* [PATCH RESEND v2 0/4] hw/dma: Align SiFive PDMA behavior with real hardware
@ 2021-09-12 13:05 frank.chang
2021-09-12 13:05 ` [PATCH RESEND v2 1/4] hw/dma: sifive_pdma: reset Next* registers when Control.claim is set frank.chang
` (3 more replies)
0 siblings, 4 replies; 6+ messages in thread
From: frank.chang @ 2021-09-12 13:05 UTC (permalink / raw)
To: qemu-devel, qemu-riscv; +Cc: Frank Chang
From: Frank Chang <frank.chang@sifive.com>
Current QEMU PDMA doesn't align with real PDMA's behavior. This would
result in Linux dmatest failed. This patchest aligns with real PDMA's
behavior we tested on the real board. The golden results are performed
in U-Boot on the Unmatched board with PDMA supported.
Changelog:
v2:
* Add comment to describe document errors.
Frank Chang (3):
hw/dma: sifive_pdma: reset Next* registers when Control.claim is set
hw/dma: sifive_pdma: claim bit must be set before DMA transactions
hw/dma: sifive_pdma: don't set Control.error if 0 bytes to transfer
Green Wan (1):
hw/dma: sifive_pdma: allow non-multiple transaction size transactions
hw/dma/sifive_pdma.c | 54 +++++++++++++++++++++++++++++++++++---------
1 file changed, 43 insertions(+), 11 deletions(-)
--
2.25.1
^ permalink raw reply [flat|nested] 6+ messages in thread
* [PATCH RESEND v2 1/4] hw/dma: sifive_pdma: reset Next* registers when Control.claim is set
2021-09-12 13:05 [PATCH RESEND v2 0/4] hw/dma: Align SiFive PDMA behavior with real hardware frank.chang
@ 2021-09-12 13:05 ` frank.chang
2021-09-12 13:05 ` [PATCH RESEND v2 2/4] hw/dma: sifive_pdma: claim bit must be set before DMA transactions frank.chang
` (2 subsequent siblings)
3 siblings, 0 replies; 6+ messages in thread
From: frank.chang @ 2021-09-12 13:05 UTC (permalink / raw)
To: qemu-devel, qemu-riscv
Cc: Frank Chang, Bin Meng, Max Hsu, Palmer Dabbelt, Alistair Francis,
Bin Meng
From: Frank Chang <frank.chang@sifive.com>
Setting Control.claim clears all of the chanel's Next registers.
This is effective only when Control.claim is set from 0 to 1.
Signed-off-by: Frank Chang <frank.chang@sifive.com>
Tested-by: Max Hsu <max.hsu@sifive.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
---
hw/dma/sifive_pdma.c | 19 +++++++++++++++++++
1 file changed, 19 insertions(+)
diff --git a/hw/dma/sifive_pdma.c b/hw/dma/sifive_pdma.c
index 9b2ac2017d9..d92e27dfdc4 100644
--- a/hw/dma/sifive_pdma.c
+++ b/hw/dma/sifive_pdma.c
@@ -54,6 +54,13 @@
#define DMA_EXEC_DST 0x110
#define DMA_EXEC_SRC 0x118
+/*
+ * FU540/FU740 docs are incorrect with NextConfig.wsize/rsize reset values.
+ * The reset values tested on Unleashed/Unmatched boards are 6 instead of 0.
