work-arounds for DMA on BIO

- writes need at least a "pipe cleaner" to set the pipe in the correct
state
- may also need a "double-write" to commit the data - this is to be
explored with a smaller test case
- DMA write only works over all corners at 700MHz when
  FCLK=ACLK (i.e. FCLK is at half rate)
This commit is contained in:
bunnie 2025-04-20 03:25:44 -07:00
parent 7bd545e95c
commit 1bad10fb21

View file

@ -200,7 +200,7 @@ pub fn dma_basic(concurrent: bool, clkmode: u8) -> usize {
const TEST_LEN: usize = 64;
let mut passing = 0;
if !concurrent {
print!("DMA basic\r");
print!("DMA basic2w\r");
} else {
print!("DMA cpu+dma concurrent\r");
}
@ -211,7 +211,7 @@ pub fn dma_basic(concurrent: bool, clkmode: u8) -> usize {
let mut bio_ss = BioSharedState::new();
// stop all the machines, so that code can be loaded
bio_ss.bio.wo(utra::bio_bdma::SFR_CTRL, 0x0);
bio_ss.load_code(dma_basic_code(), 0, BioCore::Core0);
bio_ss.load_code(dma_basic_code(), 0, BioCore::Core3);
// setup clocking mode option
bio_ss.bio.rmwf(SFR_CONFIG_CLOCKING_MODE, clkmode as u32);
@ -223,7 +223,7 @@ pub fn dma_basic(concurrent: bool, clkmode: u8) -> usize {
bio_ss.bio.wo(utra::bio_bdma::SFR_QDIV2, 0x1_0000);
bio_ss.bio.wo(utra::bio_bdma::SFR_QDIV3, 0x1_0000);
// start the machine
bio_ss.bio.wo(utra::bio_bdma::SFR_CTRL, 0x111);
bio_ss.bio.wo(utra::bio_bdma::SFR_CTRL, 0x888);
let mut main_mem_src: [u32; TEST_LEN] = [0u32; TEST_LEN];
let mut main_mem_dst: [u32; TEST_LEN] = [0u32; TEST_LEN];
@ -235,19 +235,19 @@ pub fn dma_basic(concurrent: bool, clkmode: u8) -> usize {
unsafe { core::slice::from_raw_parts_mut((utralib::HW_IFRAM1_MEM + 10000) as *mut u32, TEST_LEN) };
ifram_src.fill(0);
ifram_dst.fill(0);
passing += basic_u32(&mut bio_ss, &mut main_mem_src, &mut main_mem_dst, 0, "Main->main", concurrent);
passing += basic_u32(&mut bio_ss, &mut main_mem_src, &mut main_mem_dst, 0, "m->m", concurrent);
main_mem_src.fill(0);
main_mem_dst.fill(0);
passing += basic_u32(&mut bio_ss, ifram_src, &mut main_mem_dst, 0x40, "ifram0->main", concurrent);
passing += basic_u32(&mut bio_ss, ifram_src, &mut main_mem_dst, 0x40, "i0->m", concurrent);
ifram_src.fill(0);
main_mem_dst.fill(0);
passing += basic_u32(&mut bio_ss, &mut main_mem_src, ifram_dst, 0x80, "Main->ifram1", concurrent);
passing += basic_u32(&mut bio_ss, &mut main_mem_src, ifram_dst, 0x80, "m->i1", concurrent);
main_mem_src.fill(0);
ifram_dst.fill(0);
passing += basic_u32(&mut bio_ss, ifram_src, ifram_dst, 0xC0, "ifram0->ifram1", concurrent);
passing += basic_u32(&mut bio_ss, ifram_src, ifram_dst, 0xC0, "i0->i1", concurrent);
if !concurrent {
print!("DMA basic done.\r");
@ -281,13 +281,25 @@ fn basic_u32(
bio_ss.bio.wo(utra::bio_bdma::SFR_TXF0, (src.len() * size_of::<u32>()) as u32); // bytes to move
while bio_ss.bio.r(utra::bio_bdma::SFR_EVENT_STATUS) & 0x1 == 0 {}
cache_flush();
let mut errs = 0;
for (i, &d) in src.iter().enumerate() {
let rbk = unsafe { dst.as_ptr().add(i).read_volatile() };
