This is a note to let you know that I've just added the patch titled
usb: dwc3: gadget: Don't setup more than requested
to my usb git tree which can be found at
git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb.git
in the usb-linus branch.
The patch will show up in the next release of the linux-next tree
(usually sometime within the next 24 hours during the week.)
The patch will hopefully also be merged in Linus's tree for the
next -rc kernel release.
If you have any questions about this process, please let me know.
>From 5d187c0454ef4c5e046a81af36882d4d515922ec Mon Sep 17 00:00:00 2001
From: Thinh Nguyen <Thinh.Nguyen(a)synopsys.com>
Date: Thu, 6 Aug 2020 19:46:23 -0700
Subject: usb: dwc3: gadget: Don't setup more than requested
The SG list may be set up with entry size more than the requested
length. Check the usb_request->length and make sure that we don't setup
the TRBs to send/receive more than requested. This case may occur when
the SG entry is allocated up to a certain minimum size, but the request
length is less than that. It can also occur when the request is reused
for a different request length.
Cc: <stable(a)vger.kernel.org> # v4.18+
Fixes: a31e63b608ff ("usb: dwc3: gadget: Correct handling of scattergather lists")
Signed-off-by: Thinh Nguyen <thinhn(a)synopsys.com>
Signed-off-by: Felipe Balbi <balbi(a)kernel.org>
---
drivers/usb/dwc3/gadget.c | 51 +++++++++++++++++++++++++++------------
1 file changed, 35 insertions(+), 16 deletions(-)
diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
index e44bfc3b5096..f9231253cbed 100644
--- a/drivers/usb/dwc3/gadget.c
+++ b/drivers/usb/dwc3/gadget.c
@@ -1054,27 +1054,25 @@ static void __dwc3_prepare_one_trb(struct dwc3_ep *dep, struct dwc3_trb *trb,
* dwc3_prepare_one_trb - setup one TRB from one request
* @dep: endpoint for which this request is prepared
* @req: dwc3_request pointer
+ * @trb_length: buffer size of the TRB
* @chain: should this TRB be chained to the next?
* @node: only for isochronous endpoints. First TRB needs different type.
*/
static void dwc3_prepare_one_trb(struct dwc3_ep *dep,
- struct dwc3_request *req, unsigned chain, unsigned node)
+ struct dwc3_request *req, unsigned int trb_length,
+ unsigned chain, unsigned node)
{
struct dwc3_trb *trb;
- unsigned int length;
dma_addr_t dma;
unsigned stream_id = req->request.stream_id;
unsigned short_not_ok = req->request.short_not_ok;
unsigned no_interrupt = req->request.no_interrupt;
unsigned is_last = req->request.is_last;
- if (req->request.num_sgs > 0) {
- length = sg_dma_len(req->start_sg);
+ if (req->request.num_sgs > 0)
dma = sg_dma_address(req->start_sg);
- } else {
- length = req->request.length;
+ else
dma = req->request.dma;
- }
trb = &dep->trb_pool[dep->trb_enqueue];
@@ -1086,7 +1084,7 @@ static void dwc3_prepare_one_trb(struct dwc3_ep *dep,
req->num_trbs++;
- __dwc3_prepare_one_trb(dep, trb, dma, length, chain, node,
+ __dwc3_prepare_one_trb(dep, trb, dma, trb_length, chain, node,
stream_id, short_not_ok, no_interrupt, is_last);
}
@@ -1096,16 +1094,27 @@ static void dwc3_prepare_one_trb_sg(struct dwc3_ep *dep,
struct scatterlist *sg = req->start_sg;
struct scatterlist *s;
int i;
-
+ unsigned int length = req->request.length;
unsigned int remaining = req->request.num_mapped_sgs
- req->num_queued_sgs;
+ /*
+ * If we resume preparing the request, then get the remaining length of
+ * the request and resume where we left off.
+ */
+ for_each_sg(req->request.sg, s, req->num_queued_sgs, i)
+ length -= sg_dma_len(s);
+
for_each_sg(sg, s, remaining, i) {
- unsigned int length = req->request.length;
unsigned int maxp = usb_endpoint_maxp(dep->endpoint.desc);
unsigned int rem = length % maxp;
+ unsigned int trb_length;
unsigned chain = true;
+ trb_length = min_t(unsigned int, length, sg_dma_len(s));
+
+ length -= trb_length;
+
/*
* IOMMU driver is coalescing the list of sgs which shares a
* page boundary into one and giving it to USB driver. With
@@ -1113,7 +1122,7 @@ static void dwc3_prepare_one_trb_sg(struct dwc3_ep *dep,
* sgs passed. So mark the chain bit to false if it isthe last
* mapped sg.
*/
- if (i == remaining - 1)
+ if ((i == remaining - 1) || !length)
chain = false;
if (rem && usb_endpoint_dir_out(dep->endpoint.desc) && !chain) {
@@ -1123,7 +1132,7 @@ static void dwc3_prepare_one_trb_sg(struct dwc3_ep *dep,
req->needs_extra_trb = true;
/* prepare normal TRB */
- dwc3_prepare_one_trb(dep, req, true, i);
+ dwc3_prepare_one_trb(dep, req, trb_length, true, i);
/* Now prepare one extra TRB to align transfer size */
trb = &dep->trb_pool[dep->trb_enqueue];
@@ -1135,7 +1144,7 @@ static void dwc3_prepare_one_trb_sg(struct dwc3_ep *dep,
req->request.no_interrupt,
req->request.is_last);
} else {
- dwc3_prepare_one_trb(dep, req, chain, i);
+ dwc3_prepare_one_trb(dep, req, trb_length, chain, i);
}
/*
@@ -1150,6 +1159,16 @@ static void dwc3_prepare_one_trb_sg(struct dwc3_ep *dep,
req->num_queued_sgs++;
+ /*
+ * The number of pending SG entries may not correspond to the
+ * number of mapped SG entries. If all the data are queued, then
+ * don't include unused SG entries.
+ */
+ if (length == 0) {
+ req->num_pending_sgs -= req->request.num_mapped_sgs - req->num_queued_sgs;
+ break;
+ }
+
if (!dwc3_calc_trbs_left(dep))
break;
}
@@ -1169,7 +1188,7 @@ static void dwc3_prepare_one_trb_linear(struct dwc3_ep *dep,
req->needs_extra_trb = true;
/* prepare normal TRB */
- dwc3_prepare_one_trb(dep, req, true, 0);
+ dwc3_prepare_one_trb(dep, req, length, true, 0);
/* Now prepare one extra TRB to align transfer size */
trb = &dep->trb_pool[dep->trb_enqueue];
@@ -1187,7 +1206,7 @@ static void dwc3_prepare_one_trb_linear(struct dwc3_ep *dep,
req->needs_extra_trb = true;
/* prepare normal TRB */
- dwc3_prepare_one_trb(dep, req, true, 0);
+ dwc3_prepare_one_trb(dep, req, length, true, 0);
/* Now prepare one extra TRB to handle ZLP */
trb = &dep->trb_pool[dep->trb_enqueue];
@@ -1198,7 +1217,7 @@ static void dwc3_prepare_one_trb_linear(struct dwc3_ep *dep,
req->request.no_interrupt,
req->request.is_last);
} else {
- dwc3_prepare_one_trb(dep, req, false, 0);
+ dwc3_prepare_one_trb(dep, req, length, false, 0);
}
}
--
2.28.0
This is a note to let you know that I've just added the patch titled
usb: dwc3: gadget: Fix handling ZLP
to my usb git tree which can be found at
git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb.git
in the usb-linus branch.
The patch will show up in the next release of the linux-next tree
(usually sometime within the next 24 hours during the week.)
The patch will hopefully also be merged in Linus's tree for the
next -rc kernel release.
If you have any questions about this process, please let me know.
>From d2ee3ff79e6a3d4105e684021017d100524dc560 Mon Sep 17 00:00:00 2001
From: Thinh Nguyen <Thinh.Nguyen(a)synopsys.com>
Date: Thu, 6 Aug 2020 19:46:29 -0700
Subject: usb: dwc3: gadget: Fix handling ZLP
The usb_request->zero doesn't apply for isoc. Also, if we prepare a
0-length (ZLP) TRB for the OUT direction, we need to prepare an extra
TRB to pad up to the MPS alignment. Use the same bounce buffer for the
ZLP TRB and the extra pad TRB.
