Welcome to Jaehyek's Blog!

• eMMC UFS Issues 16

  2017-01-12

  Jaehyek

  eMMC-UFS

  eMMC  UFS  Issues

  Component Feature

  • Controller :
  • Mi

  Failure Symptom

  • After 2529 cycles of overall 5000 cycles, this bit crash happened at 1 sample among total 10 samples

  Test Scenario

  • Write a promised data pattern with cache/packed feature disabled
  • Perform a sudden power loss at random timing
  • Check a the written data after power on. If the data is different from promised pattern, the test fails

  Data Corruption after Suddern Power Off

  

  The Picture shows the data of last 3 write command was mismatched

  Root Cause

  

  • A sudden Power Loss hits the page line programming including block 67/page 512/CE3 and block 68/page 1/CE0&CE1&CE2
  • The only data in block 67/page 512/CE3 becomes uncorrectable after booting up
  • When scanning MLC Block 68, FW lack of vision on MLC block 67, which was hit by power loss. So FW would not discard the data in block 68 to avoid response host data pattern as new+old+new
  Read All

  ---

• eMMC UFS Issues 15

  2017-01-11

  Jaehyek

  eMMC-UFS

  eMMC  UFS  Issues

  Component Feature

  • Controller :
  • Hy

  Failure Symptom

  • Encryption Screen pop-up when power on.

  Root Cause

  • read time out(10ms) at RPMB ( Replay Protect Memory Block )

  Cause procedure

  1. encryption key is saved in RPMB for FDE(Full-disk Encryption)
  2. When mount the user data region , it time-over to read-out the RPMB encryption key

  • At Android M-OS, the timeout max value was set to 150mS.

  • At Android N-OS, it was failed.

  • FDE(Full-disk Encryption) : The User data partition was encripted with Encryption key for the protection of Hacking, and the Encryption key was saved to eMMC RPMB block. when mount the partition it was descripted with Encryption key.
  Read All

  ---

• eMMC UFS Issues 14

  2017-01-10

  Jaehyek

  eMMC-UFS

  eMMC  UFS  Issues

  Component Feature

  • Controller : PS8225 + 2xL94C + 2x V01M
  • Mi

  Failure Symptom

  • The boot system stops after power-on

  Root Cause

  • MLC Open Block 616 is with UECC
  • Vt distribution shows the UECC block has been double programmed in MLC mode.

  Cause procedure

  After reviewing the firmware structure, the following scenario of event sequence has been identified as suspected failure sequence
  

  1. During power up init, firmware selects the block to use, for example the BLK-A never used before and sets the dirty bit=’1’ in SRAM
  2. Firmware enters suspend mode (auto standby) after few ms. Some SRAM including dirty bit of BLK-A will be turn off during suspend mode.
  3. Write command triggers the device awakes from suspend mode and firmware uses the BLK-A without erase it. The dirty bit “i” and P/E of BLK-A will be updated into a system block when the write operation is completed
  4. If power loss occurs during write BLK-A in step 3, the information related to the BLK-A will be lost. In the next power cycle, due to FW bug, BLK-A is again recognized as never used block and by repeating step 1-3, the BLK-A is used without erase, thus double program can occur.

  Note : FW management of dirty bit of never used NAND block ( P/W=1)
  

  • To avoid the FW perform unnecessary erase of a never used NAND block, there is a dirty bit flag. When the dirty bit is set to “1”, erase is needed before program. on the contrary, if dirty bit is clean, no need to erase before program.
  • This bit is no more used on the NAND block erased at least one time (P/E>=2)

  Conclusions:

  This failure mode was caused when a sudden power-off occurs during writing some data to a never-used block.

  A firmware issue is confirmed , the firmware issue will induce NAND block double programmed after power loss.

  As this is production failure observed after 300K units processed, Micron considers it as singular case where uncontrolled power cuts occur in line ( 11 times of power loss )

  At end user side, by assuming 3/5GB x day transferred in the early usage of the phone, it is expected that within few days all NAND blocks of the device will be written once and this issue will not be triggered after that.

