support for PnP disks, part 2: adapt current caching

git-svn-id: svn://kolibrios.org@2140 a494cfbc-eb01-0410-851d-a64ba20cac60

kernel/trunk/blkdev/disk.inc

@@ -5,6 +5,8 @@
 ;;                                                              ;;
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 
+$Revision$
+
 ; =============================================================================
 ; ================================= Constants =================================
 ; =============================================================================
@@ -76,6 +78,11 @@ struct DISKFUNC
 ; Note that read/write are called by the cache manager, so a driver should not
 ; create a software cache. This function is implemented for flushing a hardware
 ; cache, if it exists.
+.adjust_cache_size      dd      ?
+; The pointer to the function which returns the cache size for this device.
+; Optional, may be NULL.
+; unsigned int adjust_cache_size(unsigned int suggested_size);
+; Return value: 0 = disable cache, otherwise = used cache size in bytes.
 ends
 
 ; This structure holds an information about a media.
@@ -90,6 +97,20 @@ struct DISKMEDIAINFO
 ; Size of the media in sectors.
 ends
 
+; This structure represents disk cache. To follow the old implementation,
+; there are two distinct caches for a disk, one for "system" data, other
+; for "application" data.
+struct DISKCACHE
+.Lock           MUTEX
+; Lock to protect the cache.
+; The following fields are inherited from data32.inc:cache_ideX.
+.pointer     rd 1
+.data_size   rd 1   ; not use
+.data        rd 1
+.sad_size    rd 1
+.search_start       rd 1
+ends
+
 ; This structure represents a disk device and its media for the kernel.
 ; This structure is allocated by the kernel in the 'disk_add' function,
 ; freed in the 'disk_dereference' function.
@@ -146,6 +167,11 @@ struct DISK
 ; Number of partitions on this media.
 .Partitions     dd      ?
 ; Pointer to array of .NumPartitions pointers to PARTITION structures.
+.cache_size     dd      ?
+; inherited from cache_ideX_size
+.SysCache       DISKCACHE
+.AppCache       DISKCACHE
+; Two caches for the disk.
 ends
 
 ; This structure represents one partition for the kernel. This is a base
@@ -156,6 +182,8 @@ struct PARTITION
 ; First sector of the partition.
 .Length         dq      ?
 ; Length of the partition in sectors.
+.Disk           dd      ?
+; Pointer to parent DISK structure.
 .FSUserFunctions        dd      ?
 ; Handlers for the sysfunction 70h. This field is a pointer to the following
 ; array. The first dword is a number of supported subfunctions, other dwords
@@ -242,12 +270,13 @@ disk_list_mutex MUTEX
 endg
 
 iglobal
-; The function 'disk_scan_partitions' needs two 512-byte buffers for
+; The function 'disk_scan_partitions' needs three 512-byte buffers for
-; MBR and bootsectors data. It can not use the static buffers always,
+; MBR, bootsector and fs-temporary sector data. It can not use the static
-; since it can be called for two or more disks in parallel. However, this
+; buffers always, since it can be called for two or more disks in parallel.
-; case is not typical. We reserve two static 512-byte buffers and a flag
+; However, this case is not typical. We reserve three static 512-byte buffers
-; that these buffers are currently used. If 'disk_scan_partitions' detects that
+; and a flag that these buffers are currently used. If 'disk_scan_partitions'
-; the buffers are currently used, it allocates buffers from the heap.
+; detects that the buffers are currently used, it allocates buffers from the
+; heap.
 ; The flag is implemented as a global dword variable. When the static buffers
 ; are not used, the value is -1. When the static buffers are used, the value
 ; is normally 0 and temporarily can become greater. The function increments
@@ -258,10 +287,11 @@ iglobal
 partition_buffer_users  dd      -1
 endg
 uglobal
-; The static buffers for MBR and bootsectors data.
+; The static buffers for MBR, bootsector and fs-temporary sector data.
 align 16
 mbr_buffer      rb      512
 bootsect_buffer rb      512
+fs_tmp_buffer   rb      512
 endg
 
 iglobal
@@ -276,6 +306,7 @@ disk_default_callbacks:
         dd      disk_default_read
         dd      disk_default_write
         dd      disk_default_flush
+        dd      disk_default_adjust_cache_size
 endg
 
