Joel Fernandes
2025-Nov-02 23:59 UTC
[PATCH v2 08/12] nova-core: sequencer: Add register opcodes
These opcodes are used for register write, modify, poll and store (save)
sequencer operations.
Signed-off-by: Joel Fernandes <joelagnelf at nvidia.com>
---
drivers/gpu/nova-core/gsp/sequencer.rs | 138 +++++++++++++++++++++++--
1 file changed, 131 insertions(+), 7 deletions(-)
diff --git a/drivers/gpu/nova-core/gsp/sequencer.rs
b/drivers/gpu/nova-core/gsp/sequencer.rs
index 48c40140876b..f429fee01f86 100644
--- a/drivers/gpu/nova-core/gsp/sequencer.rs
+++ b/drivers/gpu/nova-core/gsp/sequencer.rs
@@ -5,6 +5,7 @@
use core::mem::size_of;
use kernel::alloc::flags::GFP_KERNEL;
use kernel::device;
+use kernel::io::poll::read_poll_timeout;
use kernel::prelude::*;
use kernel::time::Delta;
use kernel::transmute::FromBytes;
@@ -40,13 +41,36 @@ struct GspSequencerInfo<'a> {
/// GSP Sequencer Command types with payload data.
/// Commands have an opcode and a opcode-dependent struct.
-#[allow(dead_code)]
-pub(crate) enum GspSeqCmd {}
+#[allow(clippy::enum_variant_names)]
+pub(crate) enum GspSeqCmd {
+ RegWrite(fw::GSP_SEQ_BUF_PAYLOAD_REG_WRITE),
+ RegModify(fw::GSP_SEQ_BUF_PAYLOAD_REG_MODIFY),
+ RegPoll(fw::GSP_SEQ_BUF_PAYLOAD_REG_POLL),
+ RegStore(fw::GSP_SEQ_BUF_PAYLOAD_REG_STORE),
+}
impl GspSeqCmd {
/// Creates a new GspSeqCmd from a firmware GSP_SEQUENCER_BUFFER_CMD.
- pub(crate) fn from_fw_cmd(_cmd: &fw::GSP_SEQUENCER_BUFFER_CMD) ->
Result<Self> {
- Err(EINVAL)
+ pub(crate) fn from_fw_cmd(cmd: &fw::GSP_SEQUENCER_BUFFER_CMD) ->
Result<Self> {
+ match cmd.opCode {
+ fw::GSP_SEQ_BUF_OPCODE_GSP_SEQ_BUF_OPCODE_REG_WRITE => {
+ // SAFETY: We're using the union field that corresponds to
the opCode.
+ Ok(GspSeqCmd::RegWrite(unsafe { cmd.payload.regWrite }))
+ }
+ fw::GSP_SEQ_BUF_OPCODE_GSP_SEQ_BUF_OPCODE_REG_MODIFY => {
+ // SAFETY: We're using the union field that corresponds to
the opCode.
+ Ok(GspSeqCmd::RegModify(unsafe { cmd.payload.regModify }))
+ }
+ fw::GSP_SEQ_BUF_OPCODE_GSP_SEQ_BUF_OPCODE_REG_POLL => {
+ // SAFETY: We're using the union field that corresponds to
the opCode.
+ Ok(GspSeqCmd::RegPoll(unsafe { cmd.payload.regPoll }))
+ }
+ fw::GSP_SEQ_BUF_OPCODE_GSP_SEQ_BUF_OPCODE_REG_STORE => {
+ // SAFETY: We're using the union field that corresponds to
the opCode.
+ Ok(GspSeqCmd::RegStore(unsafe { cmd.payload.regStore }))
+ }
+ _ => Err(EINVAL),
+ }
}
pub(crate) fn new(data: &[u8], dev:
&device::Device<device::Bound>) -> Result<Self> {
@@ -64,7 +88,16 @@ pub(crate) fn new(data: &[u8], dev:
&device::Device<device::Bound>) -> Result<Se
/// Get the size of this command in bytes, the command consists of
/// a 4-byte opcode, and a variable-sized payload.
pub(crate) fn size_bytes(&self) -> usize {
- 0
+ let opcode_size = size_of::<fw::GSP_SEQ_BUF_OPCODE>();
+ match self {
+ // For commands with payloads, add the payload size in bytes.