+ */
+#define CONFIG_WRSZ_DEFAULT 6
+#define CONFIG_RDSZ_DEFAULT 6
+
enum dma_chan_state {
DMA_CHAN_STATE_IDLE,
DMA_CHAN_STATE_STARTED,
@@ -221,6 +228,7 @@ static void sifive_pdma_write(void *opaque, hwaddr offset,
{
SiFivePDMAState *s = opaque;
int ch = SIFIVE_PDMA_CHAN_NO(offset);
+ bool claimed;
if (ch >= SIFIVE_PDMA_CHANS) {
qemu_log_mask(LOG_GUEST_ERROR, "%s: Invalid channel no %d\n",
@@ -231,6 +239,17 @@ static void sifive_pdma_write(void *opaque, hwaddr offset,
offset &= 0xfff;
switch (offset) {
case DMA_CONTROL:
+ claimed = !!s->chan[ch].control & CONTROL_CLAIM;
+
+ if (!claimed && (value & CONTROL_CLAIM)) {
+ /* reset Next* registers */
+ s->chan[ch].next_config = (CONFIG_RDSZ_DEFAULT << CONFIG_RDSZ_SHIFT) |
+ (CONFIG_WRSZ_DEFAULT << CONFIG_WRSZ_SHIFT);
+ s->chan[ch].next_bytes = 0;
+ s->chan[ch].next_dst = 0;
+ s->chan[ch].next_src = 0;
+ }
+
s->chan[ch].control = value;
if (value & CONTROL_RUN) {
--
2.25.1
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH RESEND v2 2/4] hw/dma: sifive_pdma: claim bit must be set before DMA transactions
2021-09-12 13:05 [PATCH RESEND v2 0/4] hw/dma: Align SiFive PDMA behavior with real hardware frank.chang
2021-09-12 13:05 ` [PATCH RESEND v2 1/4] hw/dma: sifive_pdma: reset Next* registers when Control.claim is set frank.chang
@ 2021-09-12 13:05 ` frank.chang
2021-09-12 13:05 ` [PATCH RESEND v2 3/4] hw/dma: sifive_pdma: allow non-multiple transaction size transactions frank.chang
2021-09-12 13:05 ` [PATCH RESEND v2 4/4] hw/dma: sifive_pdma: don't set Control.error if 0 bytes to transfer frank.chang
3 siblings, 0 replies; 6+ messages in thread
From: frank.chang @ 2021-09-12 13:05 UTC (permalink / raw)
To: qemu-devel, qemu-riscv
Cc: Frank Chang, Bin Meng, Max Hsu, Palmer Dabbelt, Alistair Francis,
Bin Meng
From: Frank Chang <frank.chang@sifive.com>
Real PDMA must have Control.claim bit to be set before
Control.run bit is set to start any DMA transactions.
Otherwise nothing will be transferred.
The following result is PDMA tested in U-Boot on Unmatched board:
=> mw.l 0x3000000 0x0 <= Disclaim channel 0
(Channel 0 is not claimed)
=> mw.l 0x3000004 0x55000000 <= wsize = rsize = 5 (2^5 = 32 bytes)
=> mw.q 0x3000008 0x2 <= NextBytes = 2
=> mw.q 0x3000010 0x84000000 <= NextDestination = 0x84000000
=> mw.q 0x3000018 0x84001000 <= NextSource = 0x84001000
=> mw.l 0x84000000 0x87654321 <= Fill test data to dst
=> mw.l 0x84001000 0x12345678 <= Fill test data to src
=> md.l 0x84000000 1; md.l 0x84001000 1 <= Dump src/dst memory contents
84000000: 87654321 !Ce.
84001000: 12345678 xV4.
=> md.l 0x3000000 8 <= Dump PDMA status
03000000: 00000000 55000000 00000002 00000000 .......U........
03000010: 84000000 00000000 84001000 00000000 ................
=> mw.l 0x3000000 0x3 <= Set channel 0 run and claim bits
=> md.l 0x3000000 8 <= Dump PDMA status
03000000: 00000001 66000000 00000000 00000000 .......f........
03000010: 00000000 00000000 00000000 00000000 ................
=> md.l 0x84000000 1; md.l 0x84001000 1 <= Dump src/dst memory contents
84000000: 87654321 !Ce.
84001000: 12345678 xV4.
