LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: Greg KH <gregkh@suse.de>
To: Barak Fargoun <barak@neocleus.com>
Cc: linux-kernel@vger.kernel.org, linux-pci@atrey.karlin.mff.cuni.cz,
	Guy Zana <guy@neocleus.com>
Subject: Re: [PATCH] Align PCI memory regions to page size (4K) - Fix
Date: Sun, 28 Oct 2007 12:31:04 -0700	[thread overview]
Message-ID: <20071028193104.GA13956@suse.de> (raw)
In-Reply-To: <9392A06CB0FDC847B3A530B3DC174E7B03C96F1D@mse10be1.mse10.exchange.ms>

On Sun, Oct 28, 2007 at 01:27:27PM -0400, Barak Fargoun wrote:
> Sorry, my mail client corrupted the previous mail containing the patch... it should be ok now...
> 
> Add a boot parameter (?pci-mem-align?) which forces PCI memory regions
> to be aligned to 4K. 
> 
> This is very useful when developing an hypervisor, since in case we
> want to let native domains direct access to specific hardware, we
> don?t want PCI devices to share their memory region page with other
> devices. In Xen for example, PCI devices mmio resources are mapped by
> remapping complete pages (by Intel VT-d & the Neocleus pass-through
> patch for Xen).
> 
> Signed-off-by: Barak Fargoun???? (barak@neocleus.com)
> ---
> Kernel version: 2.6.18	(since this is the kernel version used in Xen for Dom-0)
> ---
> diff -r 840b9df48b6a drivers/pci/bus.c
> --- a/drivers/pci/bus.c	Tue Aug 07 09:37:41 2007 +0100
> +++ b/drivers/pci/bus.c	Sun Oct 28 08:40:52 2007 -0400
> @@ -16,6 +16,8 @@
>  #include <linux/init.h>
>  
>  #include "pci.h"
> +
> +extern int pci_mem_align;

Please don't declare externs in a .c file, put it in a header file.

>  /**
>   * pci_bus_alloc_resource - allocate a resource from a parent bus
> @@ -43,6 +45,15 @@ pci_bus_alloc_resource(struct pci_bus *b
>  	int i, ret = -ENOMEM;
>  
>  	type_mask |= IORESOURCE_IO | IORESOURCE_MEM;
> +
> +	/* if the boot parameter 'pci-mem-align' was specified, then we need to 
> +		align the memory addresses, at page size alignment  */
> +	if (pci_mem_align)
> +	{
> +		/* we change only alignments which are smaller than page size */
> +		if (align < (PAGE_SIZE-1))
> +			align = PAGE_SIZE - 1;
> +	}

Please follow the proper coding style rules as documented in
Documentation/CodingStyle.  Also running the script scrips/checkpatch.pl
would have pointed this out.

>  
>  	for (i = 0; i < PCI_BUS_NUM_RESOURCES; i++) {
>  		struct resource *r = bus->resource[i];
> diff -r 840b9df48b6a drivers/pci/quirks.c
> --- a/drivers/pci/quirks.c	Tue Aug 07 09:37:41 2007 +0100
> +++ b/drivers/pci/quirks.c	Sun Oct 28 08:40:52 2007 -0400
> @@ -22,6 +22,54 @@
>  #include <linux/delay.h>
>  #include <linux/acpi.h>
>  #include "pci.h"
> +
> +/* a global flag, which signals if we should align PCI mem windows to 4K */
> +int pci_mem_align = 0;
> +
> +/* this function is called, if the 'pci-mem-align' was specified as a boot
> +		parameter. Used to force the kernel to align PCI mem windows to 4k */

All command line paramaters need to be documented in the Documentation
directory.

> +static int __init set_pci_mem_align(char *str)
> +{
> +	pci_mem_align = 1;
> +	return 1;
> +}
> +__setup("pci-mem-align", set_pci_mem_align);
> +
> +/*	
> +	This quirk function enables us to force all memory resources which are 
> +		assigned to PCI devices, to be aligned at a page size. This 
> +		fetaurs helps us in xen when running native domains

Trailing spaces are not allowed :(

> +*/
> +static void __devinit quirk_xen_align_mem_resources(struct pci_dev *dev)
> +{
> +	int i;
> +	struct resource *r;
> +
> +	/* if the boot parameter 'pci-mem-align' wasn't specified, then no need
> +		to align the memory addresses */
> +	if (pci_mem_align == 0)
> +		return;
> +
> +	for (i=0; i < DEVICE_COUNT_RESOURCE; ++i) {
> +		r = &dev->resource[i];
> +
> +		if (r == NULL)
> +			continue;
> +
> +		/* we need to adjust only memory resources */
> +		if (r->flags & IORESOURCE_MEM) {
> +			/* align the start address in page size alignment alignment (round up) */
> +			resource_size_t old_start = r->start;
> +			r->start = (r->start + (PAGE_SIZE-1))  & (~(PAGE_SIZE-1));
> +
> +			/* update the end address, so that the region size will not be 
> +				affected */
> +			r->end = r->end - (old_start - r->start);
> +		}
> +	}
> +}
> +DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, quirk_xen_align_mem_resources);

Now for the main reason.  I really don't understand why this is needed.
Is it only for virtual machines?  Are there drivers in the kernel tree
today that need this functionality?  What breaks if you do not align
these resources?

thanks,

greg k-h

  reply	other threads:[~2007-10-28 19:33 UTC|newest]

Thread overview: 15+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2007-10-28 17:27 Barak Fargoun
2007-10-28 19:31 ` Greg KH [this message]
2007-10-28 19:53   ` Barak Fargoun
2007-10-28 20:03     ` Greg KH
2007-10-28 20:44       ` Barak Fargoun
2007-10-29  1:08       ` David Miller
2007-11-13 21:17         ` Benjamin Herrenschmidt
2007-11-14  6:21           ` Grant Grundler
2007-11-14  8:16             ` Benjamin Herrenschmidt
2007-11-14 21:55               ` Grant Grundler
2007-11-14 22:16                 ` Benjamin Herrenschmidt
2007-10-29  5:52       ` Grant Grundler
2007-11-08 23:24         ` Linas Vepstas
2007-11-12 23:43           ` Grant Grundler
2007-10-28 19:48 ` Arjan van de Ven

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20071028193104.GA13956@suse.de \
    --to=gregkh@suse.de \
    --cc=barak@neocleus.com \
    --cc=guy@neocleus.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-pci@atrey.karlin.mff.cuni.cz \
    --subject='Re: [PATCH] Align PCI memory regions to page size (4K) - Fix' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).