[Suspected-Phishing]Re: [PATCH V6] nvme-pci: add SGL support

Max Gurtovoy maxg at mellanox.com
Sun Oct 8 02:57:15 PDT 2017



On 10/8/2017 12:16 PM, Max Gurtovoy wrote:
> 
> 
> On 10/5/2017 10:48 PM, Chaitanya Kulkarni wrote:
>> From: Chaitanya Kulkarni <chaitanya.kulkarni at wdc.com>
>>
>> This adds SGL support for NVMe PCIe driver, based on an earlier patch
>> from Rajiv Shanmugam Madeswaran <smrajiv15 at gmail.com>. This patch
>> refactors the original code and adds new module parameter sgl_threshold
>> to determine whether to use SGL or PRP for IOs.
>>
>> The usage of SGLs is controlled by the sgl_threshold module parameter,
>> which allows to conditionally use SGLs if average request segment size
>> (avg_seg_size) is greater than sgl_threshold. In the original patch,
>> the decision of using SGLs was dependent only on the IO size,
>> with the new approach we consider not only IO size but also the
>> number of physical segments present in the IO.
>>
>> We calculate avg_seg_size based on request payload bytes and number
>> of physical segments present in the request.
> 
> Can you add some performance numbers here to show the benefit ?
> and also please recomend the user when it's better to use sgls vs prps.
> 
>>
>> For e.g.:-
>>
>> 1. blk_rq_nr_phys_segments = 2 blk_rq_payload_bytes = 8k
>> avg_seg_size = 4K use sgl if avg_seg_size >= sgl_threshold.
>>
>> 2. blk_rq_nr_phys_segments = 2 blk_rq_payload_bytes = 64k
>> avg_seg_size = 32K use sgl if avg_seg_size >= sgl_threshold.
>>
>> 3. blk_rq_nr_phys_segments = 16 blk_rq_payload_bytes = 64k
>> avg_seg_size = 4K use sgl if avg_seg_size
> 
> typo :)
> 
>>
>> Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni at wdc.com>
>> ---
>>   drivers/nvme/host/pci.c | 222 
>> ++++++++++++++++++++++++++++++++++++++++++------
>>   1 file changed, 197 insertions(+), 25 deletions(-)
>>
>> diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
>> index cb73bc8..56db876 100644
>> --- a/drivers/nvme/host/pci.c
>> +++ b/drivers/nvme/host/pci.c
>> @@ -45,6 +45,8 @@
>>    */
>>   #define NVME_AQ_BLKMQ_DEPTH    (NVME_AQ_DEPTH - NVME_NR_AERS)
>> +#define SGES_PER_PAGE          (PAGE_SIZE / sizeof(struct 
>> nvme_sgl_desc))
> 
> should we use PAGE_SIZE or ctrl->page_size ?
> 
>> +
>>   static int use_threaded_interrupts;
>>   module_param(use_threaded_interrupts, int, 0);
>> @@ -57,6 +59,10 @@
>>   MODULE_PARM_DESC(max_host_mem_size_mb,
>>       "Maximum Host Memory Buffer (HMB) size per controller (in MiB)");
>> +static unsigned int sgl_threshold = SZ_32K;
>> +module_param(sgl_threshold, uint, 0644);
>> +MODULE_PARM_DESC(sgl_threshold, "use SGL for I/O larger or equal to 
>> this size");
>> +
> 
> please mention that 0 value disable sgl usage.
> 
>>   static int io_queue_depth_set(const char *val, const struct 
>> kernel_param *kp);
>>   static const struct kernel_param_ops io_queue_depth_ops = {
>>       .set = io_queue_depth_set,
>> @@ -178,6 +184,7 @@ struct nvme_queue {
>>   struct nvme_iod {
>>       struct nvme_request req;
>>       struct nvme_queue *nvmeq;
>> +    bool use_sgl;
>>       int aborted;
>>       int npages;        /* In the PRP list. 0 means small pool in use */
>>       int nents;        /* Used in scatterlist */
>> @@ -331,17 +338,32 @@ static int nvme_npages(unsigned size, struct 
>> nvme_dev *dev)
>>       return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8);
>>   }
>> +/*
>> + * Calculates the number of pages needed for the SGL segments. For 
>> example a 4k
>> + * page can accommodate 256 SGL descriptors.