Signed-off-by: Frank Chang <frank.chang@sifive.com>
Tested-by: Max Hsu <max.hsu@sifive.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
---
hw/dma/sifive_pdma.c | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/hw/dma/sifive_pdma.c b/hw/dma/sifive_pdma.c
index d92e27dfdc4..a8ce3e6699a 100644
--- a/hw/dma/sifive_pdma.c
+++ b/hw/dma/sifive_pdma.c
@@ -252,6 +252,15 @@ static void sifive_pdma_write(void *opaque, hwaddr offset,
s->chan[ch].control = value;
+ /*
+ * If channel was not claimed before run bit is set,
+ * DMA won't run.
+ */
+ if (!claimed) {
+ s->chan[ch].control &= ~CONTROL_RUN;
+ return;
+ }
+
if (value & CONTROL_RUN) {
sifive_pdma_run(s, ch);
}
--
2.25.1
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH RESEND v2 3/4] hw/dma: sifive_pdma: allow non-multiple transaction size transactions
2021-09-12 13:05 [PATCH RESEND v2 0/4] hw/dma: Align SiFive PDMA behavior with real hardware frank.chang
2021-09-12 13:05 ` [PATCH RESEND v2 1/4] hw/dma: sifive_pdma: reset Next* registers when Control.claim is set frank.chang
2021-09-12 13:05 ` [PATCH RESEND v2 2/4] hw/dma: sifive_pdma: claim bit must be set before DMA transactions frank.chang
@ 2021-09-12 13:05 ` frank.chang
2021-09-12 13:05 ` [PATCH RESEND v2 4/4] hw/dma: sifive_pdma: don't set Control.error if 0 bytes to transfer frank.chang
3 siblings, 0 replies; 6+ messages in thread
From: frank.chang @ 2021-09-12 13:05 UTC (permalink / raw)
To: qemu-devel, qemu-riscv
Cc: Frank Chang, Bin Meng, Green Wan, Max Hsu, Alistair Francis,
Palmer Dabbelt, Bin Meng
From: Green Wan <green.wan@sifive.com>
Real PDMA is able to deal with non-multiple transaction size transactions.
The following result is PDMA tested in U-Boot on Unmatched board:
=> mw.l 0x3000000 0x0 <= Disclaim channel 0
=> mw.l 0x3000000 0x1 <= Claim channel 0
=> mw.l 0x3000004 0x11000000 <= wsize = rsize = 1 (2^1 = 2 bytes)
=> mw.q 0x3000008 0x3 <= NextBytes = 3
=> mw.q 0x3000010 0x84000000 <= NextDestination = 0x84000000
=> mw.q 0x3000018 0x84001000 <= NextSource = 0x84001000
=> mw.l 0x84000000 0x87654321 <= Fill test data to dst
=> mw.l 0x84001000 0x12345678 <= Fill test data to src
=> md.l 0x84000000 1; md.l 0x84001000 1 <= Dump src/dst memory contents
84000000: 87654321 !Ce.
84001000: 12345678 xV4.
=> md.l 0x3000000 8 <= Dump PDMA status
03000000: 00000001 11000000 00000003 00000000 ................
03000010: 84000000 00000000 84001000 00000000 ................
=> mw.l 0x3000000 0x3 <= Set channel 0 run and claim bits
=> md.l 0x3000000 8 <= Dump PDMA status
03000000: 40000001 11000000 00000003 00000000 ...@............
03000010: 84000000 00000000 84001000 00000000 ................
=> md.l 0x84000000 1; md.l 0x84001000 1 <= Dump src/dst memory contents
84000000: 87345678 xV4.
84001000: 12345678 xV4.