 ; =============================================================================
@@ -312,45 +343,44 @@ disk_add:
         jz      .nothing
 ; 2. Copy disk name to the DISK structure.
 ; 2a. Get length of the name, including the terminating zero.
-        mov     esi, [esp+8+8]  ; esi = pointer to name
+        mov     ebx, [esp+8+8]  ; ebx = pointer to name
         push    eax             ; save allocated pointer to DISK
         xor     eax, eax        ; the argument of malloc() is in eax
 @@:
         inc     eax
-        cmp     byte [esi+eax-1], 0
+        cmp     byte [ebx+eax-1], 0
         jnz     @b
 ; 2b. Call the heap manager.
         call    malloc
 ; 2c. Check the result. If allocation failed, go to 7.
-        pop     ebx             ; restore allocated pointer to DISK
+        pop     esi             ; restore allocated pointer to DISK
         test    eax, eax
         jz      .free
 ; 2d. Store the allocated pointer to the DISK structure.
-        mov     [ebx+DISK.Name], eax
+        mov     [esi+DISK.Name], eax
 ; 2e. Copy the name.
 @@:
-        mov     dl, [esi]
+        mov     dl, [ebx]
         mov     [eax], dl
-        inc     esi
+        inc     ebx
         inc     eax
         test    dl, dl
         jnz     @b
 ; 3. Copy other arguments of the function to the DISK structure.
         mov     eax, [esp+4+8]
-        mov     [ebx+DISK.Functions], eax
+        mov     [esi+DISK.Functions], eax
         mov     eax, [esp+12+8]
-        mov     [ebx+DISK.UserData], eax
+        mov     [esi+DISK.UserData], eax
         mov     eax, [esp+16+8]
-        mov     [ebx+DISK.DriverFlags], eax
+        mov     [esi+DISK.DriverFlags], eax
 ; 4. Initialize other fields of the DISK structure.
-; Media is not inserted, initialized state of mutex is zero,
+; Media is not inserted, reference counter is 1.
-; reference counter is 1.
+        lea     ecx, [esi+DISK.MediaLock]
-        lea     ecx, [ebx+DISK.MediaLock]
         call    mutex_init
         xor     eax, eax
-        mov     dword [ebx+DISK.MediaInserted], eax
+        mov     dword [esi+DISK.MediaInserted], eax
         inc     eax
-        mov     [ebx+DISK.RefCount], eax
+        mov     [esi+DISK.RefCount], eax
 ; The DISK structure is initialized.
 ; 5. Insert the new structure to the global list.
 ; 5a. Acquire the mutex.
@@ -358,16 +388,16 @@ disk_add:
         call    mutex_lock
 ; 5b. Insert item to the tail of double-linked list.
         mov     edx, disk_list
-        list_add_tail ebx, edx     ;ebx= new edx= list head
+        list_add_tail esi, edx     ;esi= new edx= list head
 ; 5c. Release the mutex.
-        call mutex_unlock
+        call    mutex_unlock
 ; 6. Return with eax = pointer to DISK.
-        xchg    eax, ebx
+        xchg    eax, esi
         jmp     .nothing
 .free:
 ; Memory allocation for DISK structure succeeded, but for disk name failed.
 ; 7. Free the DISK structure.
-        xchg    eax, ebx
+        xchg    eax, esi
         call    free
 ; 8. Return with eax = 0.
         xor     eax, eax
@@ -474,10 +504,12 @@ lock    dec     [esi+DISK.MediaRefCount]
         jnz     .freeloop
 .nofree:
         pop     edi esi
-; 3b. Call the driver.
+; 3b. Free the cache.
+        call    disk_free_cache
+; 3c. Call the driver.
         mov     al, DISKFUNC.closemedia
         stdcall disk_call_driver
-; 3c. Clear the flag.
+; 3d. Clear the flag.
         mov     [esi+DISK.MediaUsed], 0
 .nothing:
         ret
@@ -525,12 +557,16 @@ disk_media_changed:
 ; 3b. Check the result of the callback. Abort if it failed.
         test    eax, eax
         jnz     .noinsert
-; 3c. Acquire the lifetime reference for the media object.
+; 3c. Allocate the cache unless disabled by the driver. Abort if failed.
+        call    disk_init_cache
+        test    al, al
+        jz      .noinsert
+; 3d. Acquire the lifetime reference for the media object.
         inc     [esi+DISK.MediaRefCount]
-; 3d. Scan for partitions. Ignore result; the list of partitions is valid even
+; 3e. Scan for partitions. Ignore result; the list of partitions is valid even
 ; on errors.
         call    disk_scan_partitions
-; 3e. Media is inserted and available for use.
+; 3f. Media is inserted and available for use.
         inc     [esi+DISK.MediaInserted]
 .noinsert:
 ; 4. Return.
@@ -586,6 +622,11 @@ disk_default_flush:
         xor     eax, eax
         ret     4
 
