* Re: [PATCH v1 2/3] megaraid_sas: iouring iopoll support
2020-10-15 13:37 [PATCH v1 2/3] megaraid_sas: iouring iopoll support Kashyap Desai
@ 2020-10-15 20:44 ` kernel test robot
2020-11-13 10:48 ` Kashyap Desai
` (2 subsequent siblings)
3 siblings, 0 replies; 8+ messages in thread
From: kernel test robot @ 2020-10-15 20:44 UTC (permalink / raw)
To: Kashyap Desai, linux-scsi
Cc: kbuild-all, clang-built-linux, Kashyap Desai, sumit.saxena,
chandrakanth.patil, linux-block
[-- Attachment #1: Type: text/plain, Size: 2985 bytes --]
Hi Kashyap,
I love your patch! Perhaps something to improve:
[auto build test WARNING on next-20201015]
[cannot apply to scsi/for-next mkp-scsi/for-next v5.9 v5.9-rc8 v5.9-rc7 v5.9]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]
url: https://github.com/0day-ci/linux/commits/Kashyap-Desai/io_uring-iopoll-in-scsi-layer/20201015-213946
base: 03d992bd2de6c7f2c9bbd4260ff104c0926ce3ff
config: riscv-randconfig-r026-20201015 (attached as .config)
compiler: clang version 12.0.0 (https://github.com/llvm/llvm-project e7b4feea8e1bf520b34ad8c116abab6677344b74)
reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
chmod +x ~/bin/make.cross
# install riscv cross compiling tool for clang build
# apt-get install binutils-riscv64-linux-gnu
# https://github.com/0day-ci/linux/commit/011c9682f4ab6651d3b56e7bb819b6605e1f8d32
git remote add linux-review https://github.com/0day-ci/linux
git fetch --no-tags linux-review Kashyap-Desai/io_uring-iopoll-in-scsi-layer/20201015-213946
git checkout 011c9682f4ab6651d3b56e7bb819b6605e1f8d32
# save the attached .config to linux build tree
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross ARCH=riscv
If you fix the issue, kindly add following tag as appropriate
Reported-by: kernel test robot <lkp@intel.com>
All warnings (new ones prefixed by >>):
>> drivers/scsi/megaraid/megaraid_sas_fusion.c:3652:5: warning: no previous prototype for function 'megasas_blk_mq_poll' [-Wmissing-prototypes]
int megasas_blk_mq_poll(struct Scsi_Host *shost, unsigned int queue_num)
^
drivers/scsi/megaraid/megaraid_sas_fusion.c:3652:1: note: declare 'static' if the function is not intended to be used outside of this translation unit
int megasas_blk_mq_poll(struct Scsi_Host *shost, unsigned int queue_num)
^
static
1 warning generated.
vim +/megasas_blk_mq_poll +3652 drivers/scsi/megaraid/megaraid_sas_fusion.c
3651
> 3652 int megasas_blk_mq_poll(struct Scsi_Host *shost, unsigned int queue_num)
3653 {
3654
3655 struct megasas_instance *instance;
3656 int num_entries = 0;
3657 struct fusion_context *fusion;
3658
3659 instance = (struct megasas_instance *)shost->hostdata;
3660
3661 fusion = instance->ctrl_context;
3662
3663 queue_num = queue_num + instance->low_latency_index_start;
3664
3665 if (!atomic_add_unless(&fusion->busy_mq_poll[queue_num], 1, 1))
3666 return 0;
3667
3668 num_entries = complete_cmd_fusion(instance, queue_num, NULL);
3669 atomic_dec(&fusion->busy_mq_poll[queue_num]);
3670
3671 return num_entries;
3672 }
3673
---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all@lists.01.org
[-- Attachment #2: .config.gz --]
[-- Type: application/gzip, Size: 28577 bytes --]
^ permalink raw reply [flat|nested] 8+ messages in thread
* RE: [PATCH v1 2/3] megaraid_sas: iouring iopoll support
2020-10-15 13:37 [PATCH v1 2/3] megaraid_sas: iouring iopoll support Kashyap Desai
2020-10-15 20:44 ` kernel test robot
@ 2020-11-13 10:48 ` Kashyap Desai
2020-11-13 11:56 ` chenxiang (M)
2020-11-20 7:14 ` Hannes Reinecke
3 siblings, 0 replies; 8+ messages in thread
From: Kashyap Desai @ 2020-11-13 10:48 UTC (permalink / raw)
To: linux-scsi; +Cc: Sumit Saxena, Chandrakanth Patil, linux-block
[-- Attachment #1: Type: text/plain, Size: 1369 bytes --]
> Add support of iouring iopoll interface. This feature requires shared
hosttag
> support in kernel and driver.
>
> Driver will work in non-IRQ mode = There will not be any msix vector
> associated for poll_queues and h/w can still work in this mode.
> MegaRaid h/w is single submission queue and multiple reply queue, but
using
> shared host tagset support it will enable simulated multiple hw queue.
>
> Driver allocates some extra reply queues and it will be marked as
poll_queue.
> These poll_queues will not have associated msix vectors. All the IO
> completion on this queue will be done from IOPOLL interface.
>
> megaraid_sas driver having 8 poll_queues and using io_uring hiprio=1
> settings, It can reach 3.2M IOPs and there is zero interrupt generated
by h/w.
>
> This feature can be enabled using module parameter poll_queues.
