Add the concept of a "user pointer" (similar to the void __user * in

the kernel), which encapsulates the information needed to refer to an
external buffer.  Under normal operation, this can just be a void *
equivalent, but under -DKEEP_IT_REAL it would be a segoff_t equivalent.

Use this concept to avoid the need for bounce buffers in int13.c,
which reduces memory usage and opens up the possibility of using
multi-sector reads.

Extend the block-device API and the SCSI block device implementation
to support multi-sector reads.

Update iscsi.c to use user buffers.

Move the obsolete portions of realmode.h to old_realmode.h.

MS-DOS now boots an order of magnitude faster over iSCSI (~10 seconds
from power-up to C:> prompt in bochs).

src/arch/i386/core/pic8259.c

@@ -6,7 +6,7 @@
 
 #include <etherboot.h>
 #include "pic8259.h"
-#include "realmode.h"
+#include "old_realmode.h"
 
 /* State of trivial IRQ handler */
 irq_t trivial_irq_installed_on = IRQ_NONE;

src/arch/i386/include/bits/uaccess.h

@@ -0,0 +1,6 @@
+#ifndef _BITS_UACCESS_H
+#define _BITS_UACCESS_H
+
+#include <realmode.h>
+
+#endif /* _BITS_UACCESS_H */

src/arch/i386/include/librm.h

@@ -92,6 +92,69 @@ copy_from_real_librm ( void *dest, unsigned int src_seg,
 #define put_real put_real_librm
 #define get_real get_real_librm
 
+/**
+ * A pointer to a user buffer
+ *
+ * Even though we could just use a void *, we use an intptr_t so that
+ * attempts to use normal pointers show up as compiler warnings.  Such
+ * code is actually valid for librm, but not for libkir (i.e. under
+ * KEEP_IT_REAL), so it's good to have the warnings even under librm.
+ */
+typedef intptr_t userptr_t;
+
+/**
+ * Copy data to user buffer
+ *
+ * @v buffer	User buffer
+ * @v offset	Offset within user buffer
+ * @v src	Source
+ * @v len	Length
+ */
+static inline __attribute__ (( always_inline )) void
+copy_to_user ( userptr_t buffer, off_t offset, const void *src, size_t len ) {
+	memcpy ( ( void * ) buffer + offset, src, len );
+}
+
+/**
+ * Copy data from user buffer
+ *
+ * @v dest	Destination
+ * @v buffer	User buffer
+ * @v offset	Offset within user buffer
+ * @v len	Length
+ */
+static inline __attribute__ (( always_inline )) void
+copy_from_user ( void *dest, userptr_t buffer, off_t offset, size_t len ) {
+	memcpy ( dest, ( void * ) buffer + offset, len );
+}
+
+/**
+ * Convert virtual address to user buffer
+ *
+ * @v virtual	Virtual address
+ * @ret buffer	User buffer
+ *
+ * This constructs a user buffer from an ordinary pointer.  Use it
+ * when you need to pass a pointer to an internal buffer to a function
+ * that expects a @c userptr_t.
+ */
+static inline __attribute__ (( always_inline )) userptr_t
+virt_to_user ( void * virtual ) {
+	return ( ( intptr_t ) virtual );
+}
+
+/**
+ * Convert segment:offset address to user buffer
+ *
+ * @v segment	Real-mode segment
+ * @v offset	Real-mode offset
+ * @ret buffer	User buffer
+ */
+static inline __attribute__ (( always_inline )) userptr_t
+real_to_user ( unsigned int segment, unsigned int offset ) {
+	return virt_to_user ( VIRTUAL ( segment, offset ) );
+}
+
 /* Copy to/from real-mode stack */
 extern uint16_t copy_to_rm_stack ( void *data, size_t size );
 extern void remove_from_rm_stack ( void *data, size_t size );