+; The default implementation of DISKFUNC.adjust_cache_size.
+disk_default_adjust_cache_size:
+        mov     eax, [esp+4]
+        ret     4
+
 ; This is an internal function called from 'disk_media_changed' when new media
 ; is detected. It creates the list of partitions for the media.
 ; If media is not partitioned, then the list consists of one partition which
@@ -609,7 +650,7 @@ disk_scan_partitions:
 lock    inc     [partition_buffer_users]
         jz      .buffer_acquired        ; yes, it is free
 lock    dec     [partition_buffer_users]        ; no, we must allocate
-        stdcall kernel_alloc, 1024
+        stdcall kernel_alloc, 512*3
         test    eax, eax
         jz      .nothing
         xchg    eax, ebx
@@ -712,6 +753,7 @@ lock    dec     [partition_buffer_users]        ; no, we must allocate
 ; compatibility problems.
         pop     eax     ; load extended partition
         add     ebp, eax
+        jc      .mbr_failed
 ; 12c. If extended partition has not yet started, start it.
         test    eax, eax
         jnz     @f
@@ -849,7 +891,7 @@ proc disk_add_partition stdcall uses ebx edi, start:qword, length:qword
 ; The range [.FirstSector, .FirstSector+.Length) must be either entirely to
 ; the left of [start, start+length) or entirely to the right.
 ; 2c. Subtract .FirstSector - start. The possible overflow distinguish between
-; cases "to the left" (2?) and "to the right" (2d).
+; cases "to the left" (2e) and "to the right" (2d).
         mov     eax, dword [ecx+PARTITION.FirstSector]
         mov     edx, dword [ecx+PARTITION.FirstSector+4]
         sub     eax, dword [start]