+ GspSeqCmd::RegWrite(_) => opcode_size +
size_of::<fw::GSP_SEQ_BUF_PAYLOAD_REG_WRITE>(),
+ GspSeqCmd::RegModify(_) => {
+ opcode_size +
size_of::<fw::GSP_SEQ_BUF_PAYLOAD_REG_MODIFY>()
+ }
+ GspSeqCmd::RegPoll(_) => opcode_size +
size_of::<fw::GSP_SEQ_BUF_PAYLOAD_REG_POLL>(),
+ GspSeqCmd::RegStore(_) => opcode_size +
size_of::<fw::GSP_SEQ_BUF_PAYLOAD_REG_STORE>(),
+ }
}
}
@@ -83,12 +116,103 @@ pub(crate) trait GspSeqCmdRunner {
fn run(&self, sequencer: &GspSequencer<'_>) -> Result;
}
-impl GspSeqCmdRunner for GspSeqCmd {
- fn run(&self, _seq: &GspSequencer<'_>) -> Result {
+impl GspSeqCmdRunner for fw::GSP_SEQ_BUF_PAYLOAD_REG_WRITE {
+ fn run(&self, sequencer: &GspSequencer<'_>) -> Result
{
+ dev_dbg!(
+ sequencer.dev,
+ "RegWrite: addr=0x{:x}, val=0x{:x}\n",
+ self.addr,
+ self.val
+ );
+ let addr = self.addr as usize;
+ let val = self.val;
+ let _ = sequencer.bar.try_write32(val, addr);
+ Ok(())
+ }
+}
+
+impl GspSeqCmdRunner for fw::GSP_SEQ_BUF_PAYLOAD_REG_MODIFY {
+ fn run(&self, sequencer: &GspSequencer<'_>) -> Result
{
+ dev_dbg!(
+ sequencer.dev,
+ "RegModify: addr=0x{:x}, mask=0x{:x}, val=0x{:x}\n",
+ self.addr,
+ self.mask,
+ self.val
+ );
+
+ let addr = self.addr as usize;
+ if let Ok(temp) = sequencer.bar.try_read32(addr) {
+ let _ = sequencer
+ .bar
+ .try_write32((temp & !self.mask) | self.val, addr);
+ }
Ok(())
}
}
+impl GspSeqCmdRunner for fw::GSP_SEQ_BUF_PAYLOAD_REG_POLL {
+ fn run(&self, sequencer: &GspSequencer<'_>) -> Result
{
+ dev_dbg!(
+ sequencer.dev,
+ "RegPoll: addr=0x{:x}, mask=0x{:x}, val=0x{:x},
timeout=0x{:x}, error=0x{:x}\n",
+ self.addr,
+ self.mask,
+ self.val,
+ self.timeout,
+ self.error
+ );
+
+ let addr = self.addr as usize;
+ let mut timeout_us = i64::from(self.timeout);
+
+ // Default timeout to 4 seconds.
+ timeout_us = if timeout_us == 0 { 4000000 } else { timeout_us };
+
+ // First read.
+ sequencer.bar.try_read32(addr)?;
+
+ // Poll the requested register with requested timeout.