>
> Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
> Cc: sumit.saxena@broadcom.com
> Cc: chandrakanth.patil@broadcom.com
> Cc: linux-block@vger.kernel.org
>
> ---
> drivers/scsi/megaraid/megaraid_sas.h | 2 +
> drivers/scsi/megaraid/megaraid_sas_base.c | 90 ++++++++++++++++++---
> drivers/scsi/megaraid/megaraid_sas_fusion.c | 43 +++++++++-
> drivers/scsi/megaraid/megaraid_sas_fusion.h | 3 +
> 4 files changed, 127 insertions(+), 11 deletions(-)
>
Is there any feedback/comment on this patch series ?
[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4169 bytes --]
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH v1 2/3] megaraid_sas: iouring iopoll support
2020-10-15 13:37 [PATCH v1 2/3] megaraid_sas: iouring iopoll support Kashyap Desai
2020-10-15 20:44 ` kernel test robot
2020-11-13 10:48 ` Kashyap Desai
@ 2020-11-13 11:56 ` chenxiang (M)
2020-11-20 7:14 ` Hannes Reinecke
3 siblings, 0 replies; 8+ messages in thread
From: chenxiang (M) @ 2020-11-13 11:56 UTC (permalink / raw)
To: John Garry; +Cc: Kashyap Desai, linux-scsi, linux-block
@John Garry,
If want to support iouring iopoll support for hisi_sas driver, mabe it
also can be done as megaraid does,
using some queues for poll_queues, and reserve some queues for those
internal IOs from SCSI layer.
在 2020/10/15 21:37, Kashyap Desai 写道:
> Add support of iouring iopoll interface. This feature requires shared
> hosttag support in kernel and driver.
>
> Driver will work in non-IRQ mode = There will not be any msix vector
> associated for poll_queues and h/w can still work in this mode.
> MegaRaid h/w is single submission queue and multiple reply queue, but
> using shared host tagset support it will enable simulated multiple hw queue.
>
> Driver allocates some extra reply queues and it will be marked as poll_queue.
> These poll_queues will not have associated msix vectors. All the IO
> completion on this queue will be done from IOPOLL interface.
>
> megaraid_sas driver having 8 poll_queues and using io_uring hiprio=1 settings,
> It can reach 3.2M IOPs and there is zero interrupt generated by h/w.
>
> This feature can be enabled using module parameter poll_queues.
>
> Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
> Cc: sumit.saxena@broadcom.com
> Cc: chandrakanth.patil@broadcom.com
> Cc: linux-block@vger.kernel.org
>
> ---
> drivers/scsi/megaraid/megaraid_sas.h | 2 +
> drivers/scsi/megaraid/megaraid_sas_base.c | 90 ++++++++++++++++++---
> drivers/scsi/megaraid/megaraid_sas_fusion.c | 43 +++++++++-
> drivers/scsi/megaraid/megaraid_sas_fusion.h | 3 +
> 4 files changed, 127 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/scsi/megaraid/megaraid_sas.h b/drivers/scsi/megaraid/megaraid_sas.h
> index 5e4137f10e0e..4bf6fde49565 100644
> --- a/drivers/scsi/megaraid/megaraid_sas.h
> +++ b/drivers/scsi/megaraid/megaraid_sas.h
> @@ -2212,6 +2212,7 @@ struct megasas_irq_context {
> struct irq_poll irqpoll;
> bool irq_poll_scheduled;
> bool irq_line_enable;
> + atomic_t in_used;
> };
>
> struct MR_DRV_SYSTEM_INFO {
> @@ -2446,6 +2447,7 @@ struct megasas_instance {
> bool support_pci_lane_margining;
> u8 low_latency_index_start;
> int perf_mode;
> + int iopoll_q_count;
> };
>
> struct MR_LD_VF_MAP {
> diff --git a/drivers/scsi/megaraid/megaraid_sas_base.c b/drivers/scsi/megaraid/megaraid_sas_base.c
> index 020270ce790b..e2c307523a8a 100644
> --- a/drivers/scsi/megaraid/megaraid_sas_base.c
> +++ b/drivers/scsi/megaraid/megaraid_sas_base.c
> @@ -114,6 +114,15 @@ unsigned int enable_sdev_max_qd;
> module_param(enable_sdev_max_qd, int, 0444);
> MODULE_PARM_DESC(enable_sdev_max_qd, "Enable sdev max qd as can_queue. Default: 0");
>
> +int poll_queues;
> +module_param(poll_queues, int, 0444);
> +MODULE_PARM_DESC(poll_queues, "Number of queues to be use for io_uring poll mode.\n\t\t"
> + "This parameter is effective only if host_tagset_enable=1 &\n\t\t"
> + "It is not applicable for MFI_SERIES. &\n\t\t"
> + "Driver will work in latency mode. &\n\t\t"
> + "High iops queues are not allocated &\n\t\t"
> + );
> +
> int host_tagset_enable = 1;
> module_param(host_tagset_enable, int, 0444);
> MODULE_PARM_DESC(host_tagset_enable, "Shared host tagset enable/disable Default: enable(1)");
> @@ -207,6 +216,7 @@ static bool support_pci_lane_margining;
> static spinlock_t poll_aen_lock;
>
> extern struct dentry *megasas_debugfs_root;
> +extern int megasas_blk_mq_poll(struct Scsi_Host *shost, unsigned int queue_num);
>
> void
> megasas_complete_cmd(struct megasas_instance *instance, struct megasas_cmd *cmd,
> @@ -3127,14 +3137,45 @@ megasas_bios_param(struct scsi_device *sdev, struct block_device *bdev,
> static int megasas_map_queues(struct Scsi_Host *shost)
> {
> struct megasas_instance *instance;