kernel/trunk/blkdev/disk_cache.inc

@@ -0,0 +1,592 @@
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;                                                              ;;
+;; Copyright (C) KolibriOS team 2011. All rights reserved.      ;;
+;; Distributed under terms of the GNU General Public License    ;;
+;;                                                              ;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+$Revision$
+
+; This function is intended to replace the old 'hd_read' function when
+; [hdd_appl_data] = 0, so its input/output parameters are the same, except
+; that it can't use the global variables 'hd_error' and 'hdd_appl_data'.
+; in: eax = sector, ebx = buffer, ebp = pointer to PARTITION structure
+; eax is relative to partition start
+; out: eax = error code; 0 = ok
+fs_read32_sys:
+; Compatibility hack: if PARTITION.Disk is 'old', there is no DISK structure,
+; this request should be processed by hd_read.
+        cmp     [ebp+PARTITION.Disk], 'old'
+        jnz     @f
+        mov     [hdd_appl_data], 0
+        call    hd_read
+        mov     [hdd_appl_data], 1      ; restore to default state
+        ret
+@@:
+; In the normal case, save ecx, set ecx to SysCache and let the common part
+; do its work.
+        push    ecx
+        mov     ecx, [ebp+PARTITION.Disk]
+        add     ecx, DISK.SysCache
+        jmp     fs_read32_common
+
+; This function is intended to replace the old 'hd_read' function when
+; [hdd_appl_data] = 1, so its input/output parameters are the same, except
+; that it can't use the global variables 'hd_error' and 'hdd_appl_data'.
+; in: eax = sector, ebx = buffer, ebp = pointer to PARTITION structure
+; eax is relative to partition start
+; out: eax = error code; 0 = ok
+fs_read32_app:
+; Compatibility hack: if PARTITION.Disk is 'old', there is no DISK structure,
+; this request should be processed by hd_read.
+        cmp     [ebp+PARTITION.Disk], 'old'
+        jnz     @f
+        mov     [hdd_appl_data], 1
+        jmp     hd_read
+@@:
+; In the normal case, save ecx, set ecx to AppCache and let the common part
+; do its work.
+        push    ecx
+        mov     ecx, [ebp+PARTITION.Disk]
+        add     ecx, DISK.AppCache
+
+; This label is the common part of fs_read32_sys and fs_read32_app.
+fs_read32_common:
+; 1. Check that the required sector is inside the partition. If no, return
+; DISK_STATUS_END_OF_MEDIA.
+        cmp     dword [ebp+PARTITION.Length+4], 0
+        jnz     @f
+        cmp     dword [ebp+PARTITION.Length], eax
+        ja      @f
+        mov     eax, DISK_STATUS_END_OF_MEDIA
+        pop     ecx
+        ret
+@@:
+; 2. Get the absolute sector on the disk.
+        push    edx
+        xor     edx, edx
+        add     eax, dword [ebp+PARTITION.FirstSector]
+        adc     edx, dword [ebp+PARTITION.FirstSector+4]
+; 3. If there is no cache for this disk, just pass the request to the driver.
+        cmp     [ecx+DISKCACHE.pointer], 0
+        jnz     .scancache
+        push    1
+        push    esp     ; numsectors
+        push    edx     ; startsector
+        push    eax     ; startsector
+        push    ebx     ; buffer
+        mov     al, DISKFUNC.read
+        call    disk_call_driver
+        pop     ecx
+        pop     edx
+        pop     ecx
+        ret
+.scancache:
+; 4. Scan the cache.
+        push    esi edi ecx ; scan cache
+        push    edx eax
+virtual at esp
+.sector_lo      dd      ?
+.sector_hi      dd      ?
+.cache          dd      ?
+end virtual
+; The following code is inherited from hd_read. The differences are:
+; all code is protected by the cache lock; instead of static calls
+; to hd_read_dma/hd_read_pio/bd_read the dynamic call to DISKFUNC.read is used;
+; sector is 64-bit, not 32-bit.
+        call    mutex_lock
+        mov     eax, [.sector_lo]
+        mov     edx, [.sector_hi]
+        mov     esi, [ecx+DISKCACHE.pointer]
+        mov     ecx, [ecx+DISKCACHE.sad_size]
+        add     esi, 12
+
+        mov     edi, 1
+
+.hdreadcache:
+
+        cmp     dword [esi+8], 0        ; empty
+        je      .nohdcache
+
