque, int ret)
-{
-    struct nvme_aio_format_ctx *ctx = opaque;
-    NvmeRequest *req = ctx->req;
-    NvmeNamespace *ns = ctx->ns;
-    uintptr_t *num_formats = (uintptr_t *)&req->opaque;
-    int *count = ctx->count;
-
-    g_free(ctx);
-
-    if (ret) {
-        nvme_aio_err(req, ret);
-    }
-
-    if (--(*count)) {
-        return;
-    }
-
-    g_free(count);
-    ns->status = 0x0;
-
-    if (--(*num_formats)) {
-        return;
-    }
-
-    nvme_enqueue_req_completion(nvme_cq(req), req);
-}
-
 static void nvme_verify_cb(void *opaque, int ret)
 {
     NvmeBounceContext *ctx = opaque;
@@ -2899,6 +2863,11 @@ static uint16_t nvme_compare(NvmeCtrl *n, NvmeRequest *req)
     return NVME_NO_COMPLETE;
 }
 
+static inline AioContext *nvme_get_aio_context(BlockAIOCB *acb)
+{
+    return qemu_get_aio_context();
+}
+
 typedef struct NvmeFlushAIOCB {
     BlockAIOCB common;
     BlockAIOCB *aiocb;
@@ -2923,15 +2892,10 @@ static void nvme_flush_cancel(BlockAIOCB *acb)
     iocb->ret = -ECANCELED;
 }
 
-static AioContext *nvme_flush_get_aio_context(BlockAIOCB *acb)
-{
-    return qemu_get_aio_context();
-}
-
 static const AIOCBInfo nvme_flush_aiocb_info = {
     .aiocb_size = sizeof(NvmeFlushAIOCB),
     .cancel_async = nvme_flush_cancel,
-    .get_aio_context = nvme_flush_get_aio_context,
+    .get_aio_context = nvme_get_aio_context,
 };
 
 static void nvme_flush_ns_cb(void *opaque, int ret)
@@ -5238,30 +5202,98 @@ static uint16_t nvme_ns_attachment(NvmeCtrl *n, NvmeRequest *req)
     return NVME_SUCCESS;
 }
 
-static uint16_t nvme_format_ns(NvmeCtrl *n, NvmeNamespace *ns, uint8_t lbaf,
-                               uint8_t mset, uint8_t pi, uint8_t pil,
-                               NvmeRequest *req)
-{
-    int64_t len, offset;
-    struct nvme_aio_format_ctx *ctx;
-    BlockBackend *blk = ns->blkconf.blk;
-    uint16_t ms;
-    uintptr_t *num_formats = (uintptr_t *)&req->opaque;
-    int *count;
+typedef struct NvmeFormatAIOCB {
+    BlockAIOCB common;
+    BlockAIOCB *aiocb;
+    QEMUBH *bh;
+    NvmeRequest *req;
+    int ret;
 
+    NvmeNamespace *ns;
+    uint32_t nsid;
+    bool broadcast;
+    int64_t offset;
+} NvmeFormatAIOCB;
+
+static void nvme_format_bh(void *opaque);
+
+static void nvme_format_cancel(BlockAIOCB *aiocb)
+{
+    NvmeFormatAIOCB *iocb = container_of(aiocb, NvmeFormatAIOCB, common);
+
+    if (iocb->aiocb) {
+        blk_aio_cancel_async(iocb->aiocb);
+    }
+}
+
+static const AIOCBInfo nvme_format_aiocb_info = {
+    .aiocb_size = sizeof(NvmeFormatAIOCB),
+    .cancel_async = nvme_format_cancel,
+    .get_aio_context = nvme_get_aio_context,
+};
+
+static void nvme_format_set(NvmeNamespace *ns, NvmeCmd *cmd)
+{
+    uint32_t dw10 = le32_to_cpu(cmd->cdw10);
+    uint8_t lbaf = dw10 & 0xf;
+    uint8_t pi = (dw10 >> 5) & 0x7;
+    uint8_t mset = (dw10 >> 4) & 0x1;
+    uint8_t pil = (dw10 >> 8) & 0x1;
+
+    trace_pci_nvme_format_set(ns->params.nsid, lbaf, mset, pi, pil);
+
+    ns->id_ns.dps = (pil << 3) | pi;
+    ns->id_ns.flbas = lbaf | (mset << 4);
+
+    nvme_ns_init_format(ns);
+}
+
+static void nvme_format_ns_cb(void *opaque, int ret)
+{
+    NvmeFormatAIOCB *iocb = opaque;
+    NvmeRequest *req = iocb->req;
+    NvmeNamespace *ns = iocb->ns;
+    int bytes;
+
+    if (ret < 0) {
+        iocb->ret = ret;
+        goto done;
+    }
+
+    assert(ns);
+
+    if (iocb->offset < ns->size) {
+        bytes = MIN(BDRV_REQUEST_MAX_BYTES, ns->size - iocb->offset);
+
+        iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, iocb->offset,
+                                            bytes, BDRV_REQ_MAY_UNMAP,
+                                            nvme_format_ns_cb, iocb);
+
+        iocb->offset += bytes;
+        return;
+    }
+
+    nvme_format_set(ns, &req->cmd);
+    ns->status = 0x0;
+    iocb->ns = NULL;
+    iocb->offset = 0;
+
+done:
+    iocb->aiocb = NULL;
+    qemu_bh_schedule(iocb->bh);
+}
+
+static uint16_t nvme_format_check(NvmeNamespace *ns, uint8_t lbaf, uint8_t pi)
+{
     if (ns->params.zoned) {
         return NVME_INVALID_FORMAT | NVME_DNR;
     }
 
-    trace_pci_nvme_format_ns(nvme_cid(req), nvme_nsid(ns), lbaf, mset, pi, pil);
-
     if (lbaf > ns->id_ns.nlbaf) {
         return NVME_INVALID_FORMAT | NVME_DNR;
     }
 
-    ms = ns->id_ns.lbaf[lbaf].ms;
-
-    if (pi && (ms < sizeof(NvmeDifTuple))) {
+    if (pi && (ns->id_ns.lbaf[lbaf].ms < sizeof(NvmeDifTuple))) {
         return NVME_INVALID_FORMAT | NVME_DNR;
     }
 
@@ -5269,107 +5301,96 @@ static uint16_t nvme_format_ns(NvmeCtrl *n, NvmeNamespace *ns, uint8_t lbaf,
         return NVME_INVALID_FIELD | NVME_DNR;
     }
 
