k_server_session: remove scratch buffer usage in favor of direct copy

This commit is contained in:
Liam 2023-12-17 20:22:38 -05:00
parent 35501ba41c
commit cae675343c

View file

@ -37,8 +37,6 @@ constexpr inline size_t ReceiveListDataSize =
using ThreadQueueImplForKServerSessionRequest = KThreadQueue;
static thread_local Common::ScratchBuffer<u8> temp_buffer;
class ReceiveList {
public:
static constexpr int GetEntryCount(const MessageBuffer::MessageHeader& header) {
@ -269,12 +267,20 @@ Result ProcessReceiveMessagePointerDescriptors(int& offset, int& pointer_key,
R_UNLESS(recv_pointer != 0, ResultOutOfResource);
// Perform the pointer data copy.
// TODO: KProcessPageTable::CopyMemoryFromHeapToHeapWithoutCheckDestination
// TODO: KProcessPageTable::CopyMemoryFromLinearToUser
temp_buffer.resize_destructive(recv_size);
src_page_table.GetMemory().ReadBlock(src_pointer, temp_buffer.data(), recv_size);
dst_page_table.GetMemory().WriteBlock(recv_pointer, temp_buffer.data(), recv_size);
if (dst_user) {
R_TRY(src_page_table.CopyMemoryFromHeapToHeapWithoutCheckDestination(
dst_page_table, recv_pointer, recv_size, KMemoryState::FlagReferenceCounted,
KMemoryState::FlagReferenceCounted,
KMemoryPermission::NotMapped | KMemoryPermission::KernelReadWrite,
KMemoryAttribute::Uncached | KMemoryAttribute::Locked, KMemoryAttribute::Locked,
src_pointer, KMemoryState::FlagLinearMapped, KMemoryState::FlagLinearMapped,
KMemoryPermission::UserRead, KMemoryAttribute::Uncached, KMemoryAttribute::None));
} else {
R_TRY(src_page_table.CopyMemoryFromLinearToUser(
recv_pointer, recv_size, src_pointer, KMemoryState::FlagLinearMapped,
KMemoryState::FlagLinearMapped, KMemoryPermission::UserRead,
KMemoryAttribute::Uncached, KMemoryAttribute::None));
}
}
// Set the output descriptor.
@ -303,21 +309,22 @@ constexpr Result GetMapAliasMemoryState(KMemoryState& out,
R_SUCCEED();
}
constexpr Result GetMapAliasTestStateAndAttributeMask(u32& out_state, u32& out_attr_mask,
constexpr Result GetMapAliasTestStateAndAttributeMask(KMemoryState& out_state,
KMemoryAttribute& out_attr_mask,
KMemoryState state) {
switch (state) {
case KMemoryState::Ipc:
out_state = static_cast<u32>(KMemoryState::FlagCanUseIpc);
out_attr_mask = static_cast<u32>(KMemoryAttribute::Uncached |
KMemoryAttribute::DeviceShared | KMemoryAttribute::Locked);
out_state = KMemoryState::FlagCanUseIpc;
out_attr_mask =
KMemoryAttribute::Uncached | KMemoryAttribute::DeviceShared | KMemoryAttribute::Locked;
break;
case KMemoryState::NonSecureIpc:
out_state = static_cast<u32>(KMemoryState::FlagCanUseNonSecureIpc);
out_attr_mask = static_cast<u32>(KMemoryAttribute::Uncached | KMemoryAttribute::Locked);
out_state = KMemoryState::FlagCanUseNonSecureIpc;
out_attr_mask = KMemoryAttribute::Uncached | KMemoryAttribute::Locked;
break;
case KMemoryState::NonDeviceIpc:
out_state = static_cast<u32>(KMemoryState::FlagCanUseNonDeviceIpc);
out_attr_mask = static_cast<u32>(KMemoryAttribute::Uncached | KMemoryAttribute::Locked);
out_state = KMemoryState::FlagCanUseNonDeviceIpc;
out_attr_mask = KMemoryAttribute::Uncached | KMemoryAttribute::Locked;
break;
default:
R_THROW(ResultInvalidCombination);
@ -708,13 +715,48 @@ Result ReceiveMessage(KernelCore& kernel, bool& recv_list_broken, uint64_t dst_m
if (!dst_user && !src_user) {
// Fast case is TLS -> TLS, do raw memcpy if we can.
std::memcpy(dst_msg_ptr + offset, src_msg_ptr + offset, raw_size);
} else {
} else if (dst_user) {
// Determine how much fast size we can copy.