+        cmp     [esi], eax      ; correct sector
+        jne     .nohdcache
+        cmp     [esi+4], edx    ; correct sector
+        je      .yeshdcache
+
+.nohdcache:
+
+        add     esi, 12
+        inc     edi
+        dec     ecx
+        jnz     .hdreadcache
+
+        mov     esi, [.cache]
+        call    find_empty_slot64       ; ret in edi
+        test    eax, eax
+        jnz     .read_done
+
+        push    1
+        push    esp
+        push    edx
+        push    [.sector_lo+12]
+        mov     ecx, [.cache]
+        mov     eax, edi
+        shl     eax, 9
+        add     eax, [ecx+DISKCACHE.data]
+        push    eax
+        mov     esi, [ebp+PARTITION.Disk]
+        mov     al, DISKFUNC.read
+        call    disk_call_driver
+        pop     ecx
+        dec     ecx
+        jnz     .read_done
+
+        mov     ecx, [.cache]
+        lea     eax, [edi*3]
+        mov     esi, [ecx+DISKCACHE.pointer]
+        lea     esi, [eax*4+esi]
+
+        mov     eax, [.sector_lo]
+        mov     edx, [.sector_hi]
+        mov     [esi], eax      ; sector number
+        mov     [esi+4], edx    ; sector number
+        mov     dword [esi+8], 1; hd read - mark as same as in hd
+
+.yeshdcache:
+
+        mov     esi, edi
+        mov     ecx, [.cache]
+        shl     esi, 9
+        add     esi, [ecx+DISKCACHE.data]
+
+        mov     edi, ebx
+        mov     ecx, 512/4
+        rep     movsd           ; move data
+        xor     eax, eax        ; successful read
+.read_done:
+        mov     ecx, [.cache]
+        push    eax
+        call    mutex_unlock
+        pop     eax
+        add     esp, 12
+        pop     edi esi edx ecx
+        ret
+
+; This function is intended to replace the old 'hd_write' function when
+; [hdd_appl_data] = 0, so its input/output parameters are the same, except
+; that it can't use the global variables 'hd_error' and 'hdd_appl_data'.
+; in: eax = sector, ebx = buffer, ebp = pointer to PARTITION structure
+; eax is relative to partition start
+; out: eax = error code; 0 = ok
+fs_write32_sys:
+; Compatibility hack: if PARTITION.Disk is 'old', there is no DISK structure,
+; this request should be processed by hd_write.
+        cmp     [ebp+PARTITION.Disk], 'old'
+        jnz     @f
+        mov     [hdd_appl_data], 0
+        call    hd_write
+        mov     [hdd_appl_data], 1      ; restore to default state
+        ret
+@@:
+; In the normal case, save ecx, set ecx to SysCache and let the common part
+; do its work.
+        push    ecx
+        mov     ecx, [ebp+PARTITION.Disk]
+        add     ecx, DISK.SysCache
+        jmp     fs_write32_common
+
+; This function is intended to replace the old 'hd_write' function when
+; [hdd_appl_data] = 1, so its input/output parameters are the same, except
+; that it can't use the global variables 'hd_error' and 'hdd_appl_data'.
+; in: eax = sector, ebx = buffer, ebp = pointer to PARTITION structure
+; eax is relative to partition start
+; out: eax = error code; 0 = ok
+fs_write32_app:
+; Compatibility hack: if PARTITION.Disk is 'old', there is no DISK structure,
+; this request should be processed by hd_write.
+        cmp     [ebp+PARTITION.Disk], 'old'
+        jnz     @f
+        mov     [hdd_appl_data], 1
+        jmp     hd_write
+@@:
+; In the normal case, save ecx, set ecx to AppCache and let the common part
+; do its work.
+        push    ecx
+        mov     ecx, [ebp+PARTITION.Disk]
+        add     ecx, DISK.AppCache
+
+; This label is the common part of fs_read32_sys and fs_read32_app.
+fs_write32_common:
+; 1. Check that the required sector is inside the partition. If no, return
+; DISK_STATUS_END_OF_MEDIA.
+        cmp     dword [ebp+PARTITION.Length+4], 0
+        jnz     @f
+        cmp     dword [ebp+PARTITION.Length], eax
+        ja      @f
+        mov     eax, DISK_STATUS_END_OF_MEDIA
+        pop     ecx
+        ret
+@@:
+        push    edx
+; 2. Get the absolute sector on the disk.
+        xor     edx, edx
+        add     eax, dword [ebp+PARTITION.FirstSector]
+        adc     edx, dword [ebp+PARTITION.FirstSector+4]