-    nvme_ns_drain(ns);
-    nvme_ns_shutdown(ns);
-    nvme_ns_cleanup(ns);
-
-    ns->id_ns.dps = (pil << 3) | pi;
-    ns->id_ns.flbas = lbaf | (mset << 4);
-
-    nvme_ns_init_format(ns);
-
-    ns->status = NVME_FORMAT_IN_PROGRESS;
-
-    len = ns->size;
-    offset = 0;
-
-    count = g_new(int, 1);
-    *count = 1;
-
-    (*num_formats)++;
-
-    while (len) {
-        ctx = g_new(struct nvme_aio_format_ctx, 1);
-        ctx->req = req;
-        ctx->ns = ns;
-        ctx->count = count;
-
-        size_t bytes = MIN(BDRV_REQUEST_MAX_BYTES, len);
-
-        (*count)++;
-
-        blk_aio_pwrite_zeroes(blk, offset, bytes, BDRV_REQ_MAY_UNMAP,
-                              nvme_aio_format_cb, ctx);
-
-        offset += bytes;
-        len -= bytes;
-
-    }
-
-    if (--(*count)) {
-        return NVME_NO_COMPLETE;
-    }
-
-    g_free(count);
-    ns->status = 0x0;
-    (*num_formats)--;
-
     return NVME_SUCCESS;
 }
 
-static uint16_t nvme_format(NvmeCtrl *n, NvmeRequest *req)
+static void nvme_format_bh(void *opaque)
 {
-    NvmeNamespace *ns;
+    NvmeFormatAIOCB *iocb = opaque;
+    NvmeRequest *req = iocb->req;
+    NvmeCtrl *n = nvme_ctrl(req);
     uint32_t dw10 = le32_to_cpu(req->cmd.cdw10);
-    uint32_t nsid = le32_to_cpu(req->cmd.nsid);
     uint8_t lbaf = dw10 & 0xf;
-    uint8_t mset = (dw10 >> 4) & 0x1;
     uint8_t pi = (dw10 >> 5) & 0x7;
-    uint8_t pil = (dw10 >> 8) & 0x1;
-    uintptr_t *num_formats = (uintptr_t *)&req->opaque;
     uint16_t status;
     int i;
 
-    trace_pci_nvme_format(nvme_cid(req), nsid, lbaf, mset, pi, pil);
+    if (iocb->ret < 0) {
+        goto done;
+    }
 
-    /* 1-initialize; see the comment in nvme_dsm */
-    *num_formats = 1;
-
-    if (nsid != NVME_NSID_BROADCAST) {
-        if (!nvme_nsid_valid(n, nsid)) {
-            return NVME_INVALID_NSID | NVME_DNR;
-        }
-
-        ns = nvme_ns(n, nsid);
-        if (!ns) {
-            return NVME_INVALID_FIELD | NVME_DNR;
-        }
-
-        status = nvme_format_ns(n, ns, lbaf, mset, pi, pil, req);
-        if (status && status != NVME_NO_COMPLETE) {
-            req->status = status;
-        }
-    } else {
-        for (i = 1; i <= NVME_MAX_NAMESPACES; i++) {
-            ns = nvme_ns(n, i);
-            if (!ns) {
-                continue;
-            }
-
-            status = nvme_format_ns(n, ns, lbaf, mset, pi, pil, req);
-            if (status && status != NVME_NO_COMPLETE) {
-                req->status = status;
+    if (iocb->broadcast) {
+        for (i = iocb->nsid + 1; i <= NVME_MAX_NAMESPACES; i++) {
+            iocb->ns = nvme_ns(n, i);
+            if (iocb->ns) {
+                iocb->nsid = i;
                 break;
             }
         }
     }
 
-    /* account for the 1-initialization */
-    if (--(*num_formats)) {
-        return NVME_NO_COMPLETE;
+    if (!iocb->ns) {
+        goto done;
     }
 
-    return req->status;
+    status = nvme_format_check(iocb->ns, lbaf, pi);
+    if (status) {
+        req->status = status;
+        goto done;
+    }
+
+    iocb->ns->status = NVME_FORMAT_IN_PROGRESS;
+    nvme_format_ns_cb(iocb, 0);
+    return;
+
+done:
+    qemu_bh_delete(iocb->bh);
+    iocb->bh = NULL;
+
+    iocb->common.cb(iocb->common.opaque, iocb->ret);
+
+    qemu_aio_unref(iocb);
+}
+
+static uint16_t nvme_format(NvmeCtrl *n, NvmeRequest *req)
+{
+    NvmeFormatAIOCB *iocb;
+    uint32_t nsid = le32_to_cpu(req->cmd.nsid);
+    uint16_t status;
+
+    iocb = qemu_aio_get(&nvme_format_aiocb_info, NULL, nvme_misc_cb, req);
+
+    iocb->req = req;
+    iocb->bh = qemu_bh_new(nvme_format_bh, iocb);
+    iocb->ret = 0;
+    iocb->ns = NULL;
+    iocb->nsid = 0;
+    iocb->broadcast = (nsid == NVME_NSID_BROADCAST);
+    iocb->offset = 0;
+
+    if (!iocb->broadcast) {
+        if (!nvme_nsid_valid(n, nsid)) {
+            status = NVME_INVALID_NSID | NVME_DNR;
+            goto out;
+        }
+
+        iocb->ns = nvme_ns(n, nsid);
+        if (!iocb->ns) {
+            status = NVME_INVALID_FIELD | NVME_DNR;
+            goto out;
+        }
+    }
+
+    req->aiocb = &iocb->common;
+    qemu_bh_schedule(iocb->bh);
+
+    return NVME_NO_COMPLETE;
+
+out:
+    qemu_bh_delete(iocb->bh);
+    iocb->bh = NULL;
+    qemu_aio_unref(iocb);
+    return status;
 }
 
