| Message ID | 20190123212234.32068-3-stefanha@redhat.com |
|---|---|
| State | New |
| Headers | show |
| Series | arm: microbit Non-Volatile Memory Controller | expand |
Hi Stefan, On 1/23/19 10:22 PM, Stefan Hajnoczi wrote: > pflash devices should mark the memory region dirty and invalidate TBs > after directly writing to the RAM backing the ROM device. > > Note that pflash_cfi01_get_memory() is used by several machine types to > populate ROM contents directly. Callers are untouched by this patch > because they only modify memory before the guest is started. > > Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> > --- > hw/block/pflash_cfi01.c | 5 +++++ > hw/block/pflash_cfi02.c | 8 ++++++++ > 2 files changed, 13 insertions(+) > > diff --git a/hw/block/pflash_cfi01.c b/hw/block/pflash_cfi01.c > index bffb4c40e7..5301c11c18 100644 > --- a/hw/block/pflash_cfi01.c > +++ b/hw/block/pflash_cfi01.c > @@ -446,6 +446,7 @@ static inline void pflash_data_write(pflash_t *pfl, hwaddr offset, > break; > } > > + memory_region_flush_rom_device(&pfl->mem, offset, width); > } > > static void pflash_write(pflash_t *pfl, hwaddr offset, > @@ -482,6 +483,8 @@ static void pflash_write(pflash_t *pfl, hwaddr offset, > if (!pfl->ro) { > memset(p + offset, 0xff, pfl->sector_len); > pflash_update(pfl, offset, pfl->sector_len); > + memory_region_flush_rom_device(&pfl->mem, offset, > + pfl->sector_len); > } else { > pfl->status |= 0x20; /* Block erase error */ > } > @@ -763,6 +766,8 @@ static void pflash_cfi01_realize(DeviceState *dev, Error **errp) > error_setg(errp, "failed to read the initial flash content"); > return; > } > + > + memory_region_flush_rom_device(&pfl->mem, 0, total_len); The device realize() is also executed before the guest is started, is this call really necessary? > } > > /* Default to devices being used at their maximum device width. This was > diff --git a/hw/block/pflash_cfi02.c b/hw/block/pflash_cfi02.c > index 0f8b7b8c7b..d04572eca4 100644 > --- a/hw/block/pflash_cfi02.c > +++ b/hw/block/pflash_cfi02.c > @@ -378,6 +378,8 @@ static void pflash_write (pflash_t *pfl, hwaddr offset, > pflash_update(pfl, offset, 4); > break; > } > + > + memory_region_flush_rom_device(&pfl->orig_mem, offset, width); > } > pfl->status = 0x00 | ~(value & 0x80); > /* Let's pretend write is immediate */ > @@ -426,6 +428,8 @@ static void pflash_write (pflash_t *pfl, hwaddr offset, > if (!pfl->ro) { > memset(pfl->storage, 0xFF, pfl->chip_len); > pflash_update(pfl, 0, pfl->chip_len); > + memory_region_flush_rom_device(&pfl->orig_mem, 0, > + pfl->chip_len); > } > pfl->status = 0x00; > /* Let's wait 5 seconds before chip erase is done */ > @@ -441,6 +445,8 @@ static void pflash_write (pflash_t *pfl, hwaddr offset, > if (!pfl->ro) { > memset(p + offset, 0xFF, pfl->sector_len); > pflash_update(pfl, offset, pfl->sector_len); > + memory_region_flush_rom_device(&pfl->orig_mem, offset, > + pfl->sector_len); > } > pfl->status = 0x00; > /* Let's wait 1/2 second before sector erase is done */ > @@ -590,6 +596,8 @@ static void pflash_cfi02_realize(DeviceState *dev, Error **errp) > error_setg(errp, "failed to read the initial flash content"); > return; > } > + > + memory_region_flush_rom_device(&pfl->orig_mem, 0, chip_len); Ditto. > } > > pflash_setup_mappings(pfl); >
On Thu, Jan 24, 2019 at 12:11:55PM +0100, Philippe Mathieu-Daudé wrote: > On 1/23/19 10:22 PM, Stefan Hajnoczi wrote: > > pflash devices should mark the memory region dirty and invalidate TBs > > after directly writing to the RAM backing the ROM device. > > > > Note that pflash_cfi01_get_memory() is used by several machine types to > > populate ROM contents directly. Callers are untouched by this patch > > because they only modify memory before the guest is started. ... > > @@ -763,6 +766,8 @@ static void pflash_cfi01_realize(DeviceState *dev, Error **errp) > > error_setg(errp, "failed to read the initial flash content"); > > return; > > } > > + > > + memory_region_flush_rom_device(&pfl->mem, 0, total_len); > > The device realize() is also executed before the guest is started, is > this call really necessary? My rationale was that machine init only happens before the guest is started while ->realize() is called by hotplug too. That said, can pflash devices be hotplugged? If you guys prefer not flushing from pflash ->realize() then I'll drop it. Stefan
On Fri, 25 Jan 2019 at 10:28, Stefan Hajnoczi <stefanha@gmail.com> wrote: > > On Thu, Jan 24, 2019 at 12:11:55PM +0100, Philippe Mathieu-Daudé wrote: > > The device realize() is also executed before the guest is started, is > > this call really necessary? > > My rationale was that machine init only happens before the guest is > started while ->realize() is called by hotplug too. Yes, but before realize the flash memory can't actually be visible to the guest, so there's no need to invalidate anything. > That said, can pflash devices be hotplugged? I guess you could have one inside a hotplugged device in theory; they can't be directly hotpluggged. (I bet that in practice there are leaks though if you did do that, and we don't have any devices that do so today.) > If you guys prefer not flushing from pflash ->realize() then I'll drop > it. I think it makes more sense not to do a flush in the realize method. thanks -- PMM
