| Message ID | 20190813103548.2008-1-weh@microsoft.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | video: hyperv: hyperv_fb: Support deferred IO for Hyper-V frame buffer driver | expand |
From: Wei Hu <weh@microsoft.com> Sent: Tuesday, August 13, 2019 3:37 AM > > Without deferred IO support, hyperv_fb driver informs the host to refresh > the entire guest frame buffer at fixed rate, e.g. at 20Hz, no matter there > is screen update or not. This patch supports defered IO for screens in s/defered/deferred/ > graphic mode and also enables the framme buffer on-demand refresh. The s/graphic/graphics/ s/framme/frame/ > highest refresh rate is still set at 20Hz. > > Due to limitation on Hyper-V host, we keep a shadow copy of frame buffer I think it might be worthwhile to explain exactly what the issue is so that there's a record kept. > in the guest. This means one more copy of the dirty rectangle inside > guest when doing the on-demand refresh. This can be optimized in the > future with help from host. For now the host performance gain from deferred > IO outweighs the shadow copy impact in the guest. > > Signed-off-by: Wei Hu <weh@microsoft.com> > --- > drivers/video/fbdev/Kconfig | 1 + > drivers/video/fbdev/hyperv_fb.c | 217 +++++++++++++++++++++++++++++--- > 2 files changed, 198 insertions(+), 20 deletions(-) > > diff --git a/drivers/video/fbdev/Kconfig b/drivers/video/fbdev/Kconfig > index 1b2f5f31fb6f..e781f89a1824 100644 > --- a/drivers/video/fbdev/Kconfig > +++ b/drivers/video/fbdev/Kconfig > @@ -2241,6 +2241,7 @@ config FB_HYPERV > select FB_CFB_FILLRECT > select FB_CFB_COPYAREA > select FB_CFB_IMAGEBLIT > + select FB_DEFERRED_IO > help > This framebuffer driver supports Microsoft Hyper-V Synthetic Video. > > diff --git a/drivers/video/fbdev/hyperv_fb.c b/drivers/video/fbdev/hyperv_fb.c > index 1042f3311fa2..85198a6ea8e7 100644 > --- a/drivers/video/fbdev/hyperv_fb.c > +++ b/drivers/video/fbdev/hyperv_fb.c > @@ -233,6 +233,7 @@ struct synthvid_msg { > #define RING_BUFSIZE (256 * 1024) > #define VSP_TIMEOUT (10 * HZ) > #define HVFB_UPDATE_DELAY (HZ / 20) > +#define HVFB_ONDEMAND_THROTTLE (HZ / 20) > > struct hvfb_par { > struct fb_info *info; > @@ -252,6 +253,17 @@ struct hvfb_par { > bool synchronous_fb; > > struct notifier_block hvfb_panic_nb; > + > + /* Memory for deferred IO and frame buffer itself */ > + unsigned char *dio_vp; > + unsigned char *mmio_vp; > + unsigned long mmio_pp; > + spinlock_t docopy_lock; /* Lock to protect memory copy */ > + > + /* Dirty rectangle, protected by delayed_refresh_lock */ > + int x1, y1, x2, y2; > + bool delayed_refresh; > + spinlock_t delayed_refresh_lock; > }; > > static uint screen_width = HVFB_WIDTH; > @@ -260,6 +272,7 @@ static uint screen_width_max = HVFB_WIDTH; > static uint screen_height_max = HVFB_HEIGHT; > static uint screen_depth; > static uint screen_fb_size; > +static uint dio_fb_size; /* FB size for deferred IO */ > > /* Send message to Hyper-V host */ > static inline int synthvid_send(struct hv_device *hdev, > @@ -346,28 +359,88 @@ static int synthvid_send_ptr(struct hv_device *hdev) > } > > /* Send updated screen area (dirty rectangle) location to host */ > -static int synthvid_update(struct fb_info *info) > +static int > +synthvid_update(struct fb_info *info, int x1, int y1, int x2, int y2) > { > struct hv_device *hdev = device_to_hv_device(info->device); > struct synthvid_msg msg; > > memset(&msg, 0, sizeof(struct synthvid_msg)); > + if (x2 == INT_MAX) > + x2 = info->var.xres; > + if (y2 == INT_MAX) > + y2 = info->var.yres; > > msg.vid_hdr.type = SYNTHVID_DIRT; > msg.vid_hdr.size = sizeof(struct synthvid_msg_hdr) + > sizeof(struct synthvid_dirt); > msg.dirt.video_output = 0; > msg.dirt.dirt_count = 1; > - msg.dirt.rect[0].x1 = 0; > - msg.dirt.rect[0].y1 = 0; > - msg.dirt.rect[0].x2 = info->var.xres; > - msg.dirt.rect[0].y2 = info->var.yres; > + msg.dirt.rect[0].x1 = (x1 < 0 || x1 > x2) ? 0 : x1; > + msg.dirt.rect[0].y1 = (y2 < 0 || y1 > y2) ? 0 : y1; This should be: msg.dirt.rect[0].y1 = (y1 < 0 || y1 > y2) ? 0 : y1; Also, throughout the code, I don't think there are any places where x or y coordinate values are ever negative. INT_MAX or 0 is used as the sentinel value indicating "not set". So can all the tests for less than 0 now be eliminated, both in this function and in other functions? > + msg.dirt.rect[0].x2 = > + (x2 < x1 || x2 > info->var.xres) ? info->var.xres : x2; > + msg.dirt.rect[0].y2 = > + (y2 < y1 || y2 > info->var.yres) ? info->var.yres : y2; How exactly is the dirty rectangle specified to Hyper-V? Suppose the frame buffer resolution is 100x200. If you want to specify the entire rectangle, the first coordinate is (0, 0). But what is the second coordinate? Should it be (99, 199) or (100, 200)? The above code (and original code) implies it should specified as (100, 200), which is actually a point outside the maximum resolution, which is counter-intuitive and makes me wonder if the code is