 static uint16_t nvme_admin_cmd(NvmeCtrl *n, NvmeRequest *req)
hw/nvme/trace-events +1 -3
index dc00c2860db7..48d10c36e85b 100644
@@ -10,9 +10,7 @@ pci_nvme_map_sgl(uint8_t typ, uint64_t len) "type 0x%"PRIx8" len %"PRIu64""
 pci_nvme_io_cmd(uint16_t cid, uint32_t nsid, uint16_t sqid, uint8_t opcode, const char *opname) "cid %"PRIu16" nsid 0x%"PRIx32" sqid %"PRIu16" opc 0x%"PRIx8" opname '%s'"
 pci_nvme_admin_cmd(uint16_t cid, uint16_t sqid, uint8_t opcode, const char *opname) "cid %"PRIu16" sqid %"PRIu16" opc 0x%"PRIx8" opname '%s'"
 pci_nvme_flush_ns(uint32_t nsid) "nsid 0x%"PRIx32""
-pci_nvme_format(uint16_t cid, uint32_t nsid, uint8_t lbaf, uint8_t mset, uint8_t pi, uint8_t pil) "cid %"PRIu16" nsid %"PRIu32" lbaf %"PRIu8" mset %"PRIu8" pi %"PRIu8" pil %"PRIu8""
-pci_nvme_format_ns(uint16_t cid, uint32_t nsid, uint8_t lbaf, uint8_t mset, uint8_t pi, uint8_t pil) "cid %"PRIu16" nsid %"PRIu32" lbaf %"PRIu8" mset %"PRIu8" pi %"PRIu8" pil %"PRIu8""
-pci_nvme_format_cb(uint16_t cid, uint32_t nsid) "cid %"PRIu16" nsid %"PRIu32""
+pci_nvme_format_set(uint32_t nsid, uint8_t lbaf, uint8_t mset, uint8_t pi, uint8_t pil) "nsid %"PRIu32" lbaf %"PRIu8" mset %"PRIu8" pi %"PRIu8" pil %"PRIu8""
 pci_nvme_read(uint16_t cid, uint32_t nsid, uint32_t nlb, uint64_t count, uint64_t lba) "cid %"PRIu16" nsid %"PRIu32" nlb %"PRIu32" count %"PRIu64" lba 0x%"PRIx64""
 pci_nvme_write(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'"
Signature · 1 line
2.31.1
#11 lore raw

RFC PATCH 11/11 Partially revert "hw/block/nvme: drain namespaces on sq deletion"

In reply to Klaus Jensen

From: Klaus Jensen <[email protected]>

This partially reverts commit 98f84f5a4eca5c03e32fff20f246d9b4b96d6422.

Since all "multi aio" commands are now reimplemented to properly track the nested aiocbs, we can revert the "hack" that was introduced to make sure all requests we're properly drained upon sq deletion.

The revert is partial since we keep the assert that no outstanding requests remain on the submission queue after the explicit cancellation.

---
 hw/nvme/ctrl.c | 19 ++-----------------
 1 file changed, 2 insertions(+), 17 deletions(-)
hw/nvme/ctrl.c +2 -17
index ee77abd19661..d25c5d8187e9 100644
@@ -3886,7 +3886,6 @@ static uint16_t nvme_del_sq(NvmeCtrl *n, NvmeRequest *req)
     NvmeSQueue *sq;
     NvmeCQueue *cq;
     uint16_t qid = le16_to_cpu(c->qid);
-    uint32_t nsid;
 
     if (unlikely(!qid || nvme_check_sqid(n, qid))) {
         trace_pci_nvme_err_invalid_del_sq(qid);
@@ -3898,22 +3897,8 @@ static uint16_t nvme_del_sq(NvmeCtrl *n, NvmeRequest *req)
     sq = n->sq[qid];
     while (!QTAILQ_EMPTY(&sq->out_req_list)) {
         r = QTAILQ_FIRST(&sq->out_req_list);
-        if (r->aiocb) {
-            blk_aio_cancel(r->aiocb);
-        }
-    }
-
-    /*
-     * Drain all namespaces if there are still outstanding requests that we
-     * could not cancel explicitly.
-     */
-    if (!QTAILQ_EMPTY(&sq->out_req_list)) {
-        for (nsid = 1; nsid <= NVME_MAX_NAMESPACES; nsid++) {
-            NvmeNamespace *ns = nvme_ns(n, nsid);
-            if (ns) {
-                nvme_ns_drain(ns);
-            }
-        }
+        assert(r->aiocb);
+        blk_aio_cancel(r->aiocb);
     }
 
     assert(QTAILQ_EMPTY(&sq->out_req_list));
Signature · 1 line
2.31.1
#12 lore raw
In reply to Klaus Jensen
Quoted 11 lines · Klaus Jensen

From: Klaus Jensen <[email protected]>

This series reimplements flush, dsm, copy, zone reset and format nvm to allow cancellation. I posted an RFC back in March ("hw/block/nvme:
convert ad-hoc aio tracking to aiocb") and I've applied some feedback from Stefan and reimplemented the remaining commands.

The basic idea is to define custom AIOCBs for these commands. The custom AIOCB takes care of issuing all the "nested" AIOs one by one instead of blindly sending them off simultaneously without tracking the returned aiocbs.

I'm not familiar with nvme. But intuitively, isn't it less efficient to send mutltiple commands one-by-one? Overall, wouldn't it be slower? In block layer we mostly do opposite transition: instead of doing IO operations one-by-one, run them simultaneously to make a non-empty queue on a device.. Even on one device. This way overall performance is increased.

If you need to store nested AIOCBs, you may store them in a list for example, and cancel in a loop, keeping simultaneous start for all flushes.. If you send two flushes on two different disks, what's the reason to wait for first flush finish before issuing the second?

Quoted 6 lines

I've kept the RFC since I'm still new to using the block layer like this. I was hoping that Stefan could find some time to look over this - this is a huge series, so I don't expect non-nvme folks to spend a large amount of time on it, but I would really like feedback on my approach in the reimplementation of flush and format.

If I understand your code correctly, you do stat next io operation from call-back of a previous. It works, and it is similar to haw mirror block-job was operating some time ago (still it maintained several in-flight requests simultaneously, but I'm about using _aio_ functions). Still, now mirror doesn't use _aio_ functions like this.

Better approach to call several io functions of block layer one-by-one is creating a coroutine. You may just add a coroutine function, that does all your linear logic as you want, without any callbacks like:

nvme_co_flush()
{

   for (...) {
      blk_co_flush();

}
}

(and you'll need qemu_coroutine_create() and qemu_coroutine_enter() to start a coroutine).

Still, I'm not sure that moving from simultaneous issuing several IO commands to sequential is good idea.. And this way you of course can't use blk_aio_canel.. This leads to my last doubt:

One more thing I don't understand after fast look at the series: how cancelation works? It seems to me, that you just call cancel on nested AIOCBs, produced by blk_<io_functions>, but no one of them implement cancel.. I see only four implementations of .cancel_async callback in the whole Qemu code: in iscsi, in ide/core.c, in dma-helpers.c and in thread-pool.c.. Seems no one is related to blk_aio_flush() and other blk_* functions you call in the series. Or, what I miss?

Quoted 27 lines

Those commands are special in
that may issue AIOs to multiple namespaces and thus, to multiple block backends. Since this device does not support iothreads, I've opted for simply always returning the main loop aio context, but I wonder if this is acceptable or not. It might be the case that this should contain an assert of some kind, in case someone starts adding iothread support.