> + int i, qoff, offset;
>
> instance = (struct megasas_instance *)shost->hostdata;
>
> if (shost->nr_hw_queues == 1)
> return 0;
>
> - return blk_mq_pci_map_queues(&shost->tag_set.map[HCTX_TYPE_DEFAULT],
> - instance->pdev, instance->low_latency_index_start);
> + offset = instance->low_latency_index_start;
> +
> + for (i = 0, qoff = 0; i < HCTX_MAX_TYPES; i++) {
> + struct blk_mq_queue_map *map = &shost->tag_set.map[i];
> +
> + map->nr_queues = 0;
> +
> + if (i == HCTX_TYPE_DEFAULT)
> + map->nr_queues = instance->msix_vectors - offset;
> + else if (i == HCTX_TYPE_POLL)
> + map->nr_queues = instance->iopoll_q_count;
> +
> + if (!map->nr_queues) {
> + BUG_ON(i == HCTX_TYPE_DEFAULT);
> + continue;
> + }
> +
> + /*
> + * The poll queue(s) doesn't have an IRQ (and hence IRQ
> + * affinity), so use the regular blk-mq cpu mapping
> + */
> + map->queue_offset = qoff;
> + if (i != HCTX_TYPE_POLL)
> + blk_mq_pci_map_queues(map, instance->pdev, offset);
> + else
> + blk_mq_map_queues(map);
> +
> + qoff += map->nr_queues;
> + offset += map->nr_queues;
> + }
> +
> + return 0;
> }
>
> static void megasas_aen_polling(struct work_struct *work);
> @@ -3446,6 +3487,7 @@ static struct scsi_host_template megasas_template = {
> .shost_attrs = megaraid_host_attrs,
> .bios_param = megasas_bios_param,
> .map_queues = megasas_map_queues,
> + .mq_poll = megasas_blk_mq_poll,
> .change_queue_depth = scsi_change_queue_depth,
> .max_segment_size = 0xffffffff,
> };
> @@ -5834,13 +5876,16 @@ __megasas_alloc_irq_vectors(struct megasas_instance *instance)
> irq_flags = PCI_IRQ_MSIX;
>
> if (instance->smp_affinity_enable)
> - irq_flags |= PCI_IRQ_AFFINITY;
> + irq_flags |= PCI_IRQ_AFFINITY | PCI_IRQ_ALL_TYPES;
> else
> descp = NULL;
>
> + /* Do not allocate msix vectors for poll_queues.
> + * msix_vectors is always within a range of FW supported reply queue.
> + */
> i = pci_alloc_irq_vectors_affinity(instance->pdev,
> instance->low_latency_index_start,
> - instance->msix_vectors, irq_flags, descp);
> + instance->msix_vectors - instance->iopoll_q_count, irq_flags, descp);
>
> return i;
> }
> @@ -5856,10 +5901,25 @@ megasas_alloc_irq_vectors(struct megasas_instance *instance)
> int i;
> unsigned int num_msix_req;
>
> + instance->iopoll_q_count = 0;
> + if ((instance->adapter_type != MFI_SERIES) &&
> + poll_queues) {
> +
> + instance->perf_mode = MR_LATENCY_PERF_MODE;
> + instance->low_latency_index_start = 1;
> +
> + /* reserve for default and non-mananged pre-vector. */
> + if (instance->msix_vectors > (instance->iopoll_q_count + 2))
> + instance->iopoll_q_count = poll_queues;
> + else
> + instance->iopoll_q_count = 0;
> + }
> +
> i = __megasas_alloc_irq_vectors(instance);
>
> - if ((instance->perf_mode == MR_BALANCED_PERF_MODE) &&
> - (i != instance->msix_vectors)) {
> + if (((instance->perf_mode == MR_BALANCED_PERF_MODE)
> + || instance->iopoll_q_count) &&
> + (i != (instance->msix_vectors - instance->iopoll_q_count))) {
> if (instance->msix_vectors)
> pci_free_irq_vectors(instance->pdev);
> /* Disable Balanced IOPS mode and try realloc vectors */
> @@ -5870,12 +5930,15 @@ megasas_alloc_irq_vectors(struct megasas_instance *instance)
> instance->msix_vectors = min(num_msix_req,
> instance->msix_vectors);
>
> + instance->iopoll_q_count = 0;
> i = __megasas_alloc_irq_vectors(instance);
>
> }
>
> dev_info(&instance->pdev->dev,
> - "requested/available msix %d/%d\n", instance->msix_vectors, i);
> + "requested/available msix %d/%d poll_queue %d\n",
> + instance->msix_vectors - instance->iopoll_q_count,
> + i, instance->iopoll_q_count);
>
> if (i > 0)
> instance->msix_vectors = i;
> @@ -6841,12 +6904,18 @@ static int megasas_io_attach(struct megasas_instance *instance)
> instance->smp_affinity_enable) {
> host->host_tagset = 1;
> host->nr_hw_queues = instance->msix_vectors -
> - instance->low_latency_index_start;
> + instance->low_latency_index_start + instance->iopoll_q_count;
> + if (instance->iopoll_q_count)
> + host->nr_maps = 3;
> + } else {
> + instance->iopoll_q_count = 0;
> }
>
> dev_info(&instance->pdev->dev,
> - "Max firmware commands: %d shared with nr_hw_queues = %d\n",
> - instance->max_fw_cmds, host->nr_hw_queues);
> + "Max firmware commands: %d shared with default "
> + "hw_queues = %d poll_queues %d\n", instance->max_fw_cmds,
> + host->nr_hw_queues - instance->iopoll_q_count,
> + instance->iopoll_q_count);
> /*
> * Notify the mid-layer about the new controller
> */
> @@ -8907,6 +8976,7 @@ static int __init megasas_init(void)
> msix_vectors = 1;
> rdpq_enable = 0;
> dual_qdepth_disable = 1;
> + poll_queues = 0;
> }
>
> /*
> diff --git a/drivers/scsi/megaraid/megaraid_sas_fusion.c b/drivers/scsi/megaraid/megaraid_sas_fusion.c
> index fd607287608e..49c8cc3507e0 100644