correct. > > synthvid_send(hdev, &msg); > > return 0; > } > > +static void hvfb_docopy(struct hvfb_par *par, > + unsigned long offset, > + unsigned long size) > +{ > + if (!par || !par->mmio_vp || !par->dio_vp || !par->fb_ready || > + size == 0 || offset >= dio_fb_size) > + return; > + > + if (offset + size > dio_fb_size) > + size = dio_fb_size - offset; > + > + memcpy(par->mmio_vp + offset, par->dio_vp + offset, size); > +} > + > +/* Deferred IO callback */ > +static void synthvid_deferred_io(struct fb_info *p, > + struct list_head *pagelist) > +{ > + struct hvfb_par *par = p->par; > + struct page *page; > + unsigned long start, end; > + int y1, y2, miny, maxy; > + unsigned long flags; > + > + miny = INT_MAX; > + maxy = 0; > + > + list_for_each_entry(page, pagelist, lru) { > + start = page->index << PAGE_SHIFT; > + end = start + PAGE_SIZE - 1; > + y1 = start / p->fix.line_length; > + y2 = end / p->fix.line_length; The above division rounds down because any remainder is discarded. I wondered whether rounding down is correct, which got me to thinking about how the dirty rectangle is specified. Is y2 the index of the last dirty row? If so, that's not consistent with the code in synthvid_update(), which might choose var.yres as y2, and that's the index of a row outside of the frame buffer. > + if (y2 > p->var.yres) > + y2 = p->var.yres; > + miny = min_t(int, miny, y1); > + maxy = max_t(int, maxy, y2); > + > + /* Copy from dio space to mmio address */ > + if (par->fb_ready) { > + spin_lock_irqsave(&par->docopy_lock, flags); > + hvfb_docopy(par, start, PAGE_SIZE); > + spin_unlock_irqrestore(&par->docopy_lock, flags); > + } > + } > + > + if (par->fb_ready) > + synthvid_update(p, 0, miny, p->var.xres, maxy); > +} > + > +static struct fb_deferred_io synthvid_defio = { > + .delay = HZ / 20, > + .deferred_io = synthvid_deferred_io, > +}; > > /* > * Actions on received messages from host: > @@ -597,7 +670,7 @@ static int synthvid_send_config(struct hv_device *hdev) > msg->vid_hdr.type = SYNTHVID_VRAM_LOCATION; > msg->vid_hdr.size = sizeof(struct synthvid_msg_hdr) + > sizeof(struct synthvid_vram_location); > - msg->vram.user_ctx = msg->vram.vram_gpa = info->fix.smem_start; > + msg->vram.user_ctx = msg->vram.vram_gpa = par->mmio_pp; > msg->vram.is_vram_gpa_specified = 1; > synthvid_send(hdev, msg); > > @@ -607,7 +680,7 @@ static int synthvid_send_config(struct hv_device *hdev) > ret = -ETIMEDOUT; > goto out; > } > - if (msg->vram_ack.user_ctx != info->fix.smem_start) { > + if (msg->vram_ack.user_ctx != par->mmio_pp) { > pr_err("Unable to set VRAM location\n"); > ret = -ENODEV; > goto out; > @@ -624,19 +697,85 @@ static int synthvid_send_config(struct hv_device *hdev) > > /* > * Delayed work callback: > - * It is called at HVFB_UPDATE_DELAY or longer time interval to process > - * screen updates. It is re-scheduled if further update is necessary. > + * It is scheduled to call whenever update request is received and it has > + * not been called in last HVFB_ONDEMAND_THROTTLE time interval. > */ > static void hvfb_update_work(struct work_struct *w) > { > struct hvfb_par *par = container_of(w, struct hvfb_par, dwork.work); > struct fb_info *info = par->info; > + unsigned long flags; > + int x1, x2, y1, y2; > + int j; > + > + x1 = y1 = 0; > + x2 = y2 = INT_MAX; The above two lines seem superfluous since all four values are unconditionally set below when storing the dirty rectangle to local variables. > + > + spin_lock_irqsave(&par->delayed_refresh_lock, flags); > + /* Reset the request flag */ > + par->delayed_refresh = false; > + > + /* Store the dirty rectangle to local variables */ > + x1 = par->x1; > + x2 = par->x2; > + y1 = par->y1; > + y2 = par->y2; > + > + /* Clear dirty rectangle */ > + par->x1 = par->y1 = INT_MAX; > + par->x2 = par->y2 = 0; > > + spin_unlock_irqrestore(&par->delayed_refresh_lock, flags); > + > + if (x1 < 0 || x1 > info->var.xres || x2 < 0 || > + x2 > info->var.xres || y1 < 0 || y1 > info->var.yres || > + y2 < 0 || y2 > info->var.yres) > + return; > + > + /* Copy the dirty rectangle to frame buffer memory */ > + spin_lock_irqsave(&par->docopy_lock, flags); > + for (j = y1; j <= y2 && x1 < x2; j++) { x1 < x2 doesn't seem to be needed as a loop control test as neither value is changed in the loop. > + if (j == info->var.yres) > + break; > + hvfb_docopy(par, > + j * info->fix.line_length + > + (x1 * screen_depth / 8), > + (x2 - x1 + 1) * screen_depth / 8); Whether the +1 is needed above gets back to the question I raised earlier about how to interpret the coordinates -- whether the (x2, y2) coordinate is just outside the dirty rectangle or just inside the dirty rectangle. Most of the code seems to treat it as being just outside the dirty rectangle, in which case the +1 should not be used. > + } > + spin_unlock_irqrestore(&par->docopy_lock, flags); > + > + /* Refresh */ > if (par->fb_ready) > - synthvid_update(info); > + synthvid_update(info, x1, y1, x2, y2); > +} > + > +/* > + * Control