3.1. Memory and Addressing in the Linux 2.4 Kernel

The Linux 2.4 kernel includes configuration options for specifying the amount of physical memory in the target computer. By default, the configuration is limited to the amount of memory that can be directly mapped into the kernel's virtual address space starting at PAGE_OFFSET. On i386 systems the default mapping scheme limits kernel-mode addressability to the first gigabyte (GB) of physical memory, also known as low memory. Conversely, high memory is normally the memory above 1 GB. High memory is not directly accessible or permanently mapped by the kernel. Support for high memory is an option that is enabled during configuration of the Linux kernel.

3.2. The Problem with Bounce Buffers

When DMA I/O is performed to or from high memory, an area is allocated in low memory known as a bounce buffer. When data travels between a device and high memory, it is first copied through the bounce buffer.

Systems with a large amount of high memory and intense I/O activity can create a large number of bounce buffers that can cause memory shortage problems. In addition, the excessive number of bounce buffer data copies can lead to performance degradation.

Peripheral component interface (PCI) devices normally address up to 4 GB of physical memory. When a bounce buffer is used for high memory that is below 4 GB, time and memory are wasted because the peripheral has the ability to address that memory directly. Using the bounce buffer patch can decrease, and possibly eliminate, the use of bounce buffers.

3.3. Locating the Patch

The latest version of the bounce buffer patch is block-highmem-all-18b.bz2, and it is available from Andrea Arcangeli's -aa series kernels at http://kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/.

3.4. Modifying Your Device Driver to Avoid Bounce Buffers

If your device drivers are not listed above in the Enabled Device Drivers section, and the device is capable of high-memory DMA I/O, you can modify your device driver to make use of the bounce buffer patch as follows. More information on rebuilding a Linux device driver is available at http://www.xml.com/ldd/chapter/book/index.html.

1. A.) For SCSI Adapter Drivers: set the highmem_io bit in the Scsi_Host_Template structure.

   B.) For IDE Adapter Drivers: set the highmembit in the ide_hwif_t structure.

2. Call pci_set_dma_mask(struct pci_dev *pdev, dma_addr_t mask) to specify the address bits that the device can successfully use on DMA operations.

   If DMA I/O can be supported with the specified mask, pci_set_dma_mask() will set pdev->dma_mask and return 0. For SCSI or IDE, the mask value will also be passed by the mid-level drivers to blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr) so that bounce buffers are not created for memory directly addressable by the device. Drivers other than SCSI or IDE must call blk_queue_bounce_limit() directly.

3. Use pci_map_page(dev, page, offset, size, direction), instead of pci_map_single(dev, address, size, direction) to map a memory region so that it is accessible by the peripheral device. pci_map_page() supports both high and low memory.

   The address parameter for pci_map_single() correlates to the page and offset parameters for pci_map_page(). Use the virt_to_page() macro to convert an address to a page and offset. The virt_to_page() macro is defined by including pci.h. For example:

   void *address;

   struct page *page;

   unsigned long offset;

   page = virt_to_page(address);

   offset = (unsigned long) address & ~PAGE_MASK;

   Call pci_unmap_page() after the DMA I/O transfer is complete to remove the mapping established by pci_map_page().

   pci_map_single() is implemented using virt_to_bus(). virt_to_bus() handles low memory addresses only. Drivers supporting high memory should no longer call virt_to_bus() or bus_to_virt().

4. If your driver calls pci_map_sg() to map a scatter-gather DMA operation, your driver should set the page and offset fields instead of the address field of the scatterlist structure. Refer to step 3 for converting an address to a page and offset.

   If your driver is already using the PCI DMA API, continue to use pci_map_page() or pci_map_sg() as appropriate. However, do not use the address field of the scatterlist structure.

4.1. Locating the Patch

You can download the patch from one of the following locations:

• Andrea Arcangeli has made the patch available at http://www.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.18pre7aa2/. The name of the file is 10_rawio-vary-io-1.

• Alan Cox has included the patch in the 2.4.18pre9-ac2 kernel patch. The patch is available at http://www.kernel.org/pub/linux/kernel/people/alan/linux-2.4/2.4.18/.

• The patch is available from SourceForge at http://sourceforge.net/projects/lse/io. The latest version is PAGESIZE_io-2.4.17.patch.

4.2. Modifying Your Driver for the Raw I/O Variable-Size Optimization Patch

In previous versions of this patch, changes were enabled for all drivers. However, the 2.4.17 and later versions of the patch enable the changes only for the Adaptec, Qlogic ISP1020, and IBM ServerRAID drivers. All other drivers for version 2.4.17 and later must be modified to make use of the patch by setting the can_do_varyio bit in the Scsi_Host_Template structure.

Drivers that have the raw I/O patch enabled must support buffer heads of variable sizes (b_size) in a single I/O request because hardsect_size is used until the data buffer is aligned on a 4 K boundary. Additional information is available on rebuilding Linux device drivers at http://www.xml.com/ldd/chapter/book/index.html.

5.1. Locating the Patch

You can download the I/O request patch from Sourceforge at http://sourceforge.net/projects/lse/io. The latest version is sior1-v1.2416. Patches that enable concurrent queuing for specific drivers are also available at SourceForge. The patch for the Emulex SCSI/FC is lpfc_sior1-v0.249 and the patch for Adaptec SCSI is aic_sior1-v0.249.

5.2. Modifying Your Driver for the I/O Request Lock Patch

The I/O request lock patch installs concurrent queuing capability into the SCSI midlayer. Concurrent queuing is activated for each SCSI adapter device driver. To activate the driver, the concurrent_queue field in the Scsi_Host_Template structure must be set when the driver is registered.

Drivers that activate concurrent queuing must ensure that any access of the request_queue by the driver is protected by the request_queue.queue_lock. Additional information is available on rebuilding device drivers at http://www.xml.com/ldd/chapter/book/index.html.