>> + */
>> +static int nvme_npages_sgl(unsigned int num_seg)
>> +{
>> +    return DIV_ROUND_UP(num_seg * sizeof(struct nvme_sgl_desc), 
>> PAGE_SIZE);
>> +}
>> +
>>   static unsigned int nvme_iod_alloc_size(struct nvme_dev *dev,
>> -        unsigned int size, unsigned int nseg)
>> +        unsigned int size, unsigned int nseg, bool use_sgl)
>>   {
>> -    return sizeof(__le64 *) * nvme_npages(size, dev) +
>> -            sizeof(struct scatterlist) * nseg;
>> +    size_t alloc_size;
>> +
>> +    if (use_sgl)
>> +        alloc_size = sizeof(__le64 *) * nvme_npages_sgl(nseg);
>> +    else
>> +        alloc_size = sizeof(__le64 *) * nvme_npages(size, dev);
>> +
>> +    return alloc_size + sizeof(struct scatterlist) * nseg;
>>   }
>> -static unsigned int nvme_cmd_size(struct nvme_dev *dev)
>> +static unsigned int nvme_cmd_size(struct nvme_dev *dev, bool use_sgl)
>>   {
>>       return sizeof(struct nvme_iod) +
>> -        nvme_iod_alloc_size(dev, NVME_INT_BYTES(dev), NVME_INT_PAGES);
>> +        nvme_iod_alloc_size(dev, NVME_INT_BYTES(dev), NVME_INT_PAGES, 
>> use_sgl);
>>   }
>>   static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
>> @@ -425,10 +447,10 @@ static void __nvme_submit_cmd(struct nvme_queue 
>> *nvmeq,
>>       nvmeq->sq_tail = tail;
>>   }
>> -static __le64 **iod_list(struct request *req)
>> +static void **iod_list(struct request *req)
>>   {
>>       struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
>> -    return (__le64 **)(iod->sg + blk_rq_nr_phys_segments(req));
>> +    return (void **)(iod->sg + blk_rq_nr_phys_segments(req));
>>   }
>>   static blk_status_t nvme_init_iod(struct request *rq, struct 
>> nvme_dev *dev)
>> @@ -438,7 +460,10 @@ static blk_status_t nvme_init_iod(struct request 
>> *rq, struct nvme_dev *dev)
>>       unsigned int size = blk_rq_payload_bytes(rq);
>>       if (nseg > NVME_INT_PAGES || size > NVME_INT_BYTES(dev)) {
>> -        iod->sg = kmalloc(nvme_iod_alloc_size(dev, size, nseg), 
>> GFP_ATOMIC);
>> +        size_t alloc_size = nvme_iod_alloc_size(dev, size, nseg,
>> +                iod->use_sgl);
>> +
>> +        iod->sg = kmalloc(alloc_size, GFP_ATOMIC);
>>           if (!iod->sg)
>>               return BLK_STS_RESOURCE;
>>       } else {
>> @@ -456,18 +481,29 @@ static blk_status_t nvme_init_iod(struct request 
>> *rq, struct nvme_dev *dev)
>>   static void nvme_free_iod(struct nvme_dev *dev, struct request *req)
>>   {
>>       struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
>> -    const int last_prp = dev->ctrl.page_size / 8 - 1;
>> +    const int last_prp = dev->ctrl.page_size / sizeof(__le64) - 1;
>> +    dma_addr_t dma_addr = iod->first_dma, next_dma_addr;
>>       int i;
>> -    __le64 **list = iod_list(req);
>> -    dma_addr_t prp_dma = iod->first_dma;
>>       if (iod->npages == 0)
>> -        dma_pool_free(dev->prp_small_pool, list[0], prp_dma);
>> +        dma_pool_free(dev->prp_small_pool, iod_list(req)[0], dma_addr);
>> +
>>       for (i = 0; i < iod->npages; i++) {
>> -        __le64 *prp_list = list[i];
>> -        dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]);