the on-demand refresh frequency. It schedules a delayed > + * screen update if it has not yet. > + */ > +static void hvfb_ondemand_refresh_throttle(struct hvfb_par *par, > + int x1, int y1, int w, int h) > +{ > + unsigned long flags; > + int x2 = x1 + w; > + int y2 = y1 + h; > + > + spin_lock_irqsave(&par->delayed_refresh_lock, flags); > + > + /* Merge dirty rectangle */ > + par->x1 = min_t(int, par->x1, x1); > + par->y1 = min_t(int, par->y1, y1); > + par->x2 = max_t(int, par->x2, x2); > + par->y2 = max_t(int, par->y2, y2); > + > + /* Schedule a delayed screen update if not yet */ > + if (par->delayed_refresh == false) { > + schedule_delayed_work(&par->dwork, > + HVFB_ONDEMAND_THROTTLE); > + par->delayed_refresh = true; > + } > > - if (par->update) > - schedule_delayed_work(&par->dwork, HVFB_UPDATE_DELAY); > + spin_unlock_irqrestore(&par->delayed_refresh_lock, flags); > } > > static int hvfb_on_panic(struct notifier_block *nb, > @@ -648,7 +787,8 @@ static int hvfb_on_panic(struct notifier_block *nb, > par = container_of(nb, struct hvfb_par, hvfb_panic_nb); > par->synchronous_fb = true; > info = par->info; > - synthvid_update(info); > + hvfb_docopy(par, 0, dio_fb_size); > + synthvid_update(info, 0, 0, INT_MAX, INT_MAX); > > return NOTIFY_DONE; > } > @@ -709,7 +849,10 @@ static void hvfb_cfb_fillrect(struct fb_info *p, > > cfb_fillrect(p, rect); > if (par->synchronous_fb) > - synthvid_update(p); > + synthvid_update(p, 0, 0, INT_MAX, INT_MAX); > + else > + hvfb_ondemand_refresh_throttle(par, rect->dx, rect->dy, > + rect->width, rect->height); > } > > static void hvfb_cfb_copyarea(struct fb_info *p, > @@ -719,7 +862,10 @@ static void hvfb_cfb_copyarea(struct fb_info *p, > > cfb_copyarea(p, area); > if (par->synchronous_fb) > - synthvid_update(p); > + synthvid_update(p, 0, 0, INT_MAX, INT_MAX); > + else > + hvfb_ondemand_refresh_throttle(par, area->dx, area->dy, > + area->width, area->height); > } > > static void hvfb_cfb_imageblit(struct fb_info *p, > @@ -729,7 +875,10 @@ static void hvfb_cfb_imageblit(struct fb_info *p, > > cfb_imageblit(p, image); > if (par->synchronous_fb) > - synthvid_update(p); > + synthvid_update(p, 0, 0, INT_MAX, INT_MAX); > + else > + hvfb_ondemand_refresh_throttle(par, image->dx, image->dy, > + image->width, image->height); > } > > static struct fb_ops hvfb_ops = { > @@ -788,6 +937,9 @@ static int hvfb_getmem(struct hv_device *hdev, struct fb_info > *info) > resource_size_t pot_start, pot_end; > int ret; > > + dio_fb_size = > + screen_width * screen_height * screen_depth / 8; > + > if (gen2vm) { > pot_start = 0; > pot_end = -1; > @@ -822,9 +974,15 @@ static int hvfb_getmem(struct hv_device *hdev, struct fb_info > *info) > if (!fb_virt) > goto err2; > > + /* Allocate memory for deferred IO */ > + par->dio_vp = vzalloc(((dio_fb_size >> PAGE_SHIFT) + 1) > + << PAGE_SHIFT); I'd suggest using the round_up() function so that what you are doing is explicit. > + if (par->dio_vp == NULL) > + goto err3; > + > info->apertures = alloc_apertures(1); > if (!info->apertures) > - goto err3; > + goto err4; > > if (gen2vm) { > info->apertures->ranges[0].base = screen_info.lfb_base; > @@ -836,16 +994,23 @@ static int hvfb_getmem(struct hv_device *hdev, struct fb_info > *info) > info->apertures->ranges[0].size = pci_resource_len(pdev, 0); > } > > + /* Physical address of FB device */ > + par->mmio_pp = par->mem->start; > + /* Virtual address of FB device */ > + par->mmio_vp = (unsigned char *) fb_virt; > + > info->fix.smem_start = par->mem->start; > - info->fix.smem_len = screen_fb_size; > - info->screen_base = fb_virt; > - info->screen_size = screen_fb_size; > + info->fix.smem_len = dio_fb_size; > + info->screen_base = par->dio_vp; > + info->screen_size = dio_fb_size; > > if (!gen2vm) > pci_dev_put(pdev); > > return 0; > > +err4: > + vfree(par->dio_vp); > err3: > iounmap(fb_virt); > err2: > @@ -863,6 +1028,7 @@ static void hvfb_putmem(struct fb_info *info) > { > struct hvfb_par *par = info->par; > > + vfree(par->dio_vp); > iounmap(info->screen_base); > vmbus_free_mmio(par->mem->start, screen_fb_size); > par->mem = NULL; > @@ -888,6 +1054,12 @@ static int hvfb_probe(struct hv_device *hdev, > init_completion(&par->wait); > INIT_DELAYED_WORK(&par->dwork, hvfb_update_work); > > + par->delayed_refresh = false; > + spin_lock_init(&par->delayed_refresh_lock); > + spin_lock_init(&par->docopy_lock); > + par->x1 = par->y1 = INT_MAX; > + par->x2 = par->y2 = 0; > + > /* Connect to VSP */ > hv_set_drvdata(hdev, info); > ret = synthvid_connect_vsp(hdev); > @@ -939,6 +1111,10 @@ static int hvfb_probe(struct hv_device *hdev, > info->fbops = &hvfb_ops; > info->pseudo_palette = par->pseudo_palette; > > + /* Initialize deferred IO */ > + info->fbdefio = &synthvid_defio; > + fb_deferred_io_init(info); > + > /* Send config to host */ > ret = synthvid_send_config(hdev); > if (ret) > @@ -960,6 +1136,7 @@ static int hvfb_probe(struct hv_device *hdev, > return 0; > > error: > + fb_deferred_io_cleanup(info); > hvfb_putmem(info); > error2: > vmbus_close(hdev->channel); > -- > 2.20.1
