Patchwork fix acpi regression

login
register
mail settings
Submitter Wen Congyang
Date April 12, 2011, 5:13 a.m.
Message ID <4DA3DF6B.8010308@cn.fujitsu.com>
Download mbox | patch
Permalink /patch/90705/
State New
Headers show

Comments

Wen Congyang - April 12, 2011, 5:13 a.m.
This bug is introduced by commit 23910d3f.

Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>

---
 hw/acpi.c |   10 +++-------
 1 files changed, 3 insertions(+), 7 deletions(-)
Isaku Yamahata - April 12, 2011, 8:48 a.m.
On Tue, Apr 12, 2011 at 01:13:15PM +0800, Wen Congyang wrote:
> This bug is introduced by commit 23910d3f.

Oh, Thanks. Good catch. I agree with the first hunk.
But what is the second hunk for? The GPE STS register is W1C.
(NULL check and 0xff masking is okay. it's a bit redundant, though)

From ACPI4 spec 4.7.4.1.1.
can only be cleared by software writing a "1" to its respective bit position.

thanks,

> 
> Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
> 
> ---
>  hw/acpi.c |   10 +++-------
>  1 files changed, 3 insertions(+), 7 deletions(-)
> 
> diff --git a/hw/acpi.c b/hw/acpi.c
> index e372474..790bd3b 100644
> --- a/hw/acpi.c
> +++ b/hw/acpi.c
> @@ -355,7 +355,7 @@ static uint8_t *acpi_gpe_ioport_get_ptr(ACPIGPE *gpe, uint32_t addr)
>      if (addr < gpe->len / 2) {
>          cur = gpe->sts + addr;
>      } else if (addr < gpe->len) {
> -        cur = gpe->en + addr;
> +        cur = gpe->en + addr - gpe->len / 2;
>      } else {
>          abort();
>      }
> @@ -369,12 +369,8 @@ void acpi_gpe_ioport_writeb(ACPIGPE *gpe, uint32_t addr, uint32_t val)
>  
>      addr -= gpe->blk;
>      cur = acpi_gpe_ioport_get_ptr(gpe, addr);
> -    if (addr < gpe->len / 2) {
> -        /* GPE_STS */
> -        *cur = (*cur) & ~val;
> -    } else if (addr < gpe->len) {
> -        /* GPE_EN */
> -        *cur = val;
> +    if (cur != NULL) {
> +        *cur = val & 0xff;
>      } else {
>          abort();
>      }
> -- 
> 1.7.1
>
Wen Congyang - April 12, 2011, 9:08 a.m.
At 04/12/2011 04:48 PM, Isaku Yamahata Write:
> On Tue, Apr 12, 2011 at 01:13:15PM +0800, Wen Congyang wrote:
>> This bug is introduced by commit 23910d3f.
> 
> Oh, Thanks. Good catch. I agree with the first hunk.
> But what is the second hunk for? The GPE STS register is W1C.
> (NULL check and 0xff masking is okay. it's a bit redundant, though)

I found this bug when I hotplug a nic.
I do not know about ACPI.
The second hunk is the same before commit 23910d3f.
I will read ACPI spec to confirm it and update this patch.

Thanks

> 
>>From ACPI4 spec 4.7.4.1.1.
> can only be cleared by software writing a "1" to its respective bit position.
> 
> thanks,
> 
>>
>> Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
>>
>> ---
>>  hw/acpi.c |   10 +++-------
>>  1 files changed, 3 insertions(+), 7 deletions(-)
>>
>> diff --git a/hw/acpi.c b/hw/acpi.c
>> index e372474..790bd3b 100644
>> --- a/hw/acpi.c
>> +++ b/hw/acpi.c
>> @@ -355,7 +355,7 @@ static uint8_t *acpi_gpe_ioport_get_ptr(ACPIGPE *gpe, uint32_t addr)
>>      if (addr < gpe->len / 2) {
>>          cur = gpe->sts + addr;
>>      } else if (addr < gpe->len) {
>> -        cur = gpe->en + addr;
>> +        cur = gpe->en + addr - gpe->len / 2;
>>      } else {
>>          abort();
>>      }
>> @@ -369,12 +369,8 @@ void acpi_gpe_ioport_writeb(ACPIGPE *gpe, uint32_t addr, uint32_t val)
>>  
>>      addr -= gpe->blk;
>>      cur = acpi_gpe_ioport_get_ptr(gpe, addr);
>> -    if (addr < gpe->len / 2) {
>> -        /* GPE_STS */
>> -        *cur = (*cur) & ~val;
>> -    } else if (addr < gpe->len) {
>> -        /* GPE_EN */
>> -        *cur = val;
>> +    if (cur != NULL) {
>> +        *cur = val & 0xff;
>>      } else {
>>          abort();
>>      }
>> -- 
>> 1.7.1
>>
Isaku Yamahata - April 12, 2011, 9:20 a.m.
On Tue, Apr 12, 2011 at 05:08:18PM +0800, Wen Congyang wrote:
> At 04/12/2011 04:48 PM, Isaku Yamahata Write:
> > On Tue, Apr 12, 2011 at 01:13:15PM +0800, Wen Congyang wrote:
> >> This bug is introduced by commit 23910d3f.
> > 
> > Oh, Thanks. Good catch. I agree with the first hunk.
> > But what is the second hunk for? The GPE STS register is W1C.
> > (NULL check and 0xff masking is okay. it's a bit redundant, though)
> 
> I found this bug when I hotplug a nic.
> I do not know about ACPI.
> The second hunk is the same before commit 23910d3f.
> I will read ACPI spec to confirm it and update this patch.

