1 files changed, 45 insertions, 0 deletions
diff --git a/Documentation/arm/Sharp-LH/IOBarrier b/Documentation/arm/Sharp-LH/IOBarrier
new file mode 100644
index 000000000000..c0d8853672dc
--- /dev/null
+++ b/Documentation/arm/Sharp-LH/IOBarrier
@@ -0,0 +1,45 @@
+README on the IOBARRIER for CardEngine IO
+=========================================
+
+Due to an unfortunate oversight when the Card Engines were designed,
+the signals that control access to some peripherals, most notably the
+SMC91C9111 ethernet controller, are not properly handled.
+
+The symptom is that some back to back IO with the peripheral returns
+unreliable data.  With the SMC chip, you'll see errors about the bank
+register being 'screwed'.
+
+The cause is that the AEN signal to the SMC chip does not transition
+for every memory access.  It is driven through the CPLD from the CS7
+line of the CPU's static memory controller which is optimized to
+eliminate unnecessary transitions.  Yet, the SMC requires a transition
+for every write access.  The Sharp website has more information about
+the effect this power-conserving feature has on peripheral
+interfacing.
+
+The solution is to follow every write access to the SMC chip with an
+access to another memory region that will force the CPU to release the
+chip select line.  It is important to guarantee that this access
+forces the CPU off-chip.  We map a page of SDRAM as if it were an
+uncacheable IO device and read from it after every SMC IO write
+operation.
+
+  SMC IO
+  BARRIER IO
+
+Only this sequence is important.  It does not matter that there is no
+BARRIER IO before the access to the SMC chip because the AEN latch
+only needs occurs after the SMC IO write cycle.  The routines that
+implement this work-around make an additional concession which is to
+disable interrupts during the IO sequence.  Other hardware devices
+(the LogicPD CPLD) have registers in the same the physical memory
+region as the SMC chip.  An interrupt might allow an access to one of
+those registers while SMC IO is being performed.
+
+You might be tempted to think that we have to access another device
+attached to the static memory controller, but the empirical evidence
+indicates that this is not so.  Mapping 0x00000000 (flash) and
+0xc0000000 (SDRAM) appear to have the same effect.  Using SDRAM seems
+to be faster.  Choosing to access an undecoded memory region is not
+desirable as there is no way to know how that chip select will be used
+in the future.