> --- a/drivers/scsi/megaraid/megaraid_sas_fusion.c
> +++ b/drivers/scsi/megaraid/megaraid_sas_fusion.c
> @@ -685,6 +685,8 @@ megasas_alloc_reply_fusion(struct megasas_instance *instance)
> fusion = instance->ctrl_context;
>
> count = instance->msix_vectors > 0 ? instance->msix_vectors : 1;
> + count += instance->iopoll_q_count;
> +
> fusion->reply_frames_desc_pool =
> dma_pool_create("mr_reply", &instance->pdev->dev,
> fusion->reply_alloc_sz * count, 16, 0);
> @@ -779,6 +781,7 @@ megasas_alloc_rdpq_fusion(struct megasas_instance *instance)
> }
>
> msix_count = instance->msix_vectors > 0 ? instance->msix_vectors : 1;
> + msix_count += instance->iopoll_q_count;
>
> fusion->reply_frames_desc_pool = dma_pool_create("mr_rdpq",
> &instance->pdev->dev,
> @@ -1129,7 +1132,7 @@ megasas_ioc_init_fusion(struct megasas_instance *instance)
> MPI2_IOCINIT_MSGFLAG_RDPQ_ARRAY_MODE : 0;
> IOCInitMessage->SystemRequestFrameBaseAddress = cpu_to_le64(fusion->io_request_frames_phys);
> IOCInitMessage->SenseBufferAddressHigh = cpu_to_le32(upper_32_bits(fusion->sense_phys_addr));
> - IOCInitMessage->HostMSIxVectors = instance->msix_vectors;
> + IOCInitMessage->HostMSIxVectors = instance->msix_vectors + instance->iopoll_q_count;
> IOCInitMessage->HostPageSize = MR_DEFAULT_NVME_PAGE_SHIFT;
>
> time = ktime_get_real();
> @@ -1823,6 +1826,8 @@ megasas_init_adapter_fusion(struct megasas_instance *instance)
> sizeof(union MPI2_SGE_IO_UNION))/16;
>
> count = instance->msix_vectors > 0 ? instance->msix_vectors : 1;
> + count += instance->iopoll_q_count;
> +
> for (i = 0 ; i < count; i++)
> fusion->last_reply_idx[i] = 0;
>
> @@ -1835,6 +1840,9 @@ megasas_init_adapter_fusion(struct megasas_instance *instance)
> MEGASAS_FUSION_IOCTL_CMDS);
> sema_init(&instance->ioctl_sem, MEGASAS_FUSION_IOCTL_CMDS);
>
> + for (i = 0; i < MAX_MSIX_QUEUES_FUSION; i++)
> + atomic_set(&fusion->busy_mq_poll[i], 0);
> +
> if (megasas_alloc_ioc_init_frame(instance))
> return 1;
>
> @@ -3500,6 +3508,9 @@ complete_cmd_fusion(struct megasas_instance *instance, u32 MSIxIndex,
> if (reply_descript_type == MPI2_RPY_DESCRIPT_FLAGS_UNUSED)
> return IRQ_NONE;
>
> + if (irq_context && !atomic_add_unless(&irq_context->in_used, 1, 1))
> + return 0;
> +
> num_completed = 0;
>
> while (d_val.u.low != cpu_to_le32(UINT_MAX) &&
> @@ -3613,6 +3624,7 @@ complete_cmd_fusion(struct megasas_instance *instance, u32 MSIxIndex,
> irq_context->irq_line_enable = true;
> irq_poll_sched(&irq_context->irqpoll);
> }
> + atomic_dec(&irq_context->in_used);
> return num_completed;
> }
> }
> @@ -3630,9 +3642,36 @@ complete_cmd_fusion(struct megasas_instance *instance, u32 MSIxIndex,
> instance->reply_post_host_index_addr[0]);
> megasas_check_and_restore_queue_depth(instance);
> }
> +
> + if (irq_context)
> + atomic_dec(&irq_context->in_used);
> +
> return num_completed;
> }
>
> +int megasas_blk_mq_poll(struct Scsi_Host *shost, unsigned int queue_num)
> +{
> +
> + struct megasas_instance *instance;
> + int num_entries = 0;
> + struct fusion_context *fusion;
> +
> + instance = (struct megasas_instance *)shost->hostdata;
> +
> + fusion = instance->ctrl_context;
> +
> + queue_num = queue_num + instance->low_latency_index_start;
> +
> + if (!atomic_add_unless(&fusion->busy_mq_poll[queue_num], 1, 1))
> + return 0;
> +
> + num_entries = complete_cmd_fusion(instance, queue_num, NULL);
> + atomic_dec(&fusion->busy_mq_poll[queue_num]);
> +
> + return num_entries;
> +}
> +
> +
> /**
> * megasas_enable_irq_poll() - enable irqpoll
> * @instance: Adapter soft state
> @@ -4164,6 +4203,8 @@ void megasas_reset_reply_desc(struct megasas_instance *instance)
>
> fusion = instance->ctrl_context;
> count = instance->msix_vectors > 0 ? instance->msix_vectors : 1;
> + count += instance->iopoll_q_count;
> +
> for (i = 0 ; i < count ; i++) {
> fusion->last_reply_idx[i] = 0;
> reply_desc = fusion->reply_frames_desc[i];
> diff --git a/drivers/scsi/megaraid/megaraid_sas_fusion.h b/drivers/scsi/megaraid/megaraid_sas_fusion.h
> index 30de4b01f703..242ff58a3404 100644
> --- a/drivers/scsi/megaraid/megaraid_sas_fusion.h
> +++ b/drivers/scsi/megaraid/megaraid_sas_fusion.h
> @@ -1303,6 +1303,9 @@ struct fusion_context {
> u8 *sense;
> dma_addr_t sense_phys_addr;
>
> + atomic_t busy_mq_poll[MAX_MSIX_QUEUES_FUSION];
> +
> +
> dma_addr_t reply_frames_desc_phys[MAX_MSIX_QUEUES_FUSION];
> union MPI2_REPLY_DESCRIPTORS_UNION *reply_frames_desc[MAX_MSIX_QUEUES_FUSION];
> struct rdpq_alloc_detail rdpq_tracker[RDPQ_MAX_CHUNK_COUNT];
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH v1 2/3] megaraid_sas: iouring iopoll support
2020-10-15 13:37 [PATCH v1 2/3] megaraid_sas: iouring iopoll support Kashyap Desai
` (2 preceding siblings ...)
