Files
gem5/src/dev/amdgpu/interrupt_handler.cc
Matthew Poremba 8b91ac6f8d dev-amdgpu: Refactor MMIO interface for SDMA engines
Currently the amdgpu simulated device is assumed to be a Vega10. As a
result there are a few things that are hardcoded. One of those is the
number of SDMAs. In order to add a newer device, such as MI100+, we need
to enable a flexible number of SDMAs.

In order to support a variable number of SDMAs and with the MMIO offsets
of each device being potentially different, the MMIO interface for SDMAs
is changed to use an SDMA class method dispatch table with forwards a
32-bit value from the MMIO packet to the MMIO functions in SDMA of the
format `void method(uint32_t)`. Several changes are made to enable this:

 - Allow the SDMA to have a variable MMIO base and size. These are
   configured in python.
 - An SDMA class method dispatch table which contains the MMIO offset
   relative to the SDMA's MMIO base address.
 - An updated writeMMIO method to iterate over the SDMA MMIO address
   ranges and call the appropriate SDMA MMIO method which matches the
   MMIO offset.
 - Moved all SDMA related MMIO data bit twiddling, masking, etc. into
   the MMIO methods themselves instead of in the writeMMIO method in
   SDMAEngine.

Change-Id: Ifce626f84d52f9e27e4438ba4e685e30dbf06dbc
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/70040
Maintainer: Matt Sinclair <mattdsinclair@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
2023-04-28 00:48:35 +00:00