Thanks Michael. See my reply inline to some of your comments. > -----Original Message----- > From: Michael Kelley <mikelley@microsoft.com> > Sent: Monday, August 19, 2019 6:41 AM > To: Wei Hu <weh@microsoft.com>; rdunlap@infradead.org; shc_work@mail.ru; > > - msg.dirt.rect[0].x1 = 0; > > - msg.dirt.rect[0].y1 = 0; > > - msg.dirt.rect[0].x2 = info->var.xres; > > - msg.dirt.rect[0].y2 = info->var.yres; > > + msg.dirt.rect[0].x1 = (x1 < 0 || x1 > x2) ? 0 : x1; > > + msg.dirt.rect[0].y1 = (y2 < 0 || y1 > y2) ? 0 : y1; > > This should be: > > msg.dirt.rect[0].y1 = (y1 < 0 || y1 > y2) ? 0 : y1; > > Also, throughout the code, I don't think there are any places where > x or y coordinate values are ever negative. INT_MAX or 0 is used as the > sentinel value indicating "not set". So can all the tests for less than 0 > now be eliminated, both in this function and in other functions? > > > + msg.dirt.rect[0].x2 = > > + (x2 < x1 || x2 > info->var.xres) ? info->var.xres : x2; > > + msg.dirt.rect[0].y2 = > > + (y2 < y1 || y2 > info->var.yres) ? info->var.yres : y2; > > How exactly is the dirty rectangle specified to Hyper-V? Suppose the frame > buffer resolution is 100x200. If you want to specify the entire rectangle, the > first coordinate is (0, 0). But what is the second coordinate? Should it be > (99, 199) or (100, 200)? The above code (and original code) implies it > should specified as (100, 200), which is actually a point outside the > maximum resolution, which is counter-intuitive and makes me wonder > if the code is correct. > [Wei Hu] The current code treat the entire framebuffer rectangle as (0,0) -> (var.xres, var.yres). Every time it sends refresh request, these are two points sent to host and host seems accept it. See the above (x1, y1) and (x2, y2) in the deleted lines. So in your example the second coordinate is (100, 200). > > +/* Deferred IO callback */ > > +static void synthvid_deferred_io(struct fb_info *p, > > + struct list_head *pagelist) > > +{ > > + struct hvfb_par *par = p->par; > > + struct page *page; > > + unsigned long start, end; > > + int y1, y2, miny, maxy; > > + unsigned long flags; > > + > > + miny = INT_MAX; > > + maxy = 0; > > + > > + list_for_each_entry(page, pagelist, lru) { > > + start = page->index << PAGE_SHIFT; > > + end = start + PAGE_SIZE - 1; > > + y1 = start / p->fix.line_length; > > + y2 = end / p->fix.line_length; > > The above division rounds down because any remainder is discarded. I > wondered whether rounding down is correct, which got me to thinking > about how the dirty rectangle is specified. Is y2 the index of the last > dirty row? If so, that's not consistent with the code in synthvid_update(), > which might choose var.yres as y2, and that's the index of a row outside > of the frame buffer. > [Wei Hu] In this place we try to figure out and merge all the faulted pages into one big dirty rectangle. A page in memory represents one or multiple lines in frame buffer. For example, one faulted page could represent all the linear pixels from (x, y) to (x-1, y+1). In this case we just form the dirty rectangle as (0, y) -> (var.xres, y+1). Also keep in mind we need to merge multiple pages. That's why in the end the dirty rectangle is (0, miny) -> (var.xres, maxy). > > + if (y2 > p->var.yres) > > + y2 = p->var.yres; > > + miny = min_t(int, miny, y1); > > + maxy = max_t(int, maxy, y2); > > + > > + /* Copy from dio space to mmio address */ > > + if (par->fb_ready) { > > + spin_lock_irqsave(&par->docopy_lock, flags); > > + hvfb_docopy(par, start, PAGE_SIZE); > > + spin_unlock_irqrestore(&par->docopy_lock, flags); > > + } > > + } > > + > > + if (par->fb_ready) > > + synthvid_update(p, 0, miny, p->var.xres, maxy); > > +} > > + > > + if (j == info->var.yres) > > + break; > > + hvfb_docopy(par, > > + j * info->fix.line_length + > > + (x1 * screen_depth / 8), > > + (x2 - x1 + 1) * screen_depth / 8); > > Whether the +1 is needed above gets back to the question I > raised earlier about how to interpret the coordinates -- whether > the (x2, y2) coordinate is just outside the dirty rectangle or > just inside the dirty rectangle. Most of the code seems to treat > it as being just outside the dirty rectangle, in which case the +1 > should not be used. > [Wei Hu] This dirty rectangle is not from page fault, but rather from frame buffer framework when the screen is in text mode. I am not 100% sure if the dirty rectangle given from kernel includes on extra line outside or not. Here I just play it safe by copying one extra line in the worst case. Suppose dirty rectangle only contain one pixel, for example (0,0) is the only pixel changed in the entire frame buffer. If kernel sends me dirty rectangle as (0, 0) -> (0, 0), the above function works correctly. If the kernel sends (0, 0) -> (1, 1), then the above function just copies one extra row and one extra column, which should also be fine. The hvfb_docopy() takes care of the edge cases. Thanks, Wei
