hw/nvme: fix pin-based interrupt behavior (again)
PATCH 1/1 hw/nvme: fix pin-based interrupt behavior (again)
From: Klaus Jensen <[email protected]>
Jakub noticed[1] that, when using pin-based interrupts, the device will unconditionally deasssert when any CQEs are acknowledged. However, the pin should not be deasserted if other completion queues still holds unacknowledged CQEs.
The bug is an artifact of commit ca247d35098d ("hw/block/nvme: fix pin-based interrupt behavior") which fixed one bug but introduced another. This is the third time someone tries to fix pin-based interrupts (see commit 5e9aa92eb1a5 ("hw/block: Fix pin-based interrupt behaviour of NVMe"))...
Third time's the charm, so fix it, again, by keeping track of how many CQs have unacknowledged CQEs and only deassert when all are cleared.
[1]: <[email protected]>
--- hw/nvme/nvme.h | 1 + hw/nvme/ctrl.c | 15 ++++++++++++++- 2 files changed, 15 insertions(+), 1 deletion(-)
hw/nvme/nvme.h +1 -0
@@ -405,6 +405,7 @@ typedef struct NvmeCtrl {uint32_t max_q_ents;uint8_t outstanding_aers;uint32_t irq_status; +int cq_pending;uint64_t host_timestamp;/* Timestamp sent by the host */ uint64_t timestamp_set_qemu_clock_ms;/* QEMU clock time */ uint64_t starttime_ms;
hw/nvme/ctrl.c +14 -1
@@ -473,7 +473,9 @@ static void nvme_irq_deassert(NvmeCtrl *n, NvmeCQueue *cq)return ; }else {assert (cq->vector <32 ); - n->irq_status &= ~(1 << cq->vector); +if (!n->cq_pending) { + n->irq_status &= ~(1 << cq->vector); + }nvme_irq_check (n); } }
@@ -1258,6 +1260,7 @@ static void nvme_post_cqes(void *opaque)NvmeCQueue *cq = opaque;NvmeCtrl *n = cq->ctrl;NvmeRequest *req, *next; +bool pending = cq->head != cq->tail;int ret; QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
@@ -1287,6 +1290,10 @@ static void nvme_post_cqes(void *opaque)QTAILQ_INSERT_TAIL (&sq->req_list, req, entry); }if (cq->tail != cq->head) { +if (!pending) { + n->cq_pending++; + } +nvme_irq_assert (n, cq); } }
@@ -4099,6 +4106,11 @@ static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeRequest *req)trace_pci_nvme_err_invalid_del_cq_notempty (qid);return NVME_INVALID_QUEUE_DEL ; } + +if (cq->tail != cq->head) { + n->cq_pending--; + } +nvme_irq_deassert (n, cq);trace_pci_nvme_del_cq (qid);nvme_free_cq (cq, n);
@@ -5758,6 +5770,7 @@ static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val) }if (cq->tail == cq->head) { + n->cq_pending--;nvme_irq_deassert (n, cq); } }else {
Signature · 1 line
2.32.0
Quoted 95 lines · Klaus Jensen
From: Klaus Jensen <[email protected]>
Jakub noticed[1] that, when using pin-based interrupts, the device will unconditionally deasssert when any CQEs are acknowledged. However, the pin should not be deasserted if other completion queues still holds unacknowledged CQEs.
The bug is an artifact of commit
ca247d35098d("hw/block/nvme: fix pin-based interrupt behavior") which fixed one bug but introduced another. This is the third time someone tries to fix pin-based interrupts (see commit5e9aa92eb1a5("hw/block: Fix pin-based interrupt behaviour of NVMe"))...Third time's the charm, so fix it, again, by keeping track of how many CQs have unacknowledged CQEs and only deassert when all are cleared.
[1]: <[email protected]>
Fixes ca247d35098d ("hw/block/nvme: fix pin-based interrupt behavior")Reported-by Jakub Jermář <[email protected]>Signed-off-by Klaus Jensen <[email protected]>---
hw/nvme/nvme.h| 1 +hw/nvme/ctrl.c| 15 ++++++++++++++-2 files changed, 15 insertions(+), 1 deletion(-)hw/nvme/nvme.h +1
-0@@ -405,6 +405,7 @@ typedef struct NvmeCtrl {uint32_t max_q_ents;uint8_t outstanding_aers;uint32_t irq_status; +int cq_pending;uint64_t host_timestamp;/* Timestamp sent by the host */ uint64_t timestamp_set_qemu_clock_ms;/* QEMU clock time */ uint64_t starttime_ms;hw/nvme/ctrl.c +14
-1@@ -473,7 +473,9 @@ static void nvme_irq_deassert(NvmeCtrl *n, NvmeCQueue *cq)return ; }else {assert (cq->vector <32 ); - n->irq_status &= ~(1 << cq->vector); +if (!n->cq_pending) { + n->irq_status &= ~(1 << cq->vector); + }nvme_irq_check (n); } }@@ -1258,6 +1260,7 @@ static void nvme_post_cqes(void *opaque)NvmeCQueue *cq = opaque;NvmeCtrl *n = cq->ctrl;NvmeRequest *req, *next; +bool pending = cq->head != cq->tail;int ret; QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {@@ -1287,6 +1290,10 @@ static void nvme_post_cqes(void *opaque)QTAILQ_INSERT_TAIL (&sq->req_list, req, entry); }if (cq->tail != cq->head) { +if (!pending) { + n->cq_pending++; + } +nvme_irq_assert (n, cq); } }@@ -4099,6 +4106,11 @@ static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeRequest *req)trace_pci_nvme_err_invalid_del_cq_notempty (qid);return NVME_INVALID_QUEUE_DEL ; } + +if (cq->tail != cq->head) { + n->cq_pending--; + } +nvme_irq_deassert (n, cq);trace_pci_nvme_del_cq (qid);nvme_free_cq (cq, n);@@ -5758,6 +5770,7 @@ static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val) }if (cq->tail == cq->head) { + n->cq_pending--;nvme_irq_deassert (n, cq); } }else {Signature · 1 line
2.32.0
Jakub, can you test this in your environment?
