[PATCH v2 10/17] libnvdimm, pmem: use REQ_FUA, REQ_FLUSH for nvdimm_flush()

[Dan Williams]

Given that nvdimm_flush() has higher overhead than wmb_pmem() (pointer
chasing through nd_region), and that we otherwise assume a platform has
ADR capability when flush hints are not present, move nvdimm_flush() to
REQ_FLUSH context.

Cc: Ross Zwisler <ross.zwis...@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.willi...@intel.com>
---
 drivers/nvdimm/pmem.c |   24 +++++++++++++++++-------
 1 file changed, 17 insertions(+), 7 deletions(-)

diff --git a/drivers/nvdimm/pmem.c b/drivers/nvdimm/pmem.c
index e303655f243e..18cd95719da0 100644
--- a/drivers/nvdimm/pmem.c
+++ b/drivers/nvdimm/pmem.c
@@ -113,6 +113,11 @@ static int pmem_do_bvec(struct pmem_device *pmem, struct 
page *page,
        return rc;
 }
 
+/* account for REQ_FLUSH rename, replace with REQ_PREFLUSH after v4.8-rc1 */
+#ifndef REQ_FLUSH
+#define REQ_FLUSH REQ_PREFLUSH
+#endif
+
 static blk_qc_t pmem_make_request(struct request_queue *q, struct bio *bio)
 {
        int rc = 0;
@@ -121,6 +126,10 @@ static blk_qc_t pmem_make_request(struct request_queue *q, 
struct bio *bio)
        struct bio_vec bvec;
        struct bvec_iter iter;
        struct pmem_device *pmem = q->queuedata;
+       struct nd_region *nd_region = to_region(pmem);
+
+       if (bio->bi_rw & REQ_FLUSH)
+               nvdimm_flush(nd_region);
 
        do_acct = nd_iostat_start(bio, &start);
        bio_for_each_segment(bvec, bio, iter) {
@@ -135,8 +144,8 @@ static blk_qc_t pmem_make_request(struct request_queue *q, 
struct bio *bio)
        if (do_acct)
                nd_iostat_end(bio, start);
 
-       if (bio_data_dir(bio))
-               nvdimm_flush(to_region(pmem));
+       if (bio->bi_rw & REQ_FUA)
+               nvdimm_flush(nd_region);
 
        bio_endio(bio);
        return BLK_QC_T_NONE;
@@ -149,8 +158,6 @@ static int pmem_rw_page(struct block_device *bdev, sector_t 
sector,
        int rc;
 
        rc = pmem_do_bvec(pmem, page, PAGE_SIZE, 0, rw, sector);
-       if (rw & WRITE)
-               nvdimm_flush(to_region(pmem));
 
        /*
         * The ->rw_page interface is subtle and tricky.  The core
@@ -209,9 +216,9 @@ static int pmem_attach_disk(struct device *dev,
        struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
        struct nd_region *nd_region = to_nd_region(dev->parent);
        struct vmem_altmap __altmap, *altmap = NULL;
+       int nid = dev_to_node(dev), has_flush;
        struct resource *res = &nsio->res;
        struct nd_pfn *nd_pfn = NULL;
-       int nid = dev_to_node(dev);
        struct nd_pfn_sb *pfn_sb;
        struct pmem_device *pmem;
        struct resource pfn_res;
@@ -237,8 +244,6 @@ static int pmem_attach_disk(struct device *dev,
        dev_set_drvdata(dev, pmem);
        pmem->phys_addr = res->start;
        pmem->size = resource_size(res);
-       if (nvdimm_has_flush(nd_region) < 0)
-               dev_warn(dev, "unable to guarantee persistence of writes\n");
 
        if (!devm_request_mem_region(dev, res->start, resource_size(res),
                                dev_name(dev))) {
@@ -279,6 +284,11 @@ static int pmem_attach_disk(struct device *dev,
                return PTR_ERR(addr);
        pmem->virt_addr = (void __pmem *) addr;
 
+       has_flush = nvdimm_has_flush(nd_region);
+       if (has_flush < 0)
+               dev_warn(dev, "unable to guarantee persistence of writes\n");
+       else if (has_flush > 0)
+               blk_queue_write_cache(q, true, true);
        blk_queue_make_request(q, pmem_make_request);
        blk_queue_physical_block_size(q, PAGE_SIZE);
        blk_queue_max_hw_sectors(q, UINT_MAX);