const size_t max_fast_size = std::min<size_t>(offset_words + raw_size, PageSize);
const size_t fast_size = max_fast_size - offset_words;
// Determine source state; if user buffer, we require heap, and otherwise only linear
// mapped (to enable tls use).
const auto src_state =
src_user ? KMemoryState::FlagReferenceCounted : KMemoryState::FlagLinearMapped;
// Determine the source permission. User buffer should be unmapped + read, TLS should be
// user readable.
const KMemoryPermission src_perm = static_cast<KMemoryPermission>(
src_user ? KMemoryPermission::NotMapped | KMemoryPermission::KernelRead
: KMemoryPermission::UserRead);
// Perform the fast part of the copy.
R_TRY(src_page_table.CopyMemoryFromLinearToKernel(
dst_msg_ptr + offset, fast_size, src_message_buffer + offset_words, src_state,
src_state, src_perm, KMemoryAttribute::Uncached, KMemoryAttribute::None));
// If the fast part of the copy didn't get everything, perform the slow part of the
// copy.
if (fast_size < raw_size) {
R_TRY(src_page_table.CopyMemoryFromHeapToHeap(
dst_page_table, dst_message_buffer + max_fast_size, raw_size - fast_size,
KMemoryState::FlagReferenceCounted, KMemoryState::FlagReferenceCounted,
KMemoryPermission::NotMapped | KMemoryPermission::KernelReadWrite,
KMemoryAttribute::Uncached | KMemoryAttribute::Locked, KMemoryAttribute::Locked,
src_message_buffer + max_fast_size, src_state, src_state, src_perm,
KMemoryAttribute::Uncached, KMemoryAttribute::None));
}
} else /* if (src_user) */ {
// The source is a user buffer, so it should be unmapped + readable.
constexpr KMemoryPermission SourcePermission = static_cast<KMemoryPermission>(
KMemoryPermission::NotMapped | KMemoryPermission::KernelRead);
// Copy the memory.
temp_buffer.resize_destructive(raw_size);
src_page_table.GetMemory().ReadBlock(src_message_buffer + offset_words,
temp_buffer.data(), raw_size);
dst_page_table.GetMemory().WriteBlock(dst_message_buffer + offset_words,
temp_buffer.data(), raw_size);
R_TRY(src_page_table.CopyMemoryFromLinearToUser(
dst_message_buffer + offset_words, raw_size, src_message_buffer + offset_words,
KMemoryState::FlagReferenceCounted, KMemoryState::FlagReferenceCounted,
SourcePermission, KMemoryAttribute::Uncached, KMemoryAttribute::None));
}
}
@ -731,8 +773,8 @@ Result ProcessSendMessageReceiveMapping(KProcessPageTable& src_page_table,
R_SUCCEED_IF(size == 0);
// Get the memory state and attribute mask to test.
u32 test_state;
u32 test_attr_mask;
KMemoryState test_state;
KMemoryAttribute test_attr_mask;
R_TRY(GetMapAliasTestStateAndAttributeMask(test_state, test_attr_mask, src_state));
// Determine buffer extents.
@ -749,18 +791,18 @@ Result ProcessSendMessageReceiveMapping(KProcessPageTable& src_page_table,
if (aligned_dst_start != mapping_dst_start) {
ASSERT(client_address < mapping_dst_start);
const size_t copy_size = std::min<size_t>(size, mapping_dst_start - client_address);
temp_buffer.resize_destructive(copy_size);
src_page_table.GetMemory().ReadBlock(client_address, temp_buffer.data(), copy_size);
dst_page_table.GetMemory().WriteBlock(server_address, temp_buffer.data(), copy_size);
R_TRY(dst_page_table.CopyMemoryFromUserToLinear(
client_address, copy_size, test_state, test_state, KMemoryPermission::UserReadWrite,
test_attr_mask, KMemoryAttribute::None, server_address));
}
// If the end of the buffer is unaligned, handle that.
