releases/4.8.5/cxl-prevent-adapter-reset-if-an-active-context-exists.patch

   1 From 70b565bbdb911023373e035225ab10077e4ab937 Mon Sep 17 00:00:00 2001
   2 From: Vaibhav Jain <vaibhav@linux.vnet.ibm.com>
   3 Date: Fri, 14 Oct 2016 15:08:36 +0530
   4 Subject: cxl: Prevent adapter reset if an active context exists
   5
   6 From: Vaibhav Jain <vaibhav@linux.vnet.ibm.com>
   7
   8 commit 70b565bbdb911023373e035225ab10077e4ab937 upstream.
   9
  10 This patch prevents resetting the cxl adapter via sysfs in presence of
  11 one or more active cxl_context on it. This protects against an
  12 unrecoverable error caused by PSL owning a dirty cache line even after
  13 reset and host tries to touch the same cache line. In case a force reset
  14 of the card is required irrespective of any active contexts, the int
  15 value -1 can be stored in the 'reset' sysfs attribute of the card.
  16
  17 The patch introduces a new atomic_t member named contexts_num inside
  18 struct cxl that holds the number of active context attached to the card
  19 , which is checked against '0' before proceeding with the reset. To
  20 prevent against a race condition where a context is activated just after
  21 reset check is performed, the contexts_num is atomically set to '-1'
  22 after reset-check to indicate that no more contexts can be activated on
  23 the card anymore.
  24
  25 Before activating a context we atomically test if contexts_num is
  26 non-negative and if so, increment its value by one. In case the value of
  27 contexts_num is negative then it indicates that the card is about to be
  28 reset and context activation is error-ed out at that point.
  29
  30 Fixes: 62fa19d4b4fd ("cxl: Add ability to reset the card")
  31 Acked-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com>
  32 Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
  33 Signed-off-by: Vaibhav Jain <vaibhav@linux.vnet.ibm.com>
  34 Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
  35 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  36
  37 ---
  38  Documentation/ABI/testing/sysfs-class-cxl |    7 +++--
  39  drivers/misc/cxl/api.c                    |    9 ++++++
  40  drivers/misc/cxl/context.c                |    3 ++
  41  drivers/misc/cxl/cxl.h                    |   24 +++++++++++++++++
  42  drivers/misc/cxl/file.c                   |   11 +++++++
  43  drivers/misc/cxl/guest.c                  |    3 ++
  44  drivers/misc/cxl/main.c                   |   42 +++++++++++++++++++++++++++++-
  45  drivers/misc/cxl/pci.c                    |    2 +
  46  drivers/misc/cxl/sysfs.c                  |   27 ++++++++++++++++---
  47  9 files changed, 121 insertions(+), 7 deletions(-)
  48
  49 --- a/Documentation/ABI/testing/sysfs-class-cxl
  50 +++ b/Documentation/ABI/testing/sysfs-class-cxl
  51 @@ -220,8 +220,11 @@ What:           /sys/class/cxl/<card>/re
  52  Date:           October 2014
  53  Contact:        linuxppc-dev@lists.ozlabs.org
  54  Description:    write only
  55 -                Writing 1 will issue a PERST to card which may cause the card
  56 -                to reload the FPGA depending on load_image_on_perst.
  57 +                Writing 1 will issue a PERST to card provided there are no
  58 +                contexts active on any one of the card AFUs. This may cause
  59 +                the card to reload the FPGA depending on load_image_on_perst.
  60 +                Writing -1 will do a force PERST irrespective of any active
  61 +                contexts on the card AFUs.
  62  Users:         https://github.com/ibm-capi/libcxl
  63
  64  What:          /sys/class/cxl/<card>/perst_reloads_same_image (not in a guest)
  65 --- a/drivers/misc/cxl/api.c
  66 +++ b/drivers/misc/cxl/api.c
  67 @@ -229,6 +229,14 @@ int cxl_start_context(struct cxl_context
  68         if (ctx->status == STARTED)
  69                 goto out; /* already started */
  70
  71 +       /*
  72 +        * Increment the mapped context count for adapter. This also checks
  73 +        * if adapter_context_lock is taken.