Cc: <stable(a)vger.kernel.org> # v4.5+
Fixes: d6e5a549cc4d ("usb: dwc3: simplify ZLP handling")
Fixes: 04c03d10e507 ("usb: dwc3: gadget: handle request->zero")
Signed-off-by: Thinh Nguyen <thinhn(a)synopsys.com>
Signed-off-by: Felipe Balbi <balbi(a)kernel.org>
---
drivers/usb/dwc3/gadget.c | 25 +++++++++++++++++++++++--
1 file changed, 23 insertions(+), 2 deletions(-)
diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
index f9231253cbed..df603a817a98 100644
--- a/drivers/usb/dwc3/gadget.c
+++ b/drivers/usb/dwc3/gadget.c
@@ -1199,6 +1199,7 @@ static void dwc3_prepare_one_trb_linear(struct dwc3_ep *dep,
req->request.no_interrupt,
req->request.is_last);
} else if (req->request.zero && req->request.length &&
+ !usb_endpoint_xfer_isoc(dep->endpoint.desc) &&
(IS_ALIGNED(req->request.length, maxp))) {
struct dwc3 *dwc = dep->dwc;
struct dwc3_trb *trb;
@@ -1208,14 +1209,25 @@ static void dwc3_prepare_one_trb_linear(struct dwc3_ep *dep,
/* prepare normal TRB */
dwc3_prepare_one_trb(dep, req, length, true, 0);
- /* Now prepare one extra TRB to handle ZLP */
+ /* Prepare one extra TRB to handle ZLP */
trb = &dep->trb_pool[dep->trb_enqueue];
req->num_trbs++;
__dwc3_prepare_one_trb(dep, trb, dwc->bounce_addr, 0,
- false, 1, req->request.stream_id,
+ !req->direction, 1, req->request.stream_id,
req->request.short_not_ok,
req->request.no_interrupt,
req->request.is_last);
+
+ /* Prepare one more TRB to handle MPS alignment for OUT */
+ if (!req->direction) {
+ trb = &dep->trb_pool[dep->trb_enqueue];
+ req->num_trbs++;
+ __dwc3_prepare_one_trb(dep, trb, dwc->bounce_addr, maxp,
+ false, 1, req->request.stream_id,
+ req->request.short_not_ok,
+ req->request.no_interrupt,
+ req->request.is_last);
+ }
} else {
dwc3_prepare_one_trb(dep, req, length, false, 0);
}
@@ -2690,8 +2702,17 @@ static int dwc3_gadget_ep_cleanup_completed_request(struct dwc3_ep *dep,
status);
if (req->needs_extra_trb) {
+ unsigned int maxp = usb_endpoint_maxp(dep->endpoint.desc);
+
ret = dwc3_gadget_ep_reclaim_trb_linear(dep, req, event,
status);
+
+ /* Reclaim MPS padding TRB for ZLP */
+ if (!req->direction && req->request.zero && req->request.length &&
+ !usb_endpoint_xfer_isoc(dep->endpoint.desc) &&
+ (IS_ALIGNED(req->request.length, maxp)))
+ ret = dwc3_gadget_ep_reclaim_trb_linear(dep, req, event, status);
+
req->needs_extra_trb = false;
}
--
2.28.0
The patch below does not apply to the 4.14-stable tree.
If someone wants it applied there, or to any other stable or longterm
tree, then please email the backport, including the original git commit
id to <stable(a)vger.kernel.org>.
thanks,
greg k-h
------------------ original commit in Linus's tree ------------------
>From 156abe2961601d60a8c2a60c6dc8dd6ce7adcdaf Mon Sep 17 00:00:00 2001
From: Hans de Goede <hdegoede(a)redhat.com>
Date: Sat, 6 Jun 2020 11:31:50 +0200
Subject: [PATCH] pinctrl: baytrail: Fix pin being driven low for a while on
gpiod_get(..., GPIOD_OUT_HIGH)
The pins on the Bay Trail SoC have separate input-buffer and output-buffer
enable bits and a read of the level bit of the value register will always
return the value from the input-buffer.
The BIOS of a device may configure a pin in output-only mode, only enabling
the output buffer, and write 1 to the level bit to drive the pin high.
This 1 written to the level bit will be stored inside the data-latch of the
output buffer.
But a subsequent read of the value register will return 0 for the level bit
because the input-buffer is disabled. This causes a read-modify-write as
done by byt_gpio_set_direction() to write 0 to the level bit, driving the
pin low!
Before this commit byt_gpio_direction_output() relied on
pinctrl_gpio_direction_output() to set the direction, followed by a call
to byt_gpio_set() to apply the selected value. This causes the pin to
go low between the pinctrl_gpio_direction_output() and byt_gpio_set()
calls.
Change byt_gpio_direction_output() to directly make the register
modifications itself instead. Replacing the 2 subsequent writes to the
value register with a single write.
Note that the pinctrl code does not keep track internally of the direction,
so not going through pinctrl_gpio_direction_output() is not an issue.
This issue was noticed on a Trekstor SurfTab Twin 10.1. When the panel is
already on at boot (no external monitor connected), then the i915 driver
does a gpiod_get(..., GPIOD_OUT_HIGH) for the panel-enable GPIO. The
temporarily going low of that GPIO was causing the panel to reset itself
after which it would not show an image until it was turned off and back on
again (until a full modeset was done on it). This commit fixes this.
This commit also updates the byt_gpio_direction_input() to use direct
register accesses instead of going through pinctrl_gpio_direction_input(),
to keep it consistent with byt_gpio_direction_output().
Note for backporting, this commit depends on:
commit e2b74419e5cc ("pinctrl: baytrail: Replace WARN with dev_info_once
when setting direct-irq pin to output")
Cc: stable(a)vger.kernel.org
Fixes: 86e3ef812fe3 ("pinctrl: baytrail: Update gpio chip operations")
Signed-off-by: Hans de Goede <hdegoede(a)redhat.com>
Acked-by: Mika Westerberg <mika.westerberg(a)linux.intel.com>
Signed-off-by: Andy Shevchenko <andriy.shevchenko(a)linux.intel.com>
diff --git a/drivers/pinctrl/intel/pinctrl-baytrail.c b/drivers/pinctrl/intel/pinctrl-baytrail.c
index e3ceb3dfeabe..a917a2df520e 100644
--- a/drivers/pinctrl/intel/pinctrl-baytrail.c
+++ b/drivers/pinctrl/intel/pinctrl-baytrail.c
@@ -800,6 +800,21 @@ static void byt_gpio_disable_free(struct pinctrl_dev *pctl_dev,
pm_runtime_put(vg->dev);
}
+static void byt_gpio_direct_irq_check(struct intel_pinctrl *vg,
+ unsigned int offset)
+{
+ void __iomem *conf_reg = byt_gpio_reg(vg, offset, BYT_CONF0_REG);
+
+ /*
+ * Before making any direction modifications, do a check if gpio is set
+ * for direct IRQ. On Bay Trail, setting GPIO to output does not make
+ * sense, so let's at least inform the caller before they shoot
+ * themselves in the foot.
+ */
+ if (readl(conf_reg) & BYT_DIRECT_IRQ_EN)
+ dev_info_once(vg->dev, "Potential Error: Setting GPIO with direct_irq_en to output");
+}
+
static int byt_gpio_set_direction(struct pinctrl_dev *pctl_dev,
struct pinctrl_gpio_range *range,
unsigned int offset,
@@ -807,7 +822,6 @@ static int byt_gpio_set_direction(struct pinctrl_dev *pctl_dev,
{
struct intel_pinctrl *vg = pinctrl_dev_get_drvdata(pctl_dev);
void __iomem *val_reg = byt_gpio_reg(vg, offset, BYT_VAL_REG);
- void __iomem *conf_reg = byt_gpio_reg(vg, offset, BYT_CONF0_REG);
unsigned long flags;
u32 value;
@@ -817,14 +831,8 @@ static int byt_gpio_set_direction(struct pinctrl_dev *pctl_dev,
value &= ~BYT_DIR_MASK;
if (input)
value |= BYT_OUTPUT_EN;
- else if (readl(conf_reg) & BYT_DIRECT_IRQ_EN)
- /*
- * Before making any direction modifications, do a check if gpio
- * is set for direct IRQ. On baytrail, setting GPIO to output
- * does not make sense, so let's at least inform the caller before
- * they shoot themselves in the foot.