You mean gpe_writeb(), gpe_write_val(), gpe_reset_val()?
gpe_reset_val() does W1C.
The GPE0_LEN of piix4 is 4 byte length and the first 2 registers are
sts register. From the old code,

static void gpe_reset_val(uint16_t *cur, int addr, uint32_t val)
{   
    uint16_t x1, x0 = val & 0xff;
    int shift = (addr & 1) ? 8 : 0;

    x1 = (*cur >> shift) & 0xff;

    x1 = x1 & ~x0; <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< W1C

    *cur = (*cur & (0xff << (8 - shift))) | (x1 << shift);
}


static void gpe_writeb(void *opaque, uint32_t addr, uint32_t val)
{   
    PIIX4PMState *s = opaque;
    struct gpe_regs *g = &s->gpe;

    switch (addr) {
        case GPE_BASE:
        case GPE_BASE + 1:
            gpe_reset_val(&g->sts, addr, val);
            break;

thanks,

> 
> Thanks
> 
> > 
> >>From ACPI4 spec 4.7.4.1.1.
> > can only be cleared by software writing a "1" to its respective bit position.
> > 
> > thanks,
> > 
> >>
> >> Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
> >>
> >> ---
> >>  hw/acpi.c |   10 +++-------
> >>  1 files changed, 3 insertions(+), 7 deletions(-)
> >>
> >> diff --git a/hw/acpi.c b/hw/acpi.c
> >> index e372474..790bd3b 100644
> >> --- a/hw/acpi.c
> >> +++ b/hw/acpi.c
> >> @@ -355,7 +355,7 @@ static uint8_t *acpi_gpe_ioport_get_ptr(ACPIGPE *gpe, uint32_t addr)
> >>      if (addr < gpe->len / 2) {
> >>          cur = gpe->sts + addr;
> >>      } else if (addr < gpe->len) {
> >> -        cur = gpe->en + addr;
> >> +        cur = gpe->en + addr - gpe->len / 2;
> >>      } else {
> >>          abort();
> >>      }
> >> @@ -369,12 +369,8 @@ void acpi_gpe_ioport_writeb(ACPIGPE *gpe, uint32_t addr, uint32_t val)
> >>  
> >>      addr -= gpe->blk;
> >>      cur = acpi_gpe_ioport_get_ptr(gpe, addr);
> >> -    if (addr < gpe->len / 2) {
> >> -        /* GPE_STS */
> >> -        *cur = (*cur) & ~val;
> >> -    } else if (addr < gpe->len) {
> >> -        /* GPE_EN */
> >> -        *cur = val;
> >> +    if (cur != NULL) {
> >> +        *cur = val & 0xff;
> >>      } else {
> >>          abort();
> >>      }
> >> -- 
> >> 1.7.1
> >>
> 
>
Wen Congyang - April 12, 2011, 9:26 a.m.
At 04/12/2011 05:20 PM, Isaku Yamahata Write:
> On Tue, Apr 12, 2011 at 05:08:18PM +0800, Wen Congyang wrote:
>> At 04/12/2011 04:48 PM, Isaku Yamahata Write:
>>> On Tue, Apr 12, 2011 at 01:13:15PM +0800, Wen Congyang wrote:
>>>> This bug is introduced by commit 23910d3f.
>>>
>>> Oh, Thanks. Good catch. I agree with the first hunk.
>>> But what is the second hunk for? The GPE STS register is W1C.
>>> (NULL check and 0xff masking is okay. it's a bit redundant, though)
>>
>> I found this bug when I hotplug a nic.
>> I do not know about ACPI.
>> The second hunk is the same before commit 23910d3f.
>> I will read ACPI spec to confirm it and update this patch.
> 
> You mean gpe_writeb(), gpe_write_val(), gpe_reset_val()?
> gpe_reset_val() does W1C.
> The GPE0_LEN of piix4 is 4 byte length and the first 2 registers are
> sts register. From the old code,
> 
> static void gpe_reset_val(uint16_t *cur, int addr, uint32_t val)
> {   
>     uint16_t x1, x0 = val & 0xff;
>     int shift = (addr & 1) ? 8 : 0;
> 
>     x1 = (*cur >> shift) & 0xff;
> 
>     x1 = x1 & ~x0; <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< W1C
> 
>     *cur = (*cur & (0xff << (8 - shift))) | (x1 << shift);
> }
> 
> 
> static void gpe_writeb(void *opaque, uint32_t addr, uint32_t val)
> {   
>     PIIX4PMState *s = opaque;
>     struct gpe_regs *g = &s->gpe;
> 
>     switch (addr) {
>         case GPE_BASE:
>         case GPE_BASE + 1:
>             gpe_reset_val(&g->sts, addr, val);
>             break;

