All of lore.kernel.org
 help / color / mirror / Atom feed
From: "Michael S. Tsirkin" <mst@redhat.com>
To: Marcel Apfelbaum <marcel.a@redhat.com>
Cc: kevin@koconnor.net, seabios@seabios.org, qemu-devel@nongnu.org,
	kraxel@redhat.com
Subject: Re: [Qemu-devel] [PATCH V3 2/2] hw/pci: check if pci2pci bridges implement optional limit registers
Date: Thu, 10 Apr 2014 18:46:48 +0300	[thread overview]
Message-ID: <20140410154648.GG21110@redhat.com> (raw)
In-Reply-To: <1397136581-10500-3-git-send-email-marcel.a@redhat.com>

On Thu, Apr 10, 2014 at 04:29:41PM +0300, Marcel Apfelbaum wrote:
> <I/O Base Register, I/O Limit Register> pair and
> <Prefetchable Memory Base Register, Prefetchable Memory Limit Register> pair
> are both optional.
> Do not reserve ranges if the above registers are not implemented.
> 
> Signed-off-by: Marcel Apfelbaum <marcel.a@redhat.com>


Reviewed-by: Michael S. Tsirkin <mst@redhat.com>




> ---
>  src/fw/pciinit.c |  9 ++-------
>  src/hw/pci.c     | 48 ++++++++++++++++++++++++++++++++++++++++++++++++
>  src/hw/pci.h     |  9 +++++++++
>  3 files changed, 59 insertions(+), 7 deletions(-)
> 
> diff --git a/src/fw/pciinit.c b/src/fw/pciinit.c
> index 9b5d7ad..bbaecd6 100644
> --- a/src/fw/pciinit.c
> +++ b/src/fw/pciinit.c
> @@ -26,13 +26,6 @@
>  #define PCI_BRIDGE_MEM_MIN    (1<<21)  // 2M == hugepage size
>  #define PCI_BRIDGE_IO_MIN      0x1000  // mandated by pci bridge spec
>  
> -enum pci_region_type {
> -    PCI_REGION_TYPE_IO,
> -    PCI_REGION_TYPE_MEM,
> -    PCI_REGION_TYPE_PREFMEM,
> -    PCI_REGION_TYPE_COUNT,
> -};
> -
>  static const char *region_type_name[] = {
>      [ PCI_REGION_TYPE_IO ]      = "io",
>      [ PCI_REGION_TYPE_MEM ]     = "mem",
> @@ -681,6 +674,8 @@ static int pci_bios_check_devices(struct pci_bus *busses)
>          for (type = 0; type < PCI_REGION_TYPE_COUNT; type++) {
>              u64 align = (type == PCI_REGION_TYPE_IO) ?
>                  PCI_BRIDGE_IO_MIN : PCI_BRIDGE_MEM_MIN;
> +            if (!pci_bridge_has_region(s->bus_dev, type))
> +                continue;
>              if (pci_region_align(&s->r[type]) > align)
>                   align = pci_region_align(&s->r[type]);
>              u64 sum = pci_region_sum(&s->r[type]);
> diff --git a/src/hw/pci.c b/src/hw/pci.c
> index 055353d..27e7b1c 100644
> --- a/src/hw/pci.c
> +++ b/src/hw/pci.c
> @@ -243,6 +243,54 @@ u8 pci_find_capability(struct pci_device *pci, u8 cap_id)
>      return 0;
>  }
>  
> +static int pci_config_writableb(struct pci_device *pci, u32 addr, u8 test_val)

Don't like this name.
pci_config_writable_byte or better see below.

> +{
> +    u8 val;
> +
> +    val = pci_config_readb(pci->bdf, addr);
> +    pci_config_writeb(pci->bdf, addr, test_val);
> +
> +    if (!(pci_config_readb(pci->bdf, addr)))
> +        return 0;

This only works for readonly fields which return 0
on read. Maybe name should reflect this:
how about pci_config_is_reserved() ?

There's no need to pass in test value either,
it makes the interface fragile:
passing e.g. 0x0 as value won't work.

> +
> +    pci_config_writeb(pci->bdf, addr, val);

Practically that's overkill.

Just set base > limit and there won't be
need to save and restore: range is disabled.



> +    return 1;
> +}
> +
> +static int pci_config_writablew(struct pci_device *pci, u32 addr, u16 test_val)
> +{
> +    u16 val;
> +
> +    val = pci_config_readw(pci->bdf, addr);
> +    pci_config_writew(pci->bdf, addr, test_val);
> +
> +    if (!(pci_config_readw(pci->bdf, addr)))
> +        return 0;
> +
> +    pci_config_writew(pci->bdf, addr, val);
> +    return 1;
> +}


This one isn't needed at all.

> +
> +int pci_bridge_has_region(struct pci_device *pci,
> +                          enum pci_region_type region_type)
> +{
> +    if (pci->class != PCI_CLASS_BRIDGE_PCI)
> +        return 0;

Do we really need this test here?

> +
> +    switch (region_type) {
> +    case PCI_REGION_TYPE_IO:
> +        return pci_config_writableb(pci, PCI_IO_BASE, 0xF0) &&
> +               pci_config_writableb(pci, PCI_IO_LIMIT, 0xF0);

I think we should only test base.
Testing limit like this might cause issues as bridge now
claims some resources temporarily.
Also why 0xF0 and not 0xFF? Seems like a random value.