if rbk != d {
print!("{} DMA err @{}, {:x} rbk: {:x}\r", name, i, d, rbk);
if errs < 4 {
unsafe {
print!("{} @{:x}/{:x}, {:x} rb:{:x}\r", name, src.as_ptr().add(i) as usize, dst.as_ptr().add(i) as usize, d, rbk);
}
} else {
print!("x");
}
errs += 1;
pass = 0;
}
}
if errs > 0 {
print!("\r");
// errs = 0;
}
} else {
// this flushes any read data from the cache, so that the CPU copy is forced to fetch
// the data from memory
@ -322,12 +334,16 @@ bio_code!(dma_basic_code, DMA_BASIC_START, DMA_BASIC_END,
"li x29, 0x1", // clear event done flag - just before the last parameter arrives
"mv a1, x16", // wait for # of bytes to move
"sw x0, 0(a2)", // make sure write pipeline is in a good state
"sw x0, 0(a2)", // make sure write pipeline is in a good state
"add a4, a1, a3", // a4 <- end condition based on source address increment
"30:",
"lw t0, 0(a3)", // blocks until load responds
"sw t0, 0(a2)", // blocks until store completes
"addi a3, a3, 4", // 3 cycles
"sw t0, 0(a2)", // blocks until store completes
"addi a2, a2, 4", // 3 cycles
"bne a3, a4, 30b", // 5 cycles
"li x28, 0x1", // flip event done flag
@ -687,14 +703,14 @@ pub fn dma_coincident(clkmode: u8) -> usize {
ifram_src.fill(0);
ifram_dst.fill(0);
passing += coincident_u32(&mut bio_ss, &mut main_mem_src, &mut main_mem_dst, 0x200, "DMA coincident");
print!("m->m\r");
passing += coincident_u32(&mut bio_ss, &mut main_mem_src, &mut main_mem_dst, 0x200, "m->m");
print!("m->m coincident\r");
// try other memory banks
passing += coincident_u32(&mut bio_ss, ifram_src, &mut main_mem_dst, 0x300, "ifram0->main");
passing += coincident_u32(&mut bio_ss, ifram_src, &mut main_mem_dst, 0x300, "i0->m");
print!("0->m\r");
passing += coincident_u32(&mut bio_ss, &mut main_mem_src, ifram_dst, 0x400, "main->ifram1");
passing += coincident_u32(&mut bio_ss, &mut main_mem_src, ifram_dst, 0x400, "m->i1");
print!("m->1\r");
passing += coincident_u32(&mut bio_ss, ifram_src, ifram_dst, 0x500, "ifram0->ifram1");
passing += coincident_u32(&mut bio_ss, ifram_src, ifram_dst, 0x500, "i0->i1");
print!("0->1\r");
print!("DMA coincident done.\r");
@ -722,13 +738,22 @@ fn coincident_u32(
while bio_ss.bio.r(utra::bio_bdma::SFR_EVENT_STATUS) & 0xF_0000 != 0xF_0000 {} // one bit per core is set on completion of bits 8:12
cache_flush();
let mut pass = 1;
let mut errs = 0;
for (i, &d) in src.iter().enumerate() {
let rbk = unsafe { dst.as_ptr().add(i).read_volatile() };
if rbk != d {
print!("{} err @{}, {:x} rbk: {:x}\r", name, i, d, rbk);
if errs < 4 {
print!("{} err @{}, {:x} rbk: {:x}\r", name, i, d, rbk);
} else {
print!(".");
}
errs += 1;
pass = 0;
}
}
if errs >= 4 {
print!("\r");
}
pass
}
@ -826,7 +851,7 @@ pub fn dmareq_test() -> usize {
bio_ss.bio.wo(utra::bio_bdma::SFR_TXF1, bit as u32);
print!("starting search \r");
for i in 1..15 {
print!("search iter {} \r", i);
print!("{}\r", i);
// send the bit pattern into the event-set via core 0
bio_ss.bio.wo(utra::bio_bdma::SFR_TXF0, i);