src/arch/i386/include/old_realmode.h

@@ -0,0 +1,22 @@
+#ifndef _OLD_REALMODE_H
+#define _OLD_REALMODE_H
+
+#include <realmode.h>
+
+#warning "Anything including this header is obsolete and must be rewritten"
+
+/* Just for now */
+#define SEGMENT(x) ( virt_to_phys ( x ) >> 4 )
+#define OFFSET(x) ( virt_to_phys ( x ) & 0xf )
+#define SEGOFF(x) { OFFSET(x), SEGMENT(x) }
+
+/* To make basemem.c compile */
+extern int lock_real_mode_stack;
+extern char *real_mode_stack;
+extern char real_mode_stack_size[];
+
+#define RM_FRAGMENT(name,asm) \
+	void name ( void ) {} \
+	extern char name ## _size[];
+
+#endif /* _OLD_REALMODE_H */

src/arch/i386/include/realmode.h

@@ -149,25 +149,6 @@ typedef struct segoff segoff_t;
  * "(discard)" in the above code.
  */
 
-#warning "realmode.h contains placeholders for obsolete macros"
-
-
-/* Just for now */
-#define SEGMENT(x) ( virt_to_phys ( x ) >> 4 )
-#define OFFSET(x) ( virt_to_phys ( x ) & 0xf )
-#define SEGOFF(x) { OFFSET(x), SEGMENT(x) }
-
-/* To make basemem.c compile */
-extern int lock_real_mode_stack;
-extern char *real_mode_stack;
-extern char real_mode_stack_size[];
-
-#define RM_FRAGMENT(name,asm) \
-	void name ( void ) {} \
-	extern char name ## _size[];
-
-
-
 #endif /* ASSEMBLY */
 
 #endif /* REALMODE_H */

src/arch/i386/interface/pcbios/int13.c

@@ -93,7 +93,7 @@ static unsigned long chs_to_lba ( struct int13_drive *drive,
 	lba = ( ( ( ( cylinder * drive->heads ) + head )
 		  * drive->sectors_per_track ) + sector - 1 );
 
-	DBG ( "C/H/S address %x/%x/%x -> LBA %x\n",
+	DBG ( "C/H/S address %x/%x/%x -> LBA %lx\n",
 	      cylinder, head, sector, lba );
 
 	return lba;
@@ -111,19 +111,15 @@ static unsigned long chs_to_lba ( struct int13_drive *drive,
 static int int13_read ( struct int13_drive *drive, uint64_t lba,
 			struct segoff data, unsigned long count ) {
 	struct block_device *blockdev = drive->blockdev;
-	size_t blksize = blockdev->blksize;
-	uint8_t buffer[blksize];
+	userptr_t buffer = real_to_user ( data.segment, data.offset );
 	int rc;
 
 	DBG ( "Read %lx sectors from %llx to %04x:%04x\n", count,
 	      ( unsigned long long ) lba, data.segment, data.offset );
-	while ( count-- ) {
-		if ( ( rc = blockdev->read ( blockdev, lba, buffer ) ) != 0 )
-			return INT13_STATUS_READ_ERROR;
-		copy_to_real ( data.segment, data.offset, buffer, blksize );
-		data.offset += blksize;
-		lba++;
-	}
+	
+	if ( ( rc = blockdev->read ( blockdev, lba, count, buffer ) ) != 0 )
+		return INT13_STATUS_READ_ERROR;
+
 	return 0;
 }
 
@@ -139,19 +135,15 @@ static int int13_read ( struct int13_drive *drive, uint64_t lba,
 static int int13_write ( struct int13_drive *drive, uint64_t lba,
 			 struct segoff data, unsigned long count ) {
 	struct block_device *blockdev = drive->blockdev;
-	size_t blksize = blockdev->blksize;
-	uint8_t buffer[blksize];
+	userptr_t buffer = real_to_user ( data.segment, data.offset );
 	int rc;
 
 	DBG ( "Write %lx sectors from %04x:%04x to %llx\n", count,
 	      data.segment, data.offset, ( unsigned long long ) lba );
-	while ( count-- ) {
-		copy_from_real ( buffer, data.segment, data.offset, blksize );
-		if ( ( rc = blockdev->write ( blockdev, lba, buffer ) ) != 0 )
-			return INT13_STATUS_WRITE_ERROR;
-		data.offset += blksize;
-		lba++;
-	}
+
+	if ( ( rc = blockdev->write ( blockdev, lba, count, buffer ) ) != 0 )
+		return INT13_STATUS_WRITE_ERROR;
+
 	return 0;
 }
 
