[crypto] Rework AES CTR/XTS streaming and squash heap churn (#2782)

AES Updates:
Replaced heap churn with stack scratch buffers tail handling now stays in-place, no more recursive transcode detours.
CTR/XTS modes read in larger, aligned chunks and still handle odd offsets cleanly.
XTS prefetches a few sectors ahead to reduce extra reads.
AesCtrStorage writer now uses the pooled buffer properly one stack slab, chunk forward, bump counter, repeat.
Result: less malloc noise, fewer watchdog spikes at startup (though mbedtls still sets the pace).
This should make the loading speed slightly better than before. Make sure to test.

Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/2782
Reviewed-by: MaranBr <maranbr@eden-emu.dev>
Reviewed-by: crueter <crueter@eden-emu.dev>
Co-authored-by: godpow <thesaviorsrule@yahoo.com>
Co-committed-by: godpow <thesaviorsrule@yahoo.com>

1 files changed, 38 insertions, 56 deletions

diff --git a/src/core/crypto/xts_encryption_layer.cpp b/src/core/crypto/xts_encryption_layer.cpp
index 240f930c57..5cfc67847d 100644
--- a/src/core/crypto/xts_encryption_layer.cpp
+++ b/src/core/crypto/xts_encryption_layer.cpp
@@ -20,67 +20,49 @@ std::size_t XTSEncryptionLayer::Read(u8* data, std::size_t length, std::size_t o
     if (length == 0)
         return 0;
 
+    constexpr std::size_t PrefetchSectors = 4;
+
+    auto* out = data;
+    std::size_t remaining = length;
+    std::size_t current_offset = offset;
     std::size_t total_read = 0;
-    // Handle initial unaligned part within a sector.
-    if (auto const sector_offset = offset % XTS_SECTOR_SIZE; sector_offset != 0) {
-        const std::size_t aligned_off = offset - sector_offset;
-        std::array<u8, XTS_SECTOR_SIZE> block{};
-        if (auto const got = base->Read(block.data(), XTS_SECTOR_SIZE, aligned_off); got > 0) {
-            if (got < XTS_SECTOR_SIZE)
-                std::memset(block.data() + got, 0, XTS_SECTOR_SIZE - got);
-            cipher.XTSTranscode(block.data(), XTS_SECTOR_SIZE, block.data(), aligned_off / XTS_SECTOR_SIZE,
-                                XTS_SECTOR_SIZE, Op::Decrypt);
-
-            auto const to_copy = std::min<std::size_t>(length, got > sector_offset ? got - sector_offset : 0);
-            if (to_copy > 0) {
-                std::memcpy(data, block.data() + sector_offset, to_copy);
-                data += to_copy;
-                offset += to_copy;
-                length -= to_copy;
-                total_read += to_copy;
-            }
-        } else {
-            return 0;
-        }
-    }
 
-    if (length > 0) {
-        // Process aligned middle inplace, in sector sized multiples.
-        while (length >= XTS_SECTOR_SIZE) {
-            const std::size_t req = (length / XTS_SECTOR_SIZE) * XTS_SECTOR_SIZE;
-            const std::size_t got = base->Read(data, req, offset);
-            if (got == 0) {
+    std::array<u8, XTS_SECTOR_SIZE> sector{};
+
+    while (remaining > 0) {
+        const std::size_t sector_index = current_offset / XTS_SECTOR_SIZE;
+        const std::size_t sector_offset = current_offset % XTS_SECTOR_SIZE;
+
+        const std::size_t sectors_to_read = std::min<std::size_t>(PrefetchSectors,
+                                                                  (remaining + sector_offset +
+                                                                   XTS_SECTOR_SIZE - 1) /
+                                                                      XTS_SECTOR_SIZE);
+
+        for (std::size_t s = 0; s < sectors_to_read && remaining > 0; ++s) {
+            const std::size_t index = sector_index + s;
+            const std::size_t read_offset = index * XTS_SECTOR_SIZE;
+            const std::size_t got = base->Read(sector.data(), XTS_SECTOR_SIZE, read_offset);
+            if (got == 0)
                 return total_read;
-            }
-            const std::size_t got_rounded = got - (got % XTS_SECTOR_SIZE);
-            if (got_rounded > 0) {
-                cipher.XTSTranscode(data, got_rounded, data, offset / XTS_SECTOR_SIZE, XTS_SECTOR_SIZE, Op::Decrypt);
-                data += got_rounded;
-                offset += got_rounded;
-                length -= got_rounded;
-                total_read += got_rounded;
-            }
-            // If we didn't get a full sector next, break to handle tail.
-            if (got_rounded != got) {
-                break;
-            }
-        }
-        // Handle tail within a sector, if any.
-        if (length > 0) {
-            std::array<u8, XTS_SECTOR_SIZE> block{};
-            const std::size_t got = base->Read(block.data(), XTS_SECTOR_SIZE, offset);
-            if (got > 0) {
-                if (got < XTS_SECTOR_SIZE) {
-                    std::memset(block.data() + got, 0, XTS_SECTOR_SIZE - got);
-                }
-                cipher.XTSTranscode(block.data(), XTS_SECTOR_SIZE, block.data(),
-                                    offset / XTS_SECTOR_SIZE, XTS_SECTOR_SIZE, Op::Decrypt);
-                const std::size_t to_copy = std::min<std::size_t>(length, got);
-                std::memcpy(data, block.data(), to_copy);
-                total_read += to_copy;
-            }
+
+            if (got < XTS_SECTOR_SIZE)
+                std::memset(sector.data() + got, 0, XTS_SECTOR_SIZE - got);
+
+            cipher.XTSTranscode(sector.data(), XTS_SECTOR_SIZE, sector.data(), index, XTS_SECTOR_SIZE,
+                                Op::Decrypt);
+
+            const std::size_t local_offset = (s == 0) ? sector_offset : 0;
+            const std::size_t available = XTS_SECTOR_SIZE - local_offset;
+            const std::size_t to_copy = std::min<std::size_t>(available, remaining);
+            std::memcpy(out, sector.data() + local_offset, to_copy);
+
+            out += to_copy;
+            current_offset += to_copy;
+            remaining -= to_copy;
+            total_read += to_copy;
         }
     }
+
     return total_read;
 }
 } // namespace Core::Crypto