>> -        dma_pool_free(dev->prp_page_pool, prp_list, prp_dma);
>> -        prp_dma = next_prp_dma;
>> +        void *addr = iod_list(req)[i];
>> +
>> +        if (iod->use_sgl) {
>> +            struct nvme_sgl_desc *sg_list = addr;
>> +
>> +            next_dma_addr =
>> +                le64_to_cpu((sg_list[SGES_PER_PAGE - 1]).addr);
>> +        } else {
>> +            __le64 *prp_list = addr;
>> +
>> +            next_dma_addr = le64_to_cpu(prp_list[last_prp]);
>> +        }
>> +
>> +        dma_pool_free(dev->prp_page_pool, addr, dma_addr);
>> +        dma_addr = next_dma_addr;
>>       }
>>       if (iod->sg != iod->inline_sg)
>> @@ -555,7 +591,8 @@ static void nvme_print_sgl(struct scatterlist 
>> *sgl, int nents)
>>       }
>>   }
>> -static blk_status_t nvme_setup_prps(struct nvme_dev *dev, struct 
>> request *req)
>> +static blk_status_t nvme_setup_prps(struct nvme_dev *dev, struct 
>> request *req,
>> +        struct nvme_rw_command *cmnd)
>>   {
>>       struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
>>       struct dma_pool *pool;
>> @@ -566,14 +603,16 @@ static blk_status_t nvme_setup_prps(struct 
>> nvme_dev *dev, struct request *req)
>>       u32 page_size = dev->ctrl.page_size;
>>       int offset = dma_addr & (page_size - 1);
>>       __le64 *prp_list;
>> -    __le64 **list = iod_list(req);
>> +    void **list = iod_list(req);
>>       dma_addr_t prp_dma;
>>       int nprps, i;
>> +    iod->use_sgl = false;
>> +
>>       length -= (page_size - offset);
>>       if (length <= 0) {
>>           iod->first_dma = 0;
>> -        return BLK_STS_OK;
>> +        goto done;
>>       }
>>       dma_len -= (page_size - offset);
>> @@ -587,7 +626,7 @@ static blk_status_t nvme_setup_prps(struct 
>> nvme_dev *dev, struct request *req)
>>       if (length <= page_size) {
>>           iod->first_dma = dma_addr;
>> -        return BLK_STS_OK;
>> +        goto done;
>>       }
>>       nprps = DIV_ROUND_UP(length, page_size);
>> @@ -634,6 +673,10 @@ static blk_status_t nvme_setup_prps(struct 
>> nvme_dev *dev, struct request *req)
>>           dma_len = sg_dma_len(sg);
>>       }
>> +done:
>> +    cmnd->dptr.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
>> +    cmnd->dptr.prp2 = cpu_to_le64(iod->first_dma);
>> +
>>       return BLK_STS_OK;
>>    bad_sgl:
>> @@ -643,6 +686,129 @@ static blk_status_t nvme_setup_prps(struct 
>> nvme_dev *dev, struct request *req)
>>       return BLK_STS_IOERR;
>>   }
>> +static void nvme_pci_sgl_set_data(struct nvme_sgl_desc *sge,
>> +        struct scatterlist *sg)
>> +{
>> +    sge->addr = cpu_to_le64(sg_dma_address(sg));
>> +    sge->length = cpu_to_le32(sg_dma_len(sg));
>> +    sge->type = NVME_SGL_FMT_DATA_DESC << 4;
>> +}
>> +
>> +static void nvme_pci_sgl_set_last_seg(struct nvme_sgl_desc *sge,
>> +        dma_addr_t dma_addr, int entries)
>> +{
>> +    sge->addr = cpu_to_le64(dma_addr);
>> +    sge->length = cpu_to_le32(entries * sizeof(struct nvme_sgl_desc));
>> +    sge->type = NVME_SGL_FMT_LAST_SEG_DESC << 4;
>> +}
>> +
>> +static void nvme_pci_sgl_set_seg(struct nvme_sgl_desc *sge,
>> +        dma_addr_t dma_addr)
>> +{
>> +    sge->addr = cpu_to_le64(dma_addr);