From: Wei Hu <weh@microsoft.com> Sent: Wednesday, August 21, 2019 4:59 AM > > > From: Michael Kelley <mikelley@microsoft.com> > > Sent: Monday, August 19, 2019 6:41 AM > > To: Wei Hu <weh@microsoft.com>; rdunlap@infradead.org; shc_work@mail.ru; > > > > - msg.dirt.rect[0].x1 = 0; > > > - msg.dirt.rect[0].y1 = 0; > > > - msg.dirt.rect[0].x2 = info->var.xres; > > > - msg.dirt.rect[0].y2 = info->var.yres; > > > + msg.dirt.rect[0].x1 = (x1 < 0 || x1 > x2) ? 0 : x1; > > > + msg.dirt.rect[0].y1 = (y2 < 0 || y1 > y2) ? 0 : y1; > > > > This should be: > > > > msg.dirt.rect[0].y1 = (y1 < 0 || y1 > y2) ? 0 : y1; > > > > Also, throughout the code, I don't think there are any places where > > x or y coordinate values are ever negative. INT_MAX or 0 is used as the > > sentinel value indicating "not set". So can all the tests for less than 0 > > now be eliminated, both in this function and in other functions? > > > > > + msg.dirt.rect[0].x2 = > > > + (x2 < x1 || x2 > info->var.xres) ? info->var.xres : x2; > > > + msg.dirt.rect[0].y2 = > > > + (y2 < y1 || y2 > info->var.yres) ? info->var.yres : y2; > > > > How exactly is the dirty rectangle specified to Hyper-V? Suppose the frame > > buffer resolution is 100x200. If you want to specify the entire rectangle, the > > first coordinate is (0, 0). But what is the second coordinate? Should it be > > (99, 199) or (100, 200)? The above code (and original code) implies it > > should specified as (100, 200), which is actually a point outside the > > maximum resolution, which is counter-intuitive and makes me wonder > > if the code is correct. > > > [Wei Hu] > The current code treat the entire framebuffer rectangle as (0,0) -> (var.xres, var.yres). > Every time it sends refresh request, these are two points sent to host and host > seems accept it. See the above (x1, y1) and (x2, y2) in the deleted lines. > > So in your example the second coordinate is (100, 200). OK, agreed. I ran some experiments and confirmed that this is indeed the Hyper-V behavior. > > > > > +/* Deferred IO callback */ > > > +static void synthvid_deferred_io(struct fb_info *p, > > > + struct list_head *pagelist) > > > +{ > > > + struct hvfb_par *par = p->par; > > > + struct page *page; > > > + unsigned long start, end; > > > + int y1, y2, miny, maxy; > > > + unsigned long flags; > > > + > > > + miny = INT_MAX; > > > + maxy = 0; > > > + > > > + list_for_each_entry(page, pagelist, lru) { > > > + start = page->index << PAGE_SHIFT; > > > + end = start + PAGE_SIZE - 1; > > > + y1 = start / p->fix.line_length; > > > + y2 = end / p->fix.line_length; > > > > The above division rounds down because any remainder is discarded. I > > wondered whether rounding down is correct, which got me to thinking > > about how the dirty rectangle is specified. Is y2 the index of the last > > dirty row? If so, that's not consistent with the code in synthvid_update(), > > which might choose var.yres as y2, and that's the index of a row outside > > of the frame buffer. > > > [Wei Hu] > In this place we try to figure out and merge all the faulted pages into one > big dirty rectangle. A page in memory represents one or multiple lines in > frame buffer. For example, one faulted page could represent all the linear > pixels from (x, y) to (x-1, y+1). In this case we just form the dirty rectangle > as (0, y) -> (var.xres, y+1). Also keep in mind we need to merge multiple > pages. That's why in the end the dirty rectangle is (0, miny) -> (var.xres, maxy). Let me give an example of where I think the new code doesn't work. Suppose the frame buffer resolution is 1024x768. With 4 bytes per pixel, each row is 4096 bytes, or exactly one page. So each page contains exactly one row of pixels. For simplicity in my example, let's look at the case when this function is called with only one dirty page. The calculation of y1 will identify the row that is dirty. The calculation of y2 will identify the same row. So y1 will equal y2, and miny will equal maxy. Then when synthvid_update() is called, Hyper-V will interpret the parameters as no rows needing to be updated. In a more complex case where the pagelist contains multiple dirty pages, maxy also ends up one less than it needs to be. I think passing 'maxy + 1' instead of 'maxy' to synthvid_update() will solve the problem. It certainly warrants a comment that the calculation of maxy is "inclusive", while synthvid_update() expects its parameters to be "exclusive" per Hyper-V's expectations. There's also another interesting situation. Suppose the resolution and page size is such that a page contains multiple rows. If the last page of the frame buffer is dirty, this routine could calculate a y2 value identifying a "phantom" row that is off the end of the frame buffer -- i.e., that is bigger than yres. You have synthvid_send() handling that case by clamping the y2 value to yres, but it might be worth a comment here acknowledging