Klaus Jensen (11):
hw/nvme: reimplement flush to allow cancellation
hw/nvme: add nvme_block_status_all helper
hw/nvme: reimplement dsm to allow cancellation
hw/nvme: save reftag when generating pi
hw/nvme: remove assert from nvme_get_zone_by_slba hw/nvme: use prinfo directly in nvme_check_prinfo and nvme_dif_check hw/nvme: add dw0/1 to the req completion trace event hw/nvme: reimplement the copy command to allow aio cancellation hw/nvme: reimplement zone reset to allow cancellation hw/nvme: reimplement format nvm to allow cancellation Partially revert "hw/block/nvme: drain namespaces on sq deletion"

  hw/nvme/nvme.h       |   10 +-
  include/block/nvme.h |    8 +
  hw/nvme/ctrl.c       | 1861 ++++++++++++++++++++++++------------------
  hw/nvme/dif.c        |   64 +-
  hw/nvme/trace-events |   21 +-
  5 files changed, 1102 insertions(+), 862 deletions(-)
Signature · 2 lines
Best regards,
Vladimir
#13 lore raw

Hi Vladimir,

Thanks for taking the time to look through this!

I'll try to comment on all your observations below.

Quoted 8 lines · Vladimir Sementsov-Ogievskiy
Quoted 7 lines · Klaus Jensen

From: Klaus Jensen <[email protected]>

This series reimplements flush, dsm, copy, zone reset and format nvm to allow cancellation. I posted an RFC back in March ("hw/block/nvme:
convert ad-hoc aio tracking to aiocb") and I've applied some feedback from Stefan and reimplemented the remaining commands.

Quoted 6 lines
Quoted 4 lines

The basic idea is to define custom AIOCBs for these commands. The custom AIOCB takes care of issuing all the "nested" AIOs one by one instead of blindly sending them off simultaneously without tracking the returned aiocbs.

I'm not familiar with nvme. But intuitively, isn't it less efficient to send mutltiple commands one-by-one? Overall, wouldn't it be slower?

No, you are right, it is of course slower overall.

Quoted 2 lines

In block layer we mostly do opposite transition: instead of doing IO operations one-by-one, run them simultaneously to make a non-empty queue on a device.. Even on one device. This way overall performance is increased.

Of these commands, Copy is the only one that I would consider optimizing like this. But the most obvious use of the Copy command is host-driven garbage collection in the context of zoned namespaces. And I would not consider that operation to be performance critical in terms of latency. All regular I/O commands are "one aiocb" and doesnt need any of this. And we already "parallelize" this heavily.

Quoted 2 lines

If you need to store nested AIOCBs, you may store them in a list for example, and cancel in a loop, keeping simultaneous start for all flushes.. If you send two flushes on two different disks, what's the reason to wait for first flush finish before issuing the second?

Keeping a list of returned aiocbs was my initial approach actually. But when I looked at hw/ide I got the impression that the AIOCB approach was the right one. My first approach involved adding an aiocblist to the core NvmeRequest structure, but I ended up killing that approach because I didnt want to deal with it on the normal I/O path.

But you are absolutely correct that waiting for the first flush to finish is suboptimal.

Quoted 11 lines
Quoted 6 lines

I've kept the RFC since I'm still new to using the block layer like this. I was hoping that Stefan could find some time to look over this - this is a huge series, so I don't expect non-nvme folks to spend a large amount of time on it, but I would really like feedback on my approach in the reimplementation of flush and format.

If I understand your code correctly, you do stat next io operation from call-back of a previous. It works, and it is similar to haw mirror block-job was operating some time ago (still it maintained several in-flight requests simultaneously, but I'm about using _aio_ functions). Still, now mirror doesn't use _aio_ functions like this.

Better approach to call several io functions of block layer one-by-one is creating a coroutine. You may just add a coroutine function, that does all your linear logic as you want, without any callbacks like:

Quoted 9 lines

nvme_co_flush()
{

  for (...) {
     blk_co_flush();

}
}

(and you'll need qemu_coroutine_create() and qemu_coroutine_enter() to start a coroutine).

So, this is definitely a tempting way to implement this. I must admit that I did not consider it like this because I thought this was at the wrong level of abstraction (looked to me like this was something that belonged in block/, not hw/). Again, I referred to the Trim implementation in hw/ide as the source of inspiration on the sequential AIOCB approach.

Quoted 5 lines

Still, I'm not sure that moving from simultaneous issuing several IO commands to sequential is good idea.. And this way you of course can't use blk_aio_canel.. This leads to my last doubt:

One more thing I don't understand after fast look at the series: how cancelation works? It seems to me, that you just call cancel on nested AIOCBs, produced by blk_<io_functions>, but no one of them implement cancel.. I see only four implementations of .cancel_async callback in the whole Qemu code: in iscsi, in ide/core.c, in dma-helpers.c and in thread-pool.c.. Seems no one is related to blk_aio_flush() and other blk_* functions you call in the series. Or, what I miss?

Right now, cancellation is only initiated by the device when a submission queue is deleted. This causes blk_aio_cancel() to be called on each BlockAIOCB (NvmeRequest.aiocb) for outstanding requests. In most cases this BlockAIOCB is a DMAAIOCB from softmmu/dma-helpers.c, which implements .cancel_async. Prior to this patchset, Flush, DSM, Copy and so on, did not have any BlockAIOCB to cancel since we did not keep references to the ongoing AIOs.

The blk_aio_cancel() call is synchronous, but it does call bdrv_aio_cancel_async() which calls the .cancel_async callback if implemented. This means that we can now cancel ongoing DSM or Copy commands while they are processing their individual LBA ranges. So while blk_aio_cancel() subsequently waits for the AIO to complete this may cause them to complete earlier than if they had run to full completion (i.e. if they did not implement .cancel_async).

There are two things I'd like to do subsequent to this patch series:

1. Fix the Abort command to actually do something. Currently the command is a no-op (which is allowed by the spec), but I'd like it to actually cancel the command that the host specified.

2. Make submission queue deletion asynchronous.

The infrastructure provided by this refactor should allow this if I am not mistaken.

Overall, I think this "sequentialization" makes it easier to reason about cancellation, but that might just be me ;)

Quoted 30 lines
Quoted 28 lines

Those commands are special in
that may issue AIOs to multiple namespaces and thus, to multiple block backends. Since this device does not support iothreads, I've opted for simply always returning the main loop aio context, but I wonder if this is acceptable or not. It might be the case that this should contain an assert of some kind, in case someone starts adding iothread support.