- */
- dev_info_once(vg->dev, "Potential Error: Setting GPIO with direct_irq_en to output");
+ else
+ byt_gpio_direct_irq_check(vg, offset);
writel(value, val_reg);
@@ -1165,19 +1173,50 @@ static int byt_gpio_get_direction(struct gpio_chip *chip, unsigned int offset)
static int byt_gpio_direction_input(struct gpio_chip *chip, unsigned int offset)
{
- return pinctrl_gpio_direction_input(chip->base + offset);
+ struct intel_pinctrl *vg = gpiochip_get_data(chip);
+ void __iomem *val_reg = byt_gpio_reg(vg, offset, BYT_VAL_REG);
+ unsigned long flags;
+ u32 reg;
+
+ raw_spin_lock_irqsave(&byt_lock, flags);
+
+ reg = readl(val_reg);
+ reg &= ~BYT_DIR_MASK;
+ reg |= BYT_OUTPUT_EN;
+ writel(reg, val_reg);
+
+ raw_spin_unlock_irqrestore(&byt_lock, flags);
+ return 0;
}
+/*
+ * Note despite the temptation this MUST NOT be converted into a call to
+ * pinctrl_gpio_direction_output() + byt_gpio_set() that does not work this
+ * MUST be done as a single BYT_VAL_REG register write.
+ * See the commit message of the commit adding this comment for details.
+ */
static int byt_gpio_direction_output(struct gpio_chip *chip,
unsigned int offset, int value)
{
- int ret = pinctrl_gpio_direction_output(chip->base + offset);
+ struct intel_pinctrl *vg = gpiochip_get_data(chip);
+ void __iomem *val_reg = byt_gpio_reg(vg, offset, BYT_VAL_REG);
+ unsigned long flags;
+ u32 reg;
- if (ret)
- return ret;
+ raw_spin_lock_irqsave(&byt_lock, flags);
+
+ byt_gpio_direct_irq_check(vg, offset);
- byt_gpio_set(chip, offset, value);
+ reg = readl(val_reg);
+ reg &= ~BYT_DIR_MASK;
+ if (value)
+ reg |= BYT_LEVEL;
+ else
+ reg &= ~BYT_LEVEL;
+ writel(reg, val_reg);
+
+ raw_spin_unlock_irqrestore(&byt_lock, flags);
return 0;
}
The patch below does not apply to the 4.19-stable tree.
If someone wants it applied there, or to any other stable or longterm
tree, then please email the backport, including the original git commit
id to <stable(a)vger.kernel.org>.
thanks,
greg k-h
------------------ original commit in Linus's tree ------------------
>From 156abe2961601d60a8c2a60c6dc8dd6ce7adcdaf Mon Sep 17 00:00:00 2001
From: Hans de Goede <hdegoede(a)redhat.com>
Date: Sat, 6 Jun 2020 11:31:50 +0200
Subject: [PATCH] pinctrl: baytrail: Fix pin being driven low for a while on
gpiod_get(..., GPIOD_OUT_HIGH)
The pins on the Bay Trail SoC have separate input-buffer and output-buffer
enable bits and a read of the level bit of the value register will always
return the value from the input-buffer.
The BIOS of a device may configure a pin in output-only mode, only enabling
the output buffer, and write 1 to the level bit to drive the pin high.
This 1 written to the level bit will be stored inside the data-latch of the
output buffer.
But a subsequent read of the value register will return 0 for the level bit
because the input-buffer is disabled. This causes a read-modify-write as
done by byt_gpio_set_direction() to write 0 to the level bit, driving the
pin low!
Before this commit byt_gpio_direction_output() relied on
pinctrl_gpio_direction_output() to set the direction, followed by a call
to byt_gpio_set() to apply the selected value. This causes the pin to
go low between the pinctrl_gpio_direction_output() and byt_gpio_set()
calls.
Change byt_gpio_direction_output() to directly make the register
modifications itself instead. Replacing the 2 subsequent writes to the
value register with a single write.
Note that the pinctrl code does not keep track internally of the direction,
so not going through pinctrl_gpio_direction_output() is not an issue.
This issue was noticed on a Trekstor SurfTab Twin 10.1. When the panel is
already on at boot (no external monitor connected), then the i915 driver
does a gpiod_get(..., GPIOD_OUT_HIGH) for the panel-enable GPIO. The
temporarily going low of that GPIO was causing the panel to reset itself
after which it would not show an image until it was turned off and back on
again (until a full modeset was done on it). This commit fixes this.
This commit also updates the byt_gpio_direction_input() to use direct
register accesses instead of going through pinctrl_gpio_direction_input(),
to keep it consistent with byt_gpio_direction_output().
Note for backporting, this commit depends on:
commit e2b74419e5cc ("pinctrl: baytrail: Replace WARN with dev_info_once
when setting direct-irq pin to output")
Cc: stable(a)vger.kernel.org
Fixes: 86e3ef812fe3 ("pinctrl: baytrail: Update gpio chip operations")
Signed-off-by: Hans de Goede <hdegoede(a)redhat.com>
Acked-by: Mika Westerberg <mika.westerberg(a)linux.intel.com>
Signed-off-by: Andy Shevchenko <andriy.shevchenko(a)linux.intel.com>
diff --git a/drivers/pinctrl/intel/pinctrl-baytrail.c b/drivers/pinctrl/intel/pinctrl-baytrail.c
index e3ceb3dfeabe..a917a2df520e 100644
--- a/drivers/pinctrl/intel/pinctrl-baytrail.c
+++ b/drivers/pinctrl/intel/pinctrl-baytrail.c
@@ -800,6 +800,21 @@ static void byt_gpio_disable_free(struct pinctrl_dev *pctl_dev,
pm_runtime_put(vg->dev);
}
+static void byt_gpio_direct_irq_check(struct intel_pinctrl *vg,
+ unsigned int offset)
+{
+ void __iomem *conf_reg = byt_gpio_reg(vg, offset, BYT_CONF0_REG);
+
+ /*
+ * Before making any direction modifications, do a check if gpio is set
+ * for direct IRQ. On Bay Trail, setting GPIO to output does not make
+ * sense, so let's at least inform the caller before they shoot
+ * themselves in the foot.
+ */
+ if (readl(conf_reg) & BYT_DIRECT_IRQ_EN)
+ dev_info_once(vg->dev, "Potential Error: Setting GPIO with direct_irq_en to output");
+}
+
static int byt_gpio_set_direction(struct pinctrl_dev *pctl_dev,
struct pinctrl_gpio_range *range,
unsigned int offset,
@@ -807,7 +822,6 @@ static int byt_gpio_set_direction(struct pinctrl_dev *pctl_dev,
{
struct intel_pinctrl *vg = pinctrl_dev_get_drvdata(pctl_dev);
void __iomem *val_reg = byt_gpio_reg(vg, offset, BYT_VAL_REG);
- void __iomem *conf_reg = byt_gpio_reg(vg, offset, BYT_CONF0_REG);
unsigned long flags;
u32 value;
@@ -817,14 +831,8 @@ static int byt_gpio_set_direction(struct pinctrl_dev *pctl_dev,
value &= ~BYT_DIR_MASK;
if (input)
value |= BYT_OUTPUT_EN;
- else if (readl(conf_reg) & BYT_DIRECT_IRQ_EN)
- /*
- * Before making any direction modifications, do a check if gpio
- * is set for direct IRQ. On baytrail, setting GPIO to output
- * does not make sense, so let's at least inform the caller before
- * they shoot themselves in the foot.