2020-11-13 11:56 ` chenxiang (M)
@ 2020-11-20 7:14 ` Hannes Reinecke
2020-11-20 8:32 ` John Garry
3 siblings, 1 reply; 8+ messages in thread
From: Hannes Reinecke @ 2020-11-20 7:14 UTC (permalink / raw)
To: Kashyap Desai, linux-scsi; +Cc: sumit.saxena, chandrakanth.patil, linux-block
On 10/15/20 3:37 PM, Kashyap Desai wrote:
> Add support of iouring iopoll interface. This feature requires shared
> hosttag support in kernel and driver.
>
> Driver will work in non-IRQ mode = There will not be any msix vector
> associated for poll_queues and h/w can still work in this mode.
> MegaRaid h/w is single submission queue and multiple reply queue, but
> using shared host tagset support it will enable simulated multiple hw queue.
>
> Driver allocates some extra reply queues and it will be marked as poll_queue.
> These poll_queues will not have associated msix vectors. All the IO
> completion on this queue will be done from IOPOLL interface.
>
> megaraid_sas driver having 8 poll_queues and using io_uring hiprio=1 settings,
> It can reach 3.2M IOPs and there is zero interrupt generated by h/w.
>
> This feature can be enabled using module parameter poll_queues.
>
> Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
> Cc: sumit.saxena@broadcom.com
> Cc: chandrakanth.patil@broadcom.com
> Cc: linux-block@vger.kernel.org
>
> ---
> drivers/scsi/megaraid/megaraid_sas.h | 2 +
> drivers/scsi/megaraid/megaraid_sas_base.c | 90 ++++++++++++++++++---
> drivers/scsi/megaraid/megaraid_sas_fusion.c | 43 +++++++++-
> drivers/scsi/megaraid/megaraid_sas_fusion.h | 3 +
> 4 files changed, 127 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/scsi/megaraid/megaraid_sas.h b/drivers/scsi/megaraid/megaraid_sas.h
> index 5e4137f10e0e..4bf6fde49565 100644
> --- a/drivers/scsi/megaraid/megaraid_sas.h
> +++ b/drivers/scsi/megaraid/megaraid_sas.h
> @@ -2212,6 +2212,7 @@ struct megasas_irq_context {
> struct irq_poll irqpoll;
> bool irq_poll_scheduled;
> bool irq_line_enable;
> + atomic_t in_used;
> };
>
> struct MR_DRV_SYSTEM_INFO {
> @@ -2446,6 +2447,7 @@ struct megasas_instance {
> bool support_pci_lane_margining;
> u8 low_latency_index_start;
> int perf_mode;
> + int iopoll_q_count;
> };
>
> struct MR_LD_VF_MAP {
> diff --git a/drivers/scsi/megaraid/megaraid_sas_base.c b/drivers/scsi/megaraid/megaraid_sas_base.c
> index 020270ce790b..e2c307523a8a 100644
> --- a/drivers/scsi/megaraid/megaraid_sas_base.c
> +++ b/drivers/scsi/megaraid/megaraid_sas_base.c
> @@ -114,6 +114,15 @@ unsigned int enable_sdev_max_qd;
> module_param(enable_sdev_max_qd, int, 0444);
> MODULE_PARM_DESC(enable_sdev_max_qd, "Enable sdev max qd as can_queue. Default: 0");
>
> +int poll_queues;
> +module_param(poll_queues, int, 0444);
> +MODULE_PARM_DESC(poll_queues, "Number of queues to be use for io_uring poll mode.\n\t\t"
> + "This parameter is effective only if host_tagset_enable=1 &\n\t\t"
> + "It is not applicable for MFI_SERIES. &\n\t\t"
> + "Driver will work in latency mode. &\n\t\t"
> + "High iops queues are not allocated &\n\t\t"
> + );
> +
> int host_tagset_enable = 1;
> module_param(host_tagset_enable, int, 0444);
> MODULE_PARM_DESC(host_tagset_enable, "Shared host tagset enable/disable Default: enable(1)");
> @@ -207,6 +216,7 @@ static bool support_pci_lane_margining;
> static spinlock_t poll_aen_lock;
>
> extern struct dentry *megasas_debugfs_root;
> +extern int megasas_blk_mq_poll(struct Scsi_Host *shost, unsigned int queue_num);
>
> void
> megasas_complete_cmd(struct megasas_instance *instance, struct megasas_cmd *cmd,
> @@ -3127,14 +3137,45 @@ megasas_bios_param(struct scsi_device *sdev, struct block_device *bdev,
> static int megasas_map_queues(struct Scsi_Host *shost)
> {
> struct megasas_instance *instance;
> + int i, qoff, offset;
>
> instance = (struct megasas_instance *)shost->hostdata;
>
> if (shost->nr_hw_queues == 1)
> return 0;
>
> - return blk_mq_pci_map_queues(&shost->tag_set.map[HCTX_TYPE_DEFAULT],
> - instance->pdev, instance->low_latency_index_start);
> + offset = instance->low_latency_index_start;
> +
> + for (i = 0, qoff = 0; i < HCTX_MAX_TYPES; i++) {
> + struct blk_mq_queue_map *map = &shost->tag_set.map[i];