@@ -202,7 +194,7 @@ static int int13_read_sectors ( struct int13_drive *drive,
 	};
 
 	if ( drive->blockdev->blksize != INT13_BLKSIZE ) {
-		DBG ( "Invalid blocksize (%d) for non-extended read\n",
+		DBG ( "Invalid blocksize (%zd) for non-extended read\n",
 		      drive->blockdev->blksize );
 		return INT13_STATUS_INVALID;
 	}
@@ -233,7 +225,7 @@ static int int13_write_sectors ( struct int13_drive *drive,
 	};
 
 	if ( drive->blockdev->blksize != INT13_BLKSIZE ) {
-		DBG ( "Invalid blocksize (%d) for non-extended write\n",
+		DBG ( "Invalid blocksize (%zd) for non-extended write\n",
 		      drive->blockdev->blksize );
 		return INT13_STATUS_INVALID;
 	}

src/drivers/block/scsi.c

@@ -50,11 +50,12 @@ static int scsi_command ( struct scsi_device *scsi,
  *
  * @v blockdev		Block device
  * @v block		LBA block number
+ * @v count		Block count
  * @v buffer		Data buffer
  * @ret rc		Return status code
  */
 static int scsi_read ( struct block_device *blockdev, uint64_t block,
-		       void *buffer ) {
+		       unsigned long count, userptr_t buffer ) {
 	struct scsi_device *scsi = block_to_scsi ( blockdev );
 	struct scsi_command command;
 	struct scsi_cdb_read_16 *cdb = &command.cdb.read16;
@@ -63,9 +64,9 @@ static int scsi_read ( struct block_device *blockdev, uint64_t block,
 	memset ( &command, 0, sizeof ( command ) );
 	cdb->opcode = SCSI_OPCODE_READ_16;
 	cdb->lba = cpu_to_be64 ( block );
-	cdb->len = cpu_to_be32 ( 1 ); /* always a single block */
+	cdb->len = cpu_to_be32 ( count );
 	command.data_in = buffer;
-	command.data_in_len = blockdev->blksize;
+	command.data_in_len = ( count * blockdev->blksize );
 	return scsi_command ( scsi, &command );
 }
 
@@ -74,11 +75,12 @@ static int scsi_read ( struct block_device *blockdev, uint64_t block,
  *
  * @v blockdev		Block device
  * @v block		LBA block number
+ * @v count		Block count
  * @v buffer		Data buffer
  * @ret rc		Return status code
  */
 static int scsi_write ( struct block_device *blockdev, uint64_t block,
-			const void *buffer ) {
+		        unsigned long count, userptr_t buffer ) {
 	struct scsi_device *scsi = block_to_scsi ( blockdev );
 	struct scsi_command command;
 	struct scsi_cdb_write_16 *cdb = &command.cdb.write16;
@@ -87,9 +89,9 @@ static int scsi_write ( struct block_device *blockdev, uint64_t block,
 	memset ( &command, 0, sizeof ( command ) );
 	cdb->opcode = SCSI_OPCODE_WRITE_16;
 	cdb->lba = cpu_to_be64 ( block );
-	cdb->len = cpu_to_be32 ( 1 ); /* always a single block */
+	cdb->len = cpu_to_be32 ( count );
 	command.data_out = buffer;
-	command.data_out_len = blockdev->blksize;
+	command.data_out_len = ( count * blockdev->blksize );
 	return scsi_command ( scsi, &command );
 }
 
@@ -111,7 +113,7 @@ static int scsi_read_capacity ( struct block_device *blockdev ) {
 	cdb->opcode = SCSI_OPCODE_SERVICE_ACTION_IN;
 	cdb->service_action = SCSI_SERVICE_ACTION_READ_CAPACITY_16;
 	cdb->len = cpu_to_be32 ( sizeof ( capacity ) );
-	command.data_in = &capacity;
+	command.data_in = virt_to_user ( &capacity );
 	command.data_in_len = sizeof ( capacity );
 
 	if ( ( rc = scsi_command ( scsi, &command ) ) != 0 )