>> +    sge->length = cpu_to_le32(PAGE_SIZE);
>> +    sge->type = NVME_SGL_FMT_SEG_DESC << 4;
>> +}
>> +
>> +static blk_status_t nvme_setup_sgls(struct nvme_dev *dev, struct 
>> request *req,
>> +        struct nvme_rw_command *cmd)
>> +{
>> +    struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
>> +    int length = blk_rq_payload_bytes(req);
>> +    struct dma_pool *pool;
>> +    struct nvme_sgl_desc *sg_list;
>> +    struct scatterlist *sg = iod->sg;
>> +    int entries = iod->nents, i = 0;
>> +    dma_addr_t sgl_dma;
>> +
>> +    iod->use_sgl = true;
>> +
>> +    cmd->flags = NVME_CMD_SGL_METABUF; /* setting the transfer type 
>> as SGL */
>> +
>> +    if (length == sg_dma_len(sg)) {
>> +        nvme_pci_sgl_set_data(&cmd->dptr.sgl, sg);
>> +        return BLK_STS_OK;
>> +    }
>> +
>> +    if (entries <= 256 / sizeof(struct nvme_sgl_desc)) {
> 
> where does 256 comes from ? can we use macro definition here ?
> 
>> +        pool = dev->prp_small_pool;
>> +        iod->npages = 0;
>> +    } else {
>> +        pool = dev->prp_page_pool;
>> +        iod->npages = 1;
>> +    }
>> +
>> +    sg_list = dma_pool_alloc(pool, GFP_ATOMIC, &sgl_dma);
>> +    if (!sg_list) {
>> +        iod->npages = -1;
>> +        return BLK_STS_RESOURCE;
>> +    }
>> +
>> +    iod_list(req)[0] = sg_list;
>> +    iod->first_dma = sgl_dma;
>> +
>> +    if (entries <= SGES_PER_PAGE) {
>> +        nvme_pci_sgl_set_last_seg(&cmd->dptr.sgl, sgl_dma, entries);
>> +
>> +        for (i = 0; i < entries; i++) {
>> +            nvme_pci_sgl_set_data(&sg_list[i], sg);
>> +            length -= sg_dma_len(sg);
>> +            sg = sg_next(sg);
>> +        }
>> +
>> +        WARN_ON(length > 0);
>> +        return BLK_STS_OK;
>> +    }
>> +
>> +    nvme_pci_sgl_set_seg(&cmd->dptr.sgl, sgl_dma);
>> +
>> +    do {
>> +        if (i == SGES_PER_PAGE) {
>> +            struct nvme_sgl_desc *old_sg_desc = sg_list;
>> +            struct nvme_sgl_desc *link = &old_sg_desc[i - 1];
>> +
>> +            sg_list = dma_pool_alloc(pool, GFP_ATOMIC, &sgl_dma);
>> +            if (!sg_list)
>> +                return BLK_STS_RESOURCE;
>> +
>> +            i = 0;
>> +            iod_list(req)[iod->npages++] = sg_list;
>> +            sg_list[i++] = *link;
>> +
>> +            if (entries < SGES_PER_PAGE)
>> +                nvme_pci_sgl_set_last_seg(link, sgl_dma, entries);
>> +            else
>> +                nvme_pci_sgl_set_seg(link, sgl_dma);
>> +        }
>> +
>> +        nvme_pci_sgl_set_data(&sg_list[i], sg);
>> +
>> +        length -= sg_dma_len(sg);
>> +        sg = sg_next(sg);
>> +        entries--;
>> +    } while (length > 0);
>> +
>> +    WARN_ON(entries > 0);
>> +    return BLK_STS_OK;
>> +}
>> +
>> +static inline bool nvme_pci_use_sgls(struct nvme_dev *dev, struct 
>> request *req)
>> +{
>> +    struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
>> +    unsigned int avg_seg_size;
>> +
>> +    avg_seg_size = DIV_ROUND_UP(blk_rq_payload_bytes(req),
>> +            blk_rq_nr_phys_segments(req));
>> +
>> +    if (!(dev->ctrl.sgls & ((1 << 0) | (1 << 1))))
> 
> Can you create macros for the spec definitions ?