Klaus Jensen (11):
hw/nvme: reimplement flush to allow cancellation
hw/nvme: add nvme_block_status_all helper
hw/nvme: reimplement dsm to allow cancellation
hw/nvme: save reftag when generating pi
hw/nvme: remove assert from nvme_get_zone_by_slba hw/nvme: use prinfo directly in nvme_check_prinfo and nvme_dif_check hw/nvme: add dw0/1 to the req completion trace event hw/nvme: reimplement the copy command to allow aio cancellation hw/nvme: reimplement zone reset to allow cancellation hw/nvme: reimplement format nvm to allow cancellation Partially revert "hw/block/nvme: drain namespaces on sq deletion"

 hw/nvme/nvme.h       |   10 +-
 include/block/nvme.h |    8 +
 hw/nvme/ctrl.c       | 1861 ++++++++++++++++++++++++------------------
 hw/nvme/dif.c        |   64 +-
 hw/nvme/trace-events |   21 +-
 5 files changed, 1102 insertions(+), 862 deletions(-)
#14 lore raw
Quoted 27 lines · Klaus Jensen

Hi Vladimir,

Thanks for taking the time to look through this!

I'll try to comment on all your observations below.

Quoted 8 lines · Vladimir Sementsov-Ogievskiy
Quoted 7 lines · Klaus Jensen

From: Klaus Jensen <[email protected]>

This series reimplements flush, dsm, copy, zone reset and format nvm to allow cancellation. I posted an RFC back in March ("hw/block/nvme:
convert ad-hoc aio tracking to aiocb") and I've applied some feedback from Stefan and reimplemented the remaining commands.

Quoted 6 lines
Quoted 4 lines

The basic idea is to define custom AIOCBs for these commands. The custom AIOCB takes care of issuing all the "nested" AIOs one by one instead of blindly sending them off simultaneously without tracking the returned aiocbs.

I'm not familiar with nvme. But intuitively, isn't it less efficient to send mutltiple commands one-by-one? Overall, wouldn't it be slower?

No, you are right, it is of course slower overall.

Quoted 2 lines

In block layer we mostly do opposite transition: instead of doing IO operations one-by-one, run them simultaneously to make a non-empty queue on a device.. Even on one device. This way overall performance is increased.

Quoted 9 lines

Of these commands, Copy is the only one that I would consider optimizing like this. But the most obvious use of the Copy command is host-driven garbage collection in the context of zoned namespaces. And I would not consider that operation to be performance critical in terms of latency. All regular I/O commands are "one aiocb" and doesnt need any of this. And we already "parallelize" this heavily.

Quoted 2 lines

If you need to store nested AIOCBs, you may store them in a list for example, and cancel in a loop, keeping simultaneous start for all flushes.. If you send two flushes on two different disks, what's the reason to wait for first flush finish before issuing the second?

Keeping a list of returned aiocbs was my initial approach actually. But when I looked at hw/ide I got the impression that the AIOCB approach was the right one. My first approach involved adding an aiocblist to the core NvmeRequest structure, but I ended up killing that approach because I didnt want to deal with it on the normal I/O path.

But you are absolutely correct that waiting for the first flush to finish is suboptimal.

Aha. OK, if that's not a performance critical path.

Quoted 18 lines
Quoted 18 lines
Quoted 6 lines

I've kept the RFC since I'm still new to using the block layer like this. I was hoping that Stefan could find some time to look over this - this is a huge series, so I don't expect non-nvme folks to spend a large amount of time on it, but I would really like feedback on my approach in the reimplementation of flush and format.

If I understand your code correctly, you do stat next io operation from call-back of a previous. It works, and it is similar to haw mirror block-job was operating some time ago (still it maintained several in-flight requests simultaneously, but I'm about using _aio_ functions). Still, now mirror doesn't use _aio_ functions like this.

Better approach to call several io functions of block layer one-by-one is creating a coroutine. You may just add a coroutine function, that does all your linear logic as you want, without any callbacks like:

nvme_co_flush()
{
  for (...) {
     blk_co_flush();
  }
}

Quoted 4 lines
Quoted 2 lines

(and you'll need qemu_coroutine_create() and qemu_coroutine_enter() to start a coroutine).

So, this is definitely a tempting way to implement this. I must admit that I did not consider it like this because I thought this was at the wrong level of abstraction (looked to me like this was something that belonged in block/, not hw/). Again, I referred to the Trim implementation in hw/ide as the source of inspiration on the sequential AIOCB approach.

No, I think it's OK from abstraction point of view. Everybody is welcome to use coroutines if it is appropriate and especially for doing sequential IOs :)
Actually, it's just more efficient: the way I propose, you create one coroutine, which does all your logic as you want, when blk_aio_ functions actually create a coroutine under the hood each time (I don't think that it noticeably affects performance, but logic becomes more straightforward)

The only problem is that for this way we don't have cancellation API, so you can't use it for cancellation anyway :(

Quoted 8 lines
Quoted 6 lines

Still, I'm not sure that moving from simultaneous issuing several IO commands to sequential is good idea.. And this way you of course can't use blk_aio_canel.. This leads to my last doubt:

One more thing I don't understand after fast look at the series: how cancelation works? It seems to me, that you just call cancel on nested AIOCBs, produced by blk_<io_functions>, but no one of them implement cancel.. I see only four implementations of .cancel_async callback in the whole Qemu code: in iscsi, in ide/core.c, in dma-helpers.c and in thread-pool.c.. Seems no one is related to blk_aio_flush() and other blk_* functions you call in the series. Or, what I miss?

Right now, cancellation is only initiated by the device when a submission queue is deleted. This causes blk_aio_cancel() to be called on each BlockAIOCB (NvmeRequest.aiocb) for outstanding requests. In most cases this BlockAIOCB is a DMAAIOCB from softmmu/dma-helpers.c, which implements .cancel_async. Prior to this patchset, Flush, DSM, Copy and so on, did not have any BlockAIOCB to cancel since we did not keep references to the ongoing AIOs.

Hmm. Looking at flush for example, I don't see how DMAAIOCB comes.

You do

  iocb->aiocb = blk_aio_flush(ns->blkconf.blk, nvme_flush_ns_cb, iocb);

it calls blk_aio_prwv(), it calls blk_aio_get() with blk_aio_em_aiocb_info, that doesn't implement .cancel_async..

Quoted 5 lines

The blk_aio_cancel() call is synchronous, but it does call bdrv_aio_cancel_async() which calls the .cancel_async callback if implemented. This means that we can now cancel ongoing DSM or Copy commands while they are processing their individual LBA ranges. So while blk_aio_cancel() subsequently waits for the AIO to complete this may cause them to complete earlier than if they had run to full completion (i.e. if they did not implement .cancel_async).

