edac: reduce stack pressure by using a pre-allocated buffer
authorMauro Carvalho Chehab <mchehab@redhat.com>
Thu, 21 Feb 2013 16:36:45 +0000 (13:36 -0300)
committerMauro Carvalho Chehab <mchehab@redhat.com>
Thu, 21 Feb 2013 16:48:45 +0000 (13:48 -0300)
The number of variables at the stack is too big.
Reduces the stack usage by using a pre-allocated error
buffer.

Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
drivers/edac/edac_mc.c
include/linux/edac.h

index 34eb9703ed33a52b5e83227c8b7ccf01f3a55034..4f18dd75593983d2310c9149f82a17dd43157bb2 100644 (file)
@@ -1065,7 +1065,6 @@ static void edac_ue_error(struct mem_ctl_info *mci,
        edac_inc_ue_error(mci, enable_per_layer_report, pos, error_count);
 }
 
-#define OTHER_LABEL " or "
 
 /**
  * edac_mc_handle_error - reports a memory event to userspace
@@ -1097,19 +1096,28 @@ void edac_mc_handle_error(const enum hw_event_mc_err_type type,
                          const char *msg,
                          const char *other_detail)
 {
-       /* FIXME: too much for stack: move it to some pre-alocated area */
-       char detail[80], location[80];
-       char label[(EDAC_MC_LABEL_LEN + 1 + sizeof(OTHER_LABEL)) * mci->tot_dimms];
+       char detail[80];
        char *p;
        int row = -1, chan = -1;
        int pos[EDAC_MAX_LAYERS] = { top_layer, mid_layer, low_layer };
-       int i;
-       long grain;
-       bool enable_per_layer_report = false;
+       int i, n_labels = 0;
        u8 grain_bits;
+       struct edac_raw_error_desc *e = &mci->error_desc;
 
        edac_dbg(3, "MC%d\n", mci->mc_idx);
 
+       /* Fills the error report buffer */
+       memset(e, 0, sizeof (*e));
+       e->error_count = error_count;
+       e->top_layer = top_layer;
+       e->mid_layer = mid_layer;
+       e->low_layer = low_layer;
+       e->page_frame_number = page_frame_number;
+       e->offset_in_page = offset_in_page;
+       e->syndrome = syndrome;
+       e->msg = msg;
+       e->other_detail = other_detail;
+
        /*
         * Check if the event report is consistent and if the memory
         * location is known. If it is known, enable_per_layer_report will be
@@ -1132,7 +1140,7 @@ void edac_mc_handle_error(const enum hw_event_mc_err_type type,
                        pos[i] = -1;
                }
                if (pos[i] >= 0)
-                       enable_per_layer_report = true;
+                       e->enable_per_layer_report = true;
        }
 
        /*
@@ -1146,8 +1154,7 @@ void edac_mc_handle_error(const enum hw_event_mc_err_type type,
         * where each memory belongs to a separate channel within the same
         * branch.
         */
-       grain = 0;
-       p = label;
+       p = e->label;
        *p = '\0';
 
        for (i = 0; i < mci->tot_dimms; i++) {
@@ -1161,8 +1168,8 @@ void edac_mc_handle_error(const enum hw_event_mc_err_type type,
                        continue;
 
                /* get the max grain, over the error match range */
-               if (dimm->grain > grain)
-                       grain = dimm->grain;
+               if (dimm->grain > e->grain)
+                       e->grain = dimm->grain;
 
                /*
                 * If the error is memory-controller wide, there's no need to
@@ -1170,8 +1177,13 @@ void edac_mc_handle_error(const enum hw_event_mc_err_type type,
                 * channel/memory controller/...  may be affected.
                 * Also, don't show errors for empty DIMM slots.
                 */
-               if (enable_per_layer_report && dimm->nr_pages) {
-                       if (p != label) {
+               if (e->enable_per_layer_report && dimm->nr_pages) {
+                       if (n_labels >= EDAC_MAX_LABELS) {
+                               e->enable_per_layer_report = false;
+                               break;
+                       }
+                       n_labels++;
+                       if (p != e->label) {
                                strcpy(p, OTHER_LABEL);
                                p += strlen(OTHER_LABEL);
                        }
@@ -1198,12 +1210,12 @@ void edac_mc_handle_error(const enum hw_event_mc_err_type type,
                }
        }
 
-       if (!enable_per_layer_report) {
-               strcpy(label, "any memory");
+       if (!e->enable_per_layer_report) {
+               strcpy(e->label, "any memory");
        } else {
                edac_dbg(4, "csrow/channel to increment: (%d,%d)\n", row, chan);
-               if (p == label)
-                       strcpy(label, "unknown memory");
+               if (p == e->label)
+                       strcpy(e->label, "unknown memory");
                if (type == HW_EVENT_ERR_CORRECTED) {
                        if (row >= 0) {
                                mci->csrows[row]->ce_count += error_count;
@@ -1216,7 +1228,7 @@ void edac_mc_handle_error(const enum hw_event_mc_err_type type,
        }
 
        /* Fill the RAM location data */
-       p = location;
+       p = e->location;
 
        for (i = 0; i < mci->n_layers; i++) {
                if (pos[i] < 0)
@@ -1226,32 +1238,35 @@ void edac_mc_handle_error(const enum hw_event_mc_err_type type,
                             edac_layer_name[mci->layers[i].type],
                             pos[i]);
        }
-       if (p > location)
+       if (p > e->location)
                *(p - 1) = '\0';
 