Signed-off-by: Green Wan <green.wan@sifive.com>
Reviewed-by: Frank Chang <frank.chang@sifive.com>
Tested-by: Max Hsu <max.hsu@sifive.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Signed-off-by: Frank Chang <frank.chang@sifive.com>
---
hw/dma/sifive_pdma.c | 16 ++++++++++------
1 file changed, 10 insertions(+), 6 deletions(-)
diff --git a/hw/dma/sifive_pdma.c b/hw/dma/sifive_pdma.c
index a8ce3e6699a..d7d2c53e97e 100644
--- a/hw/dma/sifive_pdma.c
+++ b/hw/dma/sifive_pdma.c
@@ -74,7 +74,7 @@ static void sifive_pdma_run(SiFivePDMAState *s, int ch)
uint64_t dst = s->chan[ch].next_dst;
uint64_t src = s->chan[ch].next_src;
uint32_t config = s->chan[ch].next_config;
- int wsize, rsize, size;
+ int wsize, rsize, size, remainder;
uint8_t buf[64];
int n;
@@ -106,11 +106,7 @@ static void sifive_pdma_run(SiFivePDMAState *s, int ch)
size = 6;
}
size = 1 << size;
-
- /* the bytes to transfer should be multiple of transaction size */
- if (bytes % size) {
- goto error;
- }
+ remainder = bytes % size;
/* indicate a DMA transfer is started */
s->chan[ch].state = DMA_CHAN_STATE_STARTED;
@@ -131,6 +127,14 @@ static void sifive_pdma_run(SiFivePDMAState *s, int ch)
s->chan[ch].exec_bytes -= size;
}
+ if (remainder) {
+ cpu_physical_memory_read(s->chan[ch].exec_src, buf, remainder);
+ cpu_physical_memory_write(s->chan[ch].exec_dst, buf, remainder);
+ s->chan[ch].exec_src += remainder;
+ s->chan[ch].exec_dst += remainder;
+ s->chan[ch].exec_bytes -= remainder;
+ }
+
/* indicate a DMA transfer is done */
s->chan[ch].state = DMA_CHAN_STATE_DONE;
s->chan[ch].control &= ~CONTROL_RUN;
--
2.25.1
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH RESEND v2 4/4] hw/dma: sifive_pdma: don't set Control.error if 0 bytes to transfer
2021-09-12 13:05 [PATCH RESEND v2 0/4] hw/dma: Align SiFive PDMA behavior with real hardware frank.chang
` (2 preceding siblings ...)
2021-09-12 13:05 ` [PATCH RESEND v2 3/4] hw/dma: sifive_pdma: allow non-multiple transaction size transactions frank.chang
@ 2021-09-12 13:05 ` frank.chang
2021-09-14 2:24 ` Alistair Francis
3 siblings, 1 reply; 6+ messages in thread
From: frank.chang @ 2021-09-12 13:05 UTC (permalink / raw)
To: qemu-devel, qemu-riscv
Cc: Frank Chang, Bin Meng, Max Hsu, Palmer Dabbelt, Alistair Francis,
Bin Meng
From: Frank Chang <frank.chang@sifive.com>
Real PDMA doesn't set Control.error if there are 0 bytes to be
transferred. The DMA transfer is still success.
The following result is PDMA tested in U-Boot on Unmatched board:
=> mw.l 0x3000000 0x0 <= Disclaim channel 0
=> mw.l 0x3000000 0x1 <= Claim channel 0
=> mw.l 0x3000004 0x55000000 <= wsize = rsize = 5 (2^5 = 32 bytes)
=> mw.q 0x3000008 0x0 <= NextBytes = 0
=> mw.q 0x3000010 0x84000000 <= NextDestination = 0x84000000
=> mw.q 0x3000018 0x84001000 <= NextSource = 0x84001000
=> mw.l 0x84000000 0x87654321 <= Fill test data to dst
=> mw.l 0x84001000 0x12345678 <= Fill test data to src
=> md.l 0x84000000 1; md.l 0x84001000 1 <= Dump src/dst memory contents
84000000: 87654321 !Ce.
84001000: 12345678 xV4.
=> md.l 0x3000000 8 <= Dump PDMA status
03000000: 00000001 55000000 00000000 00000000 .......U........
