Merge pull request #10465 from xXPBSXx/docs-agc-copyregion-over-4gib

docs(agc): note the guest buffer region copied whole over 4 GiB

docs/dev/debt/libSceAgcDriver.md

@@ -53,6 +53,7 @@
 - [Strided buffer overhang](../../../core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp) (libSceAgcDriver) - a strided buffer descriptor that ends less than one stride past the end of the readable registered allocation holding its base is bound only up to that allocation's end, even when another readable allocation starts right there. This assumes the bytes past the end are the rounded-up `num_records` of the last record (RADV rounds it up), which no fetch inside the buffer reads; a guest that really keeps a vertex split across two adjacent allocations would read zeros for the cut tail. The same holds for a read-only strided buffer that starts below the render target and runs less than one stride into it: it is bound only up to the target, as a vertex fetch whose attribute descriptor starts at the attribute's offset reaches that far past the attribute's last bytes. Taken from the RetroArch Dolphin core case in #2912, not measured on hardware.
 - [Buffer descriptor CACHE_SWIZZLE hint](../../../core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp) (libSceAgcDriver) - bit 30 of word 1 in a buffer resource descriptor (`words[1] & 0x40000000u`) is normalized to zero when binding guest buffers in `ShaderResources::addGuestBuffer`. On GFX10 (gfx1013), bit 30 of V# word 1 is `CACHE_SWIZZLE` (Mesa's `gfx10-rsrc.json`: STRIDE [29:16], CACHE_SWIZZLE [30], SWIZZLE_ENABLE [31]), a cache-bank hint with no effect on the address or data, which is dropped here.
 - [Buffer descriptors past the range limit](../../../core/libs/prx/libSceAgcDriver/Graphics/src/ShaderResources.cpp) (libSceAgcDriver) - a buffer V# whose size is over `maxStorageBufferRange` is bound only up to the end of the readable registered allocation holding its base when that allocation ends before the limit, and as the empty placeholder buffer, like a null V#, when no registered allocation holds its base; a V# over an allocation that reaches the limit still throws. Titles leave such words in descriptor slots their shaders do not reach on the path taken (PPSA10528 binds `c8010002 2c6bec20 c0053c00 04000013`, about 33 TB from an address above the user address space). What the console returns for an access past the bound range is not known; here the device's robust buffer access decides. Follows the review of #3202, not measured on hardware.
+- [Guest buffer region over 4 GiB copied whole](../../../core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp) (libSceAgcDriver) - `GuestBufferMemory::copyRegion` copies a guest buffer region over 4 GiB whole into one HOST_VISIBLE|HOST_COHERENT buffer of the region's size (0x102000000, 0x142000000 and 0x102000000 bytes in three runs of PPSA15246 on an AMD RX 6400), and `vkAllocateMemory` refuses it with -13 although the host heap (31961 MiB) had at most 5933 MiB in use; only a sparse region is cut to its backed and viewed extent. An RTX 4080 SUPER stops at the descriptor range check in `GuestBufferMemory::Descriptor` for the same shader instead; its `maxStorageBufferRange` was not measured. Which part of the region the shader reads is unknown.
 - [Vertex fetch overhang](../../../core/libs/prx/libSceAgcDriver/Graphics/src/Draw.cpp) (libSceAgcDriver) - a zero-padded vertex fetch whose valid range (`num_records * stride`) ends less than one stride past the end of the readable memory at its base copies only the readable bytes and reads the rest as zeros, instead of failing the draw; a longer overhang still fails. This extends the strided buffer overhang rule above (the last record is the rounded-up `num_records`) to the staging copy and is not measured on hardware. The observed case is a fetch of 14858 records of 44 bytes that ends 8 bytes past the end of the 0xa0000-byte mapping holding it.
 - [Zero-stride vertex fetch range](../../../core/libs/prx/libSceAgcDriver/Graphics/include/VertexInput.hpp) (libSceAgcDriver) - a draw whose highest index is at or past a nonzero NUM_RECORDS of a stride-0 `OOB_SELECT` 1 V# throws: the pipeline binding has stride 0, so vertices cannot differ. Only the highest index is checked, so a draw with every index past it throws too, where the hardware reads zero. An indirect draw is not checked. Unmeasured on gfx1013 (the console), with an instruction offset or SOFFSET, with a swizzled V# and in a vertex shader. Rules from #8299, measured on an RX 6700 XT (gfx1031), same results on gfx1010, with raw ISA ([data](https://gist.github.com/thuanquang/3b25d677c1a32d79123dd2df1819dd83)). Stride 0 reads zero for `OOB_SELECT` 0 always; for 1 when index >= NUM_RECORDS (records); for 2 when NUM_RECORDS is 0; for 3 when NUM_RECORDS is under the fetch's bytes. Untyped loads check each dword under 3; only typed fetches are used here.
 - [Resource registration](../../../core/libs/prx/libSceAgcDriver/Resource/src/Registration.cpp) (libSceAgcDriver) - registration always fails with `RESOURCE_REGISTRATION_UNAVAILABLE`, so the owner and resource queries, `sceAgcDriverSetResourceUserData`, `sceAgcDriverFindResourcesPublic` and `sceAgcDriverUnregisterAllResourcesForOwner` return the same error without reading their arguments. Their parameter lists follow the libSceGnmDriver counterparts and are unverified. `sceAgcDriverGetResourceRegistrationMaxNameLength` writes 0xfc and returns 0, as [prosper](https://github.com/mattias800/prosper) does from a title's registration configuration; the `uint32_t*` parameter matches the SDK wrapper `sce::Agc::ResourceRegistration::getMaxNameLength(unsigned int*)`. sharpemu writes 256 and KytyPS5 returns `RESOURCE_REGISTRATION_UNAVAILABLE`, so the value is unverified; a null output throws