There are two things I'd like to do subsequent to this patch series:

Quoted 8 lines

  1. Fix the Abort command to actually do something. Currently the   command is a no-op (which is allowed by the spec), but I'd like it to   actually cancel the command that the host specified.

  2. Make submission queue deletion asynchronous.

The infrastructure provided by this refactor should allow this if I am not mistaken.

Overall, I think this "sequentialization" makes it easier to reason about cancellation, but that might just be me ;)

I just don't like sequential logic simulated on top of aio-callback async API, which in turn is simulated on top of coroutine-driven sequential API (which is made on top of real async block API (thread-based or linux-aio based, etc)) :) Still I can't suggest now an alternative that supports cancellation. But I still think that cancellation doesn't work for blk_aio_flush and friends either..

Quoted 30 lines
Quoted 30 lines
Quoted 28 lines

Those commands are special in
that may issue AIOs to multiple namespaces and thus, to multiple block backends. Since this device does not support iothreads, I've opted for simply always returning the main loop aio context, but I wonder if this is acceptable or not. It might be the case that this should contain an assert of some kind, in case someone starts adding iothread support.

Klaus Jensen (11):
  hw/nvme: reimplement flush to allow cancellation
  hw/nvme: add nvme_block_status_all helper
  hw/nvme: reimplement dsm to allow cancellation
  hw/nvme: save reftag when generating pi
  hw/nvme: remove assert from nvme_get_zone_by_slba hw/nvme: use prinfo directly in nvme_check_prinfo and nvme_dif_check hw/nvme: add dw0/1 to the req completion trace event hw/nvme: reimplement the copy command to allow aio cancellation hw/nvme: reimplement zone reset to allow cancellation hw/nvme: reimplement format nvm to allow cancellation Partially revert "hw/block/nvme: drain namespaces on sq deletion"

 hw/nvme/nvme.h       |   10 +-
 include/block/nvme.h |    8 +
 hw/nvme/ctrl.c       | 1861 ++++++++++++++++++++++++------------------ hw/nvme/dif.c        |   64 +-
 hw/nvme/trace-events |   21 +-

 5 files changed, 1102 insertions(+), 862 deletions(-)
Signature · 2 lines
Best regards,
Vladimir
#15 lore raw
Quoted 24 lines · Vladimir Sementsov-Ogievskiy
Quoted 23 lines · Klaus Jensen
Quoted 21 lines · Vladimir Sementsov-Ogievskiy
Quoted 6 lines · Klaus Jensen

I've kept the RFC since I'm still new to using the block layer like this. I was hoping that Stefan could find some time to look over this - this is a huge series, so I don't expect non-nvme folks to spend a large amount of time on it, but I would really like feedback on my approach in the reimplementation of flush and format.

If I understand your code correctly, you do stat next io operation from call-back of a previous. It works, and it is similar to haw mirror block-job was operating some time ago (still it maintained several in-flight requests simultaneously, but I'm about using _aio_ functions). Still, now mirror doesn't use _aio_ functions like this.

Better approach to call several io functions of block layer one-by-one is creating a coroutine. You may just add a coroutine function, that does all your linear logic as you want, without any callbacks like:

nvme_co_flush()
{
  for (...) {
     blk_co_flush();
  }
}

(and you'll need qemu_coroutine_create() and qemu_coroutine_enter() to start a coroutine).

Quoted 7 lines
Quoted 1 line

So, this is definitely a tempting way to implement this. I must admit that I did not consider it like this because I thought this was at the wrong level of abstraction (looked to me like this was something that belonged in block/, not hw/). Again, I referred to the Trim implementation in hw/ide as the source of inspiration on the sequential AIOCB approach.

No, I think it's OK from abstraction point of view. Everybody is welcome to use coroutines if it is appropriate and especially for doing sequential IOs :)
Actually, it's just more efficient: the way I propose, you create one coroutine, which does all your logic as you want, when blk_aio_ functions actually create a coroutine under the hood each time (I don't think that it noticeably affects performance, but logic becomes more straightforward)

The only problem is that for this way we don't have cancellation API, so you can't use it for cancellation anyway :(

Yeah, I'm not really feeling up for adding that :P

Quoted 9 lines
Quoted 8 lines
Quoted 6 lines

Still, I'm not sure that moving from simultaneous issuing several IO commands to sequential is good idea.. And this way you of course can't use blk_aio_canel.. This leads to my last doubt:

One more thing I don't understand after fast look at the series: how cancelation works? It seems to me, that you just call cancel on nested AIOCBs, produced by blk_<io_functions>, but no one of them implement cancel.. I see only four implementations of .cancel_async callback in the whole Qemu code: in iscsi, in ide/core.c, in dma-helpers.c and in thread-pool.c.. Seems no one is related to blk_aio_flush() and other blk_* functions you call in the series. Or, what I miss?

Right now, cancellation is only initiated by the device when a submission queue is deleted. This causes blk_aio_cancel() to be called on each BlockAIOCB (NvmeRequest.aiocb) for outstanding requests. In most cases this BlockAIOCB is a DMAAIOCB from softmmu/dma-helpers.c, which implements .cancel_async. Prior to this patchset, Flush, DSM, Copy and so on, did not have any BlockAIOCB to cancel since we did not keep references to the ongoing AIOs.

Quoted 8 lines

Hmm. Looking at flush for example, I don't see how DMAAIOCB comes.

You do

iocb->aiocb = blk_aio_flush(ns->blkconf.blk, nvme_flush_ns_cb, iocb);

it calls blk_aio_prwv(), it calls blk_aio_get() with blk_aio_em_aiocb_info, that doesn't implement .cancel_async..

I meant that most I/O in the regular path (read/write) are using the dma helpers (since they do DMA). We might use the blk_aio_p{read,write} directly when we read from/write to memory on the device (the controller memory buffer), but it is not the common case.

You are correct that BlkAioEmAIOCB does not implement cancel, but the "wrapper" (NvmeFlushAIOCB) *does*. This means that, from the NVMe controller perspective, we can cancel the flush in between (un-cancellable blk_aio_flush-initiated) flushes to multiple namespaces.

Quoted 7 lines
Quoted 7 lines

The blk_aio_cancel() call is synchronous, but it does call bdrv_aio_cancel_async() which calls the .cancel_async callback if implemented. This means that we can now cancel ongoing DSM or Copy commands while they are processing their individual LBA ranges. So while blk_aio_cancel() subsequently waits for the AIO to complete this may cause them to complete earlier than if they had run to full completion (i.e. if they did not implement .cancel_async).