> +
> + map->nr_queues = 0;
> +
> + if (i == HCTX_TYPE_DEFAULT)
> + map->nr_queues = instance->msix_vectors - offset;
> + else if (i == HCTX_TYPE_POLL)
> + map->nr_queues = instance->iopoll_q_count;
> +
> + if (!map->nr_queues) {
> + BUG_ON(i == HCTX_TYPE_DEFAULT);
> + continue;
> + }
> +
> + /*
> + * The poll queue(s) doesn't have an IRQ (and hence IRQ
> + * affinity), so use the regular blk-mq cpu mapping
> + */
> + map->queue_offset = qoff;
> + if (i != HCTX_TYPE_POLL)
> + blk_mq_pci_map_queues(map, instance->pdev, offset);
> + else
> + blk_mq_map_queues(map);
> +
> + qoff += map->nr_queues;
> + offset += map->nr_queues;
> + }
> +
> + return 0;
> }
>
> static void megasas_aen_polling(struct work_struct *work);
> @@ -3446,6 +3487,7 @@ static struct scsi_host_template megasas_template = {
> .shost_attrs = megaraid_host_attrs,
> .bios_param = megasas_bios_param,
> .map_queues = megasas_map_queues,
> + .mq_poll = megasas_blk_mq_poll,
> .change_queue_depth = scsi_change_queue_depth,
> .max_segment_size = 0xffffffff,
> };
> @@ -5834,13 +5876,16 @@ __megasas_alloc_irq_vectors(struct megasas_instance *instance)
> irq_flags = PCI_IRQ_MSIX;
>
> if (instance->smp_affinity_enable)
> - irq_flags |= PCI_IRQ_AFFINITY;
> + irq_flags |= PCI_IRQ_AFFINITY | PCI_IRQ_ALL_TYPES;
> else
> descp = NULL;
>
> + /* Do not allocate msix vectors for poll_queues.
> + * msix_vectors is always within a range of FW supported reply queue.
> + */
> i = pci_alloc_irq_vectors_affinity(instance->pdev,
> instance->low_latency_index_start,
> - instance->msix_vectors, irq_flags, descp);
> + instance->msix_vectors - instance->iopoll_q_count, irq_flags, descp);
>
> return i;
> }
> @@ -5856,10 +5901,25 @@ megasas_alloc_irq_vectors(struct megasas_instance *instance)
> int i;
> unsigned int num_msix_req;
>
> + instance->iopoll_q_count = 0;
> + if ((instance->adapter_type != MFI_SERIES) &&
> + poll_queues) {
> +
> + instance->perf_mode = MR_LATENCY_PERF_MODE;
> + instance->low_latency_index_start = 1;
> +
> + /* reserve for default and non-mananged pre-vector. */
> + if (instance->msix_vectors > (instance->iopoll_q_count + 2))
> + instance->iopoll_q_count = poll_queues;
> + else
> + instance->iopoll_q_count = 0;
> + }
> +
> i = __megasas_alloc_irq_vectors(instance);
>
> - if ((instance->perf_mode == MR_BALANCED_PERF_MODE) &&
> - (i != instance->msix_vectors)) {
> + if (((instance->perf_mode == MR_BALANCED_PERF_MODE)
> + || instance->iopoll_q_count) &&
> + (i != (instance->msix_vectors - instance->iopoll_q_count))) {
> if (instance->msix_vectors)
> pci_free_irq_vectors(instance->pdev);
> /* Disable Balanced IOPS mode and try realloc vectors */
> @@ -5870,12 +5930,15 @@ megasas_alloc_irq_vectors(struct megasas_instance *instance)
> instance->msix_vectors = min(num_msix_req,
> instance->msix_vectors);
>
> + instance->iopoll_q_count = 0;
> i = __megasas_alloc_irq_vectors(instance);
>
> }
>
> dev_info(&instance->pdev->dev,
> - "requested/available msix %d/%d\n", instance->msix_vectors, i);
> + "requested/available msix %d/%d poll_queue %d\n",
> + instance->msix_vectors - instance->iopoll_q_count,
> + i, instance->iopoll_q_count);
>
> if (i > 0)
> instance->msix_vectors = i;
> @@ -6841,12 +6904,18 @@ static int megasas_io_attach(struct megasas_instance *instance)
> instance->smp_affinity_enable) {
> host->host_tagset = 1;
> host->nr_hw_queues = instance->msix_vectors -
> - instance->low_latency_index_start;
> + instance->low_latency_index_start + instance->iopoll_q_count;
> + if (instance->iopoll_q_count)
> + host->nr_maps = 3;
> + } else {
> + instance->iopoll_q_count = 0;
> }
>
> dev_info(&instance->pdev->dev,
> - "Max firmware commands: %d shared with nr_hw_queues = %d\n",
> - instance->max_fw_cmds, host->nr_hw_queues);
> + "Max firmware commands: %d shared with default "
> + "hw_queues = %d poll_queues %d\n", instance->max_fw_cmds,
> + host->nr_hw_queues - instance->iopoll_q_count,
> + instance->iopoll_q_count);
> /*
> * Notify the mid-layer about the new controller
> */
> @@ -8907,6 +8976,7 @@ static int __init megasas_init(void)
> msix_vectors = 1;
> rdpq_enable = 0;
> dual_qdepth_disable = 1;