+ read_poll_timeout(
+ || sequencer.bar.try_read32(addr),
+ |current| (current & self.mask) == self.val,
+ Delta::ZERO,
+ Delta::from_micros(timeout_us),
+ )
+ .map(|_| ())
+ }
+}
+
+impl GspSeqCmdRunner for fw::GSP_SEQ_BUF_PAYLOAD_REG_STORE {
+ fn run(&self, sequencer: &GspSequencer<'_>) -> Result
{
+ let addr = self.addr as usize;
+ let _index = self.index;
+
+ let val = sequencer.bar.try_read32(addr)?;
+
+ dev_dbg!(
+ sequencer.dev,
+ "RegStore: addr=0x{:x}, index=0x{:x}, value={:?}\n",
+ self.addr,
+ self.index,
+ val
+ );
+
+ Ok(())
+ }
+}
+
+impl GspSeqCmdRunner for GspSeqCmd {
+ fn run(&self, seq: &GspSequencer<'_>) -> Result {
+ match self {
+ GspSeqCmd::RegWrite(cmd) => cmd.run(seq),
+ GspSeqCmd::RegModify(cmd) => cmd.run(seq),
+ GspSeqCmd::RegPoll(cmd) => cmd.run(seq),
+ GspSeqCmd::RegStore(cmd) => cmd.run(seq),
+ }
+ }
+}
+
pub(crate) struct GspSeqIter<'a> {
cmd_data: &'a [u8],
current_offset: usize, // Tracking the current position.
--
2.34.1
John Hubbard
2025-Nov-05 02:50 UTC
[PATCH v2 08/12] nova-core: sequencer: Add register opcodes
On 11/2/25 3:59 PM, Joel Fernandes wrote:> These opcodes are used for register write, modify, poll and store (save) > sequencer operations. > > Signed-off-by: Joel Fernandes <joelagnelf at nvidia.com> > --- > drivers/gpu/nova-core/gsp/sequencer.rs | 138 +++++++++++++++++++++++-- > 1 file changed, 131 insertions(+), 7 deletions(-)...> @@ -83,12 +116,103 @@ pub(crate) trait GspSeqCmdRunner { > fn run(&self, sequencer: &GspSequencer<'_>) -> Result; > } > > -impl GspSeqCmdRunner for GspSeqCmd { > - fn run(&self, _seq: &GspSequencer<'_>) -> Result { > +impl GspSeqCmdRunner for fw::GSP_SEQ_BUF_PAYLOAD_REG_WRITE { > + fn run(&self, sequencer: &GspSequencer<'_>) -> Result { > + dev_dbg!( > + sequencer.dev, > + "RegWrite: addr=0x{:x}, val=0x{:x}\n",Hi Joel, The RegRead, RegWrite, RegPoll prints generate over 400 lines per GPU, into the logs. This is too much, especially now that it's been working for a while. I'm thinking let's delete these entirely. If we somehow get into debugging this aspect of the sequencer, we can temporarily add whatever printing we need, but I think it's one notch too far for the final product, now that you have it working. thanks, -- John Hubbard
Alexandre Courbot
2025-Nov-10 13:50 UTC
[PATCH v2 08/12] nova-core: sequencer: Add register opcodes
On Mon Nov 3, 2025 at 8:59 AM JST, Joel Fernandes wrote:> These opcodes are used for register write, modify, poll and store (save) > sequencer operations. > > Signed-off-by: Joel Fernandes <joelagnelf at nvidia.com> > --- > drivers/gpu/nova-core/gsp/sequencer.rs | 138 +++++++++++++++++++++++-- > 1 file changed, 131 insertions(+), 7 deletions(-) > > diff --git a/drivers/gpu/nova-core/gsp/sequencer.rs b/drivers/gpu/nova-core/gsp/sequencer.rs > index 48c40140876b..f429fee01f86 100644 > --- a/drivers/gpu/nova-core/gsp/sequencer.rs > +++ b/drivers/gpu/nova-core/gsp/sequencer.rs > @@ -5,6 +5,7 @@ > use core::mem::size_of; > use kernel::alloc::flags::GFP_KERNEL; > use