+; 3. If there is no cache for this disk, just pass request to the driver.
+        cmp     [ecx+DISKCACHE.pointer], 0
+        jnz     .scancache
+        push    1
+        push    esp     ; numsectors
+        push    edx     ; startsector
+        push    eax     ; startsector
+        push    ebx     ; buffer
+        mov     al, DISKFUNC.write
+        call    disk_call_driver
+        pop     ecx
+        pop     edx
+        pop     ecx
+        ret
+.scancache:
+; 4. Scan the cache.
+        push    esi edi ecx ; scan cache
+        push    edx eax
+virtual at esp
+.sector_lo      dd      ?
+.sector_hi      dd      ?
+.cache          dd      ?
+end virtual
+; The following code is inherited from hd_write. The differences are:
+; all code is protected by the cache lock;
+; sector is 64-bit, not 32-bit.
+        call    mutex_lock
+
+        ; check if the cache already has the sector and overwrite it
+        mov     eax, [.sector_lo]
+        mov     edx, [.sector_hi]
+        mov     esi, [ecx+DISKCACHE.pointer]
+        mov     ecx, [ecx+DISKCACHE.sad_size]
+        add     esi, 12
+
+        mov     edi, 1
+
+.hdwritecache:
+        cmp     dword [esi+8], 0        ; if cache slot is empty
+        je      .not_in_cache_write
+
+        cmp     [esi], eax      ; if the slot has the sector
+        jne     .not_in_cache_write
+        cmp     [esi+4], edx    ; if the slot has the sector
+        je      .yes_in_cache_write
+
+.not_in_cache_write:
+
+        add     esi, 12
+        inc     edi
+        dec     ecx
+        jnz     .hdwritecache
+
+        ; sector not found in cache
+        ; write the block to a new location
+
+        mov     esi, [.cache]
+        call    find_empty_slot64       ; ret in edi
+        test    eax, eax
+        jne     .hd_write_access_denied
+
+        mov     ecx, [.cache]
+        lea     eax, [edi*3]
+        mov     esi, [ecx+DISKCACHE.pointer]
+        lea     esi, [eax*4+esi]
+
+        mov     eax, [.sector_lo]
+        mov     edx, [.sector_hi]
+        mov     [esi], eax      ; sector number
+        mov     [esi+4], edx    ; sector number
+
+.yes_in_cache_write:
+
+        mov     dword [esi+4], 2        ; write - differs from hd
+
+        shl     edi, 9
+        mov     ecx, [.cache]
+        add     edi, [ecx+DISKCACHE.data]
+
+        mov     esi, ebx
+        mov     ecx, 512/4
+        rep     movsd           ; move data
+        xor     eax, eax        ; success
+.hd_write_access_denied:
+        mov     ecx, [.cache]
+        push    eax
+        call    mutex_unlock
+        pop     eax
+        add     esp, 12
+        pop     edi esi edx ecx
+        ret
+
+; This internal function is called from fs_read32_* and fs_write32_*. It is the
+; analogue of find_empty_slot for 64-bit sectors.
+find_empty_slot64:
+;-----------------------------------------------------------
+; find empty or read slot, flush cache if next 12.5% is used by write
+; output : edi = cache slot
+;-----------------------------------------------------------
+.search_again:
+        mov     ecx, [esi+DISKCACHE.sad_size]
+        mov     edi, [esi+DISKCACHE.search_start]
+        shr     ecx, 3
+.search_for_empty:
+        inc     edi
+        cmp     edi, [esi+DISKCACHE.sad_size]
+        jbe     .inside_cache
+        mov     edi, 1
+.inside_cache:
+        lea     eax, [edi*3]
+        shl     eax, 2
+        add     eax, [esi+DISKCACHE.pointer]
+        cmp     dword [eax+8], 2
+        jb      .found_slot             ; it's empty or read
+        dec     ecx
+        jnz     .search_for_empty
+        call    write_cache64           ; no empty slots found, write all
+        test    eax, eax
+        jne     .found_slot_access_denied
+        jmp     .search_again           ; and start again
+.found_slot:
+        mov     [esi+DISKCACHE.search_start], edi
+        xor     eax, eax        ; success
+.found_slot_access_denied:
+        ret
+
+; This function is intended to replace the old 'write_cache' function.
+proc write_cache64 uses ecx edx esi edi
+locals
+cache_chain_started     dd      ?
+cache_chain_size        dd      ?
+cache_chain_pos         dd      ?