the situation and how it is handled. I did a test, and it appears that Hyper-V does its own clamping of the values passed in, but we should not take a dependency on how Hyper-V handles incorrect inputs. > > > > > + if (y2 > p->var.yres) > > > + y2 = p->var.yres; > > > + miny = min_t(int, miny, y1); > > > + maxy = max_t(int, maxy, y2); > > > + > > > + /* Copy from dio space to mmio address */ > > > + if (par->fb_ready) { > > > + spin_lock_irqsave(&par->docopy_lock, flags); > > > + hvfb_docopy(par, start, PAGE_SIZE); > > > + spin_unlock_irqrestore(&par->docopy_lock, flags); > > > + } > > > + } > > > + > > > + if (par->fb_ready) > > > + synthvid_update(p, 0, miny, p->var.xres, maxy); > > > +} > > > > > > > + > > > + /* Copy the dirty rectangle to frame buffer memory */ > > > + spin_lock_irqsave(&par->docopy_lock, flags); > > > + for (j = y1; j <= y2 && x1 < x2; j++) { > > > + if (j == info->var.yres) > > > + break; > > > + hvfb_docopy(par, > > > + j * info->fix.line_length + > > > + (x1 * screen_depth / 8), > > > + (x2 - x1 + 1) * screen_depth / 8); > > > > Whether the +1 is needed above gets back to the question I > > raised earlier about how to interpret the coordinates -- whether > > the (x2, y2) coordinate is just outside the dirty rectangle or > > just inside the dirty rectangle. Most of the code seems to treat > > it as being just outside the dirty rectangle, in which case the +1 > > should not be used. > > > [Wei Hu] > This dirty rectangle is not from page fault, but rather from frame buffer > framework when the screen is in text mode. I am not 100% sure if the dirty > rectangle given from kernel includes on extra line outside or not. Here I > just play it safe by copying one extra line in the worst case. Got it. I was incorrectly conflating the two ways the frame buffer can get updated. I looked at the how x2 and y2 of the dirty rectangle get set in this case. It looks to me like it's always calculated as x1 + width, and y1 + height in hvfb_ondemand_refresh_throttle(). If that's correct, then the dirty rectangle coordinates are "exclusive", which is what Hyper-V wants. And indeed the call to synthvid_update() later in this function assumes the "exclusive" format. Also, the "j == info->var.yres" test is correct only if the y2 value is in the "exclusive" format. With that being the case, the "for" loop control above should have j < y2 instead of j <= y2, as we don't need to copy the y2 row. And the +1 isn't needed in the arguments to hvfb_docopy(). I really don't like the idea of copying an extra row, or an extra byte in a row, "just in case". Michael > > Suppose dirty rectangle only contain one pixel, for example (0,0) is the only > pixel changed in the entire frame buffer. If kernel sends me dirty rectangle as > (0, 0) -> (0, 0), the above function works correctly. If the kernel sends > (0, 0) -> (1, 1), then the above function just copies one extra row and one extra > column, which should also be fine. The hvfb_docopy() takes care of the > edge cases. > > Thanks, > Wei
diff --git a/drivers/video/fbdev/Kconfig b/drivers/video/fbdev/Kconfig index 1b2f5f31fb6f..e781f89a1824 100644 --- a/drivers/video/fbdev/Kconfig +++ b/drivers/video/fbdev/Kconfig @@ -2241,6 +2241,7 @@ config FB_HYPERV select FB_CFB_FILLRECT select FB_CFB_COPYAREA select FB_CFB_IMAGEBLIT + select FB_DEFERRED_IO help This framebuffer driver supports Microsoft Hyper-V Synthetic Video. diff --git a/drivers/video/fbdev/hyperv_fb.c b/drivers/video/fbdev/hyperv_fb.c index 1042f3311fa2..85198a6ea8e7 100644 --- a/drivers/video/fbdev/hyperv_fb.c +++ b/drivers/video/fbdev/hyperv_fb.c @@ -233,6 +233,7 @@ struct synthvid_msg { #define RING_BUFSIZE (256 * 1024) #define VSP_TIMEOUT (10 * HZ) #define HVFB_UPDATE_DELAY (HZ / 20) +#define HVFB_ONDEMAND_THROTTLE (HZ / 20) struct hvfb_par { struct fb_info *info; @@ -252,6 +253,17 @@ struct hvfb_par { bool synchronous_fb; struct notifier_block hvfb_panic_nb; + + /* Memory for deferred IO and frame buffer itself */ + unsigned char *dio_vp; + unsigned char *mmio_vp; + unsigned long mmio_pp; + spinlock_t docopy_lock; /* Lock to protect memory copy */ + + /* Dirty rectangle, protected by delayed_refresh_lock */ + int x1, y1, x2, y2; + bool delayed_refresh; + spinlock_t delayed_refresh_lock; }; static uint screen_width = HVFB_WIDTH; @@ -260,6 +272,7 @@ static uint screen_width_max = HVFB_WIDTH; static uint screen_height_max = HVFB_HEIGHT; static uint screen_depth; static uint screen_fb_size; +static uint dio_fb_size; /* FB size for deferred IO */ /* Send message to Hyper-V host */ static inline int synthvid_send(struct hv_device *hdev, @@ -346,28 +359,88 @@ static int synthvid_send_ptr(struct hv_device *hdev) } /* Send updated screen area (dirty rectangle) location to host */ -static int synthvid_update(struct fb_info *info) +static int +synthvid_update(struct fb_info *info, int x1, int y1, int x2, int y2) { struct hv_device *hdev = device_to_hv_device(info->device); struct synthvid_msg msg; memset(&msg, 0, sizeof(struct synthvid_msg)); + if (x2 == INT_MAX) + x2 = info->var.xres; + if (y2 == INT_MAX) + y2 = info->var.yres; msg.vid_hdr.type = SYNTHVID_DIRT; msg.vid_hdr.size = sizeof(struct synthvid_msg_hdr) + sizeof(struct synthvid_dirt); msg.dirt.video_output = 0; msg.dirt.dirt_count = 1; - msg.dirt.rect[0].x1 = 0; - msg.dirt.rect[0].y1 = 0; - msg.dirt.rect[0].x2 = info->var.xres; - msg.dirt.rect[0].y2 = info->var.yres; + msg.dirt.rect[0].x1 = (x1 < 0 || x1 > x2) ? 