> On 11 Jun 2018, at 22.53, Heiner Litz <hlitz@xxxxxxxx> wrote:
>
> In the read path, partial reads are currently performed synchronously
> which affects performance for workloads that generate many partial
> reads. This patch adds an asynchronous partial read path as well as
> the required partial read ctx.
>
> Signed-off-by: Heiner Litz <hlitz@xxxxxxxx>
> ---
> drivers/lightnvm/pblk-read.c | 179 ++++++++++++++++++++++++++++---------------
> drivers/lightnvm/pblk.h | 10 +++
> 2 files changed, 128 insertions(+), 61 deletions(-)
>
> diff --git a/drivers/lightnvm/pblk-read.c b/drivers/lightnvm/pblk-read.c
> index 7570ff6..026c708 100644
> --- a/drivers/lightnvm/pblk-read.c
> +++ b/drivers/lightnvm/pblk-read.c
> @@ -231,74 +231,36 @@ static void pblk_end_io_read(struct nvm_rq *rqd)
> __pblk_end_io_read(pblk, rqd, true);
> }
>
> -static int pblk_partial_read(struct pblk *pblk, struct nvm_rq *rqd,
> - struct bio *orig_bio, unsigned int bio_init_idx,
> - unsigned long *read_bitmap)
> +static void pblk_end_partial_read(struct nvm_rq *rqd)
> {
> - struct pblk_sec_meta *meta_list = rqd->meta_list;
> - struct bio *new_bio;
> + struct pblk *pblk = rqd->private;
> + struct pblk_g_ctx *r_ctx = nvm_rq_to_pdu(rqd);
> + struct pblk_pr_ctx *pr_ctx = r_ctx->private;
> + struct bio *new_bio = rqd->bio;
> + struct bio *bio = pr_ctx->orig_bio;
> struct bio_vec src_bv, dst_bv;
> - void *ppa_ptr = NULL;
> - void *src_p, *dst_p;
> - dma_addr_t dma_ppa_list = 0;
> - __le64 *lba_list_mem, *lba_list_media;
> - int nr_secs = rqd->nr_ppas;
> + struct pblk_sec_meta *meta_list = rqd->meta_list;
> + int bio_init_idx = pr_ctx->bio_init_idx;
> + unsigned long *read_bitmap = &pr_ctx->bitmap;
> + int nr_secs = pr_ctx->orig_nr_secs;
> int nr_holes = nr_secs - bitmap_weight(read_bitmap, nr_secs);
> - int i, ret, hole;
> -
> - /* Re-use allocated memory for intermediate lbas */
> - lba_list_mem = (((void *)rqd->ppa_list) + pblk_dma_ppa_size);
> - lba_list_media = (((void *)rqd->ppa_list) + 2 * pblk_dma_ppa_size);
> -
> - new_bio = bio_alloc(GFP_KERNEL, nr_holes);
> -
> - if (pblk_bio_add_pages(pblk, new_bio, GFP_KERNEL, nr_holes))
> - goto err;
> -
> - if (nr_holes != new_bio->bi_vcnt) {
> - pr_err("pblk: malformed bio\n");
> - goto err;
> - }
> -
> - for (i = 0; i < nr_secs; i++)
> - lba_list_mem[i] = meta_list[i].lba;
> -
> - new_bio->bi_iter.bi_sector = 0; /* internal bio */
> - bio_set_op_attrs(new_bio, REQ_OP_READ, 0);
> -
> - rqd->bio = new_bio;
> - rqd->nr_ppas = nr_holes;
> - rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
> -
> - if (unlikely(nr_holes == 1)) {
> - ppa_ptr = rqd->ppa_list;
> - dma_ppa_list = rqd->dma_ppa_list;
> - rqd->ppa_addr = rqd->ppa_list[0];
> - }
> -
> - ret = pblk_submit_io_sync(pblk, rqd);
> - if (ret) {
