Configurable RWDS sampling and clock-start delay. #21
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Acording to spec: t_DSV (data strobe valid) which is the time from CS# going low to the first hyperbus clock can be at most 2 clock periods long (12ns@166MHz). This shrinks the RWDS valid window down to one period centered on CA4 (5th data transaction). Meaning it is valid around the 3rd rising edge of CK. Problem: With additional routing delay this may cause the RWDS sample register (clocked by clk_i) to miss the stable period of RWDS. Solution: Delaying the clock is allowed and gives RWDS more time to arrive and creates a larger stable window. It is possible to set this to zero to increase throughput.
For the worst case RWDS timing (t_DSV max, t_CSS min and t_CKDS min) the window of validity for RWDS is around one clock period centered around the 3rd rising edge of CK. This ensures we sample exactly then. Other sampling may lead to improper results (from sampling high Z) and increases the risk of metastability. For long chip-to-chip delays (or slow pads) it may still be necessary to increase the CS falling edge to first CK edge time.
Decouples the clock domain better, only the rwds_sample_o signal crosses between phy and system clk.
Exact sampling edge is adjustable.
We want to give the RWDS sampler as much time as possible to get a value. So we delay the additional latency decision to the latest point possible.
The reset was not being triggered since the gated clock stops before chip select goes high. A sticky bit driven by the ungated clock is used to indicate start of transfer. The counter reaching the target value is used to reset the sticky bit. Counter only counts while it is set (when the transfer starts until the target is reached).
phsauter
requested review from
paulsc96,
thommythomaso and
luca-valente
as code owners
phsauter
force-pushed
the
phsauter/rwds_sampling
branch
from
5c79efb to
4ab8088
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
A finalized version of #20
Current State: Tested in RTL simulation
During the Command-Address block of read and writes (except for zero-latency writes), the RWDS signal must be sampled to determine the additional latency required (1x the configured value if it is LOW, 2x if it is HIGH).
RWDS is driven from the device some times after CS is driven low by the controller and it is de-asserted by the device synchronous to the CK edge during the time the last CA data is stable.
For the worst case timing (t_DSV max, t_CSS min and t_CKDS min) the RWDS signal is only valid for about one clock period (two cycles to three cycles after CS going low).
It is important to mention that according to spec, the RWDS signal from the device and the clock start from the host are both referenced to the CS going low edge but not each other. Meaning in the case of a slower than spec'd clock, RWDS may arrive earlier than expected (even before the clock starts).
This presents us with two problems:
Fixing 1:
Solved by adding a separate module that counts clock edges after CS going low. This then enables a clock gate, propagating a single clock pulse to the sampling flip-flop. The exact edge where the sampling should occur is configurable.
A second flip-flop is added to cross the signal into the clk_phy domain the main FSM is in.
Fixing 2:
Two additions were made:
Together the clock-start and RWDS sampling time are a lot more configurable and should greatly increase the control, especially when operating with out-of-spec frequencies.
See Figure 9.4 and 9.5 in the spec (read timing diagrams) as well as table 9.4 (AC parameters). Consider that the diagrams are not to scale for a worst case transaction, t_DSV in particular is way too small.
The attached image show the relevant signal of the new RWDS sampler. The chip select to clock-start time
tcs->ckcan be increased from the minimum of one clock cycle.trwdscan be changed in 1/2 clock cycles steps with the minimum being 0.5 clock cycles (corresponding to the falling edge immediately after CS going low)