- */
- dev_info_once(vg->dev, "Potential Error: Setting GPIO with direct_irq_en to output");
+ else
+ byt_gpio_direct_irq_check(vg, offset);
writel(value, val_reg);
@@ -1165,19 +1173,50 @@ static int byt_gpio_get_direction(struct gpio_chip *chip, unsigned int offset)
static int byt_gpio_direction_input(struct gpio_chip *chip, unsigned int offset)
{
- return pinctrl_gpio_direction_input(chip->base + offset);
+ struct intel_pinctrl *vg = gpiochip_get_data(chip);
+ void __iomem *val_reg = byt_gpio_reg(vg, offset, BYT_VAL_REG);
+ unsigned long flags;
+ u32 reg;
+
+ raw_spin_lock_irqsave(&byt_lock, flags);
+
+ reg = readl(val_reg);
+ reg &= ~BYT_DIR_MASK;
+ reg |= BYT_OUTPUT_EN;
+ writel(reg, val_reg);
+
+ raw_spin_unlock_irqrestore(&byt_lock, flags);
+ return 0;
}
+/*
+ * Note despite the temptation this MUST NOT be converted into a call to
+ * pinctrl_gpio_direction_output() + byt_gpio_set() that does not work this
+ * MUST be done as a single BYT_VAL_REG register write.
+ * See the commit message of the commit adding this comment for details.
+ */
static int byt_gpio_direction_output(struct gpio_chip *chip,
unsigned int offset, int value)
{
- int ret = pinctrl_gpio_direction_output(chip->base + offset);
+ struct intel_pinctrl *vg = gpiochip_get_data(chip);
+ void __iomem *val_reg = byt_gpio_reg(vg, offset, BYT_VAL_REG);
+ unsigned long flags;
+ u32 reg;
- if (ret)
- return ret;
+ raw_spin_lock_irqsave(&byt_lock, flags);
+
+ byt_gpio_direct_irq_check(vg, offset);
- byt_gpio_set(chip, offset, value);
+ reg = readl(val_reg);
+ reg &= ~BYT_DIR_MASK;
+ if (value)
+ reg |= BYT_LEVEL;
+ else
+ reg &= ~BYT_LEVEL;
+ writel(reg, val_reg);
+
+ raw_spin_unlock_irqrestore(&byt_lock, flags);
return 0;
}
The patch below does not apply to the 5.7-stable tree.
If someone wants it applied there, or to any other stable or longterm
tree, then please email the backport, including the original git commit
id to <stable(a)vger.kernel.org>.
thanks,
greg k-h
------------------ original commit in Linus's tree ------------------
>From 156abe2961601d60a8c2a60c6dc8dd6ce7adcdaf Mon Sep 17 00:00:00 2001
From: Hans de Goede <hdegoede(a)redhat.com>
Date: Sat, 6 Jun 2020 11:31:50 +0200
Subject: [PATCH] pinctrl: baytrail: Fix pin being driven low for a while on
gpiod_get(..., GPIOD_OUT_HIGH)
The pins on the Bay Trail SoC have separate input-buffer and output-buffer
enable bits and a read of the level bit of the value register will always
return the value from the input-buffer.
The BIOS of a device may configure a pin in output-only mode, only enabling
the output buffer, and write 1 to the level bit to drive the pin high.
This 1 written to the level bit will be stored inside the data-latch of the
output buffer.
But a subsequent read of the value register will return 0 for the level bit
because the input-buffer is disabled. This causes a read-modify-write as
done by byt_gpio_set_direction() to write 0 to the level bit, driving the
pin low!
Before this commit byt_gpio_direction_output() relied on
pinctrl_gpio_direction_output() to set the direction, followed by a call
to byt_gpio_set() to apply the selected value. This causes the pin to
go low between the pinctrl_gpio_direction_output() and byt_gpio_set()
calls.
Change byt_gpio_direction_output() to directly make the register
modifications itself instead. Replacing the 2 subsequent writes to the
value register with a single write.
Note that the pinctrl code does not keep track internally of the direction,
so not going through pinctrl_gpio_direction_output() is not an issue.
This issue was noticed on a Trekstor SurfTab Twin 10.1. When the panel is
already on at boot (no external monitor connected), then the i915 driver
does a gpiod_get(..., GPIOD_OUT_HIGH) for the panel-enable GPIO. The
temporarily going low of that GPIO was causing the panel to reset itself
after which it would not show an image until it was turned off and back on
again (until a full modeset was done on it). This commit fixes this.
This commit also updates the byt_gpio_direction_input() to use direct
register accesses instead of going through pinctrl_gpio_direction_input(),
to keep it consistent with byt_gpio_direction_output().
Note for backporting, this commit depends on:
commit e2b74419e5cc ("pinctrl: baytrail: Replace WARN with dev_info_once
when setting direct-irq pin to output")
Cc: stable(a)vger.kernel.org
Fixes: 86e3ef812fe3 ("pinctrl: baytrail: Update gpio chip operations")
Signed-off-by: Hans de Goede <hdegoede(a)redhat.com>
Acked-by: Mika Westerberg <mika.westerberg(a)linux.intel.com>
Signed-off-by: Andy Shevchenko <andriy.shevchenko(a)linux.intel.com>
diff --git a/drivers/pinctrl/intel/pinctrl-baytrail.c b/drivers/pinctrl/intel/pinctrl-baytrail.c
index e3ceb3dfeabe..a917a2df520e 100644
--- a/drivers/pinctrl/intel/pinctrl-baytrail.c
+++ b/drivers/pinctrl/intel/pinctrl-baytrail.c
@@ -800,6 +800,21 @@ static void byt_gpio_disable_free(struct pinctrl_dev *pctl_dev,
pm_runtime_put(vg->dev);
}
+static void byt_gpio_direct_irq_check(struct intel_pinctrl *vg,
+ unsigned int offset)
+{
+ void __iomem *conf_reg = byt_gpio_reg(vg, offset, BYT_CONF0_REG);
+
+ /*
+ * Before making any direction modifications, do a check if gpio is set
+ * for direct IRQ. On Bay Trail, setting GPIO to output does not make
+ * sense, so let's at least inform the caller before they shoot
+ * themselves in the foot.
+ */
+ if (readl(conf_reg) & BYT_DIRECT_IRQ_EN)
+ dev_info_once(vg->dev, "Potential Error: Setting GPIO with direct_irq_en to output");
+}
+
static int byt_gpio_set_direction(struct pinctrl_dev *pctl_dev,
struct pinctrl_gpio_range *range,
unsigned int offset,
@@ -807,7 +822,6 @@ static int byt_gpio_set_direction(struct pinctrl_dev *pctl_dev,
{
struct intel_pinctrl *vg = pinctrl_dev_get_drvdata(pctl_dev);
void __iomem *val_reg = byt_gpio_reg(vg, offset, BYT_VAL_REG);
- void __iomem *conf_reg = byt_gpio_reg(vg, offset, BYT_CONF0_REG);
unsigned long flags;
u32 value;
@@ -817,14 +831,8 @@ static int byt_gpio_set_direction(struct pinctrl_dev *pctl_dev,
value &= ~BYT_DIR_MASK;
if (input)
value |= BYT_OUTPUT_EN;
- else if (readl(conf_reg) & BYT_DIRECT_IRQ_EN)
- /*
- * Before making any direction modifications, do a check if gpio
- * is set for direct IRQ. On baytrail, setting GPIO to output
- * does not make sense, so let's at least inform the caller before
- * they shoot themselves in the foot.
- */
- dev_info_once(vg->dev, "Potential Error: Setting GPIO with direct_irq_en to output");
+ else
+ byt_gpio_direct_irq_check(vg, offset);
writel(value, val_reg);
@@ -1165,19 +1173,50 @@ static int byt_gpio_get_direction(struct gpio_chip *chip, unsigned int offset)
static int byt_gpio_direction_input(struct gpio_chip *chip, unsigned int offset)
{
- return pinctrl_gpio_direction_input(chip->base + offset);
+ struct intel_pinctrl *vg = gpiochip_get_data(chip);
+ void __iomem *val_reg = byt_gpio_reg(vg, offset, BYT_VAL_REG);
+ unsigned long flags;
+ u32 reg;
+
+ raw_spin_lock_irqsave(&byt_lock, flags);
+
+ reg = readl(val_reg);
+ reg &= ~BYT_DIR_MASK;
+ reg |= BYT_OUTPUT_EN;
+ writel(reg, val_reg);
+
+ raw_spin_unlock_irqrestore(&byt_lock, flags);
+ return 0;
}
+/*
+ * Note despite the temptation this MUST NOT be converted into a call to
+ * pinctrl_gpio_direction_output() + byt_gpio_set() that does not work this
+ * MUST be done as a single BYT_VAL_REG register write.