  Read All

  ---

• BOOK-TCP/IP 쉽게 더 쉽게-01

  2017-01-08

  Jaehyek

  CCNA

  TCP-IP

  대표적인 프로토콜

  

  HTTP 응답과 상태코드

  

  TCP Port 번호의 범위와 주요 port 번호

  

  TCP Header 구조 와 컨트롤 비트

  

  UCP Header 구조

  

  IPv4 Header 구조

  

  좁은 길을 지나갈 때는 작게 분할해서 지나간다.

  • 한번에 지나갈 수 있는 데이터 크기를 MTU(Maximum Transmissin Unit )
  • 데이터를 분할하고 복원하는 방법

  IPv6 및 IPv6로 갈아타기.

  

  IPv6 = network + host

  

  IPv6 할당하기

  

  퍼블릭 IP 관리

  • ICANN ( Internet Corporation for Assigned Names and Numbers )
  • KRNIC ( Korea Network Informatin Center )

  routing protocol

  • BGP, OSPF, RIP

  • 라우터는 서로 누구와 연결이 되어 있는지 정보 교환

  • default router : 라우팅 테이블에 목적지의 address가 없는 경우, 사용하는 기본 라우터

  Dynamic Routing Algorithm

  1. Distance Vector Type
     • RIP (Routing Information Protocol )을 이용, 짧은 경로 선택
     • 비교적 간단한 설정.
  2. Link Status Type
     • OSPF ( Open Shortest Path First ) 을 이용하여 네트워크의 통신 상태를 map으로 관리하고 상태가 좋은 경로 선택.
     • 복잡하고 변화가 잦은 네트워크에 적합.
  3. AS(자율시스템)내에서 OSPF을 사용
     • 네트워크를 몇개의 영역으로 나누어 OSPF을 사용.
  4. AS 내에서 BGP을 사용
     • BGP ( Border Gateway Protocol )을 AS내에서 사용. 경로 벡터형.
     • 경로 벡터형은 경로의 정보와 AS의 정보를 함께 사용하여 정보를 만든다

  AS 번호 : IP address 처럼, 16bit 혹은 32bit 으로 활당하고, 전세계적으로 중복되지 않게 관리.

  네트워크의 오류를 통보하는 프로토콜 : ICMP ( Internet Control Message Protocol )

  데이터 전송중에 문제가 생길 경우 장애 상태를 통보하기 위해 사용.

  주요 ICMP의 메세지

  

  어드레스 변환

  가정이나 사무실에서 사용하는 프라이빗 IP은 퍼블릭IP와 통신을 하지 못한다. 이 때, 어드레스의 변환이 필요한데 NAT( Network Address Translation )을 사용한다. NAT은 IP4와 IP6의 변환에도 사용한다.

  

  NAT에서 발생가능한 제약 사항

  • 내부의 호스트가 공교롭게도 같은 port 번호를 사용할 경우 라우터는 구별하지 못한다.

  

  • 외부에서 온 데이터를 전달하지 못 한다.

  

  • NAPT ( Network Address Port Translation )을 사용한다.
  • 이는 address와 port을 함께 사용하여 변환한다.

  외부에서 접속이 가능하게 하려면

  • 대부분 내부에서 메세지 도착여부를 정기적으로 확인하는 방법으로 구현된다.

  포트 포워딩 ( port forwarding )

  • LAN안에 웹서버나 FTP 처럼 외부에 서비스를 공개해야 할 때가 있다. 이 때, 라이터의 특정 port로 들어오면 내부의 특정 서버로 전달하게 한다.