kernel::device; > +use kernel::io::poll::read_poll_timeout; > use kernel::prelude::*; > use kernel::time::Delta; > use kernel::transmute::FromBytes; > @@ -40,13 +41,36 @@ struct GspSequencerInfo<'a> { > > /// GSP Sequencer Command types with payload data. > /// Commands have an opcode and a opcode-dependent struct. > -#[allow(dead_code)] > -pub(crate) enum GspSeqCmd {} > +#[allow(clippy::enum_variant_names)] > +pub(crate) enum GspSeqCmd { > + RegWrite(fw::GSP_SEQ_BUF_PAYLOAD_REG_WRITE), > + RegModify(fw::GSP_SEQ_BUF_PAYLOAD_REG_MODIFY), > + RegPoll(fw::GSP_SEQ_BUF_PAYLOAD_REG_POLL), > + RegStore(fw::GSP_SEQ_BUF_PAYLOAD_REG_STORE), > +} > > impl GspSeqCmd { > /// Creates a new GspSeqCmd from a firmware GSP_SEQUENCER_BUFFER_CMD. > - pub(crate) fn from_fw_cmd(_cmd: &fw::GSP_SEQUENCER_BUFFER_CMD) -> Result<Self> { > - Err(EINVAL) > + pub(crate) fn from_fw_cmd(cmd: &fw::GSP_SEQUENCER_BUFFER_CMD) -> Result<Self> { > + match cmd.opCode { > + fw::GSP_SEQ_BUF_OPCODE_GSP_SEQ_BUF_OPCODE_REG_WRITE => { > + // SAFETY: We're using the union field that corresponds to the opCode. > + Ok(GspSeqCmd::RegWrite(unsafe { cmd.payload.regWrite })) > + } > + fw::GSP_SEQ_BUF_OPCODE_GSP_SEQ_BUF_OPCODE_REG_MODIFY => { > + // SAFETY: We're using the union field that corresponds to the opCode. > + Ok(GspSeqCmd::RegModify(unsafe { cmd.payload.regModify })) > + } > + fw::GSP_SEQ_BUF_OPCODE_GSP_SEQ_BUF_OPCODE_REG_POLL => { > + // SAFETY: We're using the union field that corresponds to the opCode. > + Ok(GspSeqCmd::RegPoll(unsafe { cmd.payload.regPoll })) > + } > + fw::GSP_SEQ_BUF_OPCODE_GSP_SEQ_BUF_OPCODE_REG_STORE => { > + // SAFETY: We're using the union field that corresponds to the opCode. > + Ok(GspSeqCmd::RegStore(unsafe { cmd.payload.regStore })) > + }I'd rather have these `unsafe` statements in the `fw` module - guess that's all he more reason to add an abstraction layer there.> + _ => Err(EINVAL), > + } > } > > pub(crate) fn new(data: &[u8], dev: &device::Device<device::Bound>) -> Result<Self> { > @@ -64,7 +88,16 @@ pub(crate) fn new(data: &[u8], dev: &device::Device<device::Bound>) -> Result<Se > /// Get the size of this command in bytes, the command consists of > /// a 4-byte opcode, and a variable-sized payload. > pub(crate) fn size_bytes(&self) -> usize { > - 0 > + let opcode_size = size_of::<fw::GSP_SEQ_BUF_OPCODE>(); > + match self { > + // For commands with payloads, add the payload size in bytes. > + GspSeqCmd::RegWrite(_) => opcode_size + size_of::<fw::GSP_SEQ_BUF_PAYLOAD_REG_WRITE>(), > + GspSeqCmd::RegModify(_) => { > + opcode_size + size_of::<fw::GSP_SEQ_BUF_PAYLOAD_REG_MODIFY>() > + } > + GspSeqCmd::RegPoll(_) => opcode_size + size_of::<fw::GSP_SEQ_BUF_PAYLOAD_REG_POLL>(), > + GspSeqCmd::RegStore(_) => opcode_size + size_of::<fw::GSP_SEQ_BUF_PAYLOAD_REG_STORE>(), > + } > } > } > > @@ -83,12 +116,103 @@ pub(crate) trait GspSeqCmdRunner { > fn run(&self, sequencer: &GspSequencer<'_>) -> Result; > } > > -impl GspSeqCmdRunner for GspSeqCmd { > - fn run(&self, _seq: &GspSequencer<'_>) -> Result { > +impl GspSeqCmdRunner for fw::GSP_SEQ_BUF_PAYLOAD_REG_WRITE { > + fn run(&self, sequencer: &GspSequencer<'_>) -> Result { > + dev_dbg!