+ * See the commit message of the commit adding this comment for details.
+ */
static int byt_gpio_direction_output(struct gpio_chip *chip,
unsigned int offset, int value)
{
- int ret = pinctrl_gpio_direction_output(chip->base + offset);
+ struct intel_pinctrl *vg = gpiochip_get_data(chip);
+ void __iomem *val_reg = byt_gpio_reg(vg, offset, BYT_VAL_REG);
+ unsigned long flags;
+ u32 reg;
- if (ret)
- return ret;
+ raw_spin_lock_irqsave(&byt_lock, flags);
+
+ byt_gpio_direct_irq_check(vg, offset);
- byt_gpio_set(chip, offset, value);
+ reg = readl(val_reg);
+ reg &= ~BYT_DIR_MASK;
+ if (value)
+ reg |= BYT_LEVEL;
+ else
+ reg &= ~BYT_LEVEL;
+ writel(reg, val_reg);
+
+ raw_spin_unlock_irqrestore(&byt_lock, flags);
return 0;
}
The patch below does not apply to the 5.8-stable tree.
If someone wants it applied there, or to any other stable or longterm
tree, then please email the backport, including the original git commit
id to <stable(a)vger.kernel.org>.
thanks,
greg k-h
------------------ original commit in Linus's tree ------------------
>From 156abe2961601d60a8c2a60c6dc8dd6ce7adcdaf Mon Sep 17 00:00:00 2001
From: Hans de Goede <hdegoede(a)redhat.com>
Date: Sat, 6 Jun 2020 11:31:50 +0200
Subject: [PATCH] pinctrl: baytrail: Fix pin being driven low for a while on
gpiod_get(..., GPIOD_OUT_HIGH)
The pins on the Bay Trail SoC have separate input-buffer and output-buffer
enable bits and a read of the level bit of the value register will always
return the value from the input-buffer.
The BIOS of a device may configure a pin in output-only mode, only enabling
the output buffer, and write 1 to the level bit to drive the pin high.
This 1 written to the level bit will be stored inside the data-latch of the
output buffer.
But a subsequent read of the value register will return 0 for the level bit
because the input-buffer is disabled. This causes a read-modify-write as
done by byt_gpio_set_direction() to write 0 to the level bit, driving the
pin low!
Before this commit byt_gpio_direction_output() relied on
pinctrl_gpio_direction_output() to set the direction, followed by a call
to byt_gpio_set() to apply the selected value. This causes the pin to
go low between the pinctrl_gpio_direction_output() and byt_gpio_set()
calls.
Change byt_gpio_direction_output() to directly make the register
modifications itself instead. Replacing the 2 subsequent writes to the
value register with a single write.
Note that the pinctrl code does not keep track internally of the direction,
so not going through pinctrl_gpio_direction_output() is not an issue.
This issue was noticed on a Trekstor SurfTab Twin 10.1. When the panel is
already on at boot (no external monitor connected), then the i915 driver
does a gpiod_get(..., GPIOD_OUT_HIGH) for the panel-enable GPIO. The
temporarily going low of that GPIO was causing the panel to reset itself
after which it would not show an image until it was turned off and back on
again (until a full modeset was done on it). This commit fixes this.
This commit also updates the byt_gpio_direction_input() to use direct
register accesses instead of going through pinctrl_gpio_direction_input(),
to keep it consistent with byt_gpio_direction_output().
Note for backporting, this commit depends on:
commit e2b74419e5cc ("pinctrl: baytrail: Replace WARN with dev_info_once
when setting direct-irq pin to output")
Cc: stable(a)vger.kernel.org
Fixes: 86e3ef812fe3 ("pinctrl: baytrail: Update gpio chip operations")
Signed-off-by: Hans de Goede <hdegoede(a)redhat.com>
Acked-by: Mika Westerberg <mika.westerberg(a)linux.intel.com>
Signed-off-by: Andy Shevchenko <andriy.shevchenko(a)linux.intel.com>
diff --git a/drivers/pinctrl/intel/pinctrl-baytrail.c b/drivers/pinctrl/intel/pinctrl-baytrail.c
index e3ceb3dfeabe..a917a2df520e 100644
--- a/drivers/pinctrl/intel/pinctrl-baytrail.c
+++ b/drivers/pinctrl/intel/pinctrl-baytrail.c
@@ -800,6 +800,21 @@ static void byt_gpio_disable_free(struct pinctrl_dev *pctl_dev,
pm_runtime_put(vg->dev);
}
+static void byt_gpio_direct_irq_check(struct intel_pinctrl *vg,
+ unsigned int offset)
+{
+ void __iomem *conf_reg = byt_gpio_reg(vg, offset, BYT_CONF0_REG);
+
+ /*
+ * Before making any direction modifications, do a check if gpio is set
+ * for direct IRQ. On Bay Trail, setting GPIO to output does not make
+ * sense, so let's at least inform the caller before they shoot
+ * themselves in the foot.
+ */
+ if (readl(conf_reg) & BYT_DIRECT_IRQ_EN)
+ dev_info_once(vg->dev, "Potential Error: Setting GPIO with direct_irq_en to output");
+}
+
static int byt_gpio_set_direction(struct pinctrl_dev *pctl_dev,
struct pinctrl_gpio_range *range,
unsigned int offset,
@@ -807,7 +822,6 @@ static int byt_gpio_set_direction(struct pinctrl_dev *pctl_dev,
{
struct intel_pinctrl *vg = pinctrl_dev_get_drvdata(pctl_dev);
void __iomem *val_reg = byt_gpio_reg(vg, offset, BYT_VAL_REG);
- void __iomem *conf_reg = byt_gpio_reg(vg, offset, BYT_CONF0_REG);
unsigned long flags;
u32 value;
@@ -817,14 +831,8 @@ static int byt_gpio_set_direction(struct pinctrl_dev *pctl_dev,
value &= ~BYT_DIR_MASK;
if (input)
value |= BYT_OUTPUT_EN;
- else if (readl(conf_reg) & BYT_DIRECT_IRQ_EN)
- /*
- * Before making any direction modifications, do a check if gpio
- * is set for direct IRQ. On baytrail, setting GPIO to output
- * does not make sense, so let's at least inform the caller before
- * they shoot themselves in the foot.
- */
- dev_info_once(vg->dev, "Potential Error: Setting GPIO with direct_irq_en to output");
+ else
+ byt_gpio_direct_irq_check(vg, offset);
writel(value, val_reg);
@@ -1165,19 +1173,50 @@ static int byt_gpio_get_direction(struct gpio_chip *chip, unsigned int offset)
static int byt_gpio_direction_input(struct gpio_chip *chip, unsigned int offset)
{
- return pinctrl_gpio_direction_input(chip->base + offset);
+ struct intel_pinctrl *vg = gpiochip_get_data(chip);
+ void __iomem *val_reg = byt_gpio_reg(vg, offset, BYT_VAL_REG);
+ unsigned long flags;
+ u32 reg;
+
+ raw_spin_lock_irqsave(&byt_lock, flags);
+
+ reg = readl(val_reg);
+ reg &= ~BYT_DIR_MASK;
+ reg |= BYT_OUTPUT_EN;
+ writel(reg, val_reg);
+
+ raw_spin_unlock_irqrestore(&byt_lock, flags);
+ return 0;
}
+/*
+ * Note despite the temptation this MUST NOT be converted into a call to
+ * pinctrl_gpio_direction_output() + byt_gpio_set() that does not work this
+ * MUST be done as a single BYT_VAL_REG register write.
+ * See the commit message of the commit adding this comment for details.