  74 +        */
  75 +       rc = cxl_adapter_context_get(ctx->afu->adapter);
  76 +       if (rc)
  77 +               goto out;
  78 +
  79         if (task) {
  80                 ctx->pid = get_task_pid(task, PIDTYPE_PID);
  81                 ctx->glpid = get_task_pid(task->group_leader, PIDTYPE_PID);
  82 @@ -240,6 +248,7 @@ int cxl_start_context(struct cxl_context
  83
  84         if ((rc = cxl_ops->attach_process(ctx, kernel, wed, 0))) {
  85                 put_pid(ctx->pid);
  86 +               cxl_adapter_context_put(ctx->afu->adapter);
  87                 cxl_ctx_put();
  88                 goto out;
  89         }
  90 --- a/drivers/misc/cxl/context.c
  91 +++ b/drivers/misc/cxl/context.c
  92 @@ -238,6 +238,9 @@ int __detach_context(struct cxl_context
  93         put_pid(ctx->glpid);
  94
  95         cxl_ctx_put();
  96 +
  97 +       /* Decrease the attached context count on the adapter */
  98 +       cxl_adapter_context_put(ctx->afu->adapter);
  99         return 0;
 100  }
 101
 102 --- a/drivers/misc/cxl/cxl.h
 103 +++ b/drivers/misc/cxl/cxl.h
 104 @@ -615,6 +615,14 @@ struct cxl {
 105         bool perst_select_user;
 106         bool perst_same_image;
 107         bool psl_timebase_synced;
 108 +
 109 +       /*
 110 +        * number of contexts mapped on to this card. Possible values are:
 111 +        * >0: Number of contexts mapped and new one can be mapped.
 112 +        *  0: No active contexts and new ones can be mapped.
 113 +        * -1: No contexts mapped and new ones cannot be mapped.
 114 +        */
 115 +       atomic_t contexts_num;
 116  };
 117
 118  int cxl_pci_alloc_one_irq(struct cxl *adapter);
 119 @@ -940,4 +948,20 @@ bool cxl_pci_is_vphb_device(struct pci_d
 120
 121  /* decode AFU error bits in the PSL register PSL_SERR_An */
 122  void cxl_afu_decode_psl_serr(struct cxl_afu *afu, u64 serr);
 123 +
 124 +/*
 125 + * Increments the number of attached contexts on an adapter.
 126 + * In case an adapter_context_lock is taken the return -EBUSY.
 127 + */
 128 +int cxl_adapter_context_get(struct cxl *adapter);
 129 +
 130 +/* Decrements the number of attached contexts on an adapter */
 131 +void cxl_adapter_context_put(struct cxl *adapter);
 132 +
 133 +/* If no active contexts then prevents contexts from being attached */
 134 +int cxl_adapter_context_lock(struct cxl *adapter);
 135 +
 136 +/* Unlock the contexts-lock if taken. Warn and force unlock otherwise */
 137 +void cxl_adapter_context_unlock(struct cxl *adapter);
 138 +
 139  #endif
 140 --- a/drivers/misc/cxl/file.c
 141 +++ b/drivers/misc/cxl/file.c
 142 @@ -205,11 +205,22 @@ static long afu_ioctl_start_work(struct
 143         ctx->pid = get_task_pid(current, PIDTYPE_PID);
 144         ctx->glpid = get_task_pid(current->group_leader, PIDTYPE_PID);
 145
 146 +       /*
 147 +        * Increment the mapped context count for adapter. This also checks
 148 +        * if adapter_context_lock is taken.