+cache_chain_ptr         dd      ?
+endl
+; If there is no cache for this disk, nothing to do.
+        cmp     [esi+DISKCACHE.pointer], 0
+        jz      .flush
+;-----------------------------------------------------------
+; write all changed sectors to disk
+;-----------------------------------------------------------
+
+        ; write difference ( 2 ) from cache to DISK
+        mov     ecx, [esi+DISKCACHE.sad_size]
+        mov     esi, [esi+DISKCACHE.pointer]
+        add     esi, 12
+        mov     edi, 1
+.write_cache_more:
+        cmp     dword [esi+8], 2        ; if cache slot is not different
+        jne     .write_chain
+        mov     dword [esi+8], 1        ; same as in hd
+        mov     eax, [esi]
+        mov     edx, [esi+4]            ; edx:eax = sector to write
+; Îáúåäèíÿåì çàïèñü öåïî÷êè ïîñëåäîâàòåëüíûõ ñåêòîðîâ â îäíî îáðàùåíèå ê äèñêó
+        cmp     ecx, 1
+        jz      .nonext
+        cmp     dword [esi+12+8], 2
+        jnz     .nonext
+        push    eax edx
+        add     eax, 1
+        adc     edx, 0
+        cmp     eax, [esi+12]
+        jnz     @f
+        cmp     edx, [esi+12+4]
+@@:
+        pop     edx eax
+        jnz     .nonext
+        cmp     [cache_chain_started], 1
+        jz      @f
+        mov     [cache_chain_started], 1
+        mov     [cache_chain_size], 0
+        mov     [cache_chain_pos], edi
+        mov     [cache_chain_ptr], esi
+@@:
+        inc     [cache_chain_size]
+        cmp     [cache_chain_size], 16
+        jnz     .continue
+        jmp     .write_chain
+.nonext:
+        call    .flush_cache_chain
+        test    eax, eax
+        jnz     .nothing
+        mov     [cache_chain_size], 1
+        mov     [cache_chain_ptr], esi
+        call    .write_cache_sector
+        test    eax, eax
+        jnz     .nothing
+        jmp     .continue
+.write_chain:
+        call    .flush_cache_chain
+        test    eax, eax
+        jnz     .nothing
+.continue:
+        add     esi, 12
+        inc     edi
+        dec     ecx
+        jnz     .write_cache_more
+        call    .flush_cache_chain
+        test    eax, eax
+        jnz     .nothing
+.flush:
+        mov     esi, [ebp]
+        mov     esi, [esi+PARTITION.Disk]
+        mov     al, DISKFUNC.flush
+        call    disk_call_driver
+.nothing:
+        ret
+
+.flush_cache_chain:
+        xor     eax, eax
+        cmp     [cache_chain_started], eax
+        jz      @f
+        call    .write_cache_chain
+        mov     [cache_chain_started], 0
+@@:
+        retn
+
+.write_cache_sector:
+        mov     [cache_chain_size], 1
+        mov     [cache_chain_pos], edi
+.write_cache_chain:
+        pusha
+        mov     edi, [cache_chain_pos]
+        mov     ecx, [ebp-12]
+        shl     edi, 9
+        add     edi, [ecx+DISKCACHE.data]
+        mov     ecx, [cache_chain_size]
+        push    ecx
+        push    esp     ; numsectors
+        mov     eax, [cache_chain_ptr]
+        pushd   [eax+4]
+        pushd   [eax]   ; startsector
+        push    edi     ; buffer
+        mov     esi, [ebp]
+        mov     esi, [esi+PARTITION.Disk]
+        mov     al, DISKFUNC.write
+        call    disk_call_driver
+        pop     ecx
+        mov     [esp+28], eax
+        popa
+        retn
+endp
+
+; This internal function is called from disk_add to initialize the caching for
+; a new DISK.
+; The algorithm is inherited from getcache.inc: take 1/32 part of the available
+; physical memory, round down to 8 pages, limit by 128K from below and by 1M
+; from above. Reserve 1/8 part of the cache for system data and 7/8 for app
+; data.
+; After the size is calculated, but before the cache is allocated, the device
+; driver can adjust the size. In particular, setting size to zero disables
+; caching: there is no sense in a cache for a ramdisk. In fact, such action
+; is most useful example of a non-trivial adjustment.
+; esi = pointer to DISK structure
+disk_init_cache:
+; 1. Calculate the suggested cache size.
+; 1a. Get the size of free physical memory in pages.
+        mov     eax, [pg_data.pages_free]