On Fri, Jan 25, 2019 at 10:36:17AM +0000, Peter Maydell wrote: > On Fri, 25 Jan 2019 at 10:28, Stefan Hajnoczi <stefanha@gmail.com> wrote: > > > > On Thu, Jan 24, 2019 at 12:11:55PM +0100, Philippe Mathieu-Daudé wrote: > > > The device realize() is also executed before the guest is started, is > > > this call really necessary? > > > > My rationale was that machine init only happens before the guest is > > started while ->realize() is called by hotplug too. > > Yes, but before realize the flash memory can't actually be > visible to the guest, so there's no need to invalidate anything. > > > That said, can pflash devices be hotplugged? > > I guess you could have one inside a hotplugged device in theory; > they can't be directly hotpluggged. (I bet that in practice > there are leaks though if you did do that, and we don't have > any devices that do so today.) > > > If you guys prefer not flushing from pflash ->realize() then I'll drop > > it. > > I think it makes more sense not to do a flush in the realize method. Okay, I'll resend this patch. Stefan
On 1/25/19 4:37 PM, Stefan Hajnoczi wrote: > On Fri, Jan 25, 2019 at 10:36:17AM +0000, Peter Maydell wrote: >> On Fri, 25 Jan 2019 at 10:28, Stefan Hajnoczi <stefanha@gmail.com> wrote: >>> >>> On Thu, Jan 24, 2019 at 12:11:55PM +0100, Philippe Mathieu-Daudé wrote: >>>> The device realize() is also executed before the guest is started, is >>>> this call really necessary? >>> >>> My rationale was that machine init only happens before the guest is >>> started while ->realize() is called by hotplug too. >> >> Yes, but before realize the flash memory can't actually be >> visible to the guest, so there's no need to invalidate anything. >> >>> That said, can pflash devices be hotplugged? >> >> I guess you could have one inside a hotplugged device in theory; >> they can't be directly hotpluggged. (I bet that in practice >> there are leaks though if you did do that, and we don't have >> any devices that do so today.) Can we assume pflash aren't hotpluggable until someone find an use case and is willing to work on it? I can't find any practical example. >>> If you guys prefer not flushing from pflash ->realize() then I'll drop >>> it. >> >> I think it makes more sense not to do a flush in the realize method. > > Okay, I'll resend this patch. Thanks! Phil.
diff --git a/hw/block/pflash_cfi01.c b/hw/block/pflash_cfi01.c index bffb4c40e7..5301c11c18 100644 --- a/hw/block/pflash_cfi01.c +++ b/hw/block/pflash_cfi01.c @@ -446,6 +446,7 @@ static inline void pflash_data_write(pflash_t *pfl, hwaddr offset, break; } + memory_region_flush_rom_device(&pfl->mem, offset, width); } static void pflash_write(pflash_t *pfl, hwaddr offset, @@ -482,6 +483,8 @@ static void pflash_write(pflash_t *pfl, hwaddr offset, if (!pfl->ro) { memset(p + offset, 0xff, pfl->sector_len); pflash_update(pfl, offset, pfl->sector_len); + memory_region_flush_rom_device(&pfl->mem, offset, + pfl->sector_len); } else { pfl->status |= 0x20; /* Block erase error */ } @@ -763,6 +766,8 @@ static void pflash_cfi01_realize(DeviceState *dev, Error **errp) error_setg(errp, "failed to read the initial flash content"); return; } + + memory_region_flush_rom_device(&pfl->mem, 0, total_len); } /* Default to devices being used at their maximum device width. This was diff --git a/hw/block/pflash_cfi02.c b/hw/block/pflash_cfi02.c index 0f8b7b8c7b..d04572eca4 100644 --- a/hw/block/pflash_cfi02.c +++ b/hw/block/pflash_cfi02.c @@ -378,6 +378,8 @@ static void pflash_write (pflash_t *pfl, hwaddr offset, pflash_update(pfl, offset, 4); break; } + + memory_region_flush_rom_device(&pfl->orig_mem, offset, width); } pfl->status = 0x00 | ~(value & 0x80); /* Let's pretend write is immediate */ @@ -426,6 +428,8 @@ static void pflash_write (pflash_t *pfl, hwaddr offset, if (!pfl->ro) { memset(pfl->storage, 0xFF, pfl->chip_len); pflash_update(pfl, 0, pfl->chip_len); + memory_region_flush_rom_device(&pfl->orig_mem, 0, + pfl->chip_len); } pfl->status = 0x00; /* Let's wait 5 seconds before chip erase is done */ @@ -441,6 +445,8 @@ static void pflash_write (pflash_t *pfl, hwaddr offset, if (!pfl->ro) { memset(p + offset, 0xFF, pfl->sector_len); pflash_update(pfl, offset, pfl->sector_len); + memory_region_flush_rom_device(&pfl->orig_mem, offset, + pfl->sector_len); } pfl->status = 0x00; /* Let's wait 1/2 second before sector erase is done */ @@ -590,6 +596,8 @@ static void pflash_cfi02_realize(DeviceState *dev, Error **errp) error_setg(errp, "failed to read the initial flash content"); return; } + + memory_region_flush_rom_device(&pfl->orig_mem, 0, chip_len); } pflash_setup_mappings(pfl);
pflash devices should mark the memory region dirty and invalidate TBs after directly writing to the RAM backing the ROM device. Note that pflash_cfi01_get_memory() is used by several machine types to populate ROM contents directly. Callers are untouched by this patch because they only modify memory before the guest is started. Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> --- hw/block/pflash_cfi01.c | 5 +++++ hw/block/pflash_cfi02.c | 8 ++++++++ 2 files changed, 13 insertions(+)