1      问题背景

冷启动后，使用npu-smi info命令会显示“Critical”，使用npu-smi info -t health -i 0 -c 0命令会回显有报错

 2      修复方案

给Ascend-lp-source源码打补丁，然后使用修改后的源码重新构建出固件并升级修复

 2.1      固件的编译和生效参见

https://support.huawei.com/enterprise/zh/doc/EDOC1100445968/96f94ea0?idPath=23710424|251366513|254884019|261408772|263024819

 

 2.2      打patch补丁

patch -p1 -i fix_pg_for_aicore_8_9.patch

 3      patch源码（fix_pg_for_aicore_8_9.patch）

diff --git a/drivers/firmware/lp/lpr52/driver/tsensor/tsensor.c b/drivers/firmware/lp/lpr52/driver/tsensor/tsensor.c

index 4b10414..5ad0cd2 100755

--- a/drivers/firmware/lp/lpr52/driver/tsensor/tsensor.c

+++ b/drivers/firmware/lp/lpr52/driver/tsensor/tsensor.c

@@ -318,6 +318,7 @@ void TsensorInit(void)

     int32_t ret;

     uint32_t initStatus = 0;

     uint32_t errStatus = 0;

+    uint32_t aiCorePg = READ_PG(PG_AICORE);

 

 #if defined(PLATFORM_FPGA) || defined(PLATFORM_EMU)

     return;

@@ -329,6 +330,7 @@ void TsensorInit(void)

     /* clear ddr temperature space */

     CleanTempShareMem();

     uint32_t partialGood = TsensorGetPartialGoodStatus();

+    partialGood |= (~aiCorePg) & ALL_AICORE_MASK; // ALL_AICORE_MASK=0x3FFU 只判断前面十个，最后partialGood=0x20300，位8和位9为1

     WRITEL(partialGood, TsensorRuntimeAddr(INDEX_PG));

 

 #ifdef AS610V2V3