+; 1b. Use the value to calculate the size.
+        shl     eax, 12 - 5     ; 1/32 of it in bytes
+        and     eax, -8*4096    ; round down to the multiple of 8 pages
+; 1c. Force lower and upper limits.
+        cmp     eax, 1024*1024
+        jb      @f
+        mov     eax, 1024*1024
+@@:
+        cmp     eax, 128*1024
+        ja      @f
+        mov     eax, 128*1024
+@@:
+; 1d. Give a chance to the driver to adjust the size.
+        push    eax
+        mov     al, DISKFUNC.adjust_cache_size
+        call    disk_call_driver
+; Cache size calculated.
+        mov     [esi+DISK.cache_size], eax
+        test    eax, eax
+        jz      .nocache
+; 2. Allocate memory for the cache.
+; 2a. Call the allocator.
+        stdcall kernel_alloc, eax
+        test    eax, eax
+        jnz     @f
+; 2b. If it failed, say a message and return with eax = 0.
+        dbgstr 'no memory for disk cache'
+        jmp     .nothing
+@@:
+; 3. Fill two DISKCACHE structures.
+        mov     [esi+DISK.SysCache.pointer], eax
+        lea     ecx, [esi+DISK.SysCache.Lock]
+        call    mutex_init
+        lea     ecx, [esi+DISK.AppCache.Lock]
+        call    mutex_init
+; The following code is inherited from getcache.inc.
+        mov     edx, [esi+DISK.SysCache.pointer]
+        and     [esi+DISK.SysCache.search_start], 0
+        and     [esi+DISK.AppCache.search_start], 0
+        mov     eax, [esi+DISK.cache_size]
+        shr     eax, 3
+        mov     [esi+DISK.SysCache.data_size], eax
+        add     edx, eax
+        imul    eax, 7
+        mov     [esi+DISK.AppCache.data_size], eax
+        mov     [esi+DISK.AppCache.pointer], edx
+
+        mov     eax, [esi+DISK.SysCache.data_size]
+        push    ebx
+        call    calculate_for_hd
+        pop     ebx
+        add     eax, [esi+DISK.SysCache.pointer]
+        mov     [esi+DISK.SysCache.data], eax
+        mov     [esi+DISK.SysCache.sad_size], ecx
+
+        push    edi
+        mov     edi, [esi+DISK.SysCache.pointer]
+        lea     ecx, [ecx*3]
+        xor     eax, eax
+        rep     stosd
+        pop     edi
+
+        mov     eax, [esi+DISK.AppCache.data_size]
+        push    ebx
+        call    calculate_for_hd
+        pop     ebx
+        add     eax, [esi+DISK.AppCache.pointer]
+        mov     [esi+DISK.AppCache.data], eax
+        mov     [esi+DISK.AppCache.sad_size], ecx
+
+        push    edi
+        mov     edi, [esi+DISK.AppCache.pointer]
+        lea     ecx, [ecx*3]
+        xor     eax, eax
+        rep     stosd
+        pop     edi
+
+; 4. Return with nonzero al.
+        mov     al, 1
+; 5. Return.
+.nothing:
+        ret
+; No caching is required for this driver. Zero cache pointers and return with
+; nonzero al.
+.nocache:
+        mov     [esi+DISK.SysCache.pointer], eax
+        mov     [esi+DISK.AppCache.pointer], eax
+        mov     al, 1
+        ret
+
+; This internal function is called from disk_media_dereference to free the
+; allocated cache, if there is one.
+; esi = pointer to DISK structure
+disk_free_cache:
+; The algorithm is straightforward.
+        mov     eax, [esi+DISK.SysCache.pointer]
+        test    eax, eax
+        jz      .nothing
+        stdcall kernel_free, eax
+.nothing:
+        ret

kernel/trunk/docs/drivers_api.txt

@@ -53,6 +53,10 @@ struc DISKFUNC
 ; Flushes the hardware cache, if it exists. Note that a driver should not
 ; implement a software cache for read/write, since they are called from the
 ; kernel cache manager.
+  .adjust_cache_size dd ?
+; unsigned int adjust_cache_size(unsigned int suggested_size);
+; Optional.
+; Returns the cache size for this device in bytes. 0 = disable cache.
 }
 struc DISKMEDIAINFO
 {

kernel/trunk/kernel32.inc

@@ -235,7 +235,8 @@ include "gui/button.inc"
 
 ; file system
 
-include "fs/disk.inc"     ; support for plug-n-play disks
+include "blkdev/disk.inc" ; support for plug-n-play disks
+include "blkdev/disk_cache.inc" ; caching for plug-n-play disks
 include "fs/fs.inc"       ; syscall
 include "fs/fat32.inc"    ; read / write for fat32 filesystem
 include "fs/ntfs.inc"     ; read / write for ntfs filesystem