#11085 BLOC 1.5-har: XO-1.75 Memory errors

[Zarro Boogs per Child]

#11085: XO-1.75 Memory errors
-------------------------------------------+--------------------------------
           Reporter:  wad                  |       Owner:  wad            
               Type:  defect               |      Status:  assigned       
           Priority:  blocker              |   Milestone:  1.5-hardware-C 
          Component:  ofw - open firmware  |     Version:  1.75-B1        
         Resolution:                       |    Keywords:  XO-1.75, memory
        Next_action:  reproduce            |    Verified:  0              
Deployment_affected:                       |   Blockedby:                 
           Blocking:                       |  
-------------------------------------------+--------------------------------

Comment(by wad):

 Q407mb has identical DRAM timing to Q4B05 (the one that this ticket
 started with), with the exception of having auto power saving enabled.

 A new timing (q4b07rb) is being generated for testing.   The differences
 between it and Q4B07 are:

 SDRAM_CONFIG_TYPE1_CS0 (new: 00266430)

   Change RTT_nominal from 001b to 011b

   Change DRAM output drive strength from 000b to 001b

 SDRAM_CTRL2 (new: 0f100010)

   Change SDRAM line boundary from line_32B to line_64B (rec. by Marvell)

   Enabled auto-precharge (rec. by Marvell)

 SDRAM_CTRL6_ODT_CTRL

   Enable ODT0 when writing to CS1 (rec. by Marvell)

 SDRAM_CTRL7_ODT_CTRL2

   Enable ODT1 when controlled by CTRL6 (rec. by Marvell)

 SDRAM_CTRL8_ODT_CTRL2

   Enable cross-page detection and arbitration

 PHY_CTRL10

   Enabled automatic pad calibration

 PHY_CTRL7

   Increase DQ PMOS driver strength from 0111b to 1111b

   Increase DQ NMOS driver strength from 0111b to 1111b

 PHY_CTRL8

   Increase Address/Command PMOS driver strength from 0111b to 1111b

   Increase Address/Command NMOS driver strength from 0111b to 1111b

 This timing caused a data abort on a laptop that up to that point hadn't
 showed any memory problems.   It ran for 18 hours on 23 other laptops w.
 no problems.  Suspecting the automatic pad calibration (recommended by
 Marvell engineer, but the register description says that enabling this is
 not recommended).   Q4B07rc is the same timing w. the automatic pad
 calibration bit not set.

-- 
Ticket URL: <http://dev.laptop.org/ticket/11085#comment:12>
One Laptop Per Child <http://laptop.org/>
OLPC bug tracking system