src/include/gpxe/blockdev.h

@@ -8,6 +8,8 @@
  *
  */
 
+#include <gpxe/uaccess.h>
+
 /** A block device */
 struct block_device {
 	/** Block size */
@@ -19,21 +21,23 @@ struct block_device {
 	 *
 	 * @v blockdev	Block device
 	 * @v block	Block number
+	 * @v count	Block count
 	 * @v buffer	Data buffer
 	 * @ret rc	Return status code
 	 */
 	int ( * read ) ( struct block_device *blockdev, uint64_t block,
-			 void *buffer );
+			 unsigned long count, userptr_t buffer );
 	/**
 	 * Write block
 	 *
 	 * @v blockdev	Block device
 	 * @v block	Block number
+	 * @v count	Block count
 	 * @v buffer	Data buffer
 	 * @ret rc	Return status code
 	 */
 	int ( * write ) ( struct block_device *blockdev, uint64_t block,
-			  const void *buffer );
+			  unsigned long count, userptr_t buffer );
 };
 
 #endif /* _GPXE_BLOCKDEV_H */

src/include/gpxe/scsi.h

@@ -3,6 +3,7 @@
 
 #include <stdint.h>
 #include <gpxe/blockdev.h>
+#include <gpxe/uaccess.h>
 
 /**
  * @defgroup scsiops SCSI operation codes
@@ -123,14 +124,14 @@ struct scsi_command {
 	/** CDB for this command */
 	union scsi_cdb cdb;
 	/** Data-out buffer (may be NULL) */
-	const void *data_out;
+	userptr_t data_out;
 	/** Data-out buffer length
 	 *
 	 * Must be zero if @c data_out is NULL
 	 */
 	size_t data_out_len;
 	/** Data-in buffer (may be NULL) */
-	void *data_in;
+	userptr_t data_in;
 	/** Data-in buffer length
 	 *
 	 * Must be zero if @c data_in is NULL

src/include/gpxe/uaccess.h

@@ -0,0 +1,24 @@
+#ifndef _GPXE_UACCESS_H
+#define _GPXE_UACCESS_H
+
+/**
+ * @file
+ *
+ * Access to external ("user") memory
+ *
+ * gPXE often needs to transfer data between internal and external
+ * buffers.  On i386, the external buffers may require access via a
+ * different segment, and the buffer address cannot be encoded into a
+ * simple void * pointer.  The @c userptr_t type encapsulates the
+ * information needed to identify an external buffer, and the
+ * copy_to_user() and copy_from_user() functions provide methods for
+ * transferring data between internal and external buffers.
+ *
+ * Note that userptr_t is an opaque type; in particular, performing
+ * arithmetic upon a userptr_t is not allowed.
+ *
+ */
+
+#include <bits/uaccess.h>
+
+#endif /* _GPXE_UACCESS_H */

src/net/tcp/iscsi.c

@@ -24,6 +24,7 @@
 #include <byteswap.h>
 #include <gpxe/scsi.h>
 #include <gpxe/process.h>
+#include <gpxe/uaccess.h>
 #include <gpxe/iscsi.h>
 
 /** @file
@@ -130,7 +131,7 @@ static void iscsi_rx_data_in ( struct iscsi_session *iscsi, void *data,
 	assert ( iscsi->command != NULL );
 	assert ( iscsi->command->data_in != NULL );
 	assert ( ( offset + len ) <= iscsi->command->data_in_len );
-	memcpy ( ( iscsi->command->data_in + offset ), data, len );
+	copy_to_user ( iscsi->command->data_in, offset, data, len );
 
 	/* Record SCSI status, if present */
 	if ( data_in->flags & ISCSI_DATA_FLAG_STATUS )
@@ -234,7 +235,11 @@ static void iscsi_tx_data_out ( struct iscsi_session *iscsi ) {
 	assert ( iscsi->command->data_out != NULL );
 	assert ( ( offset + len ) <= iscsi->command->data_out_len );
 	
-	tcp_send ( &iscsi->tcp, iscsi->command->data_out + offset, len );
+	if ( len > tcp_buflen )
+		len = tcp_buflen;
+	copy_from_user ( tcp_buffer, iscsi->command->data_out, offset, len );
+
+	tcp_send ( &iscsi->tcp, tcp_buffer, len );
 }
 
 /****************************************************************************