Allowing for the future possibility of implementing AML-based
(i.e., firmware-triggered) access to the QEMU fw_cfg device,
acquire the global ACPI lock when accessing the device on behalf
of the guest-side sysfs driver, to prevent any potential race
conditions.
Suggested-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Gabriel Somlo <[email protected]>
---
Turns out, there *is* a way to acquire a global ACPI lock from within
a "random" kernel driver after all. Luckily I have a healthy dose of
respect for Michael's opinions :) so I kept circling back through
existing kernel sources for an example I can use, and I think this
might be it.
I'm posting as RFC because I'm not really confident about assessing
the likelihood of there ever being a race condition between the kernel
fw_cfg sysfs driver and firmware on my own. Obviously Michael has
concerns about it, but any additional opinions from the QEMU camp would
be much appreciated.
Thanks again,
--Gabriel
drivers/firmware/qemu_fw_cfg.c | 14 ++++++++++++++
1 file changed, 14 insertions(+)
diff --git a/drivers/firmware/qemu_fw_cfg.c b/drivers/firmware/qemu_fw_cfg.c
index 7bba76c..cc4c27a 100644
--- a/drivers/firmware/qemu_fw_cfg.c
+++ b/drivers/firmware/qemu_fw_cfg.c
@@ -77,12 +77,26 @@ static inline u16 fw_cfg_sel_endianness(u16 key)
static inline void fw_cfg_read_blob(u16 key,
void *buf, loff_t pos, size_t count)
{
+#ifdef CONFIG_ACPI
+ u32 glk;
+ int status;
+ status = acpi_acquire_global_lock(ACPI_WAIT_FOREVER, &glk);
+ if (ACPI_FAILURE(status)) {
+ /* Should never get here */
+ WARN(1, "fw_cfg_read_blob: Failed to lock ACPI!\n");
+ memset(buf, 0, count);
+ return;
+ }
+#endif
mutex_lock(&fw_cfg_dev_lock);
iowrite16(fw_cfg_sel_endianness(key), fw_cfg_reg_ctrl);
while (pos-- > 0)
ioread8(fw_cfg_reg_data);
ioread8_rep(fw_cfg_reg_data, buf, count);
mutex_unlock(&fw_cfg_dev_lock);
+#ifdef CONFIG_ACPI
+ acpi_release_global_lock(glk);
+#endif
}
/* clean up fw_cfg device i/o */
--
2.4.3
On Mon, Mar 07, 2016 at 06:55:43PM -0500, Gabriel Somlo wrote:
> Allowing for the future possibility of implementing AML-based
> (i.e., firmware-triggered) access to the QEMU fw_cfg device,
> acquire the global ACPI lock when accessing the device on behalf
> of the guest-side sysfs driver, to prevent any potential race
> conditions.
>
> Suggested-by: Michael S. Tsirkin <[email protected]>
> Signed-off-by: Gabriel Somlo <[email protected]>
> ---
>
> Turns out, there *is* a way to acquire a global ACPI lock from within
> a "random" kernel driver after all. Luckily I have a healthy dose of
> respect for Michael's opinions :) so I kept circling back through
> existing kernel sources for an example I can use, and I think this
> might be it.
>
> I'm posting as RFC because I'm not really confident about assessing
> the likelihood of there ever being a race condition between the kernel
> fw_cfg sysfs driver and firmware on my own. Obviously Michael has
> concerns about it, but any additional opinions from the QEMU camp would
> be much appreciated.
>
> Thanks again,
> --Gabriel
>
> drivers/firmware/qemu_fw_cfg.c | 14 ++++++++++++++
> 1 file changed, 14 insertions(+)
>
> diff --git a/drivers/firmware/qemu_fw_cfg.c b/drivers/firmware/qemu_fw_cfg.c
> index 7bba76c..cc4c27a 100644
> --- a/drivers/firmware/qemu_fw_cfg.c
> +++ b/drivers/firmware/qemu_fw_cfg.c
> @@ -77,12 +77,26 @@ static inline u16 fw_cfg_sel_endianness(u16 key)
> static inline void fw_cfg_read_blob(u16 key,
> void *buf, loff_t pos, size_t count)
> {
> +#ifdef CONFIG_ACPI
> + u32 glk;
> + int status;
> + status = acpi_acquire_global_lock(ACPI_WAIT_FOREVER, &glk);
Why not fix up the ACPI api to not require #ifdef here? It should be
added to a .h file somewhere instead.