        /* Report the error via the trace interface */
-       grain_bits = fls_long(grain) + 1;
-       trace_mc_event(type, msg, label, error_count,
-                      mci->mc_idx, top_layer, mid_layer, low_layer,
-                      PAGES_TO_MiB(page_frame_number) | offset_in_page,
-                      grain_bits, syndrome, other_detail);
+       grain_bits = fls_long(e->grain) + 1;
+       trace_mc_event(type, e->msg, e->label, e->error_count,
+                      mci->mc_idx, e->top_layer, e->mid_layer, e->low_layer,
+                      PAGES_TO_MiB(e->page_frame_number) | e->offset_in_page,
+                      grain_bits, e->syndrome, other_detail);
 
        /* Memory type dependent details about the error */
        if (type == HW_EVENT_ERR_CORRECTED) {
                snprintf(detail, sizeof(detail),
                        "page:0x%lx offset:0x%lx grain:%ld syndrome:0x%lx",
-                       page_frame_number, offset_in_page,
-                       grain, syndrome);
-               edac_ce_error(mci, error_count, pos, msg, location, label,
-                             detail, other_detail, enable_per_layer_report,
-                             page_frame_number, offset_in_page, grain);
+                       e->page_frame_number, e->offset_in_page,
+                       e->grain, e->syndrome);
+               edac_ce_error(mci, e->error_count, pos, e->msg, e->location,
+                             e->label, detail, other_detail,
+                             e->enable_per_layer_report,
+                             e->page_frame_number, e->offset_in_page,
+                             e->grain);
        } else {
                snprintf(detail, sizeof(detail),
                        "page:0x%lx offset:0x%lx grain:%ld",
-                       page_frame_number, offset_in_page, grain);
+                       page_frame_number, offset_in_page, e->grain);
 
-               edac_ue_error(mci, error_count, pos, msg, location, label,
-                             detail, other_detail, enable_per_layer_report);
+               edac_ue_error(mci, e->error_count, pos, e->msg, e->location,
+                             e->label, detail, other_detail,
+                             e->enable_per_layer_report);
        }
 }
 EXPORT_SYMBOL_GPL(edac_mc_handle_error);
index ff18efc754f3bd33e46ba273b3dd469b1603d885..096b7fcdf484e292296ab3fd86216c026a188eb3 100644 (file)
@@ -47,8 +47,18 @@ static inline void opstate_init(void)
        return;
 }
 
+/* Max length of a DIMM label*/
 #define EDAC_MC_LABEL_LEN      31
 
+/* Maximum size of the location string */
+#define LOCATION_SIZE 80
+
+/* Defines the maximum number of labels that can be reported */
+#define EDAC_MAX_LABELS                8
+
+/* String used to join two or more labels */
+#define OTHER_LABEL " or "
+
 /**
  * enum dev_type - describe the type of memory DRAM chips used at the stick
  * @DEV_UNKNOWN:       Can't be determined, or MC doesn't support detect it
@@ -553,6 +563,46 @@ struct errcount_attribute_data {
        int layer0, layer1, layer2;
 };
 
+/**
+ * edac_raw_error_desc - Raw error report structure
+ * @grain:                     minimum granularity for an error report, in bytes
+ * @error_count:               number of errors of the same type
+ * @top_layer:                 top layer of the error (layer[0])
+ * @mid_layer:                 middle layer of the error (layer[1])
+ * @low_layer:                 low layer of the error (layer[2])
+ * @page_frame_number:         page where the error happened
+ * @offset_in_page:            page offset
+ * @syndrome:                  syndrome of the error (or 0 if unknown or if
+ *                             the syndrome is not applicable)
+ * @msg:                       error message
+ * @location:                  location of the error
+ * @label:                     label of the affected DIMM(s)
+ * @other_detail:              other driver-specific detail about the error
+ * @enable_per_layer_report:   if false, the error affects all layers
+ *                             (typically, a memory controller error)
+ */
+struct edac_raw_error_desc {
+       /*
+        * NOTE: everything before grain won't be cleaned by
+        * edac_raw_error_desc_clean()
+        */
+       char location[LOCATION_SIZE];
+       char label[(EDAC_MC_LABEL_LEN + 1 + sizeof(OTHER_LABEL)) * EDAC_MAX_LABELS];
+       long grain;
+
+       /* the vars below and grain will be cleaned on every new error report */
+       u16 error_count;
+       int top_layer;
+       int mid_layer;
+       int low_layer;
+       unsigned long page_frame_number;
+       unsigned long offset_in_page;
+       unsigned long syndrome;
+       const char *msg;
+       const char *other_detail;
+       bool enable_per_layer_report;
+};
+
 /* MEMORY controller information structure
  */
 struct mem_ctl_info {
@@ -660,6 +710,12 @@ struct mem_ctl_info {
        /* work struct for this MC */
        struct delayed_work work;
 
+       /*
+        * Used to report an error - by being at the global struct
+        * makes the memory allocated by the EDAC core
+        */
+       struct edac_raw_error_desc error_desc;
+
        /* the internal state of this controller instance */
        int op_state;