> + poll_queues = 0;
> }
>
> /*
> diff --git a/drivers/scsi/megaraid/megaraid_sas_fusion.c b/drivers/scsi/megaraid/megaraid_sas_fusion.c
> index fd607287608e..49c8cc3507e0 100644
> --- a/drivers/scsi/megaraid/megaraid_sas_fusion.c
> +++ b/drivers/scsi/megaraid/megaraid_sas_fusion.c
> @@ -685,6 +685,8 @@ megasas_alloc_reply_fusion(struct megasas_instance *instance)
> fusion = instance->ctrl_context;
>
> count = instance->msix_vectors > 0 ? instance->msix_vectors : 1;
> + count += instance->iopoll_q_count;
> +
> fusion->reply_frames_desc_pool =
> dma_pool_create("mr_reply", &instance->pdev->dev,
> fusion->reply_alloc_sz * count, 16, 0);
> @@ -779,6 +781,7 @@ megasas_alloc_rdpq_fusion(struct megasas_instance *instance)
> }
>
> msix_count = instance->msix_vectors > 0 ? instance->msix_vectors : 1;
> + msix_count += instance->iopoll_q_count;
>
> fusion->reply_frames_desc_pool = dma_pool_create("mr_rdpq",
> &instance->pdev->dev,
> @@ -1129,7 +1132,7 @@ megasas_ioc_init_fusion(struct megasas_instance *instance)
> MPI2_IOCINIT_MSGFLAG_RDPQ_ARRAY_MODE : 0;
> IOCInitMessage->SystemRequestFrameBaseAddress = cpu_to_le64(fusion->io_request_frames_phys);
> IOCInitMessage->SenseBufferAddressHigh = cpu_to_le32(upper_32_bits(fusion->sense_phys_addr));
> - IOCInitMessage->HostMSIxVectors = instance->msix_vectors;
> + IOCInitMessage->HostMSIxVectors = instance->msix_vectors + instance->iopoll_q_count;
> IOCInitMessage->HostPageSize = MR_DEFAULT_NVME_PAGE_SHIFT;
>
> time = ktime_get_real();
> @@ -1823,6 +1826,8 @@ megasas_init_adapter_fusion(struct megasas_instance *instance)
> sizeof(union MPI2_SGE_IO_UNION))/16;
>
> count = instance->msix_vectors > 0 ? instance->msix_vectors : 1;
> + count += instance->iopoll_q_count;
> +
> for (i = 0 ; i < count; i++)
> fusion->last_reply_idx[i] = 0;
>
> @@ -1835,6 +1840,9 @@ megasas_init_adapter_fusion(struct megasas_instance *instance)
> MEGASAS_FUSION_IOCTL_CMDS);
> sema_init(&instance->ioctl_sem, MEGASAS_FUSION_IOCTL_CMDS);
>
> + for (i = 0; i < MAX_MSIX_QUEUES_FUSION; i++)
> + atomic_set(&fusion->busy_mq_poll[i], 0);
> +
> if (megasas_alloc_ioc_init_frame(instance))
> return 1;
>
> @@ -3500,6 +3508,9 @@ complete_cmd_fusion(struct megasas_instance *instance, u32 MSIxIndex,
> if (reply_descript_type == MPI2_RPY_DESCRIPT_FLAGS_UNUSED)
> return IRQ_NONE;
>
> + if (irq_context && !atomic_add_unless(&irq_context->in_used, 1, 1))
> + return 0;
> +
> num_completed = 0;
>
> while (d_val.u.low != cpu_to_le32(UINT_MAX) &&
> @@ -3613,6 +3624,7 @@ complete_cmd_fusion(struct megasas_instance *instance, u32 MSIxIndex,
> irq_context->irq_line_enable = true;
> irq_poll_sched(&irq_context->irqpoll);
> }
> + atomic_dec(&irq_context->in_used);
> return num_completed;
> }
> }
> @@ -3630,9 +3642,36 @@ complete_cmd_fusion(struct megasas_instance *instance, u32 MSIxIndex,
> instance->reply_post_host_index_addr[0]);
> megasas_check_and_restore_queue_depth(instance);
> }
> +
> + if (irq_context)
> + atomic_dec(&irq_context->in_used);
> +
> return num_completed;
> }
>
> +int megasas_blk_mq_poll(struct Scsi_Host *shost, unsigned int queue_num)
> +{
> +
> + struct megasas_instance *instance;
> + int num_entries = 0;
> + struct fusion_context *fusion;
> +
> + instance = (struct megasas_instance *)shost->hostdata;
> +
> + fusion = instance->ctrl_context;
> +
> + queue_num = queue_num + instance->low_latency_index_start;
> +
> + if (!atomic_add_unless(&fusion->busy_mq_poll[queue_num], 1, 1))
> + return 0;
> +
> + num_entries = complete_cmd_fusion(instance, queue_num, NULL);
> + atomic_dec(&fusion->busy_mq_poll[queue_num]);
> +
> + return num_entries;
> +}
> +
> +
> /**
> * megasas_enable_irq_poll() - enable irqpoll
> * @instance: Adapter soft state
> @@ -4164,6 +4203,8 @@ void megasas_reset_reply_desc(struct megasas_instance *instance)
>
> fusion = instance->ctrl_context;
> count = instance->msix_vectors > 0 ? instance->msix_vectors : 1;
> + count += instance->iopoll_q_count;
> +
> for (i = 0 ; i < count ; i++) {
> fusion->last_reply_idx[i] = 0;
> reply_desc = fusion->reply_frames_desc[i];
> diff --git a/drivers/scsi/megaraid/megaraid_sas_fusion.h b/drivers/scsi/megaraid/megaraid_sas_fusion.h
