[Accel-config] Re: [PATCH 1/2] accel-config: Add APIs to get details of last error and associated devices
Dave Jiang <dave.jiang at intel.com>
| Newsgroups | dev.linux.lists.accel-config |
|---|---|
| Message-ID | <[email protected]> |
On 7/12/2021 1:40 PM, Thomas, Ramesh wrote:
> On Mon, Jul 12, 2021 at 01:26:52PM -0700, Dave Jiang wrote:
>> On 7/9/2021 10:47 AM, ramesh.thomas(a)intel.com wrote:
>>> From: Ramesh Thomas <ramesh.thomas(a)intel.com>
>>>
>>> Capture a snapshot of error condition containing cmd_status and the
>>> associated devices. Provide APIs to return the captured details. Handles
>>> both hw and sw error codes stored in sysfs cmd_status field of devices.
>>> Following APIs are added:
>>> - accfg_ctx_get_last_error;
>>> - accfg_ctx_get_last_error_str;
>>> - accfg_ctx_get_last_error_device;
>>> - accfg_ctx_get_last_error_wq;
>>> - accfg_ctx_get_last_error_group;
>>> - accfg_ctx_get_last_error_engine;
>>>
>>> Signed-off-by: Ramesh Thomas <ramesh.thomas(a)intel.com>
>>> ---
>>> accfg/lib/libaccel-config.sym | 10 +++
>>> accfg/lib/libaccfg.c | 156 ++++++++++++++++++++++++++++++----
>>> accfg/lib/private.h | 9 ++
>>> accfg/libaccel_config.h | 9 +-
>>> 4 files changed, 164 insertions(+), 20 deletions(-)
>>>
>>> diff --git a/accfg/lib/libaccel-config.sym b/accfg/lib/libaccel-config.sym
>>> index aecc98b..679c597 100644
>>> --- a/accfg/lib/libaccel-config.sym
>>> +++ b/accfg/lib/libaccel-config.sym
>>> @@ -149,3 +149,13 @@ LIBACCFG_9 {
>>> global:
>>> accfg_device_get_op_cap;
>>> } LIBACCFG_8;
>>> +
>>> +LIBACCFG_10 {
>>> +global:
>>> + accfg_ctx_get_last_error;
>>> + accfg_ctx_get_last_error_str;
>>> + accfg_ctx_get_last_error_device;
>>> + accfg_ctx_get_last_error_wq;
>>> + accfg_ctx_get_last_error_group;
>>> + accfg_ctx_get_last_error_engine;
>>> +} LIBACCFG_9;
>>> diff --git a/accfg/lib/libaccfg.c b/accfg/lib/libaccfg.c
>>> index f53751e..00314b0 100644
>>> --- a/accfg/lib/libaccfg.c
>>> +++ b/accfg/lib/libaccfg.c
>>> @@ -70,6 +70,7 @@ ACCFG_EXPORT char *accfg_mdev_basenames[] = {
>>>
>>> enum {
>>> ACCFG_CMD_STATUS_MAX = 0x45,
>>> + ACCFG_CMD_STATUS_ERROR = 0x80010000,
>>> };
>>>
>>> const char *accfg_device_cmd_status[] = {
>>> @@ -105,9 +106,43 @@ const char *accfg_device_cmd_status[] = {
>>> [0x42] = "No interrupt handle available",
>>> [0x43] = "No interrupt handles associated with the index",
>>> [0x44] = "No revoked handles associted with the index",
>>> - [ACCFG_CMD_STATUS_MAX] = "",
>>> + [ACCFG_CMD_STATUS_MAX] = "Unknown error",
>>> };
>>>
>>> +const char *accfg_sw_cmd_status[] = {
>>> + [0x01] = "Error reading cmd_status",
>>> + [0x02] = "DMA device registration error",
>>> + [0x03] = "wq error - group not set",
>>> + [0x04] = "wq error - name not set",
>>> + [0x05] = "wq error - no SVM support and needed",
>>> + [0x06] = "wq error - threshold not set",
>>> + [0x07] = "wq portal mapping error",
>>> + [0x08] = "wq resource allocation error",
>>> + [0x09] = "wq percpu counter allocation error",
>>> + [0x0a] = "wq DMA channel registration error",
>>> + [0x0b] = "wq char dev registration error",
>>> + [0x0c] = "wq error - no shared wq support",
>>> + [0x0d] = "wq error - no wqs configured",
>>> + [0x0e] = "wq error - size not set",
>>> +};
>>> +
>>> +static void save_last_error(struct accfg_device *device, struct accfg_wq *wq,
>>> + struct accfg_group *group, struct accfg_engine *engine)
>>> +{
>>> + struct accfg_ctx *ctx = device->ctx;
>>> +
>>> + if (!ctx->error_ctx)
>>> + ctx->error_ctx = calloc(1, sizeof(struct accfg_error_ctx));
>>> +
>>> + if (ctx->error_ctx) {
>>> + ctx->error_ctx->cmd_status = accfg_device_get_cmd_status(device);
>>> + ctx->error_ctx->device = device;
>>> + ctx->error_ctx->wq = wq;
>>> + ctx->error_ctx->group = group;
>>> + ctx->error_ctx->engine = engine;
>>> + }
>>> +}
>>> +
>>> static inline bool is_mdev_registered(struct accfg_device *device)
>>> {
>>> return device->mdev_path && !access(device->mdev_path, R_OK);
>>> @@ -251,6 +286,7 @@ static void free_context(struct accfg_ctx *ctx)
>>>
>>> list_for_each_safe(&ctx->devices, device, _b, list)
>>> free_device(device, &ctx->devices);