0 : x1; + msg.dirt.rect[0].y1 = (y2 < 0 || y1 > y2) ? 0 : y1; + msg.dirt.rect[0].x2 = + (x2 < x1 || x2 > info->var.xres) ? info->var.xres : x2; + msg.dirt.rect[0].y2 = + (y2 < y1 || y2 > info->var.yres) ? info->var.yres : y2; synthvid_send(hdev, &msg); return 0; } +static void hvfb_docopy(struct hvfb_par *par, + unsigned long offset, + unsigned long size) +{ + if (!par || !par->mmio_vp || !par->dio_vp || !par->fb_ready || + size == 0 || offset >= dio_fb_size) + return; + + if (offset + size > dio_fb_size) + size = dio_fb_size - offset; + + memcpy(par->mmio_vp + offset, par->dio_vp + offset, size); +} + +/* Deferred IO callback */ +static void synthvid_deferred_io(struct fb_info *p, + struct list_head *pagelist) +{ + struct hvfb_par *par = p->par; + struct page *page; + unsigned long start, end; + int y1, y2, miny, maxy; + unsigned long flags; + + miny = INT_MAX; + maxy = 0; + + list_for_each_entry(page, pagelist, lru) { + start = page->index << PAGE_SHIFT; + end = start + PAGE_SIZE - 1; + y1 = start / p->fix.line_length; + y2 = end / p->fix.line_length; + if (y2 > p->var.yres) + y2 = p->var.yres; + miny = min_t(int, miny, y1); + maxy = max_t(int, maxy, y2); + + /* Copy from dio space to mmio address */ + if (par->fb_ready) { + spin_lock_irqsave(&par->docopy_lock, flags); + hvfb_docopy(par, start, PAGE_SIZE); + spin_unlock_irqrestore(&par->docopy_lock, flags); + } + } + + if (par->fb_ready) + synthvid_update(p, 0, miny, p->var.xres, maxy); +} + +static struct fb_deferred_io synthvid_defio = { + .delay = HZ / 20, + .deferred_io = synthvid_deferred_io, +}; /* * Actions on received messages from host: @@ -597,7 +670,7 @@ static int synthvid_send_config(struct hv_device *hdev) msg->vid_hdr.type = SYNTHVID_VRAM_LOCATION; msg->vid_hdr.size = sizeof(struct synthvid_msg_hdr) + sizeof(struct synthvid_vram_location); - msg->vram.user_ctx = msg->vram.vram_gpa = info->fix.smem_start; + msg->vram.user_ctx = msg->vram.vram_gpa = par->mmio_pp; msg->vram.is_vram_gpa_specified = 1; synthvid_send(hdev, msg); @@ -607,7 +680,7 @@ static int synthvid_send_config(struct hv_device *hdev) ret = -ETIMEDOUT; goto out; } - if (msg->vram_ack.user_ctx != info->fix.smem_start) { + if (msg->vram_ack.user_ctx != par->mmio_pp) { pr_err("Unable to set VRAM location\n"); ret = -ENODEV; goto out; @@ -624,19 +697,85 @@ static int synthvid_send_config(struct hv_device *hdev) /* * Delayed work callback: - * It is called at HVFB_UPDATE_DELAY or longer time interval to process - * screen updates. It is re-scheduled if further update is necessary. + * It is scheduled to call whenever update request is received and it has + * not been called in last HVFB_ONDEMAND_THROTTLE time interval. */ static void hvfb_update_work(struct work_struct *w) { struct hvfb_par *par = container_of(w, struct hvfb_par, dwork.work); struct fb_info *info = par->info; + unsigned long flags; + int x1, x2, y1, y2; + int j; + + x1 = y1 = 0; + x2 = y2 = INT_MAX; + + spin_lock_irqsave(&par->delayed_refresh_lock, flags); + /* Reset the request flag */ + par->delayed_refresh = false; + + /* Store the dirty rectangle to local variables */ + x1 = par->x1; + x2 = par->x2; + y1 = par->y1; + y2 = par->y2; + + /* Clear dirty rectangle */ + par->x1 = par->y1 = INT_MAX; + par->x2 = par->y2 = 0; + spin_unlock_irqrestore(&par->delayed_refresh_lock, flags); + + if (x1 < 0 || x1 > info->var.xres || x2 < 0 || + x2 > info->var.xres || y1 < 0 || y1 > info->var.yres || + y2 < 0 || y2 > info->var.yres) + return; + + /* Copy the dirty rectangle to frame buffer memory */ + spin_lock_irqsave(&par->docopy_lock, flags); + for (j = y1; j <= y2 && x1 < x2; j++) { + if (j == info->var.yres) + break; + hvfb_docopy(par, + j * info->fix.line_length + + (x1 * screen_depth / 8), + (x2 - x1 + 1) * screen_depth / 8); + } + spin_unlock_irqrestore(&par->docopy_lock, flags); + + /* Refresh */ if (par->fb_ready) - synthvid_update(info); + synthvid_update(info, x1, y1, x2, y2); +} + +/* + * Control the on-demand refresh frequency. It schedules a delayed + * screen update if it has not yet. + */ +static void hvfb_ondemand_refresh_throttle(struct hvfb_par *par, + int x1, int y1, int w, int h) +{ + unsigned long flags; + int x2 = x1 + w; + int y2 = y1 + h; + + spin_lock_irqsave(&par->delayed_refresh_lock, flags); + + /* Merge dirty rectangle */ + par->x1 = min_t(int, par->x1, x1); + par->y1 = min_t(int, par->y1, y1); + par->x2 = max_t(int, par->x2, x2); + par->y2 = max_t(int, par->y2, y2); + + /* Schedule