03000010: 84000000 00000000 84001000 00000000 ................
=> mw.l 0x3000000 0x3 <= Set channel 0 run and claim bits
=> md.l 0x3000000 8 <= Dump PDMA status
03000000: 40000001 55000000 00000000 00000000 ...@...U........
03000010: 84000000 00000000 84001000 00000000 ................
=> md.l 0x84000000 1; md.l 0x84001000 1 <= Dump src/dst memory contents
84000000: 87654321 !Ce.
84001000: 12345678 xV4.
Signed-off-by: Frank Chang <frank.chang@sifive.com>
Tested-by: Max Hsu <max.hsu@sifive.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
---
hw/dma/sifive_pdma.c | 12 ++++++------
1 file changed, 6 insertions(+), 6 deletions(-)
diff --git a/hw/dma/sifive_pdma.c b/hw/dma/sifive_pdma.c
index d7d2c53e97e..b4fd40573a5 100644
--- a/hw/dma/sifive_pdma.c
+++ b/hw/dma/sifive_pdma.c
@@ -80,7 +80,7 @@ static void sifive_pdma_run(SiFivePDMAState *s, int ch)
/* do nothing if bytes to transfer is zero */
if (!bytes) {
- goto error;
+ goto done;
}
/*
@@ -135,11 +135,6 @@ static void sifive_pdma_run(SiFivePDMAState *s, int ch)
s->chan[ch].exec_bytes -= remainder;
}
- /* indicate a DMA transfer is done */
- s->chan[ch].state = DMA_CHAN_STATE_DONE;
- s->chan[ch].control &= ~CONTROL_RUN;
- s->chan[ch].control |= CONTROL_DONE;
-
/* reload exec_ registers if repeat is required */
if (s->chan[ch].next_config & CONFIG_REPEAT) {
s->chan[ch].exec_bytes = bytes;
@@ -147,6 +142,11 @@ static void sifive_pdma_run(SiFivePDMAState *s, int ch)
s->chan[ch].exec_src = src;
}
+done:
+ /* indicate a DMA transfer is done */
+ s->chan[ch].state = DMA_CHAN_STATE_DONE;
+ s->chan[ch].control &= ~CONTROL_RUN;
+ s->chan[ch].control |= CONTROL_DONE;
return;
error:
--
2.25.1
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH RESEND v2 4/4] hw/dma: sifive_pdma: don't set Control.error if 0 bytes to transfer
2021-09-12 13:05 ` [PATCH RESEND v2 4/4] hw/dma: sifive_pdma: don't set Control.error if 0 bytes to transfer frank.chang
@ 2021-09-14 2:24 ` Alistair Francis
0 siblings, 0 replies; 6+ messages in thread
From: Alistair Francis @ 2021-09-14 2:24 UTC (permalink / raw)
To: Frank Chang
Cc: open list:RISC-V, Bin Meng, qemu-devel@nongnu.org Developers,
Max Hsu, Palmer Dabbelt, Alistair Francis, Bin Meng
On Sun, Sep 12, 2021 at 11:11 PM <frank.chang@sifive.com> wrote:
>
> From: Frank Chang <frank.chang@sifive.com>
>
> Real PDMA doesn't set Control.error if there are 0 bytes to be
> transferred. The DMA transfer is still success.
>
> The following result is PDMA tested in U-Boot on Unmatched board:
>
> => mw.l 0x3000000 0x0 <= Disclaim channel 0
> => mw.l 0x3000000 0x1 <= Claim channel 0
> => mw.l 0x3000004 0x55000000 <= wsize = rsize = 5 (2^5 = 32 bytes)
> => mw.q 0x3000008 0x0 <= NextBytes = 0
> => mw.q 0x3000010 0x84000000 <= NextDestination = 0x84000000
> => mw.q 0x3000018 0x84001000 <= NextSource = 0x84001000
> => mw.l 0x84000000 0x87654321 <= Fill test data to dst
> => mw.l 0x84001000 0x12345678 <= Fill test data to src
> => md.l 0x84000000 1; md.l 0x84001000 1 <= Dump src/dst memory contents
> 84000000: 87654321 !Ce.