323 lines
10 KiB
C++

/*
* Copyright (c) 2021 Advanced Micro Devices, Inc.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* 3. Neither the name of the copyright holder nor the names of its
* contributors may be used to endorse or promote products derived from this
* software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*
*/
#include "dev/amdgpu/interrupt_handler.hh"
#include "debug/AMDGPUDevice.hh"
#include "dev/amdgpu/ih_mmio.hh"
#include "mem/packet_access.hh"
// For generating interrupts, the object causing interrupt communicates with
// the Interrupt Handler (IH), which submits a 256-bit Interrupt packet to the
// system memory. The location where the IH submits the packet is the
// IH Ring buffer in the system memory. The IH updates the Write Pointer
// and the host consumes the ring buffer and once done, updates the Read
// Pointer through the doorbell.
// IH_RB_BaseAddr, IH_RB_WptrAddr (Lo/Hi), IH_RB_RptrAddr (Lo/Hi), etc. are
// not GART addresses but system dma addresses and thus don't require
// translations through the GART table.
namespace gem5
{
AMDGPUInterruptHandler::AMDGPUInterruptHandler(
const AMDGPUInterruptHandlerParams &p)
: DmaDevice(p)
{
memset(&regs, 0, sizeof(AMDGPUIHRegs));
}
AddrRangeList
AMDGPUInterruptHandler::getAddrRanges() const
{
AddrRangeList ranges;
return ranges;
}
void
AMDGPUInterruptHandler::intrPost()
{
if (gpuDevice)
gpuDevice->intrPost();
}
void
AMDGPUInterruptHandler::prepareInterruptCookie(ContextID cntxt_id,
uint32_t ring_id,
uint32_t client_id,
uint32_t source_id)
{
assert(client_id == SOC15_IH_CLIENTID_RLC ||
client_id == SOC15_IH_CLIENTID_SDMA0 ||
client_id == SOC15_IH_CLIENTID_SDMA1 ||
client_id == SOC15_IH_CLIENTID_SDMA2 ||
client_id == SOC15_IH_CLIENTID_SDMA3 ||
client_id == SOC15_IH_CLIENTID_SDMA4 ||
client_id == SOC15_IH_CLIENTID_SDMA5 ||
client_id == SOC15_IH_CLIENTID_SDMA6 ||
client_id == SOC15_IH_CLIENTID_SDMA7 ||
client_id == SOC15_IH_CLIENTID_GRBM_CP);
assert(source_id == CP_EOP || source_id == TRAP_ID);
/**
* Setup the fields in the interrupt cookie (see header file for more
* detail on the fields). The timestamp here is a bogus value. It seems
* the driver does not really care what this value is. Additionally the
* model does not currently have anything to keep track of time. It is
* possible that tick/cycle count can be used in the future if this ends
* up being important. The remaining fields are passed from whichever
* block is sending the interrupt.
*/
AMDGPUInterruptCookie *cookie = new AMDGPUInterruptCookie();
memset(cookie, 0, sizeof(AMDGPUInterruptCookie));
// Currently only one process is supported and the first pasid from driver
// is always 0x8000. In the future this can be obtained from the PM4
// MAP_PROCESS packet and may need to be passed to this function.
//
// On a related note, leave vmid fields alone as they are only used for
// memory exceptions. Memory exceptions are not supported on gfx900.
cookie->pasid = 0x8000;
cookie->timestamp_Lo = 0x40;
cookie->clientId = client_id;
cookie->sourceId = source_id;
cookie->ringId = ring_id;
cookie->source_data_dw1 = cntxt_id;
interruptQueue.push(cookie);
}
void
AMDGPUInterruptHandler::DmaEvent::process()
{
if (data == 1) {
DPRINTF(AMDGPUDevice, "Completed interrupt cookie write\n");
deviceIh->submitWritePointer();
} else if (data == 2) {
DPRINTF(AMDGPUDevice, "Completed interrupt write pointer update\n");
deviceIh->intrPost();
} else {
fatal("Interrupt Handler DMA event returned bad value: %d\n", data);
}
}
void
AMDGPUInterruptHandler::submitWritePointer()
{
uint8_t *dataPtr = new uint8_t[sizeof(uint32_t)];
regs.IH_Wptr += sizeof(AMDGPUInterruptCookie);
Addr paddr = regs.WptrAddr;
std::memcpy(dataPtr, &regs.IH_Wptr, sizeof(uint32_t));
dmaEvent = new AMDGPUInterruptHandler::DmaEvent(this, 2);
dmaWrite(paddr, sizeof(uint32_t), dmaEvent, dataPtr);
}
void
AMDGPUInterruptHandler::submitInterruptCookie()
{
assert(!interruptQueue.empty());