Answering to myself - I saw that you discussed this already so I guess 
you can stay as is.

> In case of Dword alignment and granularity, where do we check that 
> requirement ?
> 
>> +        return false;
>> +    if (!iod->nvmeq->qid)
>> +        return false;
>> +    if (!sgl_threshold || avg_seg_size < sgl_threshold)
>> +        return false;
>> +    return true;
>> +}
>> +
>>   static blk_status_t nvme_map_data(struct nvme_dev *dev, struct 
>> request *req,
>>           struct nvme_command *cmnd)
>>   {
>> @@ -662,7 +828,11 @@ static blk_status_t nvme_map_data(struct nvme_dev 
>> *dev, struct request *req,
>>                   DMA_ATTR_NO_WARN))
>>           goto out;
>> -    ret = nvme_setup_prps(dev, req);
>> +    if (nvme_pci_use_sgls(dev, req))
>> +        ret = nvme_setup_sgls(dev, req, &cmnd->rw);
>> +    else
>> +        ret = nvme_setup_prps(dev, req, &cmnd->rw);
>> +
>>       if (ret != BLK_STS_OK)
>>           goto out_unmap;
>> @@ -682,8 +852,6 @@ static blk_status_t nvme_map_data(struct nvme_dev 
>> *dev, struct request *req,
>>               goto out_unmap;
>>       }
>> -    cmnd->rw.dptr.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
>> -    cmnd->rw.dptr.prp2 = cpu_to_le64(iod->first_dma);
>>       if (blk_integrity_rq(req))
>>           cmnd->rw.metadata = cpu_to_le64(sg_dma_address(&iod->meta_sg));
>>       return BLK_STS_OK;
>> @@ -1379,7 +1547,7 @@ static int nvme_alloc_admin_tags(struct nvme_dev 
>> *dev)
>>           dev->admin_tagset.queue_depth = NVME_AQ_BLKMQ_DEPTH - 1;
>>           dev->admin_tagset.timeout = ADMIN_TIMEOUT;
>>           dev->admin_tagset.numa_node = dev_to_node(dev->dev);
>> -        dev->admin_tagset.cmd_size = nvme_cmd_size(dev);
>> +        dev->admin_tagset.cmd_size = nvme_cmd_size(dev, false);
>>           dev->admin_tagset.flags = BLK_MQ_F_NO_SCHED;
>>           dev->admin_tagset.driver_data = dev;
>> @@ -1906,7 +2074,11 @@ static int nvme_dev_add(struct nvme_dev *dev)
>>           dev->tagset.numa_node = dev_to_node(dev->dev);
>>           dev->tagset.queue_depth =
>>                   min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1;
>> -        dev->tagset.cmd_size = nvme_cmd_size(dev);
>> +        dev->tagset.cmd_size = nvme_cmd_size(dev, false);
>> +        if ((dev->ctrl.sgls & ((1 << 0) | (1 << 1))) && sgl_threshold) {
>> +            dev->tagset.cmd_size = max(dev->tagset.cmd_size,
>> +                    nvme_cmd_size(dev, true));
>> +        }
>>           dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE;
>>           dev->tagset.driver_data = dev;
>>
> 
> _______________________________________________
> Linux-nvme mailing list
> Linux-nvme at lists.infradead.org
> https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Flists.infradead.org%2Fmailman%2Flistinfo%2Flinux-nvme&data=02%7C01%7Cmaxg%40mellanox.com%7Cb2e5265cf8e94d7008f808d50e2d6e8a%7Ca652971c7d2e4d9ba6a4d149256f461b%7C0%7C0%7C636430510665543336&sdata=1hsl4jawK03SzkUe3AFEGgF3UAeFaAaAWbn2VtTyUEQ%3D&reserved=0 
> 



More information about the Linux-nvme mailing list