a delayed screen update if not yet */ + if (par->delayed_refresh == false) { + schedule_delayed_work(&par->dwork, + HVFB_ONDEMAND_THROTTLE); + par->delayed_refresh = true; + } - if (par->update) - schedule_delayed_work(&par->dwork, HVFB_UPDATE_DELAY); + spin_unlock_irqrestore(&par->delayed_refresh_lock, flags); } static int hvfb_on_panic(struct notifier_block *nb, @@ -648,7 +787,8 @@ static int hvfb_on_panic(struct notifier_block *nb, par = container_of(nb, struct hvfb_par, hvfb_panic_nb); par->synchronous_fb = true; info = par->info; - synthvid_update(info); + hvfb_docopy(par, 0, dio_fb_size); + synthvid_update(info, 0, 0, INT_MAX, INT_MAX); return NOTIFY_DONE; } @@ -709,7 +849,10 @@ static void hvfb_cfb_fillrect(struct fb_info *p, cfb_fillrect(p, rect); if (par->synchronous_fb) - synthvid_update(p); + synthvid_update(p, 0, 0, INT_MAX, INT_MAX); + else + hvfb_ondemand_refresh_throttle(par, rect->dx, rect->dy, + rect->width, rect->height); } static void hvfb_cfb_copyarea(struct fb_info *p, @@ -719,7 +862,10 @@ static void hvfb_cfb_copyarea(struct fb_info *p, cfb_copyarea(p, area); if (par->synchronous_fb) - synthvid_update(p); + synthvid_update(p, 0, 0, INT_MAX, INT_MAX); + else + hvfb_ondemand_refresh_throttle(par, area->dx, area->dy, + area->width, area->height); } static void hvfb_cfb_imageblit(struct fb_info *p, @@ -729,7 +875,10 @@ static void hvfb_cfb_imageblit(struct fb_info *p, cfb_imageblit(p, image); if (par->synchronous_fb) - synthvid_update(p); + synthvid_update(p, 0, 0, INT_MAX, INT_MAX); + else + hvfb_ondemand_refresh_throttle(par, image->dx, image->dy, + image->width, image->height); } static struct fb_ops hvfb_ops = { @@ -788,6 +937,9 @@ static int hvfb_getmem(struct hv_device *hdev, struct fb_info *info) resource_size_t pot_start, pot_end; int ret; + dio_fb_size = + screen_width * screen_height * screen_depth / 8; + if (gen2vm) { pot_start = 0; pot_end = -1; @@ -822,9 +974,15 @@ static int hvfb_getmem(struct hv_device *hdev, struct fb_info *info) if (!fb_virt) goto err2; + /* Allocate memory for deferred IO */ + par->dio_vp = vzalloc(((dio_fb_size >> PAGE_SHIFT) + 1) + << PAGE_SHIFT); + if (par->dio_vp == NULL) + goto err3; + info->apertures = alloc_apertures(1); if (!info->apertures) - goto err3; + goto err4; if (gen2vm) { info->apertures->ranges[0].base = screen_info.lfb_base; @@ -836,16 +994,23 @@ static int hvfb_getmem(struct hv_device *hdev, struct fb_info *info) info->apertures->ranges[0].size = pci_resource_len(pdev, 0); } + /* Physical address of FB device */ + par->mmio_pp = par->mem->start; + /* Virtual address of FB device */ + par->mmio_vp = (unsigned char *) fb_virt; + info->fix.smem_start = par->mem->start; - info->fix.smem_len = screen_fb_size; - info->screen_base = fb_virt; - info->screen_size = screen_fb_size; + info->fix.smem_len = dio_fb_size; + info->screen_base = par->dio_vp; + info->screen_size = dio_fb_size; if (!gen2vm) pci_dev_put(pdev); return 0; +err4: + vfree(par->dio_vp); err3: iounmap(fb_virt); err2: @@ -863,6 +1028,7 @@ static void hvfb_putmem(struct fb_info *info) { struct hvfb_par *par = info->par; + vfree(par->dio_vp); iounmap(info->screen_base); vmbus_free_mmio(par->mem->start, screen_fb_size); par->mem = NULL; @@ -888,6 +1054,12 @@ static int hvfb_probe(struct hv_device *hdev, init_completion(&par->wait); INIT_DELAYED_WORK(&par->dwork, hvfb_update_work); + par->delayed_refresh = false; + spin_lock_init(&par->delayed_refresh_lock); + spin_lock_init(&par->docopy_lock); + par->x1 = par->y1 = INT_MAX; + par->x2 = par->y2 = 0; + /* Connect to VSP */ hv_set_drvdata(hdev, info); ret = synthvid_connect_vsp(hdev); @@ -939,6 +1111,10 @@ static int hvfb_probe(struct hv_device *hdev, info->fbops = &hvfb_ops; info->pseudo_palette = par->pseudo_palette; + /* Initialize deferred IO */ + info->fbdefio = &synthvid_defio; + fb_deferred_io_init(info); + /* Send config to host */ ret = synthvid_send_config(hdev); if (ret) @@ -960,6 +1136,7 @@ static int hvfb_probe(struct hv_device *hdev, return 0; error: + fb_deferred_io_cleanup(info); hvfb_putmem(info); error2: vmbus_close(hdev->channel);
Without deferred IO support, hyperv_fb driver informs the host to refresh the entire guest frame buffer at fixed rate, e.g. at 20Hz, no matter there is screen update or not. This patch supports defered IO for screens in graphic mode and also enables the framme buffer on-demand refresh. The highest refresh rate is still set at 20Hz. Due to limitation on Hyper-V host, we keep a shadow copy of frame buffer in the guest. This means one more copy of the dirty rectangle inside guest when doing the on-demand refresh. This can be optimized in the future with help from host. For now the host performance gain from deferred IO outweighs the shadow copy impact in the guest. Signed-off-by: Wei Hu <weh@microsoft.com> --- drivers/video/fbdev/Kconfig | 1 + drivers/video/fbdev/hyperv_fb.c | 217 +++++++++++++++++++++++++++++--- 2 files changed, 198 insertions(+), 20 deletions(-)