  도메인명

  

  • DNS 서버에 질의하기

  DNS 서버의 계층 구조

  

  도메인 등록하기

  

  PING 으로 수신자의 통신상태 확인.

  ping은 지정한 IP address으로 ICMP의 8번 메세지를 보내는 것이다. 수신자는 ICMP 0번 메세지로 응답을 한다.

  tracert 명령은 통신경로 확인하기 ( unix 에서는 traceroute )

  ping 명령어와 비슷하게 ICMP 8번 메세지를 보낸다. 이 때, 생존기간을 1 부터 시작해서 1씩 증가시키면서 ICMP의 메세지를 확인하고 경로를 추적한다.

  snlookup 명령으로 IP address을 알아내기

  nslookup 명령은 DNS 서버에게 도메인명을 주고 IP 어드레스를 물어 보거나 거꾸로 IP 어드레스를 주고 도메인명을 물어 볼 때 사용한다.

  Read All

  ---

• eMMC UFS Issues 13

  2017-01-03

  Jaehyek

  eMMC-UFS

  eMMC  UFS  Issues

  Component Feature

  • Controller :
  • Hy

  Failure Symptom

  • When assert, Bit Flip samples FA status

  Root Cause

  • System Block Consistency FAIL
  • The FTL Meta Data for Map table got a bit flip, and then FW ASSERT
  • While GC for Map Block , the bit flip cause the number of vaild page mismatch not consistency
  • and then produced the ASSERT
  • and then FW ASSERT while booting

  

  Cause of bit flip

  • suspection of SRAM defect
  • After code review, There is no code to change the bit flip
  • No defect in the Map Table Meta Data
  Read All

  ---

• eMMC UFS Issues 12

  2017-01-02

  Jaehyek

  eMMC-UFS

  eMMC  UFS  Issues

  Component Feature

  • eMMC5.1
  • Controller :
  • Qual

  Failure Symptom

  • CodeAurora Released Patch for Kernel msm-3.18
  • Reference : http://source.codeaurora.org/quic/la/kernel/msm-3.18/commit/drivers/mmc?h=LV.HB.1.1.2-09110-8x96.0&id=e1b3c352dbf1c132a24dbbd176339ceec5d4862c

  Root Cause

  • At Qualcomm chip emmc 5.1
  • From kernel 3.18 , Auto BKOPS enable bit is saved to system block ( RPMB )
  • eMMC Driver write the BKOPS enable bit to EXT_CSD register when suspen/resume
  • Frequent suspend/resume makes the frequent writing to system block, and then caused EOL ( end of life )
  • The patch prevent the enable/disable function of BKOPS when suspen/resume

  

  at drivers/mmc/core/mmc.c

  diff --git a/drivers/mmc/core/mmc.c b/drivers/mmc/core/mmc.c
  index 73c17db..f90e297 100644
  --- a/drivers/mmc/core/mmc.c
  +++ b/drivers/mmc/core/mmc.c
  @@ -2378,12 +2378,6 @@ static int _mmc_suspend(struct mmc_host *host, bool is_suspend)
   			goto out;
   	}
   
  -	if (mmc_card_doing_auto_bkops(host->card)) {
  -		err = mmc_set_auto_bkops(host->card, false);
  -		if (err)
  -			goto out;
  -	}
  -
   	err = mmc_flush_cache(host->card);
   	if (err)
   		goto out;
  @@ -2463,9 +2457,6 @@ static int mmc_partial_init(struct mmc_host *host)
   	pr_debug("%s: %s: reading and comparing ext_csd successful\n",
   		mmc_hostname(host), __func__);
   
  -	if (mmc_card_support_auto_bkops(host->card))
  -		(void)mmc_set_auto_bkops(host->card, true);
  -
   	if (card->ext_csd.cmdq_support && (card->host->caps2 &
   					   MMC_CAP2_CMD_QUEUE)) {
   		err = mmc_select_cmdq(card);
  

  Mechanism of BKOPS Operation

  

  Failure Mode estimation of this issue

  [AUTO_EN Update usage @ A Customer system] (Chipset : MSM8953 / Kernel version : 3.18.24)
  In worst case, this issue makes the emmc life to be 1 year

  Read All

  ---