Ah, you are right.
Thanks.

> 
> thanks,
> 
>>
>> Thanks
>>
>>>
>>> >From ACPI4 spec 4.7.4.1.1.
>>> can only be cleared by software writing a "1" to its respective bit position.
>>>
>>> thanks,
>>>
>>>>
>>>> Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
>>>>
>>>> ---
>>>>  hw/acpi.c |   10 +++-------
>>>>  1 files changed, 3 insertions(+), 7 deletions(-)
>>>>
>>>> diff --git a/hw/acpi.c b/hw/acpi.c
>>>> index e372474..790bd3b 100644
>>>> --- a/hw/acpi.c
>>>> +++ b/hw/acpi.c
>>>> @@ -355,7 +355,7 @@ static uint8_t *acpi_gpe_ioport_get_ptr(ACPIGPE *gpe, uint32_t addr)
>>>>      if (addr < gpe->len / 2) {
>>>>          cur = gpe->sts + addr;
>>>>      } else if (addr < gpe->len) {
>>>> -        cur = gpe->en + addr;
>>>> +        cur = gpe->en + addr - gpe->len / 2;
>>>>      } else {
>>>>          abort();
>>>>      }
>>>> @@ -369,12 +369,8 @@ void acpi_gpe_ioport_writeb(ACPIGPE *gpe, uint32_t addr, uint32_t val)
>>>>  
>>>>      addr -= gpe->blk;
>>>>      cur = acpi_gpe_ioport_get_ptr(gpe, addr);
>>>> -    if (addr < gpe->len / 2) {
>>>> -        /* GPE_STS */
>>>> -        *cur = (*cur) & ~val;
>>>> -    } else if (addr < gpe->len) {
>>>> -        /* GPE_EN */
>>>> -        *cur = val;
>>>> +    if (cur != NULL) {
>>>> +        *cur = val & 0xff;
>>>>      } else {
>>>>          abort();
>>>>      }
>>>> -- 
>>>> 1.7.1
>>>>
>>
>>
>

Patch

diff --git a/hw/acpi.c b/hw/acpi.c
index e372474..790bd3b 100644
--- a/hw/acpi.c
+++ b/hw/acpi.c
@@ -355,7 +355,7 @@  static uint8_t *acpi_gpe_ioport_get_ptr(ACPIGPE *gpe, uint32_t addr)
     if (addr < gpe->len / 2) {
         cur = gpe->sts + addr;
     } else if (addr < gpe->len) {
-        cur = gpe->en + addr;
+        cur = gpe->en + addr - gpe->len / 2;
     } else {
         abort();
     }
@@ -369,12 +369,8 @@  void acpi_gpe_ioport_writeb(ACPIGPE *gpe, uint32_t addr, uint32_t val)
 
     addr -= gpe->blk;
     cur = acpi_gpe_ioport_get_ptr(gpe, addr);
-    if (addr < gpe->len / 2) {
-        /* GPE_STS */
-        *cur = (*cur) & ~val;
-    } else if (addr < gpe->len) {
-        /* GPE_EN */
-        *cur = val;
+    if (cur != NULL) {
+        *cur = val & 0xff;
     } else {
         abort();
     }