+ */
static int byt_gpio_direction_output(struct gpio_chip *chip,
unsigned int offset, int value)
{
- int ret = pinctrl_gpio_direction_output(chip->base + offset);
+ struct intel_pinctrl *vg = gpiochip_get_data(chip);
+ void __iomem *val_reg = byt_gpio_reg(vg, offset, BYT_VAL_REG);
+ unsigned long flags;
+ u32 reg;
- if (ret)
- return ret;
+ raw_spin_lock_irqsave(&byt_lock, flags);
+
+ byt_gpio_direct_irq_check(vg, offset);
- byt_gpio_set(chip, offset, value);
+ reg = readl(val_reg);
+ reg &= ~BYT_DIR_MASK;
+ if (value)
+ reg |= BYT_LEVEL;
+ else
+ reg &= ~BYT_LEVEL;
+ writel(reg, val_reg);
+
+ raw_spin_unlock_irqrestore(&byt_lock, flags);
return 0;
}
The patch below does not apply to the 4.14-stable tree.
If someone wants it applied there, or to any other stable or longterm
tree, then please email the backport, including the original git commit
id to <stable(a)vger.kernel.org>.
thanks,
greg k-h
------------------ original commit in Linus's tree ------------------
>From 63dee5df43a31f3844efabc58972f0a206ca4534 Mon Sep 17 00:00:00 2001
From: Michael Ellerman <mpe(a)ellerman.id.au>
Date: Fri, 24 Jul 2020 19:25:25 +1000
Subject: [PATCH] powerpc: Allow 4224 bytes of stack expansion for the signal
frame
We have powerpc specific logic in our page fault handling to decide if
an access to an unmapped address below the stack pointer should expand
the stack VMA.
The code was originally added in 2004 "ported from 2.4". The rough
logic is that the stack is allowed to grow to 1MB with no extra
checking. Over 1MB the access must be within 2048 bytes of the stack
pointer, or be from a user instruction that updates the stack pointer.
The 2048 byte allowance below the stack pointer is there to cover the
288 byte "red zone" as well as the "about 1.5kB" needed by the signal
delivery code.
Unfortunately since then the signal frame has expanded, and is now
4224 bytes on 64-bit kernels with transactional memory enabled. This
means if a process has consumed more than 1MB of stack, and its stack
pointer lies less than 4224 bytes from the next page boundary, signal
delivery will fault when trying to expand the stack and the process
will see a SEGV.
The total size of the signal frame is the size of struct rt_sigframe
(which includes the red zone) plus __SIGNAL_FRAMESIZE (128 bytes on
64-bit).
The 2048 byte allowance was correct until 2008 as the signal frame
was:
struct rt_sigframe {
struct ucontext uc; /* 0 1440 */
/* --- cacheline 11 boundary (1408 bytes) was 32 bytes ago --- */
long unsigned int _unused[2]; /* 1440 16 */
unsigned int tramp[6]; /* 1456 24 */
struct siginfo * pinfo; /* 1480 8 */
void * puc; /* 1488 8 */
struct siginfo info; /* 1496 128 */
/* --- cacheline 12 boundary (1536 bytes) was 88 bytes ago --- */
char abigap[288]; /* 1624 288 */
/* size: 1920, cachelines: 15, members: 7 */
/* padding: 8 */
};
1920 + 128 = 2048
Then in commit ce48b2100785 ("powerpc: Add VSX context save/restore,
ptrace and signal support") (Jul 2008) the signal frame expanded to
2304 bytes:
struct rt_sigframe {
struct ucontext uc; /* 0 1696 */ <--
/* --- cacheline 13 boundary (1664 bytes) was 32 bytes ago --- */
long unsigned int _unused[2]; /* 1696 16 */
unsigned int tramp[6]; /* 1712 24 */
struct siginfo * pinfo; /* 1736 8 */
void * puc; /* 1744 8 */
struct siginfo info; /* 1752 128 */
/* --- cacheline 14 boundary (1792 bytes) was 88 bytes ago --- */
char abigap[288]; /* 1880 288 */
/* size: 2176, cachelines: 17, members: 7 */
/* padding: 8 */
};
2176 + 128 = 2304
At this point we should have been exposed to the bug, though as far as
I know it was never reported. I no longer have a system old enough to
easily test on.
Then in 2010 commit 320b2b8de126 ("mm: keep a guard page below a
grow-down stack segment") caused our stack expansion code to never
trigger, as there was always a VMA found for a write up to PAGE_SIZE
below r1.
That meant the bug was hidden as we continued to expand the signal
frame in commit 2b0a576d15e0 ("powerpc: Add new transactional memory
state to the signal context") (Feb 2013):
struct rt_sigframe {
struct ucontext uc; /* 0 1696 */
/* --- cacheline 13 boundary (1664 bytes) was 32 bytes ago --- */
struct ucontext uc_transact; /* 1696 1696 */ <--
/* --- cacheline 26 boundary (3328 bytes) was 64 bytes ago --- */
long unsigned int _unused[2]; /* 3392 16 */
unsigned int tramp[6]; /* 3408 24 */
struct siginfo * pinfo; /* 3432 8 */
void * puc; /* 3440 8 */
struct siginfo info; /* 3448 128 */
/* --- cacheline 27 boundary (3456 bytes) was 120 bytes ago --- */
char abigap[288]; /* 3576 288 */
/* size: 3872, cachelines: 31, members: 8 */
/* padding: 8 */
/* last cacheline: 32 bytes */
};
3872 + 128 = 4000
And commit 573ebfa6601f ("powerpc: Increase stack redzone for 64-bit
userspace to 512 bytes") (Feb 2014):
struct rt_sigframe {
struct ucontext uc; /* 0 1696 */
/* --- cacheline 13 boundary (1664 bytes) was 32 bytes ago --- */
struct ucontext uc_transact; /* 1696 1696 */
/* --- cacheline 26 boundary (3328 bytes) was 64 bytes ago --- */
long unsigned int _unused[2]; /* 3392 16 */
unsigned int tramp[6]; /* 3408 24 */
struct siginfo * pinfo; /* 3432 8 */
void * puc; /* 3440 8 */
struct siginfo info; /* 3448 128 */
/* --- cacheline 27 boundary (3456 bytes) was 120 bytes ago --- */
char abigap[512]; /* 3576 512 */ <--
/* size: 4096, cachelines: 32, members: 8 */
/* padding: 8 */
};
4096 + 128 = 4224
Then finally in 2017, commit 1be7107fbe18 ("mm: larger stack guard
gap, between vmas") exposed us to the existing bug, because it changed
the stack VMA to be the correct/real size, meaning our stack expansion
code is now triggered.
Fix it by increasing the allowance to 4224 bytes.
Hard-coding 4224 is obviously unsafe against future expansions of the
signal frame in the same way as the existing code. We can't easily use
sizeof() because the signal frame structure is not in a header. We
will either fix that, or rip out all the custom stack expansion
checking logic entirely.
Fixes: ce48b2100785 ("powerpc: Add VSX context save/restore, ptrace and signal support")
Cc: stable(a)vger.kernel.org # v2.6.27+
Reported-by: Tom Lane <tgl(a)sss.pgh.pa.us>
Tested-by: Daniel Axtens <dja(a)axtens.net>
Signed-off-by: Michael Ellerman <mpe(a)ellerman.id.au>
Link: https://lore.kernel.org/r/20200724092528.1578671-2-mpe@ellerman.id.au
diff --git a/arch/powerpc/mm/fault.c b/arch/powerpc/mm/fault.c
index 641fc5f3d7dd..3ebb1792e636 100644
--- a/arch/powerpc/mm/fault.c
+++ b/arch/powerpc/mm/fault.c
@@ -267,6 +267,9 @@ static bool bad_kernel_fault(struct pt_regs *regs, unsigned long error_code,
return false;
}
+// This comes from 64-bit struct rt_sigframe + __SIGNAL_FRAMESIZE
+#define SIGFRAME_MAX_SIZE (4096 + 128)
+
static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
struct vm_area_struct *vma, unsigned int flags,
bool *must_retry)
@@ -274,7 +277,7 @@ static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
/*
* N.B. The POWER/Open ABI allows programs to access up to
* 288 bytes below the stack pointer.
- * The kernel signal delivery code writes up to about 1.5kB
+ * The kernel signal delivery code writes a bit over 4KB
* below the stack pointer (r1) before decrementing it.