> + if (ACPI_FAILURE(status)) {
> + /* Should never get here */
> + WARN(1, "fw_cfg_read_blob: Failed to lock ACPI!\n");
> + memset(buf, 0, count);
> + return;
> + }
> +#endif
> mutex_lock(&fw_cfg_dev_lock);
> iowrite16(fw_cfg_sel_endianness(key), fw_cfg_reg_ctrl);
> while (pos-- > 0)
> ioread8(fw_cfg_reg_data);
> ioread8_rep(fw_cfg_reg_data, buf, count);
> mutex_unlock(&fw_cfg_dev_lock);
> +#ifdef CONFIG_ACPI
> + acpi_release_global_lock(glk);
Same here.
thanks,
greg k-h
On Mon, Mar 07, 2016 at 04:02:15PM -0800, Greg KH wrote:
> On Mon, Mar 07, 2016 at 06:55:43PM -0500, Gabriel Somlo wrote:
> > Allowing for the future possibility of implementing AML-based
> > (i.e., firmware-triggered) access to the QEMU fw_cfg device,
> > acquire the global ACPI lock when accessing the device on behalf
> > of the guest-side sysfs driver, to prevent any potential race
> > conditions.
> >
> > Suggested-by: Michael S. Tsirkin <[email protected]>
> > Signed-off-by: Gabriel Somlo <[email protected]>
ACK, except for Gerd's comments below.
> > ---
> >
> > Turns out, there *is* a way to acquire a global ACPI lock from within
> > a "random" kernel driver after all. Luckily I have a healthy dose of
> > respect for Michael's opinions :) so I kept circling back through
> > existing kernel sources for an example I can use, and I think this
> > might be it.
> >
> > I'm posting as RFC because I'm not really confident about assessing
> > the likelihood of there ever being a race condition between the kernel
> > fw_cfg sysfs driver and firmware on my own. Obviously Michael has
> > concerns about it, but any additional opinions from the QEMU camp would
> > be much appreciated.
> >
> > Thanks again,
> > --Gabriel
> >
> > drivers/firmware/qemu_fw_cfg.c | 14 ++++++++++++++
> > 1 file changed, 14 insertions(+)
> >
> > diff --git a/drivers/firmware/qemu_fw_cfg.c b/drivers/firmware/qemu_fw_cfg.c
> > index 7bba76c..cc4c27a 100644
> > --- a/drivers/firmware/qemu_fw_cfg.c
> > +++ b/drivers/firmware/qemu_fw_cfg.c
> > @@ -77,12 +77,26 @@ static inline u16 fw_cfg_sel_endianness(u16 key)
> > static inline void fw_cfg_read_blob(u16 key,
> > void *buf, loff_t pos, size_t count)
> > {
> > +#ifdef CONFIG_ACPI
You should include linux/acpi.h if you use this,
do not rely on some other header to pull this in.
> > + u32 glk;
> > + int status;
> > + status = acpi_acquire_global_lock(ACPI_WAIT_FOREVER, &glk);
>
> Why not fix up the ACPI api to not require #ifdef here? It should be
> added to a .h file somewhere instead.
Indeed. In fact include/acpi/platform/aclinux.h already has
this code:
#ifndef CONFIG_ACPI
/* External globals for __KERNEL__, stubs is needed */
#define ACPI_GLOBAL(t,a)
#define ACPI_INIT_GLOBAL(t,a,b)
/* Generating stubs for configurable ACPICA macros */
#define ACPI_NO_MEM_ALLOCATIONS
/* Generating stubs for configurable ACPICA functions */
#define ACPI_NO_ERROR_MESSAGES
#undef ACPI_DEBUG_OUTPUT
/* External interface for __KERNEL__, stub is needed */
#define ACPI_EXTERNAL_RETURN_STATUS(prototype) \
static ACPI_INLINE prototype {return(AE_NOT_CONFIGURED);}
#define ACPI_EXTERNAL_RETURN_OK(prototype) \
static ACPI_INLINE prototype {return(AE_OK);}
#define ACPI_EXTERNAL_RETURN_VOID(prototype) \
static ACPI_INLINE prototype {return;}
#define ACPI_EXTERNAL_RETURN_UINT32(prototype) \
static ACPI_INLINE prototype {return(0);}
#define ACPI_EXTERNAL_RETURN_PTR(prototype) \
static ACPI_INLINE prototype {return(NULL);}
#endif /* CONFIG_ACPI */
so it's just a question of testing the return code for AE_NOT_CONFIGURED.
--
MST