if (mapping_dst_end < aligned_dst_end &&
(aligned_dst_start == mapping_dst_start || aligned_dst_start < mapping_dst_end)) {
const size_t copy_size = client_address + size - mapping_dst_end;
temp_buffer.resize_destructive(copy_size);
src_page_table.GetMemory().ReadBlock(mapping_src_end, temp_buffer.data(), copy_size);
dst_page_table.GetMemory().WriteBlock(mapping_dst_end, temp_buffer.data(), copy_size);
R_TRY(dst_page_table.CopyMemoryFromUserToLinear(
mapping_dst_end, copy_size, test_state, test_state, KMemoryPermission::UserReadWrite,
test_attr_mask, KMemoryAttribute::None, mapping_src_end));
}
R_SUCCEED();
@ -796,9 +838,15 @@ Result ProcessSendMessagePointerDescriptors(int& offset, int& pointer_key,
R_UNLESS(recv_pointer != 0, ResultOutOfResource);
// Perform the pointer data copy.
temp_buffer.resize_destructive(recv_size);
src_page_table.GetMemory().ReadBlock(src_pointer, temp_buffer.data(), recv_size);
dst_page_table.GetMemory().WriteBlock(recv_pointer, temp_buffer.data(), recv_size);
const bool dst_heap = dst_user && dst_recv_list.IsToMessageBuffer();
const auto dst_state =
dst_heap ? KMemoryState::FlagReferenceCounted : KMemoryState::FlagLinearMapped;
const KMemoryPermission dst_perm =
dst_heap ? KMemoryPermission::NotMapped | KMemoryPermission::KernelReadWrite
: KMemoryPermission::UserReadWrite;
R_TRY(dst_page_table.CopyMemoryFromUserToLinear(
recv_pointer, recv_size, dst_state, dst_state, dst_perm, KMemoryAttribute::Uncached,
KMemoryAttribute::None, src_pointer));
}
// Set the output descriptor.
@ -964,13 +1012,50 @@ Result SendMessage(KernelCore& kernel, uint64_t src_message_buffer, size_t src_b
if (!dst_user && !src_user) {
// Fast case is TLS -> TLS, do raw memcpy if we can.
std::memcpy(dst_msg_ptr + offset, src_msg_ptr + offset, raw_size);
} else {
} else if (src_user) {
// Determine how much fast size we can copy.
const size_t max_fast_size = std::min<size_t>(offset_words + raw_size, PageSize);
const size_t fast_size = max_fast_size - offset_words;
// Determine dst state; if user buffer, we require heap, and otherwise only linear
// mapped (to enable tls use).
const auto dst_state =
dst_user ? KMemoryState::FlagReferenceCounted : KMemoryState::FlagLinearMapped;
// Determine the dst permission. User buffer should be unmapped + read, TLS should
// be user readable.
const KMemoryPermission dst_perm =
dst_user ? KMemoryPermission::NotMapped | KMemoryPermission::KernelReadWrite
: KMemoryPermission::UserReadWrite;
// Perform the fast part of the copy.
R_TRY(dst_page_table.CopyMemoryFromKernelToLinear(
dst_message_buffer + offset_words, fast_size, dst_state, dst_state, dst_perm,
KMemoryAttribute::Uncached, KMemoryAttribute::None, src_msg_ptr + offset));
// If the fast part of the copy didn't get everything, perform the slow part of the
// copy.
if (fast_size < raw_size) {
R_TRY(dst_page_table.CopyMemoryFromHeapToHeap(
dst_page_table, dst_message_buffer + max_fast_size, raw_size - fast_size,
dst_state, dst_state, dst_perm, KMemoryAttribute::Uncached,
KMemoryAttribute::None, src_message_buffer + max_fast_size,
KMemoryState::FlagReferenceCounted, KMemoryState::FlagReferenceCounted,
KMemoryPermission::NotMapped | KMemoryPermission::KernelRead,
KMemoryAttribute::Uncached | KMemoryAttribute::Locked,
KMemoryAttribute::Locked));
}
} else /* if (dst_user) */ {
// The destination is a user buffer, so it should be unmapped + readable.
constexpr KMemoryPermission DestinationPermission =
KMemoryPermission::NotMapped | KMemoryPermission::KernelReadWrite;
// Copy the memory.
temp_buffer.resize_destructive(raw_size);
src_page_table.GetMemory().ReadBlock(src_message_buffer + offset_words,
temp_buffer.data(), raw_size);
dst_page_table.GetMemory().WriteBlock(dst_message_buffer + offset_words,
temp_buffer.data(), raw_size);
R_TRY(dst_page_table.CopyMemoryFromUserToLinear(
dst_message_buffer + offset_words, raw_size, KMemoryState::FlagReferenceCounted,
KMemoryState::FlagReferenceCounted, DestinationPermission,
KMemoryAttribute::Uncached, KMemoryAttribute::None,
src_message_buffer + offset_words));
}
}
}