implement allocating and freeing PRP lists

drivers/nvme/nvme.asm

@@ -450,27 +450,38 @@ proc get_log_page stdcall, pci:dword, dptr:dword, lid:byte
 
 endp
 
-proc free_prp_list stdcall, prp_list_ptr:dword, nprps:dword
+proc free_prp_list stdcall, prp_list_ptr:dword, nprps:dword, recursive:byte
 
-		push 	ebx edi
+		push 	edi
 		mov 	edi, [prp_list_ptr]
-		xor 	ebx, ebx
+		mov 	ecx, [nprps]
+		cmp 	ecx, PAGE_SIZE / 8 - 1
+		jbe 	@f
+		mov 	eax, dword [edi + PAGE_SIZE - 8]
+		sub 	ecx, PAGE_SIZE / 8 - 1
+		stdcall free_prp_list, eax, ecx, TRUE
 
 @@:
-		invoke  KernelFree, dword [edi + ebx * 4]
-		inc 	ebx
-		cmp 	ebx, [nprps]
-		jb 		@b
-		pop 	edi ebx
+		cmp 	[recursive], 0
+		jz 		@f
+		invoke  FreePage, edi
+		jmp 	.exit
+		
+@@:
+		invoke  KernelFree, edi
+
+.exit:
 		xor 	eax, eax
+		pop 	edi
 		ret
 
 endp
 
 proc build_prp_list stdcall, nprps:dword
 
-		push 	ebx edi
+		push 	esi ebx edi
 		xor 	edi, edi
+		xor 	esi, esi
 		mov 	ecx, [nprps]
 		mov 	ebx, ecx
 		shl 	ecx, 3
@@ -478,34 +489,61 @@ proc build_prp_list stdcall, nprps:dword
 		test 	eax, eax
 		jz 		.err
 		mov 	edi, eax
-		mov 	ecx, [nprps]
+		mov 	ecx, ebx
 		imul 	ecx, PAGE_SIZE
 		invoke  KernelAlloc, ecx
 		test 	eax, eax
 		jz 		.err
-		mov 	edx, eax
+		mov 	esi, eax
 		xor 	ecx, ecx
+		push 	esi
 
-@@:
-		push 	edx ecx
-		invoke  GetPhysAddr, edx
-		pop 	ecx edx
+.loop:
+		; check if we need to build another PRP list
+		mov 	eax, ebx
+		sub 	eax, PAGE_SIZE
+		; is this the last entry in the list? If so, build yet another PRP list here
+		cmp 	ecx, PAGE_SIZE / 8 - 1
+		jne 	@f
+		push 	ecx
+		stdcall build_prp_list, eax
+		pop 	ecx
+		test 	eax, eax
+		jz 	    .cleanup_prp_list
+		invoke  GetPhysAddr
 		mov 	dword [edi + ecx * 4], eax
 		mov 	dword [edi + ecx * 4 + 4], 0
-		add 	edx, PAGE_SIZE
+		jmp 	.success
+
+@@:
+		mov 	eax, esi
+		push 	ecx
+		invoke  GetPhysAddr
+		pop 	ecx
+		mov 	dword [edi + ecx * 4], eax
+		mov 	dword [edi + ecx * 4 + 4], 0
+		add 	esi, PAGE_SIZE
 		inc 	ecx
 		cmp 	ecx, ebx
-		jb 		@b
-		pop 	edi ebx
+		jb 		.loop
+
+.success:
+		mov 	eax, edi
+		pop 	esi
+		pop 	edi ebx esi
 		ret
-		
+
+.cleanup_prp_list:
+		pop 	esi
+		invoke  KernelFree, esi
+
 .err:
 		test 	edi, edi
-		jnz 	@f
+		jz 		@f
 		invoke  KernelFree, edi
 
 @@:
-		pop 	edi ebx
+		pop 	edi ebx esi
 		xor 	eax, eax
 		ret
 		
@@ -549,7 +587,7 @@ proc alloc_dptr stdcall, ns:dword, prps_ptr:dword, start_sector:qword, numsector
 		jmp 	.success
 		
 .free_prp_list:
-		invoke  KernelFree, eax
+		stdcall free_prp_list, edi, eax
 
 .err:
 		mov 	eax, dword [edi]
@@ -574,7 +612,7 @@ proc nvme_read ns:dword, dst:dword, start_sector:qword, numsectors_ptr:dword
 
 		push 	ebx esi edi
 		sub 	esp, 8
-		stdcall alloc_dptr, [ns], dword [esp], dword [esp + 4], [start_sector], [numsectors_ptr]
+		stdcall alloc_dptr, [ns], esp, [start_sector], [numsectors_ptr]
 		test 	eax, eax
 		jz 		.end
 		mov 	edi, eax
@@ -618,7 +656,7 @@ proc nvme_read ns:dword, dst:dword, start_sector:qword, numsectors_ptr:dword
 endp
 
 ; See page 258-261 (read) and 269-271 (write) of the NVMe 1.4 specification for reference
-proc nvme_io_rw stdcall, pci:dword, qid:word, nsid:dword, prp1:dword, prp2:dword, slba_lo:dword, slba_hi:dword, nlb:word, opcode:dword
+proc nvme_io_rw stdcall, pci:dword, qid:word, nsid:dword, prps:qword, slba_lo:dword, slba_hi:dword, nlb:word, opcode:dword
 		
 		; TODO: Use IDENTC.NOIOB to construct read/write commands that don't
 		; cross the I/O boundary to achieve optimal performance
@@ -629,9 +667,9 @@ proc nvme_io_rw stdcall, pci:dword, qid:word, nsid:dword, prp1:dword, prp2:dword
 		movzx 	ecx, [qid]
 		stdcall set_cdw0, [pci], ecx, [opcode]
 		mov 	dword [esp + SQ_ENTRY.cdw0], eax ; CDW0
-		mov 	eax, [prp1]
+		mov 	eax, dword [prps]
 		mov 	dword [esp + SQ_ENTRY.dptr], eax
-		mov 	eax, [prp2]
+		mov 	eax, dword [prps + 4]
 		mov 	dword [esp + SQ_ENTRY.dptr + 8], eax
 		mov 	eax, [nsid]
 		mov 	dword [esp + SQ_ENTRY.nsid], eax