> index 30de4b01f703..242ff58a3404 100644
> --- a/drivers/scsi/megaraid/megaraid_sas_fusion.h
> +++ b/drivers/scsi/megaraid/megaraid_sas_fusion.h
> @@ -1303,6 +1303,9 @@ struct fusion_context {
> u8 *sense;
> dma_addr_t sense_phys_addr;
>
> + atomic_t busy_mq_poll[MAX_MSIX_QUEUES_FUSION];
> +
> +
> dma_addr_t reply_frames_desc_phys[MAX_MSIX_QUEUES_FUSION];
> union MPI2_REPLY_DESCRIPTORS_UNION *reply_frames_desc[MAX_MSIX_QUEUES_FUSION];
> struct rdpq_alloc_detail rdpq_tracker[RDPQ_MAX_CHUNK_COUNT];
>
The usage of atomic_add_unless() is a bit unusual; one might consider
using atomic_test_and_set(). Also it seems to be a bit of a waste using
an atomic counter here, seeing that the only values ever used are 0 and
1. But this is largely cosmetic, so:
Reviewed-by: Hannes Reinecke <hare@suse.de>
Cheers,
Hannes
--
Dr. Hannes Reinecke Kernel Storage Architect
hare@suse.de +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH v1 2/3] megaraid_sas: iouring iopoll support
2020-11-20 7:14 ` Hannes Reinecke
@ 2020-11-20 8:32 ` John Garry
2020-11-30 9:17 ` Kashyap Desai
0 siblings, 1 reply; 8+ messages in thread
From: John Garry @ 2020-11-20 8:32 UTC (permalink / raw)
To: Hannes Reinecke, Kashyap Desai, linux-scsi
Cc: sumit.saxena, chandrakanth.patil, linux-block
On 20/11/2020 07:14, Hannes Reinecke wrote:
>>
>> struct rdpq_alloc_detail rdpq_tracker[RDPQ_MAX_CHUNK_COUNT];
>>
> The usage of atomic_add_unless() is a bit unusual; one might consider
> using atomic_test_and_set(). Also it seems to be a bit of a waste using
> an atomic counter here, seeing that the only values ever used are 0 and
> 1. But this is largely cosmetic, so:
atomic_add_unless() is generally implemented with a cmpxchg(), which can
be inefficient also.
John
^ permalink raw reply [flat|nested] 8+ messages in thread
* RE: [PATCH v1 2/3] megaraid_sas: iouring iopoll support
2020-11-20 8:32 ` John Garry
@ 2020-11-30 9:17 ` Kashyap Desai
2020-11-30 9:40 ` John Garry
0 siblings, 1 reply; 8+ messages in thread
From: Kashyap Desai @ 2020-11-30 9:17 UTC (permalink / raw)
To: John Garry, Hannes Reinecke, linux-scsi
Cc: Sumit Saxena, Chandrakanth Patil, linux-block
[-- Attachment #1: Type: text/plain, Size: 1111 bytes --]
> -----Original Message-----
> From: John Garry [mailto:john.garry@huawei.com]
> Sent: Friday, November 20, 2020 2:03 PM
> To: Hannes Reinecke <hare@suse.de>; Kashyap Desai
> <kashyap.desai@broadcom.com>; linux-scsi@vger.kernel.org
> Cc: sumit.saxena@broadcom.com; chandrakanth.patil@broadcom.com;
> linux-block@vger.kernel.org
> Subject: Re: [PATCH v1 2/3] megaraid_sas: iouring iopoll support
>
> On 20/11/2020 07:14, Hannes Reinecke wrote:
> >>
> >> struct rdpq_alloc_detail rdpq_tracker[RDPQ_MAX_CHUNK_COUNT];
> >>
> > The usage of atomic_add_unless() is a bit unusual; one might consider
> > using atomic_test_and_set(). Also it seems to be a bit of a waste
> > using an atomic counter here, seeing that the only values ever used
> > are 0 and 1. But this is largely cosmetic, so:
>
> atomic_add_unless() is generally implemented with a cmpxchg(), which can
> be inefficient also.
John -
Noted. I was able to get expected performance on multiple host using
atomic_add_() API.
We can improve based on any real time issue. Is it fine ?
@Hannes - I will add Reviewed-by Tag in V2 series.
>
> John
[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4169 bytes --]
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH v1 2/3] megaraid_sas: iouring iopoll support
2020-11-30 9:17 ` Kashyap Desai
@ 2020-11-30 9:40 ` John Garry
0 siblings, 0 replies; 8+ messages in thread
From: John Garry @ 2020-11-30 9:40 UTC (permalink / raw)
To: Kashyap Desai, Hannes Reinecke, linux-scsi
Cc: Sumit Saxena, Chandrakanth Patil, linux-block
On 30/11/2020 09:17, Kashyap Desai wrote:
>> be inefficient also.
>
> John -
>
> Noted. I was able to get expected performance on multiple host using
> atomic_add_() API.
> We can improve based on any real time issue. Is it fine ?
sure, it's your driver :)
>
> @Hannes - I will add Reviewed-by Tag in V2 series.
>
^ permalink raw reply [flat|nested] 8+ messages in thread