| Message ID | 20191016064204.GA30249@nox.fritz.box |
|---|---|
| State | Accepted |
| Commit | 2f83481dff9c4f253a6ac341911d78d4984ca07b |
| Delegated to: | Tom Rini |
| Headers | show |
| Series | [U-Boot,v2] nvme: use page-aligned buffer for identify command | expand |
On Wed, Oct 16, 2019 at 2:42 PM Patrick Wildt <patrick@blueri.se> wrote: > > Change the stack-allocated buffer for the identification command > to explicitly allocate page-aligned buffers. Even though the spec > seems to allow having admin queue commands on non page-aligned > buffers, it seems to not be possible on my i.MX8MQ board with a > a Silicon Power P34A80. Since all of the NVMe drivers I have seen > always do admin commands on a page-aligned buffer, which does work > on my system, it makes sense for us to do that as well. > > Signed-off-by: Patrick Wildt <patrick@blueri.se> > --- > Changes for v2: > - use dev->page_size instead of hardcoded value > > drivers/nvme/nvme.c | 24 ++++++++++++++++++------ > 1 file changed, 18 insertions(+), 6 deletions(-) > Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
On Wed, Oct 16, 2019 at 08:42:04AM +0200, Patrick Wildt wrote: > Change the stack-allocated buffer for the identification command > to explicitly allocate page-aligned buffers. Even though the spec > seems to allow having admin queue commands on non page-aligned > buffers, it seems to not be possible on my i.MX8MQ board with a > a Silicon Power P34A80. Since all of the NVMe drivers I have seen > always do admin commands on a page-aligned buffer, which does work > on my system, it makes sense for us to do that as well. > > Signed-off-by: Patrick Wildt <patrick@blueri.se> > Reviewed-by: Bin Meng <bmeng.cn@gmail.com> Applied to u-boot/master, thanks!
diff --git a/drivers/nvme/nvme.c b/drivers/nvme/nvme.c index ee6b581d9e..dee92b613d 100644 --- a/drivers/nvme/nvme.c +++ b/drivers/nvme/nvme.c @@ -580,14 +580,19 @@ static int nvme_setup_io_queues(struct nvme_dev *dev) static int nvme_get_info_from_identify(struct nvme_dev *dev) { - ALLOC_CACHE_ALIGN_BUFFER(char, buf, sizeof(struct nvme_id_ctrl)); - struct nvme_id_ctrl *ctrl = (struct nvme_id_ctrl *)buf; + struct nvme_id_ctrl *ctrl; int ret; int shift = NVME_CAP_MPSMIN(dev->cap) + 12; + ctrl = memalign(dev->page_size, sizeof(struct nvme_id_ctrl)); + if (!ctrl) + return -ENOMEM; + ret = nvme_identify(dev, 0, 1, (dma_addr_t)(long)ctrl); - if (ret) + if (ret) { + free(ctrl); return -EIO; + } dev->nn = le32_to_cpu(ctrl->nn); dev->vwc = ctrl->vwc; @@ -618,6 +623,7 @@ static int nvme_get_info_from_identify(struct nvme_dev *dev) dev->max_transfer_shift = 20; } + free(ctrl); return 0; } @@ -658,16 +664,21 @@ static int nvme_blk_probe(struct udevice *udev) struct blk_desc *desc = dev_get_uclass_platdata(udev); struct nvme_ns *ns = dev_get_priv(udev); u8 flbas; - ALLOC_CACHE_ALIGN_BUFFER(char, buf, sizeof(struct nvme_id_ns)); - struct nvme_id_ns *id = (struct nvme_id_ns *)buf; struct pci_child_platdata *pplat; + struct nvme_id_ns *id; + + id = memalign(ndev->page_size, sizeof(struct nvme_id_ns)); + if (!id) + return -ENOMEM; memset(ns, 0, sizeof(*ns)); ns->dev = ndev; /* extract the namespace id from the block device name */ ns->ns_id = trailing_strtol(udev->name) + 1; - if (nvme_identify(ndev, ns->ns_id, 0, (dma_addr_t)(long)id)) + if (nvme_identify(ndev, ns->ns_id, 0, (dma_addr_t)(long)id)) { + free(id); return -EIO; + } memcpy(&ns->eui64, &id->eui64, sizeof(id->eui64)); flbas = id->flbas & NVME_NS_FLBAS_LBA_MASK; @@ -686,6 +697,7 @@ static int nvme_blk_probe(struct udevice *udev) memcpy(desc->product, ndev->serial, sizeof(ndev->serial)); memcpy(desc->revision, ndev->firmware_rev, sizeof(ndev->firmware_rev)); + free(id); return 0; }
Change the stack-allocated buffer for the identification command to explicitly allocate page-aligned buffers. Even though the spec seems to allow having admin queue commands on non page-aligned buffers, it seems to not be possible on my i.MX8MQ board with a a Silicon Power P34A80. Since all of the NVMe drivers I have seen always do admin commands on a page-aligned buffer, which does work on my system, it makes sense for us to do that as well. Signed-off-by: Patrick Wildt <patrick@blueri.se> --- Changes for v2: - use dev->page_size instead of hardcoded value drivers/nvme/nvme.c | 24 ++++++++++++++++++------ 1 file changed, 18 insertions(+), 6 deletions(-)