> 84001000: 12345678 xV4.
> => md.l 0x3000000 8 <= Dump PDMA status
> 03000000: 00000001 55000000 00000000 00000000 .......U........
> 03000010: 84000000 00000000 84001000 00000000 ................
> => mw.l 0x3000000 0x3 <= Set channel 0 run and claim bits
> => md.l 0x3000000 8 <= Dump PDMA status
> 03000000: 40000001 55000000 00000000 00000000 ...@...U........
> 03000010: 84000000 00000000 84001000 00000000 ................
> => md.l 0x84000000 1; md.l 0x84001000 1 <= Dump src/dst memory contents
> 84000000: 87654321 !Ce.
> 84001000: 12345678 xV4.
>
> Signed-off-by: Frank Chang <frank.chang@sifive.com>
> Tested-by: Max Hsu <max.hsu@sifive.com>
> Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
> Tested-by: Bin Meng <bmeng.cn@gmail.com>
Thanks!
Applied to riscv-to-apply.next
Alistair
> ---
> hw/dma/sifive_pdma.c | 12 ++++++------
> 1 file changed, 6 insertions(+), 6 deletions(-)
>
> diff --git a/hw/dma/sifive_pdma.c b/hw/dma/sifive_pdma.c
> index d7d2c53e97e..b4fd40573a5 100644
> --- a/hw/dma/sifive_pdma.c
> +++ b/hw/dma/sifive_pdma.c
> @@ -80,7 +80,7 @@ static void sifive_pdma_run(SiFivePDMAState *s, int ch)
>
> /* do nothing if bytes to transfer is zero */
> if (!bytes) {
> - goto error;
> + goto done;
> }
>
> /*
> @@ -135,11 +135,6 @@ static void sifive_pdma_run(SiFivePDMAState *s, int ch)
> s->chan[ch].exec_bytes -= remainder;
> }
>
> - /* indicate a DMA transfer is done */
> - s->chan[ch].state = DMA_CHAN_STATE_DONE;
> - s->chan[ch].control &= ~CONTROL_RUN;
> - s->chan[ch].control |= CONTROL_DONE;
> -
> /* reload exec_ registers if repeat is required */
> if (s->chan[ch].next_config & CONFIG_REPEAT) {
> s->chan[ch].exec_bytes = bytes;
> @@ -147,6 +142,11 @@ static void sifive_pdma_run(SiFivePDMAState *s, int ch)
> s->chan[ch].exec_src = src;
> }
>
> +done:
> + /* indicate a DMA transfer is done */
> + s->chan[ch].state = DMA_CHAN_STATE_DONE;
> + s->chan[ch].control &= ~CONTROL_RUN;
> + s->chan[ch].control |= CONTROL_DONE;
> return;
>
> error:
> --
> 2.25.1
>
>
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2021-09-14 2:26 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-09-12 13:05 [PATCH RESEND v2 0/4] hw/dma: Align SiFive PDMA behavior with real hardware frank.chang
2021-09-12 13:05 ` [PATCH RESEND v2 1/4] hw/dma: sifive_pdma: reset Next* registers when Control.claim is set frank.chang
2021-09-12 13:05 ` [PATCH RESEND v2 2/4] hw/dma: sifive_pdma: claim bit must be set before DMA transactions frank.chang
2021-09-12 13:05 ` [PATCH RESEND v2 3/4] hw/dma: sifive_pdma: allow non-multiple transaction size transactions frank.chang
2021-09-12 13:05 ` [PATCH RESEND v2 4/4] hw/dma: sifive_pdma: don't set Control.error if 0 bytes to transfer frank.chang
2021-09-14 2:24 ` Alistair Francis
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).