There are two things I'd like to do subsequent to this patch series:

  1. Fix the Abort command to actually do something. Currently the command is a no-op (which is allowed by the spec), but I'd like it to actually cancel the command that the host specified.

Quoted 8 lines
Quoted 6 lines

  2. Make submission queue deletion asynchronous.

The infrastructure provided by this refactor should allow this if I am not mistaken.

Overall, I think this "sequentialization" makes it easier to reason about cancellation, but that might just be me ;)

I just don't like sequential logic simulated on top of aio-callback async API, which in turn is simulated on top of coroutine-driven sequential API (which is made on top of real async block API (thread-based or linux-aio based, etc)) :)

Ha! Yes, we are not exactly improving on that layering here ;)

Quoted 2 lines

Still I can't suggest now an alternative that supports cancellation. But I still think that cancellation doesn't work for blk_aio_flush and friends either..

The aiocb from blk_aio_flush is considered "un-cancellable" I guess (by design from the block layer), but the NVMe command "Flush" is cancellable from the perspective of the NVMe controller. Or at least, that's what I am intending to do here.

#16 lore raw
Quoted 48 lines · Klaus Jensen
Quoted 24 lines · Vladimir Sementsov-Ogievskiy
Quoted 23 lines · Klaus Jensen
Quoted 21 lines · Vladimir Sementsov-Ogievskiy
Quoted 6 lines · Klaus Jensen

I've kept the RFC since I'm still new to using the block layer like this. I was hoping that Stefan could find some time to look over this - this is a huge series, so I don't expect non-nvme folks to spend a large amount of time on it, but I would really like feedback on my approach in the reimplementation of flush and format.

If I understand your code correctly, you do stat next io operation from call-back of a previous. It works, and it is similar to haw mirror block-job was operating some time ago (still it maintained several in-flight requests simultaneously, but I'm about using _aio_ functions). Still, now mirror doesn't use _aio_ functions like this.

Better approach to call several io functions of block layer one-by-one is creating a coroutine. You may just add a coroutine function, that does all your linear logic as you want, without any callbacks like:

nvme_co_flush()
{
  for (...) {
     blk_co_flush();
  }
}

(and you'll need qemu_coroutine_create() and qemu_coroutine_enter() to start a coroutine).

Quoted 7 lines
Quoted 1 line

So, this is definitely a tempting way to implement this. I must admit that I did not consider it like this because I thought this was at the wrong level of abstraction (looked to me like this was something that belonged in block/, not hw/). Again, I referred to the Trim implementation in hw/ide as the source of inspiration on the sequential AIOCB approach.

No, I think it's OK from abstraction point of view. Everybody is welcome to use coroutines if it is appropriate and especially for doing sequential IOs :)
Actually, it's just more efficient: the way I propose, you create one coroutine, which does all your logic as you want, when blk_aio_ functions actually create a coroutine under the hood each time (I don't think that it noticeably affects performance, but logic becomes more straightforward)

The only problem is that for this way we don't have cancellation API, so you can't use it for cancellation anyway :(

Yeah, I'm not really feeling up for adding that :P

Quoted 13 lines
Quoted 8 lines
Quoted 6 lines

Still, I'm not sure that moving from simultaneous issuing several IO commands to sequential is good idea.. And this way you of course can't use blk_aio_canel.. This leads to my last doubt:

One more thing I don't understand after fast look at the series: how cancelation works? It seems to me, that you just call cancel on nested AIOCBs, produced by blk_<io_functions>, but no one of them implement cancel.. I see only four implementations of .cancel_async callback in the whole Qemu code: in iscsi, in ide/core.c, in dma-helpers.c and in thread-pool.c.. Seems no one is related to blk_aio_flush() and other blk_* functions you call in the series. Or, what I miss?

Right now, cancellation is only initiated by the device when a submission queue is deleted. This causes blk_aio_cancel() to be called on each BlockAIOCB (NvmeRequest.aiocb) for outstanding requests. In most cases this BlockAIOCB is a DMAAIOCB from softmmu/dma-helpers.c, which implements .cancel_async. Prior to this patchset, Flush, DSM, Copy and so on, did not have any BlockAIOCB to cancel since we did not keep references to the ongoing AIOs.

Hmm. Looking at flush for example, I don't see how DMAAIOCB comes.

You do

Quoted 8 lines
Quoted 4 lines

 iocb->aiocb = blk_aio_flush(ns->blkconf.blk, nvme_flush_ns_cb, iocb);

it calls blk_aio_prwv(), it calls blk_aio_get() with blk_aio_em_aiocb_info, that doesn't implement .cancel_async..

I meant that most I/O in the regular path (read/write) are using the dma helpers (since they do DMA). We might use the blk_aio_p{read,write} directly when we read from/write to memory on the device (the controller memory buffer), but it is not the common case.

You are correct that BlkAioEmAIOCB does not implement cancel, but the "wrapper" (NvmeFlushAIOCB) *does*. This means that, from the NVMe controller perspective, we can cancel the flush in between (un-cancellable blk_aio_flush-initiated) flushes to multiple namespaces.

Hm. But it will work the same way if you just not implement .cancel_async in nvme_flush_aiocb_info.

Quoted 23 lines
Quoted 8 lines
Quoted 8 lines

The blk_aio_cancel() call is synchronous, but it does call bdrv_aio_cancel_async() which calls the .cancel_async callback if implemented. This means that we can now cancel ongoing DSM or Copy commands while they are processing their individual LBA ranges. So while blk_aio_cancel() subsequently waits for the AIO to complete this may cause them to complete earlier than if they had run to full completion (i.e. if they did not implement .cancel_async).

There are two things I'd like to do subsequent to this patch series:

  1. Fix the Abort command to actually do something. Currently the command is a no-op (which is allowed by the spec), but I'd like it to actually cancel the command that the host specified.

Quoted 8 lines
Quoted 6 lines

  2. Make submission queue deletion asynchronous.

The infrastructure provided by this refactor should allow this if I am not mistaken.

Overall, I think this "sequentialization" makes it easier to reason about cancellation, but that might just be me ;)

I just don't like sequential logic simulated on top of aio-callback async API, which in turn is simulated on top of coroutine-driven sequential API (which is made on top of real async block API (thread-based or linux-aio based, etc)) :)

Ha! Yes, we are not exactly improving on that layering here ;)

Quoted 2 lines

Still I can't suggest now an alternative that supports cancellation. But I still think that cancellation doesn't work for blk_aio_flush and friends either..