auto cookie = interruptQueue.front();
size_t cookieSize = sizeof(AMDGPUInterruptCookie);
uint8_t *dataPtr = new uint8_t[cookieSize];
std::memcpy(dataPtr, cookie, cookieSize);
Addr paddr = regs.baseAddr + regs.IH_Wptr;
DPRINTF(AMDGPUDevice, "InterruptHandler rptr: 0x%x wptr: 0x%x\n",
regs.IH_Rptr, regs.IH_Wptr);
dmaEvent = new AMDGPUInterruptHandler::DmaEvent(this, 1);
dmaWrite(paddr, cookieSize, dmaEvent, dataPtr);
interruptQueue.pop();
}
void
AMDGPUInterruptHandler::writeMMIO(PacketPtr pkt, Addr mmio_offset)
{
switch (mmio_offset) {
case mmIH_RB_CNTL:
setCntl(pkt->getLE<uint32_t>());
break;
case mmIH_RB_BASE:
setBase(pkt->getLE<uint32_t>());
break;
case mmIH_RB_BASE_HI:
setBaseHi(pkt->getLE<uint32_t>());
break;
case mmIH_RB_RPTR:
setRptr(pkt->getLE<uint32_t>());
break;
case mmIH_RB_WPTR:
setWptr(pkt->getLE<uint32_t>());
break;
case mmIH_RB_WPTR_ADDR_LO:
setWptrAddrLo(pkt->getLE<uint32_t>());
break;
case mmIH_RB_WPTR_ADDR_HI:
setWptrAddrHi(pkt->getLE<uint32_t>());
break;
case mmIH_DOORBELL_RPTR:
setDoorbellOffset(pkt->getLE<uint32_t>());
if (bits(pkt->getLE<uint32_t>(), 28, 28)) {
gpuDevice->setDoorbellType(getDoorbellOffset() << 2,
InterruptHandler);
}
break;
default:
DPRINTF(AMDGPUDevice, "IH Unknown MMIO %#x\n", mmio_offset);
break;
}
}
void
AMDGPUInterruptHandler::setCntl(const uint32_t &data)
{
regs.IH_Cntl = data;
}
void
AMDGPUInterruptHandler::setBase(const uint32_t &data)
{
regs.baseAddr = data;
regs.baseAddr <<= 8;
}
void
AMDGPUInterruptHandler::setBaseHi(const uint32_t &data)
{
regs.baseAddr |= static_cast<uint64_t>(data) << 40;
}
void
AMDGPUInterruptHandler::setRptr(const uint32_t &data)
{
regs.IH_Rptr = data;
}
void
AMDGPUInterruptHandler::setWptr(const uint32_t &data)
{
regs.IH_Wptr = data;
}
void
AMDGPUInterruptHandler::setWptrAddrLo(const uint32_t &data)
{
regs.IH_Wptr_Addr_Lo = data;
regs.WptrAddr |= regs.IH_Wptr_Addr_Lo;
}
void
AMDGPUInterruptHandler::setWptrAddrHi(const uint32_t &data)
{
regs.IH_Wptr_Addr_Hi = data;
regs.WptrAddr |= ((uint64_t)regs.IH_Wptr_Addr_Hi) << 32;
}
void
AMDGPUInterruptHandler::setDoorbellOffset(const uint32_t &data)
{
regs.IH_Doorbell = data & 0x3ffffff;
}
void
AMDGPUInterruptHandler::updateRptr(const uint32_t &data)
{
regs.IH_Rptr = data; // update ring buffer rptr offset
}
void
AMDGPUInterruptHandler::serialize(CheckpointOut &cp) const
{
uint32_t ih_cntl = regs.IH_Cntl;
uint32_t ih_base = regs.IH_Base;
uint32_t ih_base_hi = regs.IH_Base_Hi;
Addr ih_baseAddr = regs.baseAddr;
uint32_t ih_rptr = regs.IH_Rptr;
uint32_t ih_wptr = regs.IH_Wptr;
uint32_t ih_wptr_addr_lo = regs.IH_Wptr_Addr_Lo;
uint32_t ih_wptr_addr_hi = regs.IH_Wptr_Addr_Hi;
Addr ih_wptrAddr = regs.WptrAddr;
uint32_t ih_doorbellOffset = regs.IH_Doorbell;
SERIALIZE_SCALAR(ih_cntl);
SERIALIZE_SCALAR(ih_base);
SERIALIZE_SCALAR(ih_base_hi);
SERIALIZE_SCALAR(ih_baseAddr);
SERIALIZE_SCALAR(ih_rptr);
SERIALIZE_SCALAR(ih_wptr);
SERIALIZE_SCALAR(ih_wptr_addr_lo);
SERIALIZE_SCALAR(ih_wptr_addr_hi);
SERIALIZE_SCALAR(ih_wptrAddr);
SERIALIZE_SCALAR(ih_doorbellOffset);
}
void
AMDGPUInterruptHandler::unserialize(CheckpointIn &cp)
{
uint32_t ih_cntl;
uint32_t ih_base;
uint32_t ih_base_hi;
Addr ih_baseAddr;
uint32_t ih_rptr;
uint32_t ih_wptr;
uint32_t ih_wptr_addr_lo;
uint32_t ih_wptr_addr_hi;
Addr ih_wptrAddr;
uint32_t ih_doorbellOffset;
UNSERIALIZE_SCALAR(ih_cntl);
UNSERIALIZE_SCALAR(ih_base);
UNSERIALIZE_SCALAR(ih_base_hi);
UNSERIALIZE_SCALAR(ih_baseAddr);
UNSERIALIZE_SCALAR(ih_rptr);
UNSERIALIZE_SCALAR(ih_wptr);
UNSERIALIZE_SCALAR(ih_wptr_addr_lo);
UNSERIALIZE_SCALAR(ih_wptr_addr_hi);
UNSERIALIZE_SCALAR(ih_wptrAddr);
UNSERIALIZE_SCALAR(ih_doorbellOffset);
regs.IH_Cntl = ih_cntl;
regs.IH_Base = ih_base;
regs.IH_Base_Hi = ih_base_hi;
regs.baseAddr = ih_baseAddr;
regs.IH_Rptr = ih_rptr;
regs.IH_Wptr = ih_wptr;
regs.IH_Wptr_Addr_Lo = ih_wptr_addr_lo;
regs.IH_Wptr_Addr_Hi = ih_wptr_addr_hi;
regs.WptrAddr = ih_wptrAddr;
regs.IH_Doorbell = ih_doorbellOffset;
}
} // namespace gem5