( > + sequencer.dev, > + "RegWrite: addr=0x{:x}, val=0x{:x}\n", > + self.addr, > + self.val > + );Per the other feedback I believe you are going to remove these `dev_dbg` anyway, but since the opcodes derive a `Debug` implementation, you could have just printed that from the caller for a similar result and no extra code.> + let addr = self.addr as usize; > + let val = self.val; > + let _ = sequencer.bar.try_write32(val, addr); > + Ok(()) > + } > +} > + > +impl GspSeqCmdRunner for fw::GSP_SEQ_BUF_PAYLOAD_REG_MODIFY { > + fn run(&self, sequencer: &GspSequencer<'_>) -> Result { > + dev_dbg!( > + sequencer.dev, > + "RegModify: addr=0x{:x}, mask=0x{:x}, val=0x{:x}\n", > + self.addr, > + self.mask, > + self.val > + ); > + > + let addr = self.addr as usize; > + if let Ok(temp) = sequencer.bar.try_read32(addr) { > + let _ = sequencer > + .bar > + .try_write32((temp & !self.mask) | self.val, addr); > + } > Ok(()) > } > } > > +impl GspSeqCmdRunner for fw::GSP_SEQ_BUF_PAYLOAD_REG_POLL { > + fn run(&self, sequencer: &GspSequencer<'_>) -> Result { > + dev_dbg!( > + sequencer.dev, > + "RegPoll: addr=0x{:x}, mask=0x{:x}, val=0x{:x}, timeout=0x{:x}, error=0x{:x}\n", > + self.addr, > + self.mask, > + self.val, > + self.timeout, > + self.error > + ); > + > + let addr = self.addr as usize; > + let mut timeout_us = i64::from(self.timeout); > + > + // Default timeout to 4 seconds. > + timeout_us = if timeout_us == 0 { 4000000 } else { timeout_us };`let timeout_us = if self.timeout == 0 { 4_000_000 } else { i64::from(self.timeout)` (removes the `mut` on `timeout_us`)> + > + // First read. > + sequencer.bar.try_read32(addr)?; > + > + // Poll the requested register with requested timeout. > + read_poll_timeout( > + || sequencer.bar.try_read32(addr), > + |current| (current & self.mask) == self.val, > + Delta::ZERO, > + Delta::from_micros(timeout_us), > + ) > + .map(|_| ()) > + } > +} > + > +impl GspSeqCmdRunner for fw::GSP_SEQ_BUF_PAYLOAD_REG_STORE { > + fn run(&self, sequencer: &GspSequencer<'_>) -> Result { > + let addr = self.addr as usize; > + let _index = self.index; > + > + let val = sequencer.bar.try_read32(addr)?; > + > + dev_dbg!( > + sequencer.dev, > + "RegStore: addr=0x{:x}, index=0x{:x}, value={:?}\n", > + self.addr, > + self.index, > + val > + ); > + > + Ok(()) > + } > +} > + > +impl GspSeqCmdRunner for GspSeqCmd { > + fn run(&self, seq: &GspSequencer<'_>) -> Result { > + match self { > + GspSeqCmd::RegWrite(cmd) => cmd.run(seq), > + GspSeqCmd::RegModify(cmd) => cmd.run(seq), > + GspSeqCmd::RegPoll(cmd) => cmd.run(seq), > + GspSeqCmd::RegStore(cmd) => cmd.run(seq), > + } > + } > +}This makes me wonder: do we need to store the deserialized version of these operands, and make a second `match` on them? How about passing the `bar` to the deserialization command and have it run the operand immediately?