The aiocb from blk_aio_flush is considered "un-cancellable" I guess (by design from the block layer), but the NVMe command "Flush" is cancellable from the perspective of the NVMe controller. Or at least, that's what I am intending to do here.

Signature · 2 lines
Best regards,
Vladimir
#17 lore raw
Quoted 51 lines · Vladimir Sementsov-Ogievskiy
Quoted 50 lines · Klaus Jensen
Quoted 24 lines · Vladimir Sementsov-Ogievskiy
Quoted 23 lines · Klaus Jensen
Quoted 21 lines · Vladimir Sementsov-Ogievskiy
Quoted 6 lines · Klaus Jensen

I've kept the RFC since I'm still new to using the block layer like this. I was hoping that Stefan could find some time to look over this - this is a huge series, so I don't expect non-nvme folks to spend a large amount of time on it, but I would really like feedback on my approach in the reimplementation of flush and format.

If I understand your code correctly, you do stat next io operation from call-back of a previous. It works, and it is similar to haw mirror block-job was operating some time ago (still it maintained several in-flight requests simultaneously, but I'm about using _aio_ functions). Still, now mirror doesn't use _aio_ functions like this.

Better approach to call several io functions of block layer one-by-one is creating a coroutine. You may just add a coroutine function, that does all your linear logic as you want, without any callbacks like:

nvme_co_flush()
{
  for (...) {
     blk_co_flush();
  }
}

(and you'll need qemu_coroutine_create() and qemu_coroutine_enter() to start a coroutine).

Quoted 7 lines
Quoted 1 line

So, this is definitely a tempting way to implement this. I must admit that I did not consider it like this because I thought this was at the wrong level of abstraction (looked to me like this was something that belonged in block/, not hw/). Again, I referred to the Trim implementation in hw/ide as the source of inspiration on the sequential AIOCB approach.

No, I think it's OK from abstraction point of view. Everybody is welcome to use coroutines if it is appropriate and especially for doing sequential IOs :)
Actually, it's just more efficient: the way I propose, you create one coroutine, which does all your logic as you want, when blk_aio_ functions actually create a coroutine under the hood each time (I don't think that it noticeably affects performance, but logic becomes more straightforward)

The only problem is that for this way we don't have cancellation API, so you can't use it for cancellation anyway :(

Yeah, I'm not really feeling up for adding that :P

Quoted 15 lines
Quoted 8 lines
Quoted 6 lines

Still, I'm not sure that moving from simultaneous issuing several IO commands to sequential is good idea.. And this way you of course can't use blk_aio_canel.. This leads to my last doubt:

One more thing I don't understand after fast look at the series: how cancelation works? It seems to me, that you just call cancel on nested AIOCBs, produced by blk_<io_functions>, but no one of them implement cancel.. I see only four implementations of .cancel_async callback in the whole Qemu code: in iscsi, in ide/core.c, in dma-helpers.c and in thread-pool.c.. Seems no one is related to blk_aio_flush() and other blk_* functions you call in the series. Or, what I miss?

Right now, cancellation is only initiated by the device when a submission queue is deleted. This causes blk_aio_cancel() to be called on each BlockAIOCB (NvmeRequest.aiocb) for outstanding requests. In most cases this BlockAIOCB is a DMAAIOCB from softmmu/dma-helpers.c, which implements .cancel_async. Prior to this patchset, Flush, DSM, Copy and so on, did not have any BlockAIOCB to cancel since we did not keep references to the ongoing AIOs.

Hmm. Looking at flush for example, I don't see how DMAAIOCB comes.

You do

 iocb->aiocb = blk_aio_flush(ns->blkconf.blk, nvme_flush_ns_cb, iocb);

Quoted 9 lines
Quoted 6 lines
Quoted 2 lines

it calls blk_aio_prwv(), it calls blk_aio_get() with blk_aio_em_aiocb_info, that doesn't implement .cancel_async..

I meant that most I/O in the regular path (read/write) are using the dma helpers (since they do DMA). We might use the blk_aio_p{read,write} directly when we read from/write to memory on the device (the controller memory buffer), but it is not the common case.

You are correct that BlkAioEmAIOCB does not implement cancel, but the "wrapper" (NvmeFlushAIOCB) *does*. This means that, from the NVMe controller perspective, we can cancel the flush in between (un-cancellable blk_aio_flush-initiated) flushes to multiple namespaces.

Hm. But it will work the same way if you just not implement .cancel_async in nvme_flush_aiocb_info.

Oh. Crap, I see. I screwed this up in flush.

blk_aio_cancel_async(iocb->aiocb) will be a no-op and ret will never be -ECANCELED. I think I do this correctly in the other reimplementations (explicitly set iocb->ret), but not here in flush.

#18 lore raw
In reply to Klaus Jensen
Quoted 12 lines · Klaus Jensen

From: Klaus Jensen <[email protected]>

This series reimplements flush, dsm, copy, zone reset and format nvm to allow cancellation. I posted an RFC back in March ("hw/block/nvme:
convert ad-hoc aio tracking to aiocb") and I've applied some feedback from Stefan and reimplemented the remaining commands.

The basic idea is to define custom AIOCBs for these commands. The custom AIOCB takes care of issuing all the "nested" AIOs one by one instead of blindly sending them off simultaneously without tracking the returned aiocbs.

Quoted 10 lines

I've kept the RFC since I'm still new to using the block layer like this. I was hoping that Stefan could find some time to look over this - this is a huge series, so I don't expect non-nvme folks to spend a large amount of time on it, but I would really like feedback on my approach in the reimplementation of flush and format. Those commands are special in that may issue AIOs to multiple namespaces and thus, to multiple block backends. Since this device does not support iothreads, I've opted for simply always returning the main loop aio context, but I wonder if this is acceptable or not. It might be the case that this should contain an assert of some kind, in case someone starts adding iothread support.

This approach looks fine to me. Vladimir mentioned coroutines, which have simpler code for sequential I/O, but don't support cancellation. Since cancellation is the point of this series I think sticking to the aio approach makes sense.

Regarding coroutine cancellation, it's a hard to add since there is already a lot of coroutine code that's not written with cancellation in mind.

I think I would approach it by adding a .cancel_cb() field to Coroutine that does nothing by default (because existing code doesn't support cancellation and we must wait for the operation to complete). Cases the do support cancellation would install a .cancel_cb() across yield that causes the operation that coroutine is waiting on to complete early.

An alternative approach is to re-enter the coroutine, but this requires all yield points in QEMU to check for cancellation. I don't think this is practical because converting all the code would be hard.

Anyway, the aio approach looks fine.

Stefan