 149 +        */
 150 +       rc = cxl_adapter_context_get(ctx->afu->adapter);
 151 +       if (rc) {
 152 +               afu_release_irqs(ctx, ctx);
 153 +               goto out;
 154 +       }
 155 +
 156         trace_cxl_attach(ctx, work.work_element_descriptor, work.num_interrupts, amr);
 157
 158         if ((rc = cxl_ops->attach_process(ctx, false, work.work_element_descriptor,
 159                                                         amr))) {
 160                 afu_release_irqs(ctx, ctx);
 161 +               cxl_adapter_context_put(ctx->afu->adapter);
 162                 goto out;
 163         }
 164
 165 --- a/drivers/misc/cxl/guest.c
 166 +++ b/drivers/misc/cxl/guest.c
 167 @@ -1152,6 +1152,9 @@ struct cxl *cxl_guest_init_adapter(struc
 168         if ((rc = cxl_sysfs_adapter_add(adapter)))
 169                 goto err_put1;
 170
 171 +       /* release the context lock as the adapter is configured */
 172 +       cxl_adapter_context_unlock(adapter);
 173 +
 174         return adapter;
 175
 176  err_put1:
 177 --- a/drivers/misc/cxl/main.c
 178 +++ b/drivers/misc/cxl/main.c
 179 @@ -243,8 +243,10 @@ struct cxl *cxl_alloc_adapter(void)
 180         if (dev_set_name(&adapter->dev, "card%i", adapter->adapter_num))
 181                 goto err2;
 182
 183 -       return adapter;
 184 +       /* start with context lock taken */
 185 +       atomic_set(&adapter->contexts_num, -1);
 186
 187 +       return adapter;
 188  err2:
 189         cxl_remove_adapter_nr(adapter);
 190  err1:
 191 @@ -286,6 +288,44 @@ int cxl_afu_select_best_mode(struct cxl_
 192         return 0;
 193  }
 194
 195 +int cxl_adapter_context_get(struct cxl *adapter)
 196 +{
 197 +       int rc;
 198 +
 199 +       rc = atomic_inc_unless_negative(&adapter->contexts_num);
 200 +       return rc >= 0 ? 0 : -EBUSY;
 201 +}
 202 +
 203 +void cxl_adapter_context_put(struct cxl *adapter)
 204 +{
 205 +       atomic_dec_if_positive(&adapter->contexts_num);
 206 +}
 207 +
 208 +int cxl_adapter_context_lock(struct cxl *adapter)
 209 +{
 210 +       int rc;
 211 +       /* no active contexts -> contexts_num == 0 */
 212 +       rc = atomic_cmpxchg(&adapter->contexts_num, 0, -1);
 213 +       return rc ? -EBUSY : 0;
 214 +}
 215 +
 216 +void cxl_adapter_context_unlock(struct cxl *adapter)
 217 +{
 218 +       int val = atomic_cmpxchg(&adapter->contexts_num, -1, 0);
 219 +
 220 +       /*
 221 +        * contexts lock taken -> contexts_num == -1
 222 +        * If not true then show a warning and force reset the lock.
 223 +        * This will happen when context_unlock was requested without
 224 +        * doing a context_lock.
 225 +        */
 226 +       if (val != -1) {
 227 +               atomic_set(&adapter->contexts_num, 0);
 228 +               WARN(1, "Adapter context unlocked with %d active contexts",
 229 +                    val);
 230 +       }
 231 +}
 232 +
 233  static int __init init_cxl(void)
 234  {
 235         int rc = 0;
 236 --- a/drivers/misc/cxl/pci.c
 237 +++ b/drivers/misc/cxl/pci.c
 238 @@ -1484,6 +1484,8 @@ static int cxl_configure_adapter(struct
 239         if ((rc = cxl_native_register_psl_err_irq(adapter)))
 240                 goto err;
 241
 242 +       /* Release the context lock as adapter is configured */
 243 +       cxl_adapter_context_unlock(adapter);
 244         return 0;
 245
 246  err:
 247 --- a/drivers/misc/cxl/sysfs.c
 248 +++ b/drivers/misc/cxl/sysfs.c
 249 @@ -75,12 +75,31 @@ static ssize_t reset_adapter_store(struc
 250         int val;
 251
 252         rc = sscanf(buf, "%i", &val);
 253 -       if ((rc != 1) || (val != 1))
 254 +       if ((rc != 1) || (val != 1 && val != -1))
 255                 return -EINVAL;
 256
 257 -       if ((rc = cxl_ops->adapter_reset(adapter)))
 258 -               return rc;
 259 -       return count;
 260 +       /*
 261 +        * See if we can lock the context mapping that's only allowed
 262 +        * when there are no contexts attached to the adapter. Once
 263 +        * taken this will also prevent any context from getting activated.
 264 +        */
 265 +       if (val == 1) {
 266 +               rc =  cxl_adapter_context_lock(adapter);
 267 +               if (rc)
 268 +                       goto out;
 269 +
 270 +               rc = cxl_ops->adapter_reset(adapter);
 271 +               /* In case reset failed release context lock */
 272 +               if (rc)
 273 +                       cxl_adapter_context_unlock(adapter);
 274 +
 275 +       } else if (val == -1) {
 276 +               /* Perform a forced adapter reset */
 277 +               rc = cxl_ops->adapter_reset(adapter);
 278 +       }
 279 +
 280 +out:
 281 +       return rc ? rc : count;
 282  }
 283
 284  static ssize_t load_image_on_perst_show(struct device *device,