* The exec code can write slightly over 640kB to the stack
* before setting the user r1. Thus we allow the stack to
@@ -299,7 +302,7 @@ static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
* between the last mapped region and the stack will
* expand the stack rather than segfaulting.
*/
- if (address + 2048 >= uregs->gpr[1])
+ if (address + SIGFRAME_MAX_SIZE >= uregs->gpr[1])
return false;
if ((flags & FAULT_FLAG_WRITE) && (flags & FAULT_FLAG_USER) &&
The patch below does not apply to the 4.9-stable tree.
If someone wants it applied there, or to any other stable or longterm
tree, then please email the backport, including the original git commit
id to <stable(a)vger.kernel.org>.
thanks,
greg k-h
------------------ original commit in Linus's tree ------------------
>From 63dee5df43a31f3844efabc58972f0a206ca4534 Mon Sep 17 00:00:00 2001
From: Michael Ellerman <mpe(a)ellerman.id.au>
Date: Fri, 24 Jul 2020 19:25:25 +1000
Subject: [PATCH] powerpc: Allow 4224 bytes of stack expansion for the signal
frame
We have powerpc specific logic in our page fault handling to decide if
an access to an unmapped address below the stack pointer should expand
the stack VMA.
The code was originally added in 2004 "ported from 2.4". The rough
logic is that the stack is allowed to grow to 1MB with no extra
checking. Over 1MB the access must be within 2048 bytes of the stack
pointer, or be from a user instruction that updates the stack pointer.
The 2048 byte allowance below the stack pointer is there to cover the
288 byte "red zone" as well as the "about 1.5kB" needed by the signal
delivery code.
Unfortunately since then the signal frame has expanded, and is now
4224 bytes on 64-bit kernels with transactional memory enabled. This
means if a process has consumed more than 1MB of stack, and its stack
pointer lies less than 4224 bytes from the next page boundary, signal
delivery will fault when trying to expand the stack and the process
will see a SEGV.
The total size of the signal frame is the size of struct rt_sigframe
(which includes the red zone) plus __SIGNAL_FRAMESIZE (128 bytes on
64-bit).
The 2048 byte allowance was correct until 2008 as the signal frame
was:
struct rt_sigframe {
struct ucontext uc; /* 0 1440 */
/* --- cacheline 11 boundary (1408 bytes) was 32 bytes ago --- */
long unsigned int _unused[2]; /* 1440 16 */
unsigned int tramp[6]; /* 1456 24 */
struct siginfo * pinfo; /* 1480 8 */
void * puc; /* 1488 8 */
struct siginfo info; /* 1496 128 */
/* --- cacheline 12 boundary (1536 bytes) was 88 bytes ago --- */
char abigap[288]; /* 1624 288 */
/* size: 1920, cachelines: 15, members: 7 */
/* padding: 8 */
};
1920 + 128 = 2048
Then in commit ce48b2100785 ("powerpc: Add VSX context save/restore,
ptrace and signal support") (Jul 2008) the signal frame expanded to
2304 bytes:
struct rt_sigframe {
struct ucontext uc; /* 0 1696 */ <--
/* --- cacheline 13 boundary (1664 bytes) was 32 bytes ago --- */
long unsigned int _unused[2]; /* 1696 16 */
unsigned int tramp[6]; /* 1712 24 */
struct siginfo * pinfo; /* 1736 8 */
void * puc; /* 1744 8 */
struct siginfo info; /* 1752 128 */
/* --- cacheline 14 boundary (1792 bytes) was 88 bytes ago --- */
char abigap[288]; /* 1880 288 */
/* size: 2176, cachelines: 17, members: 7 */
/* padding: 8 */
};
2176 + 128 = 2304
At this point we should have been exposed to the bug, though as far as
I know it was never reported. I no longer have a system old enough to
easily test on.
Then in 2010 commit 320b2b8de126 ("mm: keep a guard page below a
grow-down stack segment") caused our stack expansion code to never
trigger, as there was always a VMA found for a write up to PAGE_SIZE
below r1.
That meant the bug was hidden as we continued to expand the signal
frame in commit 2b0a576d15e0 ("powerpc: Add new transactional memory
state to the signal context") (Feb 2013):
struct rt_sigframe {
struct ucontext uc; /* 0 1696 */
/* --- cacheline 13 boundary (1664 bytes) was 32 bytes ago --- */
struct ucontext uc_transact; /* 1696 1696 */ <--
/* --- cacheline 26 boundary (3328 bytes) was 64 bytes ago --- */
long unsigned int _unused[2]; /* 3392 16 */
unsigned int tramp[6]; /* 3408 24 */
struct siginfo * pinfo; /* 3432 8 */
void * puc; /* 3440 8 */
struct siginfo info; /* 3448 128 */
/* --- cacheline 27 boundary (3456 bytes) was 120 bytes ago --- */
char abigap[288]; /* 3576 288 */
/* size: 3872, cachelines: 31, members: 8 */
/* padding: 8 */
/* last cacheline: 32 bytes */
};
3872 + 128 = 4000
And commit 573ebfa6601f ("powerpc: Increase stack redzone for 64-bit
userspace to 512 bytes") (Feb 2014):
struct rt_sigframe {
struct ucontext uc; /* 0 1696 */
/* --- cacheline 13 boundary (1664 bytes) was 32 bytes ago --- */
struct ucontext uc_transact; /* 1696 1696 */
/* --- cacheline 26 boundary (3328 bytes) was 64 bytes ago --- */
long unsigned int _unused[2]; /* 3392 16 */
unsigned int tramp[6]; /* 3408 24 */
struct siginfo * pinfo; /* 3432 8 */
void * puc; /* 3440 8 */
struct siginfo info; /* 3448 128 */
/* --- cacheline 27 boundary (3456 bytes) was 120 bytes ago --- */
char abigap[512]; /* 3576 512 */ <--
/* size: 4096, cachelines: 32, members: 8 */
/* padding: 8 */
};
4096 + 128 = 4224
Then finally in 2017, commit 1be7107fbe18 ("mm: larger stack guard
gap, between vmas") exposed us to the existing bug, because it changed
the stack VMA to be the correct/real size, meaning our stack expansion
code is now triggered.
Fix it by increasing the allowance to 4224 bytes.
Hard-coding 4224 is obviously unsafe against future expansions of the
signal frame in the same way as the existing code. We can't easily use
sizeof() because the signal frame structure is not in a header. We
will either fix that, or rip out all the custom stack expansion
checking logic entirely.
Fixes: ce48b2100785 ("powerpc: Add VSX context save/restore, ptrace and signal support")
Cc: stable(a)vger.kernel.org # v2.6.27+
Reported-by: Tom Lane <tgl(a)sss.pgh.pa.us>
Tested-by: Daniel Axtens <dja(a)axtens.net>
Signed-off-by: Michael Ellerman <mpe(a)ellerman.id.au>
Link: https://lore.kernel.org/r/20200724092528.1578671-2-mpe@ellerman.id.au
diff --git a/arch/powerpc/mm/fault.c b/arch/powerpc/mm/fault.c
index 641fc5f3d7dd..3ebb1792e636 100644
--- a/arch/powerpc/mm/fault.c
+++ b/arch/powerpc/mm/fault.c
@@ -267,6 +267,9 @@ static bool bad_kernel_fault(struct pt_regs *regs, unsigned long error_code,
return false;
}
+// This comes from 64-bit struct rt_sigframe + __SIGNAL_FRAMESIZE
+#define SIGFRAME_MAX_SIZE (4096 + 128)
+
static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
struct vm_area_struct *vma, unsigned int flags,
bool *must_retry)
@@ -274,7 +277,7 @@ static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
/*
* N.B. The POWER/Open ABI allows programs to access up to
* 288 bytes below the stack pointer.
- * The kernel signal delivery code writes up to about 1.5kB
+ * The kernel signal delivery code writes a bit over 4KB
* below the stack pointer (r1) before decrementing it.
* The exec code can write slightly over 640kB to the stack
* before setting the user r1. Thus we allow the stack to
@@ -299,7 +302,7 @@ static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
* between the last mapped region and the stack will
* expand the stack rather than segfaulting.