> - bio_put(rqd->bio);
> - pr_err("pblk: sync read IO submission failed\n");
> - goto err;
> - }
> -
> - if (rqd->error) {
> - atomic_long_inc(&pblk->read_failed);
> -#ifdef CONFIG_NVM_DEBUG
> - pblk_print_failed_rqd(pblk, rqd, rqd->error);
> -#endif
> - }
> + __le64 *lba_list_mem, *lba_list_media;
> + void *src_p, *dst_p;
> + int hole, i;
>
> if (unlikely(nr_holes == 1)) {
> struct ppa_addr ppa;
>
> ppa = rqd->ppa_addr;
> - rqd->ppa_list = ppa_ptr;
> - rqd->dma_ppa_list = dma_ppa_list;
> + rqd->ppa_list = pr_ctx->ppa_ptr;
> + rqd->dma_ppa_list = pr_ctx->dma_ppa_list;
> rqd->ppa_list[0] = ppa;
> }
>
> + /* Re-use allocated memory for intermediate lbas */
> + lba_list_mem = (((void *)rqd->ppa_list) + pblk_dma_ppa_size);
> + lba_list_media = (((void *)rqd->ppa_list) + 2 * pblk_dma_ppa_size);
> +
> for (i = 0; i < nr_secs; i++) {
> lba_list_media[i] = meta_list[i].lba;
> meta_list[i].lba = lba_list_mem[i];
> @@ -316,7 +278,7 @@ static int pblk_partial_read(struct pblk *pblk, struct nvm_rq *rqd,
> meta_list[hole].lba = lba_list_media[i];
>
> src_bv = new_bio->bi_io_vec[i++];
> - dst_bv = orig_bio->bi_io_vec[bio_init_idx + hole];
> + dst_bv = bio->bi_io_vec[bio_init_idx + hole];
>
> src_p = kmap_atomic(src_bv.bv_page);
> dst_p = kmap_atomic(dst_bv.bv_page);
> @@ -334,19 +296,107 @@ static int pblk_partial_read(struct pblk *pblk, struct nvm_rq *rqd,
> } while (hole < nr_secs);
>
> bio_put(new_bio);
> + kfree(pr_ctx);
>
> /* restore original request */
> rqd->bio = NULL;
> rqd->nr_ppas = nr_secs;
>
> + bio_endio(bio);
> __pblk_end_io_read(pblk, rqd, false);
> - return NVM_IO_DONE;
> +}
> +
> +static int pblk_setup_partial_read(struct pblk *pblk, struct nvm_rq *rqd,
> + unsigned int bio_init_idx,
> + unsigned long *read_bitmap,
> + int nr_holes)
> +{
> + struct pblk_sec_meta *meta_list = rqd->meta_list;
> + struct pblk_g_ctx *r_ctx = nvm_rq_to_pdu(rqd);
> + struct pblk_pr_ctx *pr_ctx;
> + struct bio *new_bio, *bio = r_ctx->private;
> + __le64 *lba_list_mem;
> + int nr_secs = rqd->nr_ppas;
> + int i;
> +
> + /* Re-use allocated memory for intermediate lbas */
> + lba_list_mem = (((void *)rqd->ppa_list) + pblk_dma_ppa_size);
> +
> + new_bio = bio_alloc(GFP_KERNEL, nr_holes);
> +
> + if (pblk_bio_add_pages(pblk, new_bio, GFP_KERNEL, nr_holes))
> + goto fail;
> +
> + if (nr_holes != new_bio->bi_vcnt) {
> + pr_err("pblk: malformed bio\n");
> + goto fail;
> + }
> +
> + pr_ctx = kmalloc(sizeof(struct pblk_pr_ctx), GFP_KERNEL);
> + if (!pr_ctx)
> + goto fail_pages;
> +
> + for (i = 0; i < nr_secs; i++)
> + lba_list_mem[i] = meta_list[i].lba;
> +
> + new_bio->bi_iter.bi_sector = 0; /* internal bio */
> + bio_set_op_attrs(new_bio, REQ_OP_READ, 0);
> +
> + rqd->bio = new_bio;
> + rqd->nr_ppas = nr_holes;
> + rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
> +
> + pr_ctx->ppa_ptr = NULL;
> + pr_ctx->orig_bio = bio;
> + pr_ctx->bitmap = *read_bitmap;
> + pr_ctx->bio_init_idx = bio_init_idx;
> + pr_ctx->orig_nr_secs = nr_secs;
> + r_ctx->private = pr_ctx;
> +
> + if (unlikely(nr_holes == 1)) {
> + pr_ctx->ppa_ptr = rqd->ppa_list;
> + pr_ctx->dma_ppa_list = rqd->dma_ppa_list;
> + rqd->ppa_addr = rqd->ppa_list[0];
> + }
> + return 0;
> +
> +fail_pages:
> + pblk_bio_free_pages(pblk, new_bio, 0, new_bio->bi_vcnt);
> +fail:
> + bio_put(new_bio);
> +
> + return -ENOMEM;
> +}
> +
> +static int pblk_partial_read_bio(struct pblk *pblk, struct nvm_rq *rqd,
> + unsigned int bio_init_idx,
> + unsigned long *read_bitmap, int nr_secs)
> +{
> + int nr_holes;
> + int ret;
> +
> + nr_holes = nr_secs - bitmap_weight(read_bitmap, nr_secs);
> +
> + if (pblk_setup_partial_read(pblk, rqd, bio_init_idx, read_bitmap,
> + nr_holes))
> + return NVM_IO_ERR;
> +
> + rqd->end_io = pblk_end_partial_read;
> +
> + ret = pblk_submit_io(pblk, rqd);
> + if (ret) {
> + bio_put(rqd->bio);
> + pr_err("pblk: partial read IO submission failed\n");
> + goto err;
> + }
> +
> + return NVM_IO_OK;
>
> err:
> pr_err("pblk: failed to perform partial read\n");
>
> /* Free allocated pages in new bio */
> - pblk_bio_free_pages(pblk, orig_bio, 0, new_bio->bi_vcnt);
> + pblk_bio_free_pages(pblk, rqd->bio, 0, rqd->bio->bi_vcnt);
> __pblk_end_io_read(pblk, rqd, false);
> return NVM_IO_ERR;
> }
> @@ -480,8 +530,15 @@ int pblk_submit_read(struct pblk *pblk, struct bio *bio)
> /* The read bio request could be partially filled by the write buffer,
> * but there are some holes that need to be read from the drive.
> */
> - return pblk_partial_read(pblk, rqd, bio, bio_init_idx, &read_bitmap);
> + ret = pblk_partial_read_bio(pblk, rqd, bio_init_idx, &read_bitmap,
> + nr_secs);
> + if (ret)
> + goto fail_meta_free;
> +
> + return NVM_IO_OK;
>
> +fail_meta_free:
> + nvm_dev_dma_free(dev->parent, rqd->meta_list, rqd->dma_meta_list);
> fail_rqd_free:
> pblk_free_rqd(pblk, rqd, PBLK_READ);
> return ret;
> diff --git a/drivers/lightnvm/pblk.h b/drivers/lightnvm/pblk.h
> index 25ad026..4b28900 100644
> --- a/drivers/lightnvm/pblk.h
> +++ b/drivers/lightnvm/pblk.h
> @@ -119,6 +119,16 @@ struct pblk_g_ctx {
> u64 lba;
> };
>
> +/* partial read context */
> +struct pblk_pr_ctx {
> + struct bio *orig_bio;
> + unsigned long bitmap;
> + unsigned int orig_nr_secs;
> + unsigned int bio_init_idx;
> + void *ppa_ptr;
> + dma_addr_t dma_ppa_list;
> +};
> +
> /* Pad context */
> struct pblk_pad_rq {
> struct pblk *pblk;
> --
> 2.7.4
Thanks Heiner. The patch looks good.
Reviewed-by: Javier González <javier@xxxxxxxxxxxx>
Attachment:
signature.asc
Description: Message signed with OpenPGP