>>> + free(ctx->error_ctx);
>>> free(ctx);
>>> }
>>>
>>> @@ -1082,6 +1118,7 @@ ACCFG_EXPORT int accfg_create_mdev(struct accfg_device *device,
>>> rc = sysfs_write_attr(ctx, mdev_path, uuid_str);
>>> if (rc < 0) {
>>> err(ctx, "create mdev failed %d\n", rc);
>>> + save_last_error(device, NULL, NULL, NULL);
>>> goto create_err;
>>> }
>>>
>>> @@ -1105,8 +1142,10 @@ static int accfg_device_mdev_remove(struct accfg_device *device,
>>> uuid_unparse(mdev->uuid, uuid_str);
>>> sprintf(mdev_path, "%s/%s/remove", device->mdev_path, uuid_str);
>>> rc = sysfs_write_attr(ctx, mdev_path, "1");
>>> - if (rc < 0)
>>> + if (rc < 0) {
>>> + save_last_error(device, NULL, NULL, NULL);
>>> return rc;
>>> + }
>>>
>>> list_del(&mdev->list);
>>> free(mdev);
>>> @@ -1355,6 +1394,7 @@ ACCFG_EXPORT enum accfg_device_state accfg_device_get_state(
>>> err(ctx, "%s: sysfs_read_attr failed '%s': %s\n",
>>> __func__, device->device_path,
>>> strerror(errno));
>>> + save_last_error(device, NULL, NULL, NULL);
>>> return ACCFG_DEVICE_UNKNOWN;
>>> }
>>>
>>> @@ -1414,6 +1454,7 @@ ACCFG_EXPORT int accfg_device_get_clients(struct accfg_device *device)
>>> if (sysfs_read_attr(ctx, path, buf) < 0) {
>>> err(ctx, "%s: retrieve clients failed '%s': %s\n",
>>> __func__, path, strerror(errno));
>>> + save_last_error(device, NULL, NULL, NULL);
>>> return -errno;
>>> }
>>>
>>> @@ -1448,6 +1489,7 @@ ACCFG_EXPORT int accfg_device_set_token_limit(struct accfg_device
>> *dev, int val)
>>> if (sysfs_write_attr(ctx, path, buf) < 0) {
>>> err(ctx, "%s: write failed: %s\n",
>>> accfg_device_get_devname(dev), strerror(errno));
>>> + save_last_error(dev, NULL, NULL, NULL);
>>> return -errno;
>>> }
>>>
>>> @@ -1476,8 +1518,10 @@ ACCFG_EXPORT int accfg_device_is_active(struct accfg_device *device)
>>> return 0;
>>> }
>>>
>>> - if (sysfs_read_attr(ctx, path, buf) < 0)
>>> + if (sysfs_read_attr(ctx, path, buf) < 0) {
>>> + save_last_error(device, NULL, NULL, NULL);
>>> return 0;
>>> + }
>>>
>>> if (strcmp(buf, "enabled") == 0)
>>> return 1;
>>> @@ -1485,17 +1529,16 @@ ACCFG_EXPORT int accfg_device_is_active(struct accfg_device *device)
>>> return 0;
>>> }
>>>
>>> -ACCFG_EXPORT int accfg_device_get_cmd_status(struct accfg_device *device)
>>> +ACCFG_EXPORT unsigned int accfg_device_get_cmd_status(struct accfg_device *device)
>>> {
>>> struct accfg_ctx *ctx;
>>> long status;
>>> char *path;
>>> int len;
>>> char buf[SYSFS_ATTR_SIZE], *end_ptr;
>>> - int rc;
>>>
>>> if (!device)
>>> - return -EINVAL;
>>> + return ACCFG_CMD_STATUS_ERROR;
>>>
>>> ctx = accfg_device_get_ctx(device);
>>> path = device->device_buf;
>>> @@ -1504,31 +1547,96 @@ ACCFG_EXPORT int accfg_device_get_cmd_status(struct accfg_device
>> *device)
>>> if (snprintf(path, len, "%s/cmd_status", device->device_path) >= len) {
>>> err(ctx, "%s: buffer too small!\n",
>>> accfg_device_get_devname(device));
>>> - return -ENOMEM;
>>> + return ACCFG_CMD_STATUS_ERROR;
>>> }
>>>
>>> - rc = sysfs_read_attr(ctx, path, buf);
>>> - if (rc < 0)
>>> - return rc;
>>> + if (sysfs_read_attr(ctx, path, buf))
>>> + return ACCFG_CMD_STATUS_ERROR;
>>>
>>> status = strtol(buf, &end_ptr, 0);
>>> if (errno == ERANGE || end_ptr == buf)
>>> - return -EIO;
>>> + return ACCFG_CMD_STATUS_ERROR;
>>>
>>> - return (int)status;
>>> + return (unsigned int)status;
>>> }
>>>
>>> -ACCFG_EXPORT const char * accfg_device_get_cmd_status_str(struct accfg_device *device)
>>> +static const char *get_cmd_status_str(unsigned int status)
>>> {
>>> - int status;
>>> + unsigned int sw_status;
>>>
>>> - status = accfg_device_get_cmd_status(device);
>>> - if (status < 0 || status >= ACCFG_CMD_STATUS_MAX)
>>> - return NULL;
>>> + if (status & 0x80000000) {
>>> + sw_status = (status & ~0x80000000) >> 16;
>>
>> Magic number. Do I need to add a mask define to the header?
> Or add a comment stating device error codes are in the lower 16 bits and
> sw error codes are in the higher bits with the highest bit set.
Looks like I already defined the mask in the user header:
IDXD_SCMD_SOFTERR_MASK
I've also added a shift define. Just fetch djiang5/idxd-upstream-next again.