*/
- if (address + 2048 >= uregs->gpr[1])
+ if (address + SIGFRAME_MAX_SIZE >= uregs->gpr[1])
return false;
if ((flags & FAULT_FLAG_WRITE) && (flags & FAULT_FLAG_USER) &&
The patch below does not apply to the 4.4-stable tree.
If someone wants it applied there, or to any other stable or longterm
tree, then please email the backport, including the original git commit
id to <stable(a)vger.kernel.org>.
thanks,
greg k-h
------------------ original commit in Linus's tree ------------------
>From 63dee5df43a31f3844efabc58972f0a206ca4534 Mon Sep 17 00:00:00 2001
From: Michael Ellerman <mpe(a)ellerman.id.au>
Date: Fri, 24 Jul 2020 19:25:25 +1000
Subject: [PATCH] powerpc: Allow 4224 bytes of stack expansion for the signal
frame
We have powerpc specific logic in our page fault handling to decide if
an access to an unmapped address below the stack pointer should expand
the stack VMA.
The code was originally added in 2004 "ported from 2.4". The rough
logic is that the stack is allowed to grow to 1MB with no extra
checking. Over 1MB the access must be within 2048 bytes of the stack
pointer, or be from a user instruction that updates the stack pointer.
The 2048 byte allowance below the stack pointer is there to cover the
288 byte "red zone" as well as the "about 1.5kB" needed by the signal
delivery code.
Unfortunately since then the signal frame has expanded, and is now
4224 bytes on 64-bit kernels with transactional memory enabled. This
means if a process has consumed more than 1MB of stack, and its stack
pointer lies less than 4224 bytes from the next page boundary, signal
delivery will fault when trying to expand the stack and the process
will see a SEGV.
The total size of the signal frame is the size of struct rt_sigframe
(which includes the red zone) plus __SIGNAL_FRAMESIZE (128 bytes on
64-bit).
The 2048 byte allowance was correct until 2008 as the signal frame
was:
struct rt_sigframe {
struct ucontext uc; /* 0 1440 */
/* --- cacheline 11 boundary (1408 bytes) was 32 bytes ago --- */
long unsigned int _unused[2]; /* 1440 16 */
unsigned int tramp[6]; /* 1456 24 */
struct siginfo * pinfo; /* 1480 8 */
void * puc; /* 1488 8 */
struct siginfo info; /* 1496 128 */
/* --- cacheline 12 boundary (1536 bytes) was 88 bytes ago --- */
char abigap[288]; /* 1624 288 */
/* size: 1920, cachelines: 15, members: 7 */
/* padding: 8 */
};
1920 + 128 = 2048
Then in commit ce48b2100785 ("powerpc: Add VSX context save/restore,
ptrace and signal support") (Jul 2008) the signal frame expanded to
2304 bytes:
struct rt_sigframe {
struct ucontext uc; /* 0 1696 */ <--
/* --- cacheline 13 boundary (1664 bytes) was 32 bytes ago --- */
long unsigned int _unused[2]; /* 1696 16 */
unsigned int tramp[6]; /* 1712 24 */
struct siginfo * pinfo; /* 1736 8 */
void * puc; /* 1744 8 */
struct siginfo info; /* 1752 128 */
/* --- cacheline 14 boundary (1792 bytes) was 88 bytes ago --- */
char abigap[288]; /* 1880 288 */
/* size: 2176, cachelines: 17, members: 7 */
/* padding: 8 */
};
2176 + 128 = 2304
At this point we should have been exposed to the bug, though as far as
I know it was never reported. I no longer have a system old enough to
easily test on.
Then in 2010 commit 320b2b8de126 ("mm: keep a guard page below a
grow-down stack segment") caused our stack expansion code to never
trigger, as there was always a VMA found for a write up to PAGE_SIZE
below r1.
That meant the bug was hidden as we continued to expand the signal
frame in commit 2b0a576d15e0 ("powerpc: Add new transactional memory
state to the signal context") (Feb 2013):
struct rt_sigframe {
struct ucontext uc; /* 0 1696 */
/* --- cacheline 13 boundary (1664 bytes) was 32 bytes ago --- */
struct ucontext uc_transact; /* 1696 1696 */ <--
/* --- cacheline 26 boundary (3328 bytes) was 64 bytes ago --- */
long unsigned int _unused[2]; /* 3392 16 */
unsigned int tramp[6]; /* 3408 24 */
struct siginfo * pinfo; /* 3432 8 */
void * puc; /* 3440 8 */
struct siginfo info; /* 3448 128 */
/* --- cacheline 27 boundary (3456 bytes) was 120 bytes ago --- */
char abigap[288]; /* 3576 288 */
/* size: 3872, cachelines: 31, members: 8 */
/* padding: 8 */
/* last cacheline: 32 bytes */
};
3872 + 128 = 4000
And commit 573ebfa6601f ("powerpc: Increase stack redzone for 64-bit
userspace to 512 bytes") (Feb 2014):
struct rt_sigframe {
struct ucontext uc; /* 0 1696 */
/* --- cacheline 13 boundary (1664 bytes) was 32 bytes ago --- */
struct ucontext uc_transact; /* 1696 1696 */
/* --- cacheline 26 boundary (3328 bytes) was 64 bytes ago --- */
long unsigned int _unused[2]; /* 3392 16 */
unsigned int tramp[6]; /* 3408 24 */
struct siginfo * pinfo; /* 3432 8 */
void * puc; /* 3440 8 */
struct siginfo info; /* 3448 128 */
/* --- cacheline 27 boundary (3456 bytes) was 120 bytes ago --- */
char abigap[512]; /* 3576 512 */ <--
/* size: 4096, cachelines: 32, members: 8 */
/* padding: 8 */
};
4096 + 128 = 4224
Then finally in 2017, commit 1be7107fbe18 ("mm: larger stack guard
gap, between vmas") exposed us to the existing bug, because it changed
the stack VMA to be the correct/real size, meaning our stack expansion
code is now triggered.
Fix it by increasing the allowance to 4224 bytes.
Hard-coding 4224 is obviously unsafe against future expansions of the
signal frame in the same way as the existing code. We can't easily use
sizeof() because the signal frame structure is not in a header. We
will either fix that, or rip out all the custom stack expansion
checking logic entirely.
Fixes: ce48b2100785 ("powerpc: Add VSX context save/restore, ptrace and signal support")
Cc: stable(a)vger.kernel.org # v2.6.27+
Reported-by: Tom Lane <tgl(a)sss.pgh.pa.us>
Tested-by: Daniel Axtens <dja(a)axtens.net>
Signed-off-by: Michael Ellerman <mpe(a)ellerman.id.au>
Link: https://lore.kernel.org/r/20200724092528.1578671-2-mpe@ellerman.id.au
diff --git a/arch/powerpc/mm/fault.c b/arch/powerpc/mm/fault.c
index 641fc5f3d7dd..3ebb1792e636 100644
--- a/arch/powerpc/mm/fault.c
+++ b/arch/powerpc/mm/fault.c
@@ -267,6 +267,9 @@ static bool bad_kernel_fault(struct pt_regs *regs, unsigned long error_code,
return false;
}
+// This comes from 64-bit struct rt_sigframe + __SIGNAL_FRAMESIZE
+#define SIGFRAME_MAX_SIZE (4096 + 128)
+
static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
struct vm_area_struct *vma, unsigned int flags,
bool *must_retry)
@@ -274,7 +277,7 @@ static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
/*
* N.B. The POWER/Open ABI allows programs to access up to
* 288 bytes below the stack pointer.
- * The kernel signal delivery code writes up to about 1.5kB
+ * The kernel signal delivery code writes a bit over 4KB
* below the stack pointer (r1) before decrementing it.
* The exec code can write slightly over 640kB to the stack
* before setting the user r1. Thus we allow the stack to
@@ -299,7 +302,7 @@ static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
* between the last mapped region and the stack will
* expand the stack rather than segfaulting.
*/
- if (address + 2048 >= uregs->gpr[1])
+ if (address + SIGFRAME_MAX_SIZE >= uregs->gpr[1])
return false;
if ((flags & FAULT_FLAG_WRITE) && (flags & FAULT_FLAG_USER) &&