> +    case PCI_REGION_TYPE_PREFMEM:
> +        return pci_config_writablew(pci, PCI_PREF_MEMORY_BASE, 0xFFF0) &&
> +               pci_config_writablew(pci, PCI_PREF_MEMORY_LIMIT, 0xFFF0);
> +    case PCI_REGION_TYPE_MEM:    /* fall through */
> +    default:
> +        return 1;
> +    }
> +
> +    return 1;
> +}
>  
>  void
>  pci_reboot(void)


Same comments apply to prefetch and writeableb,
but also - there is no need to write all bytes.
So really

Basically:

int pci_bridge_has_region(struct pci_device *pci,
> +                          enum pci_region_type region_type)
> +{
> +    if (pci->class != PCI_CLASS_BRIDGE_PCI)
> +        return 0;

Do we really need this test here?

> +
> +    switch (region_type) {
> +    case PCI_REGION_TYPE_IO:
> +        return pci_config_writableb(pci, PCI_IO_BASE, 0xF0) &&
> +               pci_config_writableb(pci, PCI_IO_LIMIT, 0xF0);

I think we should only test base.
Testing limit like this might cause issues as bridge now
claims some resources temporarily.
Also why 0xF0 and not 0xFF? Seems like a random value.


> +    case PCI_REGION_TYPE_PREFMEM:
> +        return pci_config_writablew(pci, PCI_PREF_MEMORY_BASE, 0xFFF0) &&
> +               pci_config_writablew(pci, PCI_PREF_MEMORY_LIMIT, 0xFFF0);
> +    case PCI_REGION_TYPE_MEM:    /* fall through */
> +    default:
> +        return 1;
> +    }
> +
> +    return 1;

	unreacheable code.

> +}




Here's a simpler implementation.

int pci_bridge_has_region(struct pci_device *pci,
                          enum pci_region_type region_type)
{
    u8 base;
  
    switch (region_type) {
    case PCI_REGION_TYPE_IO:
	base = PCI_IO_BASE;
	break;
    case PCI_REGION_TYPE_PREFMEM:
	base = PCI_PREF_MEMORY_BASE;
	break;
    default:
	/* Regular memory support is mandatory */
        return 1;
    }

    /* Check that base is writeable. Safe as this increases base so it
     * won't cause bridge to claim new resources.
     */
    pci_config_writeb(pci->bdf, base, 0xFF);
    return !!pci_config_readb(pci->bdf, base);
}




> diff --git a/src/hw/pci.h b/src/hw/pci.h
> index e828225..0aaa84c 100644
> --- a/src/hw/pci.h
> +++ b/src/hw/pci.h
> @@ -12,6 +12,13 @@
>  #define PCI_NUM_REGIONS 7
>  #define PCI_BRIDGE_NUM_REGIONS 2
>  
> +enum pci_region_type {
> +    PCI_REGION_TYPE_IO,
> +    PCI_REGION_TYPE_MEM,
> +    PCI_REGION_TYPE_PREFMEM,
> +    PCI_REGION_TYPE_COUNT,
> +};
> +
>  static inline u8 pci_bdf_to_bus(u16 bdf) {
>      return bdf >> 8;
>  }
> @@ -117,6 +124,8 @@ int pci_init_device(const struct pci_device_id *ids
>  struct pci_device *pci_find_init_device(const struct pci_device_id *ids
>                                          , void *arg);
>  u8 pci_find_capability(struct pci_device *pci, u8 cap_id);
> +int pci_bridge_has_region(struct pci_device *pci,
> +                          enum pci_region_type region_type);
>  void pci_reboot(void);
>  
>  #endif
> -- 
> 1.8.3.1

  reply	other threads:[~2014-04-10 15:46 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-04-10 13:29 [Qemu-devel] [SeaBIOS] [PATCH V3 0/2] hw/pci: reserve IO and mem for pci-2-pci bridges with no devices attached Marcel Apfelbaum
2014-04-10 13:29 ` [Qemu-devel] [PATCH V3 1/2] " Marcel Apfelbaum
2014-04-10 15:48   ` Michael S. Tsirkin
2014-04-10 16:45   ` Kevin O'Connor
2014-04-10 16:58     ` Marcel Apfelbaum
2014-04-10 13:29 ` [Qemu-devel] [PATCH V3 2/2] hw/pci: check if pci2pci bridges implement optional limit registers Marcel Apfelbaum
2014-04-10 15:46   ` Michael S. Tsirkin [this message]
2014-04-10 16:29     ` Marcel Apfelbaum
2014-04-10 15:49 ` [Qemu-devel] [SeaBIOS] [PATCH V3 0/2] hw/pci: reserve IO and mem for pci-2-pci bridges with no devices attached Michael S. Tsirkin

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=20140410154648.GG21110@redhat.com \
    --to=mst@redhat.com \
    --cc=kevin@koconnor.net \
    --cc=kraxel@redhat.com \
    --cc=marcel.a@redhat.com \
    --cc=qemu-devel@nongnu.org \
    --cc=seabios@seabios.org \
    /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
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.