Quoted 91 lines · Klaus Jensen
Quoted 90 lines · Klaus Jensen
From: Klaus Jensen <[email protected]>
Jakub noticed[1] that, when using pin-based interrupts, the device will unconditionally deasssert when any CQEs are acknowledged. However, the pin should not be deasserted if other completion queues still holds unacknowledged CQEs.
The bug is an artifact of commit
ca247d35098d("hw/block/nvme: fix pin-based interrupt behavior") which fixed one bug but introduced another. This is the third time someone tries to fix pin-based interrupts (see commit5e9aa92eb1a5("hw/block: Fix pin-based interrupt behaviour of NVMe"))...Third time's the charm, so fix it, again, by keeping track of how many CQs have unacknowledged CQEs and only deassert when all are cleared.
[1]: <[email protected]>
Fixes ca247d35098d ("hw/block/nvme: fix pin-based interrupt behavior")Reported-by Jakub Jermář <[email protected]>Signed-off-by Klaus Jensen <[email protected]>---
hw/nvme/nvme.h| 1 +hw/nvme/ctrl.c| 15 ++++++++++++++-2 files changed, 15 insertions(+), 1 deletion(-)hw/nvme/nvme.h +1
-0@@ -405,6 +405,7 @@ typedef struct NvmeCtrl { uint32_t max_q_ents; uint8_t outstanding_aers; uint32_t irq_status; + int cq_pending;uint64_t host_timestamp; /* Timestamp sent by the host */ uint64_t timestamp_set_qemu_clock_ms; /* QEMU clock time */ uint64_t
starttime_ms;hw/nvme/ctrl.c +3
-1@@ -473,7 +473,9 @@ static void nvme_irq_deassert(NvmeCtrl *n, NvmeCQueue *cq) return; } else { assert(cq->vector < 32); - n->irq_status &= ~(1 << cq->vector); + if (!n->cq_pending) { + n->irq_status &= ~(1 << cq->vector); + } nvme_irq_check(n); }}
hw/nvme/ctrl.c +10
-0@@ -1258,6 +1260,7 @@ static void nvme_post_cqes(void *opaque) NvmeCQueue *cq = opaque; NvmeCtrl *n = cq->ctrl; NvmeRequest *req, *next; + bool pending = cq->head != cq->tail; int ret; QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {@@ -1287,6 +1290,10 @@ static void nvme_post_cqes(void *opaque) QTAILQ_INSERT_TAIL(&sq->req_list, req, entry); } if (cq->tail != cq->head) { + if (!pending) { + n->cq_pending++; + } + nvme_irq_assert(n, cq); } }@@ -4099,6 +4106,11 @@ static uint16_t nvme_del_cq(NvmeCtrl *n, NvmeRequest *req) trace_pci_nvme_err_invalid_del_cq_notempty(qid); return NVME_INVALID_QUEUE_DEL; } + + if (cq->tail != cq->head) { + n->cq_pending--; + } + nvme_irq_deassert(n, cq); trace_pci_nvme_del_cq(qid);
nvme_free_cq(cq, n);hw/nvme/ctrl.c +0
-0@@ -5758,6 +5770,7 @@ static void nvme_process_db(NvmeCtrl *n, hwaddr addr, int val) }
Quoted 10 lines
Quoted 8 lines
hw/nvme/ctrl.c +1
-0if (cq->tail == cq->head) { + n->cq_pending--; nvme_irq_deassert(n, cq); }} else {
Signature · 1 line
2.32.0Jakub, can you test this in your environment?
Yep, still tests fine.
Jakub
hw/nvme/ctrl.c static void nvme_post_cqes(void *opaque) · line 1295
Quoted 4 lines · Klaus Jensen
if (cq->tail != cq->head) {hw/nvme/ctrl.c +3
-0+if (!pending) { + n->cq_pending++; + }
You should check cq->irq_enabled before incrementing cq_pending. You don't want to leave the irq asserted when polled queues have outsanding CQEs.
Quoted 9 lines · Keith Busch
hw/nvme/ctrl.c static void nvme_post_cqes(void *opaque) · line 1295
Quoted 4 lines · Klaus Jensen
if (cq->tail != cq->head) {hw/nvme/ctrl.c +3
-0+if (!pending) { + n->cq_pending++; + }You should check
cq->irq_enabledbefore incrementingcq_pending. You don't want to leave the irq asserted when polled queues have outsanding CQEs.
Good catch. Thanks!