releases/4.4.75/nvme-quirk-add-a-delay-before-checking-for-adapter-readiness.patch

   1 From 54adc01055b75ec8769c5a36574c7a0895c0c0b2 Mon Sep 17 00:00:00 2001
   2 From: "Guilherme G. Piccoli" <gpiccoli@linux.vnet.ibm.com>
   3 Date: Tue, 14 Jun 2016 18:22:41 -0300
   4 Subject: nvme/quirk: Add a delay before checking for adapter readiness
   5
   6 From: Guilherme G. Piccoli <gpiccoli@linux.vnet.ibm.com>
   7
   8 commit 54adc01055b75ec8769c5a36574c7a0895c0c0b2 upstream.
   9
  10 When disabling the controller, the specification says the register
  11 NVME_REG_CC should be written and then driver needs to wait the
  12 adapter to be ready, which is checked by reading another register
  13 bit (NVME_CSTS_RDY). There's a timeout validation in this checking,
  14 so in case this timeout is reached the driver gives up and removes
  15 the adapter from the system.
  16
  17 After a firmware activation procedure, the PCI_DEVICE(0x1c58, 0x0003)
  18 (HGST adapter) end up being removed if we issue a reset_controller,
  19 because driver keeps verifying the NVME_REG_CSTS until the timeout is
  20 reached. This patch adds a necessary quirk for this adapter, by
  21 introducing a delay before nvme_wait_ready(), so the reset procedure
  22 is able to be completed. This quirk is needed because just increasing
  23 the timeout is not enough in case of this adapter - the driver must
  24 wait before start reading NVME_REG_CSTS register on this specific
  25 device.
  26
  27 Signed-off-by: Guilherme G. Piccoli <gpiccoli@linux.vnet.ibm.com>
  28 Reviewed-by: Christoph Hellwig <hch@lst.de>
  29 Signed-off-by: Jens Axboe <axboe@fb.com>
  30 [mauricfo: backport to v4.4.70 without nvme quirk handling & nvme_ctrl]
  31 Signed-off-by: Mauricio Faria de Oliveira <mauricfo@linux.vnet.ibm.com>
  32 Tested-by: Narasimhan Vaidyanathan <vnarasimhan@in.ibm.com>
  33 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  34
  35 ---
  36  drivers/nvme/host/nvme.h |    7 +++++++
  37  drivers/nvme/host/pci.c  |   10 ++++++++++
  38  2 files changed, 17 insertions(+)
  39
  40 --- a/drivers/nvme/host/nvme.h
  41 +++ b/drivers/nvme/host/nvme.h
  42 @@ -27,6 +27,13 @@ enum {
  43         NVME_NS_LIGHTNVM        = 1,
  44  };
  45
  46 +/* The below value is the specific amount of delay needed before checking
  47 + * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
  48 + * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
  49 + * found empirically.
  50 + */
  51 +#define NVME_QUIRK_DELAY_AMOUNT                2000
  52 +
  53  /*
  54   * Represents an NVM Express device.  Each nvme_dev is a PCI function.
  55   */
  56 --- a/drivers/nvme/host/pci.c
  57 +++ b/drivers/nvme/host/pci.c
  58 @@ -1633,10 +1633,20 @@ static int nvme_wait_ready(struct nvme_d
  59   */
  60  static int nvme_disable_ctrl(struct nvme_dev *dev, u64 cap)
  61  {
  62 +       struct pci_dev *pdev = to_pci_dev(dev->dev);
  63 +
  64         dev->ctrl_config &= ~NVME_CC_SHN_MASK;
  65         dev->ctrl_config &= ~NVME_CC_ENABLE;
  66         writel(dev->ctrl_config, &dev->bar->cc);
  67
  68 +       /* Checking for dev->tagset is a trick to avoid sleeping on module
  69 +        * load, since we only need the quirk on reset_controller. Notice
  70 +        * that the HGST device needs this delay only in firmware activation
  71 +        * procedure; unfortunately we have no (easy) way to verify this.
  72 +        */
  73 +       if (pdev->vendor == 0x1c58 && pdev->device == 0x0003 && dev->tagset)
  74 +               msleep(NVME_QUIRK_DELAY_AMOUNT);
  75 +
  76         return nvme_wait_ready(dev, cap, false);
  77  }
  78